Merge remote branch 'airlied/drm-core-next' into drm-intel-next

114 files changed, 10988 insertions, 5145 deletions

diff --git a/drivers/gpu/drm/drm_crtc_helper.c b/drivers/gpu/drm/drm_crtc_helper.c
index bede10a03407..c6b2e27a446a 100644
--- a/drivers/gpu/drm/drm_crtc_helper.c
+++ b/drivers/gpu/drm/drm_crtc_helper.c
@@ -336,7 +336,7 @@ bool drm_crtc_helper_set_mode(struct drm_crtc *crtc,
 			      struct drm_framebuffer *old_fb)
 {
 	struct drm_device *dev = crtc->dev;
-	struct drm_display_mode *adjusted_mode, saved_mode;
+	struct drm_display_mode *adjusted_mode, saved_mode, saved_hwmode;
 	struct drm_crtc_helper_funcs *crtc_funcs = crtc->helper_private;
 	struct drm_encoder_helper_funcs *encoder_funcs;
 	int saved_x, saved_y;
@@ -350,6 +350,7 @@ bool drm_crtc_helper_set_mode(struct drm_crtc *crtc,
 	if (!crtc->enabled)
 		return true;
 
+	saved_hwmode = crtc->hwmode;
 	saved_mode = crtc->mode;
 	saved_x = crtc->x;
 	saved_y = crtc->y;
@@ -427,11 +428,21 @@ bool drm_crtc_helper_set_mode(struct drm_crtc *crtc,
 
 	}
 
+	/* Store real post-adjustment hardware mode. */
+	crtc->hwmode = *adjusted_mode;
+
+	/* Calculate and store various constants which
+	 * are later needed by vblank and swap-completion
+	 * timestamping. They are derived from true hwmode.
+	 */
+	drm_calc_timestamping_constants(crtc);
+
 	/* XXX free adjustedmode */
 	drm_mode_destroy(dev, adjusted_mode);
 	/* FIXME: add subpixel order */
 done:
 	if (!ret) {
+		crtc->hwmode = saved_hwmode;
 		crtc->mode = saved_mode;
 		crtc->x = saved_x;
 		crtc->y = saved_y;
diff --git a/drivers/gpu/drm/drm_irq.c b/drivers/gpu/drm/drm_irq.c
index 722700d5d73e..77c875db5aae 100644
--- a/drivers/gpu/drm/drm_irq.c
+++ b/drivers/gpu/drm/drm_irq.c
@@ -40,6 +40,22 @@
 #include <linux/slab.h>
 
 #include <linux/vgaarb.h>
+
+/* Access macro for slots in vblank timestamp ringbuffer. */
+#define vblanktimestamp(dev, crtc, count) ( \
+	(dev)->_vblank_time[(crtc) * DRM_VBLANKTIME_RBSIZE + \
+	((count) % DRM_VBLANKTIME_RBSIZE)])
+
+/* Retry timestamp calculation up to 3 times to satisfy
+ * drm_timestamp_precision before giving up.
+ */
+#define DRM_TIMESTAMP_MAXRETRIES 3
+
+/* Threshold in nanoseconds for detection of redundant
+ * vblank irq in drm_handle_vblank(). 1 msec should be ok.
+ */
+#define DRM_REDUNDANT_VBLIRQ_THRESH_NS 1000000
+
 /**
  * Get interrupt from bus id.
  *
@@ -77,6 +93,87 @@ int drm_irq_by_busid(struct drm_device *dev, void *data,
 	return 0;
 }
 
+/*
+ * Clear vblank timestamp buffer for a crtc.
+ */
+static void clear_vblank_timestamps(struct drm_device *dev, int crtc)
+{
+	memset(&dev->_vblank_time[crtc * DRM_VBLANKTIME_RBSIZE], 0,
+		DRM_VBLANKTIME_RBSIZE * sizeof(struct timeval));
+}
+
+/*
+ * Disable vblank irq's on crtc, make sure that last vblank count
+ * of hardware and corresponding consistent software vblank counter
+ * are preserved, even if there are any spurious vblank irq's after
+ * disable.
+ */
+static void vblank_disable_and_save(struct drm_device *dev, int crtc)
+{
+	unsigned long irqflags;
+	u32 vblcount;
+	s64 diff_ns;
+	int vblrc;
+	struct timeval tvblank;
+
+	/* Prevent vblank irq processing while disabling vblank irqs,
+	 * so no updates of timestamps or count can happen after we've
+	 * disabled. Needed to prevent races in case of delayed irq's.
+	 * Disable preemption, so vblank_time_lock is held as short as
+	 * possible, even under a kernel with PREEMPT_RT patches.
+	 */
+	preempt_disable();
+	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
+
+	dev->driver->disable_vblank(dev, crtc);
+	dev->vblank_enabled[crtc] = 0;
+
+	/* No further vblank irq's will be processed after
+	 * this point. Get current hardware vblank count and
+	 * vblank timestamp, repeat until they are consistent.
+	 *
+	 * FIXME: There is still a race condition here and in
+	 * drm_update_vblank_count() which can cause off-by-one
+	 * reinitialization of software vblank counter. If gpu
+	 * vblank counter doesn't increment exactly at the leading
+	 * edge of a vblank interval, then we can lose 1 count if
+	 * we happen to execute between start of vblank and the
+	 * delayed gpu counter increment.
+	 */
+	do {
+		dev->last_vblank[crtc] = dev->driver->get_vblank_counter(dev, crtc);
+		vblrc = drm_get_last_vbltimestamp(dev, crtc, &tvblank, 0);
+	} while (dev->last_vblank[crtc] != dev->driver->get_vblank_counter(dev, crtc));
+
+	/* Compute time difference to stored timestamp of last vblank
+	 * as updated by last invocation of drm_handle_vblank() in vblank irq.
+	 */
+	vblcount = atomic_read(&dev->_vblank_count[crtc]);
+	diff_ns = timeval_to_ns(&tvblank) -
+		  timeval_to_ns(&vblanktimestamp(dev, crtc, vblcount));
+
+	/* If there is at least 1 msec difference between the last stored
+	 * timestamp and tvblank, then we are currently executing our
+	 * disable inside a new vblank interval, the tvblank timestamp
+	 * corresponds to this new vblank interval and the irq handler
+	 * for this vblank didn't run yet and won't run due to our disable.
+	 * Therefore we need to do the job of drm_handle_vblank() and
+	 * increment the vblank counter by one to account for this vblank.
+	 *
+	 * Skip this step if there isn't any high precision timestamp
+	 * available. In that case we can't account for this and just
+	 * hope for the best.
+	 */
+	if ((vblrc > 0) && (abs(diff_ns) > 1000000))
+		atomic_inc(&dev->_vblank_count[crtc]);
+
+	/* Invalidate all timestamps while vblank irq's are off. */
+	clear_vblank_timestamps(dev, crtc);
+
+	spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags);
+	preempt_enable();
+}
+
 static void vblank_disable_fn(unsigned long arg)
 {
 	struct drm_device *dev = (struct drm_device *)arg;
@@ -91,10 +188,7 @@ static void vblank_disable_fn(unsigned long arg)
 		if (atomic_read(&dev->vblank_refcount[i]) == 0 &&
 		    dev->vblank_enabled[i]) {
 			DRM_DEBUG("disabling vblank on crtc %d\n", i);
-			dev->last_vblank[i] =
-				dev->driver->get_vblank_counter(dev, i);
-			dev->driver->disable_vblank(dev, i);
-			dev->vblank_enabled[i] = 0;
+			vblank_disable_and_save(dev, i);
 		}
 		spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
 	}
@@ -117,6 +211,7 @@ void drm_vblank_cleanup(struct drm_device *dev)
 	kfree(dev->last_vblank);
 	kfree(dev->last_vblank_wait);
 	kfree(dev->vblank_inmodeset);
+	kfree(dev->_vblank_time);
 
 	dev->num_crtcs = 0;
 }
@@ -129,6 +224,8 @@ int drm_vblank_init(struct drm_device *dev, int num_crtcs)
 	setup_timer(&dev->vblank_disable_timer, vblank_disable_fn,
 		    (unsigned long)dev);
 	spin_lock_init(&dev->vbl_lock);
+	spin_lock_init(&dev->vblank_time_lock);
+
 	dev->num_crtcs = num_crtcs;
 
 	dev->vbl_queue = kmalloc(sizeof(wait_queue_head_t) * num_crtcs,
@@ -161,6 +258,19 @@ int drm_vblank_init(struct drm_device *dev, int num_crtcs)
 	if (!dev->vblank_inmodeset)
 		goto err;
 
+	dev->_vblank_time = kcalloc(num_crtcs * DRM_VBLANKTIME_RBSIZE,
+				    sizeof(struct timeval), GFP_KERNEL);
+	if (!dev->_vblank_time)
+		goto err;
+
+	DRM_INFO("Supports vblank timestamp caching Rev 1 (10.10.2010).\n");
+
+	/* Driver specific high-precision vblank timestamping supported? */
+	if (dev->driver->get_vblank_timestamp)
+		DRM_INFO("Driver supports precise vblank timestamp query.\n");
+	else
+		DRM_INFO("No driver support for vblank timestamp query.\n");
+
 	/* Zero per-crtc vblank stuff */
 	for (i = 0; i < num_crtcs; i++) {
 		init_waitqueue_head(&dev->vbl_queue[i]);
@@ -279,7 +389,7 @@ EXPORT_SYMBOL(drm_irq_install);
  *
  * Calls the driver's \c drm_driver_irq_uninstall() function, and stops the irq.
  */
-int drm_irq_uninstall(struct drm_device * dev)
+int drm_irq_uninstall(struct drm_device *dev)
 {
 	unsigned long irqflags;
 	int irq_enabled, i;
@@ -335,7 +445,9 @@ int drm_control(struct drm_device *dev, void *data,
 {
 	struct drm_control *ctl = data;
 
-	/* if we haven't irq we fallback for compatibility reasons - this used to be a separate function in drm_dma.h */
+	/* if we haven't irq we fallback for compatibility reasons -
+	 * this used to be a separate function in drm_dma.h
+	 */
 
 
 	switch (ctl->func) {
@@ -360,6 +472,287 @@ int drm_control(struct drm_device *dev, void *data,
 }
 
 /**
+ * drm_calc_timestamping_constants - Calculate and
+ * store various constants which are later needed by
+ * vblank and swap-completion timestamping, e.g, by
+ * drm_calc_vbltimestamp_from_scanoutpos().
+ * They are derived from crtc's true scanout timing,
+ * so they take things like panel scaling or other
+ * adjustments into account.
+ *
+ * @crtc drm_crtc whose timestamp constants should be updated.
+ *
+ */
+void drm_calc_timestamping_constants(struct drm_crtc *crtc)
+{
+	s64 linedur_ns = 0, pixeldur_ns = 0, framedur_ns = 0;
+	u64 dotclock;
+
+	/* Dot clock in Hz: */
+	dotclock = (u64) crtc->hwmode.clock * 1000;
+
+	/* Valid dotclock? */
+	if (dotclock > 0) {
+		/* Convert scanline length in pixels and video dot clock to
+		 * line duration, frame duration and pixel duration in
+		 * nanoseconds:
+		 */
+		pixeldur_ns = (s64) div64_u64(1000000000, dotclock);
+		linedur_ns  = (s64) div64_u64(((u64) crtc->hwmode.crtc_htotal *
+					      1000000000), dotclock);
+		framedur_ns = (s64) crtc->hwmode.crtc_vtotal * linedur_ns;
+	} else
+		DRM_ERROR("crtc %d: Can't calculate constants, dotclock = 0!\n",
+			  crtc->base.id);
+
+	crtc->pixeldur_ns = pixeldur_ns;
+	crtc->linedur_ns  = linedur_ns;
+	crtc->framedur_ns = framedur_ns;
+
+	DRM_DEBUG("crtc %d: hwmode: htotal %d, vtotal %d, vdisplay %d\n",
+		  crtc->base.id, crtc->hwmode.crtc_htotal,
+		  crtc->hwmode.crtc_vtotal, crtc->hwmode.crtc_vdisplay);
+	DRM_DEBUG("crtc %d: clock %d kHz framedur %d linedur %d, pixeldur %d\n",
+		  crtc->base.id, (int) dotclock/1000, (int) framedur_ns,
+		  (int) linedur_ns, (int) pixeldur_ns);
+}
+EXPORT_SYMBOL(drm_calc_timestamping_constants);
+
+/**
+ * drm_calc_vbltimestamp_from_scanoutpos - helper routine for kms
+ * drivers. Implements calculation of exact vblank timestamps from
+ * given drm_display_mode timings and current video scanout position
+ * of a crtc. This can be called from within get_vblank_timestamp()
+ * implementation of a kms driver to implement the actual timestamping.
+ *
+ * Should return timestamps conforming to the OML_sync_control OpenML
+ * extension specification. The timestamp corresponds to the end of
+ * the vblank interval, aka start of scanout of topmost-leftmost display
+ * pixel in the following video frame.
+ *
+ * Requires support for optional dev->driver->get_scanout_position()
+ * in kms driver, plus a bit of setup code to provide a drm_display_mode
+ * that corresponds to the true scanout timing.
+ *
+ * The current implementation only handles standard video modes. It
+ * returns as no operation if a doublescan or interlaced video mode is
+ * active. Higher level code is expected to handle this.
+ *
+ * @dev: DRM device.
+ * @crtc: Which crtc's vblank timestamp to retrieve.
+ * @max_error: Desired maximum allowable error in timestamps (nanosecs).
+ *             On return contains true maximum error of timestamp.
+ * @vblank_time: Pointer to struct timeval which should receive the timestamp.
+ * @flags: Flags to pass to driver:
+ *         0 = Default.
+ *         DRM_CALLED_FROM_VBLIRQ = If function is called from vbl irq handler.
+ * @refcrtc: drm_crtc* of crtc which defines scanout timing.
+ *
+ * Returns negative value on error, failure or if not supported in current
+ * video mode:
+ *
+ * -EINVAL   - Invalid crtc.
+ * -EAGAIN   - Temporary unavailable, e.g., called before initial modeset.
+ * -ENOTSUPP - Function not supported in current display mode.
+ * -EIO      - Failed, e.g., due to failed scanout position query.
+ *
+ * Returns or'ed positive status flags on success:
+ *
+ * DRM_VBLANKTIME_SCANOUTPOS_METHOD - Signal this method used for timestamping.
+ * DRM_VBLANKTIME_INVBL - Timestamp taken while scanout was in vblank interval.
+ *
+ */
+int drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev, int crtc,
+					  int *max_error,
+					  struct timeval *vblank_time,
+					  unsigned flags,
+					  struct drm_crtc *refcrtc)
+{
+	struct timeval stime, raw_time;
+	struct drm_display_mode *mode;
+	int vbl_status, vtotal, vdisplay;
+	int vpos, hpos, i;
+	s64 framedur_ns, linedur_ns, pixeldur_ns, delta_ns, duration_ns;
+	bool invbl;
+
+	if (crtc < 0 || crtc >= dev->num_crtcs) {
+		DRM_ERROR("Invalid crtc %d\n", crtc);
+		return -EINVAL;
+	}
+
+	/* Scanout position query not supported? Should not happen. */
+	if (!dev->driver->get_scanout_position) {
+		DRM_ERROR("Called from driver w/o get_scanout_position()!?\n");
+		return -EIO;
+	}
+
+	mode = &refcrtc->hwmode;
+	vtotal = mode->crtc_vtotal;
+	vdisplay = mode->crtc_vdisplay;
+
+	/* Durations of frames, lines, pixels in nanoseconds. */
+	framedur_ns = refcrtc->framedur_ns;
+	linedur_ns  = refcrtc->linedur_ns;
+	pixeldur_ns = refcrtc->pixeldur_ns;
+
+	/* If mode timing undefined, just return as no-op:
+	 * Happens during initial modesetting of a crtc.
+	 */
+	if (vtotal <= 0 || vdisplay <= 0 || framedur_ns == 0) {
+		DRM_DEBUG("crtc %d: Noop due to uninitialized mode.\n", crtc);
+		return -EAGAIN;
+	}
+
+	/* Don't know yet how to handle interlaced or
+	 * double scan modes. Just no-op for now.
+	 */
+	if (mode->flags & (DRM_MODE_FLAG_INTERLACE | DRM_MODE_FLAG_DBLSCAN)) {
+		DRM_DEBUG("crtc %d: Noop due to unsupported mode.\n", crtc);
+		return -ENOTSUPP;
+	}
+
+	/* Get current scanout position with system timestamp.
+	 * Repeat query up to DRM_TIMESTAMP_MAXRETRIES times
+	 * if single query takes longer than max_error nanoseconds.
+	 *
+	 * This guarantees a tight bound on maximum error if
+	 * code gets preempted or delayed for some reason.
+	 */
+	for (i = 0; i < DRM_TIMESTAMP_MAXRETRIES; i++) {
+		/* Disable preemption to make it very likely to
+		 * succeed in the first iteration even on PREEMPT_RT kernel.
+		 */
+		preempt_disable();
+
+		/* Get system timestamp before query. */
+		do_gettimeofday(&stime);
+
+		/* Get vertical and horizontal scanout pos. vpos, hpos. */
+		vbl_status = dev->driver->get_scanout_position(dev, crtc, &vpos, &hpos);
+
+		/* Get system timestamp after query. */
+		do_gettimeofday(&raw_time);
+
+		preempt_enable();
+
+		/* Return as no-op if scanout query unsupported or failed. */
+		if (!(vbl_status & DRM_SCANOUTPOS_VALID)) {
+			DRM_DEBUG("crtc %d : scanoutpos query failed [%d].\n",
+				  crtc, vbl_status);
+			return -EIO;
+		}
+
+		duration_ns = timeval_to_ns(&raw_time) - timeval_to_ns(&stime);
+
+		/* Accept result with <  max_error nsecs timing uncertainty. */
+		if (duration_ns <= (s64) *max_error)
+			break;
+	}
+
+	/* Noisy system timing? */
+	if (i == DRM_TIMESTAMP_MAXRETRIES) {
+		DRM_DEBUG("crtc %d: Noisy timestamp %d us > %d us [%d reps].\n",
+			  crtc, (int) duration_ns/1000, *max_error/1000, i);
+	}
+
+	/* Return upper bound of timestamp precision error. */
+	*max_error = (int) duration_ns;
+
+	/* Check if in vblank area:
+	 * vpos is >=0 in video scanout area, but negative
+	 * within vblank area, counting down the number of lines until
+	 * start of scanout.
+	 */
+	invbl = vbl_status & DRM_SCANOUTPOS_INVBL;
+
+	/* Convert scanout position into elapsed time at raw_time query
+	 * since start of scanout at first display scanline. delta_ns
+	 * can be negative if start of scanout hasn't happened yet.
+	 */
+	delta_ns = (s64) vpos * linedur_ns + (s64) hpos * pixeldur_ns;
+
+	/* Is vpos outside nominal vblank area, but less than
+	 * 1/100 of a frame height away from start of vblank?
+	 * If so, assume this isn't a massively delayed vblank
+	 * interrupt, but a vblank interrupt that fired a few
+	 * microseconds before true start of vblank. Compensate
+	 * by adding a full frame duration to the final timestamp.
+	 * Happens, e.g., on ATI R500, R600.
+	 *
+	 * We only do this if DRM_CALLED_FROM_VBLIRQ.
+	 */
+	if ((flags & DRM_CALLED_FROM_VBLIRQ) && !invbl &&
+	    ((vdisplay - vpos) < vtotal / 100)) {
+		delta_ns = delta_ns - framedur_ns;
+
+		/* Signal this correction as "applied". */
+		vbl_status |= 0x8;
+	}
+
+	/* Subtract time delta from raw timestamp to get final
+	 * vblank_time timestamp for end of vblank.
+	 */
+	*vblank_time = ns_to_timeval(timeval_to_ns(&raw_time) - delta_ns);
+
+	DRM_DEBUG("crtc %d : v %d p(%d,%d)@ %d.%d -> %d.%d [e %d us, %d rep]\n",
+		  crtc, (int) vbl_status, hpos, vpos, raw_time.tv_sec,
+		  raw_time.tv_usec, vblank_time->tv_sec, vblank_time->tv_usec,
+		  (int) duration_ns/1000, i);
+
+	vbl_status = DRM_VBLANKTIME_SCANOUTPOS_METHOD;
+	if (invbl)
+		vbl_status |= DRM_VBLANKTIME_INVBL;
+
+	return vbl_status;
+}
+EXPORT_SYMBOL(drm_calc_vbltimestamp_from_scanoutpos);
+
+/**
+ * drm_get_last_vbltimestamp - retrieve raw timestamp for the most recent
+ * vblank interval.
+ *
+ * @dev: DRM device
+ * @crtc: which crtc's vblank timestamp to retrieve
+ * @tvblank: Pointer to target struct timeval which should receive the timestamp
+ * @flags: Flags to pass to driver:
+ *         0 = Default.
+ *         DRM_CALLED_FROM_VBLIRQ = If function is called from vbl irq handler.
+ *
+ * Fetches the system timestamp corresponding to the time of the most recent
+ * vblank interval on specified crtc. May call into kms-driver to
+ * compute the timestamp with a high-precision GPU specific method.
+ *
+ * Returns zero if timestamp originates from uncorrected do_gettimeofday()
+ * call, i.e., it isn't very precisely locked to the true vblank.
+ *
+ * Returns non-zero if timestamp is considered to be very precise.
+ */
+u32 drm_get_last_vbltimestamp(struct drm_device *dev, int crtc,
+			      struct timeval *tvblank, unsigned flags)
+{
+	int ret = 0;
+
+	/* Define requested maximum error on timestamps (nanoseconds). */
+	int max_error = (int) drm_timestamp_precision * 1000;
+
+	/* Query driver if possible and precision timestamping enabled. */
+	if (dev->driver->get_vblank_timestamp && (max_error > 0)) {
+		ret = dev->driver->get_vblank_timestamp(dev, crtc, &max_error,
+							tvblank, flags);
+		if (ret > 0)
+			return (u32) ret;
+	}
+
+	/* GPU high precision timestamp query unsupported or failed.
+	 * Return gettimeofday timestamp as best estimate.
+	 */
+	do_gettimeofday(tvblank);
+
+	return 0;
+}
+EXPORT_SYMBOL(drm_get_last_vbltimestamp);
+
+/**
  * drm_vblank_count - retrieve "cooked" vblank counter value
  * @dev: DRM device
  * @crtc: which counter to retrieve
@@ -375,6 +768,40 @@ u32 drm_vblank_count(struct drm_device *dev, int crtc)
 EXPORT_SYMBOL(drm_vblank_count);
 
 /**
+ * drm_vblank_count_and_time - retrieve "cooked" vblank counter value
+ * and the system timestamp corresponding to that vblank counter value.
+ *
+ * @dev: DRM device
+ * @crtc: which counter to retrieve
+ * @vblanktime: Pointer to struct timeval to receive the vblank timestamp.
+ *
+ * Fetches the "cooked" vblank count value that represents the number of
+ * vblank events since the system was booted, including lost events due to
+ * modesetting activity. Returns corresponding system timestamp of the time
+ * of the vblank interval that corresponds to the current value vblank counter
+ * value.
+ */
+u32 drm_vblank_count_and_time(struct drm_device *dev, int crtc,
+			      struct timeval *vblanktime)
+{
+	u32 cur_vblank;
+
+	/* Read timestamp from slot of _vblank_time ringbuffer
+	 * that corresponds to current vblank count. Retry if
+	 * count has incremented during readout. This works like
+	 * a seqlock.
+	 */
+	do {
+		cur_vblank = atomic_read(&dev->_vblank_count[crtc]);
+		*vblanktime = vblanktimestamp(dev, crtc, cur_vblank);
+		smp_rmb();
+	} while (cur_vblank != atomic_read(&dev->_vblank_count[crtc]));
+
+	return cur_vblank;
+}
+EXPORT_SYMBOL(drm_vblank_count_and_time);
+
+/**
  * drm_update_vblank_count - update the master vblank counter
  * @dev: DRM device
  * @crtc: counter to update
@@ -392,7 +819,8 @@ EXPORT_SYMBOL(drm_vblank_count);
  */
 static void drm_update_vblank_count(struct drm_device *dev, int crtc)
 {
-	u32 cur_vblank, diff;
+	u32 cur_vblank, diff, tslot, rc;
+	struct timeval t_vblank;
 
 	/*
 	 * Interrupts were disabled prior to this call, so deal with counter
@@ -400,8 +828,18 @@ static void drm_update_vblank_count(struct drm_device *dev, int crtc)
 	 * NOTE!  It's possible we lost a full dev->max_vblank_count events
 	 * here if the register is small or we had vblank interrupts off for
 	 * a long time.
+	 *
+	 * We repeat the hardware vblank counter & timestamp query until
+	 * we get consistent results. This to prevent races between gpu
+	 * updating its hardware counter while we are retrieving the
+	 * corresponding vblank timestamp.
 	 */
-	cur_vblank = dev->driver->get_vblank_counter(dev, crtc);
+	do {
+		cur_vblank = dev->driver->get_vblank_counter(dev, crtc);
+		rc = drm_get_last_vbltimestamp(dev, crtc, &t_vblank, 0);
+	} while (cur_vblank != dev->driver->get_vblank_counter(dev, crtc));
+
+	/* Deal with counter wrap */
 	diff = cur_vblank - dev->last_vblank[crtc];
 	if (cur_vblank < dev->last_vblank[crtc]) {
 		diff += dev->max_vblank_count;
@@ -413,6 +851,16 @@ static void drm_update_vblank_count(struct drm_device *dev, int crtc)
 	DRM_DEBUG("enabling vblank interrupts on crtc %d, missed %d\n",
 		  crtc, diff);
 
+	/* Reinitialize corresponding vblank timestamp if high-precision query
+	 * available. Skip this step if query unsupported or failed. Will
+	 * reinitialize delayed at next vblank interrupt in that case.
+	 */
+	if (rc) {
+		tslot = atomic_read(&dev->_vblank_count[crtc]) + diff;
+		vblanktimestamp(dev, crtc, tslot) = t_vblank;
+		smp_wmb();
+	}
+
 	atomic_add(diff, &dev->_vblank_count[crtc]);
 }
 
@@ -429,15 +877,27 @@ static void drm_update_vblank_count(struct drm_device *dev, int crtc)
  */
 int drm_vblank_get(struct drm_device *dev, int crtc)
 {
-	unsigned long irqflags;
+	unsigned long irqflags, irqflags2;
 	int ret = 0;
 
 	spin_lock_irqsave(&dev->vbl_lock, irqflags);
 	/* Going from 0->1 means we have to enable interrupts again */
 	if (atomic_add_return(1, &dev->vblank_refcount[crtc]) == 1) {
+		/* Disable preemption while holding vblank_time_lock. Do
+		 * it explicitely to guard against PREEMPT_RT kernel.
+		 */
+		preempt_disable();
+		spin_lock_irqsave(&dev->vblank_time_lock, irqflags2);
 		if (!dev->vblank_enabled[crtc]) {
+			/* Enable vblank irqs under vblank_time_lock protection.
+			 * All vblank count & timestamp updates are held off
+			 * until we are done reinitializing master counter and
+			 * timestamps. Filtercode in drm_handle_vblank() will
+			 * prevent double-accounting of same vblank interval.
+			 */
 			ret = dev->driver->enable_vblank(dev, crtc);
-			DRM_DEBUG("enabling vblank on crtc %d, ret: %d\n", crtc, ret);
+			DRM_DEBUG("enabling vblank on crtc %d, ret: %d\n",
+				  crtc, ret);
 			if (ret)
 				atomic_dec(&dev->vblank_refcount[crtc]);
 			else {
@@ -445,6 +905,8 @@ int drm_vblank_get(struct drm_device *dev, int crtc)
 				drm_update_vblank_count(dev, crtc);
 			}
 		}
+		spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags2);
+		preempt_enable();
 	} else {
 		if (!dev->vblank_enabled[crtc]) {
 			atomic_dec(&dev->vblank_refcount[crtc]);
@@ -463,15 +925,17 @@ EXPORT_SYMBOL(drm_vblank_get);
  * @crtc: which counter to give up
  *
  * Release ownership of a given vblank counter, turning off interrupts
- * if possible.
+ * if possible. Disable interrupts after drm_vblank_offdelay milliseconds.
  */
 void drm_vblank_put(struct drm_device *dev, int crtc)
 {
-	BUG_ON (atomic_read (&dev->vblank_refcount[crtc]) == 0);
+	BUG_ON(atomic_read(&dev->vblank_refcount[crtc]) == 0);
 
 	/* Last user schedules interrupt disable */
-	if (atomic_dec_and_test(&dev->vblank_refcount[crtc]))
-		mod_timer(&dev->vblank_disable_timer, jiffies + 5*DRM_HZ);
+	if (atomic_dec_and_test(&dev->vblank_refcount[crtc]) &&
+	    (drm_vblank_offdelay > 0))
+		mod_timer(&dev->vblank_disable_timer,
+			  jiffies + ((drm_vblank_offdelay * DRM_HZ)/1000));
 }
 EXPORT_SYMBOL(drm_vblank_put);
 
@@ -480,10 +944,8 @@ void drm_vblank_off(struct drm_device *dev, int crtc)
 	unsigned long irqflags;
 
 	spin_lock_irqsave(&dev->vbl_lock, irqflags);
-	dev->driver->disable_vblank(dev, crtc);
+	vblank_disable_and_save(dev, crtc);
 	DRM_WAKEUP(&dev->vbl_queue[crtc]);
-	dev->vblank_enabled[crtc] = 0;
-	dev->last_vblank[crtc] = dev->driver->get_vblank_counter(dev, crtc);
 	spin_unlock_irqrestore(&dev->vbl_lock, irqflags);
 }
 EXPORT_SYMBOL(drm_vblank_off);
@@ -602,7 +1064,6 @@ static int drm_queue_vblank_event(struct drm_device *dev, int pipe,
 	e->base.file_priv = file_priv;
 	e->base.destroy = (void (*) (struct drm_pending_event *)) kfree;
 
-	do_gettimeofday(&now);
 	spin_lock_irqsave(&dev->event_lock, flags);
 
 	if (file_priv->event_space < sizeof e->event) {
@@ -611,7 +1072,8 @@ static int drm_queue_vblank_event(struct drm_device *dev, int pipe,
 	}
 
 	file_priv->event_space -= sizeof e->event;
-	seq = drm_vblank_count(dev, pipe);
+	seq = drm_vblank_count_and_time(dev, pipe, &now);
+
 	if ((vblwait->request.type & _DRM_VBLANK_NEXTONMISS) &&
 	    (seq - vblwait->request.sequence) <= (1 << 23)) {
 		vblwait->request.sequence = seq + 1;
@@ -727,11 +1189,10 @@ int drm_wait_vblank(struct drm_device *dev, void *data,
 	if (ret != -EINTR) {
 		struct timeval now;
 
-		do_gettimeofday(&now);
-
+		vblwait->reply.sequence = drm_vblank_count_and_time(dev, crtc, &now);
 		vblwait->reply.tval_sec = now.tv_sec;
 		vblwait->reply.tval_usec = now.tv_usec;
-		vblwait->reply.sequence = drm_vblank_count(dev, crtc);
+
 		DRM_DEBUG("returning %d to client\n",
 			  vblwait->reply.sequence);
 	} else {
@@ -750,8 +1211,7 @@ void drm_handle_vblank_events(struct drm_device *dev, int crtc)
 	unsigned long flags;
 	unsigned int seq;
 
-	do_gettimeofday(&now);
-	seq = drm_vblank_count(dev, crtc);
+	seq = drm_vblank_count_and_time(dev, crtc, &now);
 
 	spin_lock_irqsave(&dev->event_lock, flags);
 
@@ -789,11 +1249,64 @@ void drm_handle_vblank_events(struct drm_device *dev, int crtc)
  */
 void drm_handle_vblank(struct drm_device *dev, int crtc)
 {
+	u32 vblcount;
+	s64 diff_ns;
+	struct timeval tvblank;
+	unsigned long irqflags;
+
 	if (!dev->num_crtcs)
 		return;
 
-	atomic_inc(&dev->_vblank_count[crtc]);
+	/* Need timestamp lock to prevent concurrent execution with
+	 * vblank enable/disable, as this would cause inconsistent
+	 * or corrupted timestamps and vblank counts.
+	 */
+	spin_lock_irqsave(&dev->vblank_time_lock, irqflags);
+
+	/* Vblank irq handling disabled. Nothing to do. */
+	if (!dev->vblank_enabled[crtc]) {
+		spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags);
+		return;
+	}
+
+	/* Fetch corresponding timestamp for this vblank interval from
+	 * driver and store it in proper slot of timestamp ringbuffer.
+	 */
+
+	/* Get current timestamp and count. */
+	vblcount = atomic_read(&dev->_vblank_count[crtc]);
+	drm_get_last_vbltimestamp(dev, crtc, &tvblank, DRM_CALLED_FROM_VBLIRQ);
+
+	/* Compute time difference to timestamp of last vblank */
+	diff_ns = timeval_to_ns(&tvblank) -
+		  timeval_to_ns(&vblanktimestamp(dev, crtc, vblcount));
+
+	/* Update vblank timestamp and count if at least
+	 * DRM_REDUNDANT_VBLIRQ_THRESH_NS nanoseconds
+	 * difference between last stored timestamp and current
+	 * timestamp. A smaller difference means basically
+	 * identical timestamps. Happens if this vblank has
+	 * been already processed and this is a redundant call,
+	 * e.g., due to spurious vblank interrupts. We need to
+	 * ignore those for accounting.
+	 */
+	if (abs(diff_ns) > DRM_REDUNDANT_VBLIRQ_THRESH_NS) {
+		/* Store new timestamp in ringbuffer. */
+		vblanktimestamp(dev, crtc, vblcount + 1) = tvblank;
+		smp_wmb();
+
+		/* Increment cooked vblank count. This also atomically commits
+		 * the timestamp computed above.
+		 */
+		atomic_inc(&dev->_vblank_count[crtc]);
+	} else {
+		DRM_DEBUG("crtc %d: Redundant vblirq ignored. diff_ns = %d\n",
+			  crtc, (int) diff_ns);
+	}
+
 	DRM_WAKEUP(&dev->vbl_queue[crtc]);
 	drm_handle_vblank_events(dev, crtc);
+
+	spin_unlock_irqrestore(&dev->vblank_time_lock, irqflags);
 }
 EXPORT_SYMBOL(drm_handle_vblank);
diff --git a/drivers/gpu/drm/drm_stub.c b/drivers/gpu/drm/drm_stub.c
index cdc89ee042cc..d59edc18301f 100644
--- a/drivers/gpu/drm/drm_stub.c
+++ b/drivers/gpu/drm/drm_stub.c
@@ -40,12 +40,22 @@
 unsigned int drm_debug = 0;	/* 1 to enable debug output */
 EXPORT_SYMBOL(drm_debug);
 
+unsigned int drm_vblank_offdelay = 5000;    /* Default to 5000 msecs. */
+EXPORT_SYMBOL(drm_vblank_offdelay);
+
+unsigned int drm_timestamp_precision = 20;  /* Default to 20 usecs. */
+EXPORT_SYMBOL(drm_timestamp_precision);
+
 MODULE_AUTHOR(CORE_AUTHOR);
 MODULE_DESCRIPTION(CORE_DESC);
 MODULE_LICENSE("GPL and additional rights");
 MODULE_PARM_DESC(debug, "Enable debug output");
+MODULE_PARM_DESC(vblankoffdelay, "Delay until vblank irq auto-disable [msecs]");
+MODULE_PARM_DESC(timestamp_precision_usec, "Max. error on timestamps [usecs]");
 
 module_param_named(debug, drm_debug, int, 0600);
+module_param_named(vblankoffdelay, drm_vblank_offdelay, int, 0600);
+module_param_named(timestamp_precision_usec, drm_timestamp_precision, int, 0600);
 
 struct idr drm_minors_idr;
 
diff --git a/drivers/gpu/drm/nouveau/Makefile b/drivers/gpu/drm/nouveau/Makefile
index 23fa82d667d6..b1d8941e04d8 100644
--- a/drivers/gpu/drm/nouveau/Makefile
+++ b/drivers/gpu/drm/nouveau/Makefile
@@ -5,12 +5,13 @@
 ccflags-y := -Iinclude/drm
 nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nouveau_object.o nouveau_irq.o nouveau_notifier.o \
-             nouveau_sgdma.o nouveau_dma.o \
+             nouveau_sgdma.o nouveau_dma.o nouveau_util.o \
              nouveau_bo.o nouveau_fence.o nouveau_gem.o nouveau_ttm.o \
              nouveau_hw.o nouveau_calc.o nouveau_bios.o nouveau_i2c.o \
              nouveau_display.o nouveau_connector.o nouveau_fbcon.o \
              nouveau_dp.o nouveau_ramht.o \
 	     nouveau_pm.o nouveau_volt.o nouveau_perf.o nouveau_temp.o \
+	     nouveau_mm.o nouveau_vm.o \
              nv04_timer.o \
              nv04_mc.o nv40_mc.o nv50_mc.o \
              nv04_fb.o nv10_fb.o nv30_fb.o nv40_fb.o nv50_fb.o nvc0_fb.o \
@@ -18,14 +19,16 @@ nouveau-y := nouveau_drv.o nouveau_state.o nouveau_channel.o nouveau_mem.o \
              nv04_graph.o nv10_graph.o nv20_graph.o \
              nv40_graph.o nv50_graph.o nvc0_graph.o \
              nv40_grctx.o nv50_grctx.o \
+             nv84_crypt.o \
              nv04_instmem.o nv50_instmem.o nvc0_instmem.o \
-             nv50_crtc.o nv50_dac.o nv50_sor.o \
+             nv50_evo.o nv50_crtc.o nv50_dac.o nv50_sor.o \
              nv50_cursor.o nv50_display.o nv50_fbcon.o \
              nv04_dac.o nv04_dfp.o nv04_tv.o nv17_tv.o nv17_tv_modes.o \
              nv04_crtc.o nv04_display.o nv04_cursor.o nv04_fbcon.o \
              nv10_gpio.o nv50_gpio.o \
 	     nv50_calc.o \
-	     nv04_pm.o nv50_pm.o nva3_pm.o
+	     nv04_pm.o nv50_pm.o nva3_pm.o \
+	     nv50_vram.o nv50_vm.o
 
 nouveau-$(CONFIG_DRM_NOUVEAU_DEBUG) += nouveau_debugfs.o
 nouveau-$(CONFIG_COMPAT) += nouveau_ioc32.o
diff --git a/drivers/gpu/drm/nouveau/nouveau_bios.c b/drivers/gpu/drm/nouveau/nouveau_bios.c
index b2293576f278..d3046559bf05 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bios.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bios.c
@@ -6053,52 +6053,17 @@ static struct dcb_entry *new_dcb_entry(struct dcb_table *dcb)
 	return entry;
 }
 
-static void fabricate_vga_output(struct dcb_table *dcb, int i2c, int heads)
+static void fabricate_dcb_output(struct dcb_table *dcb, int type, int i2c,
+				 int heads, int or)
 {
 	struct dcb_entry *entry = new_dcb_entry(dcb);
 
-	entry->type = 0;
+	entry->type = type;
 	entry->i2c_index = i2c;
 	entry->heads = heads;
-	entry->location = DCB_LOC_ON_CHIP;
-	entry->or = 1;
-}
-
-static void fabricate_dvi_i_output(struct dcb_table *dcb, bool twoHeads)
-{
-	struct dcb_entry *entry = new_dcb_entry(dcb);
-
-	entry->type = 2;
-	entry->i2c_index = LEGACY_I2C_PANEL;
-	entry->heads = twoHeads ? 3 : 1;
-	entry->location = !DCB_LOC_ON_CHIP;	/* ie OFF CHIP */
-	entry->or = 1;	/* means |0x10 gets set on CRE_LCD__INDEX */
-	entry->duallink_possible = false; /* SiI164 and co. are single link */
-
-#if 0
-	/*
-	 * For dvi-a either crtc probably works, but my card appears to only
-	 * support dvi-d.  "nvidia" still attempts to program it for dvi-a,
-	 * doing the full fp output setup (program 0x6808.. fp dimension regs,
-	 * setting 0x680848 to 0x10000111 to enable, maybe setting 0x680880);
-	 * the monitor picks up the mode res ok and lights up, but no pixel
-	 * data appears, so the board manufacturer probably connected up the
-	 * sync lines, but missed the video traces / components
-	 *
-	 * with this introduction, dvi-a left as an exercise for the reader.
-	 */
-	fabricate_vga_output(dcb, LEGACY_I2C_PANEL, entry->heads);
-#endif
-}
-
-static void fabricate_tv_output(struct dcb_table *dcb, bool twoHeads)
-{
-	struct dcb_entry *entry = new_dcb_entry(dcb);
-
-	entry->type = 1;
-	entry->i2c_index = LEGACY_I2C_TV;
-	entry->heads = twoHeads ? 3 : 1;
-	entry->location = !DCB_LOC_ON_CHIP;	/* ie OFF CHIP */
+	if (type != OUTPUT_ANALOG)
+		entry->location = !DCB_LOC_ON_CHIP; /* ie OFF CHIP */
+	entry->or = or;
 }
 
 static bool
@@ -6365,8 +6330,36 @@ apply_dcb_encoder_quirks(struct drm_device *dev, int idx, u32 *conn, u32 *conf)
 	return true;
 }
 
+static void
+fabricate_dcb_encoder_table(struct drm_device *dev, struct nvbios *bios)
+{
+	struct dcb_table *dcb = &bios->dcb;
+	int all_heads = (nv_two_heads(dev) ? 3 : 1);
+
+#ifdef __powerpc__
+	/* Apple iMac G4 NV17 */
+	if (of_machine_is_compatible("PowerMac4,5")) {
+		fabricate_dcb_output(dcb, OUTPUT_TMDS, 0, all_heads, 1);
+		fabricate_dcb_output(dcb, OUTPUT_ANALOG, 1, all_heads, 2);
+		return;
+	}
+#endif
+
+	/* Make up some sane defaults */
+	fabricate_dcb_output(dcb, OUTPUT_ANALOG, LEGACY_I2C_CRT, 1, 1);
+
+	if (nv04_tv_identify(dev, bios->legacy.i2c_indices.tv) >= 0)
+		fabricate_dcb_output(dcb, OUTPUT_TV, LEGACY_I2C_TV,
+				     all_heads, 0);
+
+	else if (bios->tmds.output0_script_ptr ||
+		 bios->tmds.output1_script_ptr)
+		fabricate_dcb_output(dcb, OUTPUT_TMDS, LEGACY_I2C_PANEL,
+				     all_heads, 1);
+}
+
 static int
-parse_dcb_table(struct drm_device *dev, struct nvbios *bios, bool twoHeads)
+parse_dcb_table(struct drm_device *dev, struct nvbios *bios)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct dcb_table *dcb = &bios->dcb;
@@ -6386,12 +6379,7 @@ parse_dcb_table(struct drm_device *dev, struct nvbios *bios, bool twoHeads)
 
 	/* this situation likely means a really old card, pre DCB */
 	if (dcbptr == 0x0) {
-		NV_INFO(dev, "Assuming a CRT output exists\n");
-		fabricate_vga_output(dcb, LEGACY_I2C_CRT, 1);
-
-		if (nv04_tv_identify(dev, bios->legacy.i2c_indices.tv) >= 0)
-			fabricate_tv_output(dcb, twoHeads);
-
+		fabricate_dcb_encoder_table(dev, bios);
 		return 0;
 	}
 
@@ -6451,21 +6439,7 @@ parse_dcb_table(struct drm_device *dev, struct nvbios *bios, bool twoHeads)
 		 */
 		NV_TRACEWARN(dev, "No useful information in BIOS output table; "
 				  "adding all possible outputs\n");
-		fabricate_vga_output(dcb, LEGACY_I2C_CRT, 1);
-
-		/*
-		 * Attempt to detect TV before DVI because the test
-		 * for the former is more accurate and it rules the
-		 * latter out.
-		 */
-		if (nv04_tv_identify(dev,
-				     bios->legacy.i2c_indices.tv) >= 0)
-			fabricate_tv_output(dcb, twoHeads);
-
-		else if (bios->tmds.output0_script_ptr ||
-			 bios->tmds.output1_script_ptr)
-			fabricate_dvi_i_output(dcb, twoHeads);
-
+		fabricate_dcb_encoder_table(dev, bios);
 		return 0;
 	}
 
@@ -6859,7 +6833,7 @@ nouveau_bios_init(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	ret = parse_dcb_table(dev, bios, nv_two_heads(dev));
+	ret = parse_dcb_table(dev, bios);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_bo.c b/drivers/gpu/drm/nouveau/nouveau_bo.c
index c41e1c200ef5..42d1ad62b381 100644
--- a/drivers/gpu/drm/nouveau/nouveau_bo.c
+++ b/drivers/gpu/drm/nouveau/nouveau_bo.c
@@ -32,6 +32,8 @@
 #include "nouveau_drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_dma.h"
+#include "nouveau_mm.h"
+#include "nouveau_vm.h"
 
 #include <linux/log2.h>
 #include <linux/slab.h>
@@ -46,82 +48,51 @@ nouveau_bo_del_ttm(struct ttm_buffer_object *bo)
 	if (unlikely(nvbo->gem))
 		DRM_ERROR("bo %p still attached to GEM object\n", bo);
 
-	if (nvbo->tile)
-		nv10_mem_expire_tiling(dev, nvbo->tile, NULL);
-
+	nv10_mem_put_tile_region(dev, nvbo->tile, NULL);
+	nouveau_vm_put(&nvbo->vma);
 	kfree(nvbo);
 }
 
 static void
-nouveau_bo_fixup_align(struct drm_device *dev,
-		       uint32_t tile_mode, uint32_t tile_flags,
-		       int *align, int *size)
+nouveau_bo_fixup_align(struct nouveau_bo *nvbo, int *align, int *size,
+		       int *page_shift)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	/*
-	 * Some of the tile_flags have a periodic structure of N*4096 bytes,
-	 * align to to that as well as the page size. Align the size to the
-	 * appropriate boundaries. This does imply that sizes are rounded up
-	 * 3-7 pages, so be aware of this and do not waste memory by allocating
-	 * many small buffers.
-	 */
-	if (dev_priv->card_type == NV_50) {
-		uint32_t block_size = dev_priv->vram_size >> 15;
-		int i;
-
-		switch (tile_flags) {
-		case 0x1800:
-		case 0x2800:
-		case 0x4800:
-		case 0x7a00:
-			if (is_power_of_2(block_size)) {
-				for (i = 1; i < 10; i++) {
-					*align = 12 * i * block_size;
-					if (!(*align % 65536))
-						break;
-				}
-			} else {
-				for (i = 1; i < 10; i++) {
-					*align = 8 * i * block_size;
-					if (!(*align % 65536))
-						break;
-				}
-			}
-			*size = roundup(*size, *align);
-			break;
-		default:
-			break;
-		}
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(nvbo->bo.bdev);
 
-	} else {
-		if (tile_mode) {
+	if (dev_priv->card_type < NV_50) {
+		if (nvbo->tile_mode) {
 			if (dev_priv->chipset >= 0x40) {
 				*align = 65536;
-				*size = roundup(*size, 64 * tile_mode);
+				*size = roundup(*size, 64 * nvbo->tile_mode);
 
 			} else if (dev_priv->chipset >= 0x30) {
 				*align = 32768;
-				*size = roundup(*size, 64 * tile_mode);
+				*size = roundup(*size, 64 * nvbo->tile_mode);
 
 			} else if (dev_priv->chipset >= 0x20) {
 				*align = 16384;
-				*size = roundup(*size, 64 * tile_mode);
+				*size = roundup(*size, 64 * nvbo->tile_mode);
 
 			} else if (dev_priv->chipset >= 0x10) {
 				*align = 16384;
-				*size = roundup(*size, 32 * tile_mode);
+				*size = roundup(*size, 32 * nvbo->tile_mode);
 			}
 		}
+	} else {
+		if (likely(dev_priv->chan_vm)) {
+			if (*size > 256 * 1024)
+				*page_shift = dev_priv->chan_vm->lpg_shift;
+			else
+				*page_shift = dev_priv->chan_vm->spg_shift;
+		} else {
+			*page_shift = 12;
+		}
+
+		*size = roundup(*size, (1 << *page_shift));
+		*align = max((1 << *page_shift), *align);
 	}
 
-	/* ALIGN works only on powers of two. */
 	*size = roundup(*size, PAGE_SIZE);
-
-	if (dev_priv->card_type == NV_50) {
-		*size = roundup(*size, 65536);
-		*align = max(65536, *align);
-	}
 }
 
 int
@@ -132,7 +103,7 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan,
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_bo *nvbo;
-	int ret = 0;
+	int ret = 0, page_shift = 0;
 
 	nvbo = kzalloc(sizeof(struct nouveau_bo), GFP_KERNEL);
 	if (!nvbo)
@@ -145,10 +116,18 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan,
 	nvbo->tile_flags = tile_flags;
 	nvbo->bo.bdev = &dev_priv->ttm.bdev;
 
-	nouveau_bo_fixup_align(dev, tile_mode, nouveau_bo_tile_layout(nvbo),
-			       &align, &size);
+	nouveau_bo_fixup_align(nvbo, &align, &size, &page_shift);
 	align >>= PAGE_SHIFT;
 
+	if (!nvbo->no_vm && dev_priv->chan_vm) {
+		ret = nouveau_vm_get(dev_priv->chan_vm, size, page_shift,
+				     NV_MEM_ACCESS_RW, &nvbo->vma);
+		if (ret) {
+			kfree(nvbo);
+			return ret;
+		}
+	}
+
 	nouveau_bo_placement_set(nvbo, flags, 0);
 
 	nvbo->channel = chan;
@@ -161,6 +140,11 @@ nouveau_bo_new(struct drm_device *dev, struct nouveau_channel *chan,
 	}
 	nvbo->channel = NULL;
 
+	if (nvbo->vma.node) {
+		if (nvbo->bo.mem.mem_type == TTM_PL_VRAM)
+			nvbo->bo.offset = nvbo->vma.offset;
+	}
+
 	*pnvbo = nvbo;
 	return 0;
 }
@@ -244,7 +228,7 @@ nouveau_bo_pin(struct nouveau_bo *nvbo, uint32_t memtype)
 
 	nouveau_bo_placement_set(nvbo, memtype, 0);
 
-	ret = ttm_bo_validate(bo, &nvbo->placement, false, false, false);
+	ret = nouveau_bo_validate(nvbo, false, false, false);
 	if (ret == 0) {
 		switch (bo->mem.mem_type) {
 		case TTM_PL_VRAM:
@@ -280,7 +264,7 @@ nouveau_bo_unpin(struct nouveau_bo *nvbo)
 
 	nouveau_bo_placement_set(nvbo, bo->mem.placement, 0);
 
-	ret = ttm_bo_validate(bo, &nvbo->placement, false, false, false);
+	ret = nouveau_bo_validate(nvbo, false, false, false);
 	if (ret == 0) {
 		switch (bo->mem.mem_type) {
 		case TTM_PL_VRAM:
@@ -319,6 +303,25 @@ nouveau_bo_unmap(struct nouveau_bo *nvbo)
 		ttm_bo_kunmap(&nvbo->kmap);
 }
 
+int
+nouveau_bo_validate(struct nouveau_bo *nvbo, bool interruptible,
+		    bool no_wait_reserve, bool no_wait_gpu)
+{
+	int ret;
+
+	ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement, interruptible,
+			      no_wait_reserve, no_wait_gpu);
+	if (ret)
+		return ret;
+
+	if (nvbo->vma.node) {
+		if (nvbo->bo.mem.mem_type == TTM_PL_VRAM)
+			nvbo->bo.offset = nvbo->vma.offset;
+	}
+
+	return 0;
+}
+
 u16
 nouveau_bo_rd16(struct nouveau_bo *nvbo, unsigned index)
 {
@@ -410,37 +413,40 @@ nouveau_bo_init_mem_type(struct ttm_bo_device *bdev, uint32_t type,
 		man->default_caching = TTM_PL_FLAG_CACHED;
 		break;
 	case TTM_PL_VRAM:
-		man->func = &ttm_bo_manager_func;
+		if (dev_priv->card_type == NV_50) {
+			man->func = &nouveau_vram_manager;
+			man->io_reserve_fastpath = false;
+			man->use_io_reserve_lru = true;
+		} else {
+			man->func = &ttm_bo_manager_func;
+		}
 		man->flags = TTM_MEMTYPE_FLAG_FIXED |
 			     TTM_MEMTYPE_FLAG_MAPPABLE;
 		man->available_caching = TTM_PL_FLAG_UNCACHED |
 					 TTM_PL_FLAG_WC;
 		man->default_caching = TTM_PL_FLAG_WC;
-		if (dev_priv->card_type == NV_50)
-			man->gpu_offset = 0x40000000;
-		else
-			man->gpu_offset = 0;
 		break;
 	case TTM_PL_TT:
 		man->func = &ttm_bo_manager_func;
 		switch (dev_priv->gart_info.type) {
 		case NOUVEAU_GART_AGP:
 			man->flags = TTM_MEMTYPE_FLAG_MAPPABLE;
-			man->available_caching = TTM_PL_FLAG_UNCACHED;
-			man->default_caching = TTM_PL_FLAG_UNCACHED;
+			man->available_caching = TTM_PL_FLAG_UNCACHED |
+				TTM_PL_FLAG_WC;
+			man->default_caching = TTM_PL_FLAG_WC;
 			break;
 		case NOUVEAU_GART_SGDMA:
 			man->flags = TTM_MEMTYPE_FLAG_MAPPABLE |
 				     TTM_MEMTYPE_FLAG_CMA;
 			man->available_caching = TTM_PL_MASK_CACHING;
 			man->default_caching = TTM_PL_FLAG_CACHED;
+			man->gpu_offset = dev_priv->gart_info.aper_base;
 			break;
 		default:
 			NV_ERROR(dev, "Unknown GART type: %d\n",
 				 dev_priv->gart_info.type);
 			return -EINVAL;
 		}
-		man->gpu_offset = dev_priv->vm_gart_base;
 		break;
 	default:
 		NV_ERROR(dev, "Unsupported memory type %u\n", (unsigned)type);
@@ -485,16 +491,9 @@ nouveau_bo_move_accel_cleanup(struct nouveau_channel *chan,
 	if (ret)
 		return ret;
 
-	if (nvbo->channel) {
-		ret = nouveau_fence_sync(fence, nvbo->channel);
-		if (ret)
-			goto out;
-	}
-
 	ret = ttm_bo_move_accel_cleanup(&nvbo->bo, fence, NULL, evict,
 					no_wait_reserve, no_wait_gpu, new_mem);
-out:
-	nouveau_fence_unref((void *)&fence);
+	nouveau_fence_unref(&fence);
 	return ret;
 }
 
@@ -529,14 +528,14 @@ nv50_bo_move_m2mf(struct nouveau_channel *chan, struct ttm_buffer_object *bo,
 	dst_offset = new_mem->start << PAGE_SHIFT;
 	if (!nvbo->no_vm) {
 		if (old_mem->mem_type == TTM_PL_VRAM)
-			src_offset += dev_priv->vm_vram_base;
+			src_offset  = nvbo->vma.offset;
 		else
-			src_offset += dev_priv->vm_gart_base;
+			src_offset += dev_priv->gart_info.aper_base;
 
 		if (new_mem->mem_type == TTM_PL_VRAM)
-			dst_offset += dev_priv->vm_vram_base;
+			dst_offset  = nvbo->vma.offset;
 		else
-			dst_offset += dev_priv->vm_gart_base;
+			dst_offset += dev_priv->gart_info.aper_base;
 	}
 
 	ret = RING_SPACE(chan, 3);
@@ -683,17 +682,24 @@ nouveau_bo_move_m2mf(struct ttm_buffer_object *bo, int evict, bool intr,
 	int ret;
 
 	chan = nvbo->channel;
-	if (!chan || nvbo->no_vm)
+	if (!chan || nvbo->no_vm) {
 		chan = dev_priv->channel;
+		mutex_lock_nested(&chan->mutex, NOUVEAU_KCHANNEL_MUTEX);
+	}
 
 	if (dev_priv->card_type < NV_50)
 		ret = nv04_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
 	else
 		ret = nv50_bo_move_m2mf(chan, bo, &bo->mem, new_mem);
-	if (ret)
-		return ret;
+	if (ret == 0) {
+		ret = nouveau_bo_move_accel_cleanup(chan, nvbo, evict,
+						    no_wait_reserve,
+						    no_wait_gpu, new_mem);
+	}
 
-	return nouveau_bo_move_accel_cleanup(chan, nvbo, evict, no_wait_reserve, no_wait_gpu, new_mem);
+	if (chan == dev_priv->channel)
+		mutex_unlock(&chan->mutex);
+	return ret;
 }
 
 static int
@@ -771,7 +777,6 @@ nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem,
 	struct drm_device *dev = dev_priv->dev;
 	struct nouveau_bo *nvbo = nouveau_bo(bo);
 	uint64_t offset;
-	int ret;
 
 	if (nvbo->no_vm || new_mem->mem_type != TTM_PL_VRAM) {
 		/* Nothing to do. */
@@ -781,18 +786,12 @@ nouveau_bo_vm_bind(struct ttm_buffer_object *bo, struct ttm_mem_reg *new_mem,
 
 	offset = new_mem->start << PAGE_SHIFT;
 
-	if (dev_priv->card_type == NV_50) {
-		ret = nv50_mem_vm_bind_linear(dev,
-					      offset + dev_priv->vm_vram_base,
-					      new_mem->size,
-					      nouveau_bo_tile_layout(nvbo),
-					      offset);
-		if (ret)
-			return ret;
-
+	if (dev_priv->chan_vm) {
+		nouveau_vm_map(&nvbo->vma, new_mem->mm_node);
 	} else if (dev_priv->card_type >= NV_10) {
 		*new_tile = nv10_mem_set_tiling(dev, offset, new_mem->size,
-						nvbo->tile_mode);
+						nvbo->tile_mode,
+						nvbo->tile_flags);
 	}
 
 	return 0;
@@ -808,9 +807,7 @@ nouveau_bo_vm_cleanup(struct ttm_buffer_object *bo,
 
 	if (dev_priv->card_type >= NV_10 &&
 	    dev_priv->card_type < NV_50) {
-		if (*old_tile)
-			nv10_mem_expire_tiling(dev, *old_tile, bo->sync_obj);
-
+		nv10_mem_put_tile_region(dev, *old_tile, bo->sync_obj);
 		*old_tile = new_tile;
 	}
 }
@@ -879,6 +876,7 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
 	struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
 	struct drm_device *dev = dev_priv->dev;
+	int ret;
 
 	mem->bus.addr = NULL;
 	mem->bus.offset = 0;
@@ -901,9 +899,32 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 #endif
 		break;
 	case TTM_PL_VRAM:
-		mem->bus.offset = mem->start << PAGE_SHIFT;
+	{
+		struct nouveau_vram *vram = mem->mm_node;
+
+		if (!dev_priv->bar1_vm) {
+			mem->bus.offset = mem->start << PAGE_SHIFT;
+			mem->bus.base = pci_resource_start(dev->pdev, 1);
+			mem->bus.is_iomem = true;
+			break;
+		}
+
+		ret = nouveau_vm_get(dev_priv->bar1_vm, mem->bus.size, 12,
+				     NV_MEM_ACCESS_RW, &vram->bar_vma);
+		if (ret)
+			return ret;
+
+		nouveau_vm_map(&vram->bar_vma, vram);
+		if (ret) {
+			nouveau_vm_put(&vram->bar_vma);
+			return ret;
+		}
+
+		mem->bus.offset  = vram->bar_vma.offset;
+		mem->bus.offset -= 0x0020000000ULL;
 		mem->bus.base = pci_resource_start(dev->pdev, 1);
 		mem->bus.is_iomem = true;
+	}
 		break;
 	default:
 		return -EINVAL;
@@ -914,6 +935,17 @@ nouveau_ttm_io_mem_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 static void
 nouveau_ttm_io_mem_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 {
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(bdev);
+	struct nouveau_vram *vram = mem->mm_node;
+
+	if (!dev_priv->bar1_vm || mem->mem_type != TTM_PL_VRAM)
+		return;
+
+	if (!vram->bar_vma.node)
+		return;
+
+	nouveau_vm_unmap(&vram->bar_vma);
+	nouveau_vm_put(&vram->bar_vma);
 }
 
 static int
@@ -939,7 +971,23 @@ nouveau_ttm_fault_reserve_notify(struct ttm_buffer_object *bo)
 	nvbo->placement.fpfn = 0;
 	nvbo->placement.lpfn = dev_priv->fb_mappable_pages;
 	nouveau_bo_placement_set(nvbo, TTM_PL_VRAM, 0);
-	return ttm_bo_validate(bo, &nvbo->placement, false, true, false);
+	return nouveau_bo_validate(nvbo, false, true, false);
+}
+
+void
+nouveau_bo_fence(struct nouveau_bo *nvbo, struct nouveau_fence *fence)
+{
+	struct nouveau_fence *old_fence;
+
+	if (likely(fence))
+		nouveau_fence_ref(fence);
+
+	spin_lock(&nvbo->bo.bdev->fence_lock);
+	old_fence = nvbo->bo.sync_obj;
+	nvbo->bo.sync_obj = fence;
+	spin_unlock(&nvbo->bo.bdev->fence_lock);
+
+	nouveau_fence_unref(&old_fence);
 }
 
 struct ttm_bo_driver nouveau_bo_driver = {
@@ -949,11 +997,11 @@ struct ttm_bo_driver nouveau_bo_driver = {
 	.evict_flags = nouveau_bo_evict_flags,
 	.move = nouveau_bo_move,
 	.verify_access = nouveau_bo_verify_access,
-	.sync_obj_signaled = nouveau_fence_signalled,
-	.sync_obj_wait = nouveau_fence_wait,
-	.sync_obj_flush = nouveau_fence_flush,
-	.sync_obj_unref = nouveau_fence_unref,
-	.sync_obj_ref = nouveau_fence_ref,
+	.sync_obj_signaled = __nouveau_fence_signalled,
+	.sync_obj_wait = __nouveau_fence_wait,
+	.sync_obj_flush = __nouveau_fence_flush,
+	.sync_obj_unref = __nouveau_fence_unref,
+	.sync_obj_ref = __nouveau_fence_ref,
 	.fault_reserve_notify = &nouveau_ttm_fault_reserve_notify,
 	.io_mem_reserve = &nouveau_ttm_io_mem_reserve,
 	.io_mem_free = &nouveau_ttm_io_mem_free,
diff --git a/drivers/gpu/drm/nouveau/nouveau_channel.c b/drivers/gpu/drm/nouveau/nouveau_channel.c
index 373950e34814..6f37995aee2d 100644
--- a/drivers/gpu/drm/nouveau/nouveau_channel.c
+++ b/drivers/gpu/drm/nouveau/nouveau_channel.c
@@ -39,22 +39,22 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 
 	if (dev_priv->card_type >= NV_50) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
-					     dev_priv->vm_end, NV_DMA_ACCESS_RO,
-					     NV_DMA_TARGET_AGP, &pushbuf);
+					     (1ULL << 40), NV_MEM_ACCESS_RO,
+					     NV_MEM_TARGET_VM, &pushbuf);
 		chan->pushbuf_base = pb->bo.offset;
 	} else
 	if (pb->bo.mem.mem_type == TTM_PL_TT) {
-		ret = nouveau_gpuobj_gart_dma_new(chan, 0,
-						  dev_priv->gart_info.aper_size,
-						  NV_DMA_ACCESS_RO, &pushbuf,
-						  NULL);
+		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
+					     dev_priv->gart_info.aper_size,
+					     NV_MEM_ACCESS_RO,
+					     NV_MEM_TARGET_GART, &pushbuf);
 		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
 	} else
 	if (dev_priv->card_type != NV_04) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, 0,
 					     dev_priv->fb_available_size,
-					     NV_DMA_ACCESS_RO,
-					     NV_DMA_TARGET_VIDMEM, &pushbuf);
+					     NV_MEM_ACCESS_RO,
+					     NV_MEM_TARGET_VRAM, &pushbuf);
 		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
 	} else {
 		/* NV04 cmdbuf hack, from original ddx.. not sure of it's
@@ -62,11 +62,10 @@ nouveau_channel_pushbuf_ctxdma_init(struct nouveau_channel *chan)
 		 * VRAM.
 		 */
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
-					     pci_resource_start(dev->pdev,
-					     1),
+					     pci_resource_start(dev->pdev, 1),
 					     dev_priv->fb_available_size,
-					     NV_DMA_ACCESS_RO,
-					     NV_DMA_TARGET_PCI, &pushbuf);
+					     NV_MEM_ACCESS_RO,
+					     NV_MEM_TARGET_PCI, &pushbuf);
 		chan->pushbuf_base = pb->bo.mem.start << PAGE_SHIFT;
 	}
 
@@ -107,74 +106,60 @@ nouveau_channel_user_pushbuf_alloc(struct drm_device *dev)
 int
 nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 		      struct drm_file *file_priv,
-		      uint32_t vram_handle, uint32_t tt_handle)
+		      uint32_t vram_handle, uint32_t gart_handle)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_channel *chan;
-	int channel, user;
+	unsigned long flags;
 	int ret;
 
-	/*
-	 * Alright, here is the full story
-	 * Nvidia cards have multiple hw fifo contexts (praise them for that,
-	 * no complicated crash-prone context switches)
-	 * We allocate a new context for each app and let it write to it
-	 * directly (woo, full userspace command submission !)
-	 * When there are no more contexts, you lost
-	 */
-	for (channel = 0; channel < pfifo->channels; channel++) {
-		if (dev_priv->fifos[channel] == NULL)
+	/* allocate and lock channel structure */
+	chan = kzalloc(sizeof(*chan), GFP_KERNEL);
+	if (!chan)
+		return -ENOMEM;
+	chan->dev = dev;
+	chan->file_priv = file_priv;
+	chan->vram_handle = vram_handle;
+	chan->gart_handle = gart_handle;
+
+	kref_init(&chan->ref);
+	atomic_set(&chan->users, 1);
+	mutex_init(&chan->mutex);
+	mutex_lock(&chan->mutex);
+
+	/* allocate hw channel id */
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	for (chan->id = 0; chan->id < pfifo->channels; chan->id++) {
+		if (!dev_priv->channels.ptr[chan->id]) {
+			nouveau_channel_ref(chan, &dev_priv->channels.ptr[chan->id]);
 			break;
+		}
 	}
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
 
-	/* no more fifos. you lost. */
-	if (channel == pfifo->channels)
-		return -EINVAL;
+	if (chan->id == pfifo->channels) {
+		mutex_unlock(&chan->mutex);
+		kfree(chan);
+		return -ENODEV;
+	}
 
-	dev_priv->fifos[channel] = kzalloc(sizeof(struct nouveau_channel),
-					   GFP_KERNEL);
-	if (!dev_priv->fifos[channel])
-		return -ENOMEM;
-	chan = dev_priv->fifos[channel];
+	NV_DEBUG(dev, "initialising channel %d\n", chan->id);
 	INIT_LIST_HEAD(&chan->nvsw.vbl_wait);
+	INIT_LIST_HEAD(&chan->nvsw.flip);
 	INIT_LIST_HEAD(&chan->fence.pending);
-	chan->dev = dev;
-	chan->id = channel;
-	chan->file_priv = file_priv;
-	chan->vram_handle = vram_handle;
-	chan->gart_handle = tt_handle;
-
-	NV_INFO(dev, "Allocating FIFO number %d\n", channel);
 
 	/* Allocate DMA push buffer */
 	chan->pushbuf_bo = nouveau_channel_user_pushbuf_alloc(dev);
 	if (!chan->pushbuf_bo) {
 		ret = -ENOMEM;
 		NV_ERROR(dev, "pushbuf %d\n", ret);
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
 	nouveau_dma_pre_init(chan);
-
-	/* Locate channel's user control regs */
-	if (dev_priv->card_type < NV_40)
-		user = NV03_USER(channel);
-	else
-	if (dev_priv->card_type < NV_50)
-		user = NV40_USER(channel);
-	else
-		user = NV50_USER(channel);
-
-	chan->user = ioremap(pci_resource_start(dev->pdev, 0) + user,
-								PAGE_SIZE);
-	if (!chan->user) {
-		NV_ERROR(dev, "ioremap of regs failed.\n");
-		nouveau_channel_free(chan);
-		return -ENOMEM;
-	}
 	chan->user_put = 0x40;
 	chan->user_get = 0x44;
 
@@ -182,15 +167,15 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	ret = nouveau_notifier_init_channel(chan);
 	if (ret) {
 		NV_ERROR(dev, "ntfy %d\n", ret);
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
 	/* Setup channel's default objects */
-	ret = nouveau_gpuobj_channel_init(chan, vram_handle, tt_handle);
+	ret = nouveau_gpuobj_channel_init(chan, vram_handle, gart_handle);
 	if (ret) {
 		NV_ERROR(dev, "gpuobj %d\n", ret);
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
@@ -198,7 +183,7 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	ret = nouveau_channel_pushbuf_ctxdma_init(chan);
 	if (ret) {
 		NV_ERROR(dev, "pbctxdma %d\n", ret);
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
@@ -206,16 +191,18 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	pfifo->reassign(dev, false);
 
 	/* Create a graphics context for new channel */
-	ret = pgraph->create_context(chan);
-	if (ret) {
-		nouveau_channel_free(chan);
-		return ret;
+	if (dev_priv->card_type < NV_50) {
+		ret = pgraph->create_context(chan);
+		if (ret) {
+			nouveau_channel_put(&chan);
+			return ret;
+		}
 	}
 
 	/* Construct inital RAMFC for new channel */
 	ret = pfifo->create_context(chan);
 	if (ret) {
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
@@ -225,83 +212,108 @@ nouveau_channel_alloc(struct drm_device *dev, struct nouveau_channel **chan_ret,
 	if (!ret)
 		ret = nouveau_fence_channel_init(chan);
 	if (ret) {
-		nouveau_channel_free(chan);
+		nouveau_channel_put(&chan);
 		return ret;
 	}
 
 	nouveau_debugfs_channel_init(chan);
 
-	NV_INFO(dev, "%s: initialised FIFO %d\n", __func__, channel);
+	NV_DEBUG(dev, "channel %d initialised\n", chan->id);
 	*chan_ret = chan;
 	return 0;
 }
 
-/* stops a fifo */
+struct nouveau_channel *
+nouveau_channel_get_unlocked(struct nouveau_channel *ref)
+{
+	struct nouveau_channel *chan = NULL;
+
+	if (likely(ref && atomic_inc_not_zero(&ref->users)))
+		nouveau_channel_ref(ref, &chan);
+
+	return chan;
+}
+
+struct nouveau_channel *
+nouveau_channel_get(struct drm_device *dev, struct drm_file *file_priv, int id)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	chan = nouveau_channel_get_unlocked(dev_priv->channels.ptr[id]);
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+
+	if (unlikely(!chan))
+		return ERR_PTR(-EINVAL);
+
+	if (unlikely(file_priv && chan->file_priv != file_priv)) {
+		nouveau_channel_put_unlocked(&chan);
+		return ERR_PTR(-EINVAL);
+	}
+
+	mutex_lock(&chan->mutex);
+	return chan;
+}
+
 void
-nouveau_channel_free(struct nouveau_channel *chan)
+nouveau_channel_put_unlocked(struct nouveau_channel **pchan)
 {
+	struct nouveau_channel *chan = *pchan;
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
 	unsigned long flags;
-	int ret;
 
-	NV_INFO(dev, "%s: freeing fifo %d\n", __func__, chan->id);
+	/* decrement the refcount, and we're done if there's still refs */
+	if (likely(!atomic_dec_and_test(&chan->users))) {
+		nouveau_channel_ref(NULL, pchan);
+		return;
+	}
 
+	/* noone wants the channel anymore */
+	NV_DEBUG(dev, "freeing channel %d\n", chan->id);
 	nouveau_debugfs_channel_fini(chan);
 
-	/* Give outstanding push buffers a chance to complete */
-	nouveau_fence_update(chan);
-	if (chan->fence.sequence != chan->fence.sequence_ack) {
-		struct nouveau_fence *fence = NULL;
-
-		ret = nouveau_fence_new(chan, &fence, true);
-		if (ret == 0) {
-			ret = nouveau_fence_wait(fence, NULL, false, false);
-			nouveau_fence_unref((void *)&fence);
-		}
-
-		if (ret)
-			NV_ERROR(dev, "Failed to idle channel %d.\n", chan->id);
-	}
+	/* give it chance to idle */
+	nouveau_channel_idle(chan);
 
-	/* Ensure all outstanding fences are signaled.  They should be if the
+	/* ensure all outstanding fences are signaled.  they should be if the
 	 * above attempts at idling were OK, but if we failed this'll tell TTM
 	 * we're done with the buffers.
 	 */
 	nouveau_fence_channel_fini(chan);
 
-	/* This will prevent pfifo from switching channels. */
+	/* boot it off the hardware */
 	pfifo->reassign(dev, false);
 
-	/* We want to give pgraph a chance to idle and get rid of all potential
-	 * errors. We need to do this before the lock, otherwise the irq handler
-	 * is unable to process them.
+	/* We want to give pgraph a chance to idle and get rid of all
+	 * potential errors. We need to do this without the context
+	 * switch lock held, otherwise the irq handler is unable to
+	 * process them.
 	 */
 	if (pgraph->channel(dev) == chan)
 		nouveau_wait_for_idle(dev);
 
-	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-
-	pgraph->fifo_access(dev, false);
-	if (pgraph->channel(dev) == chan)
-		pgraph->unload_context(dev);
-	pgraph->destroy_context(chan);
-	pgraph->fifo_access(dev, true);
-
-	if (pfifo->channel_id(dev) == chan->id) {
-		pfifo->disable(dev);
-		pfifo->unload_context(dev);
-		pfifo->enable(dev);
-	}
+	/* destroy the engine specific contexts */
 	pfifo->destroy_context(chan);
+	pgraph->destroy_context(chan);
+	if (pcrypt->destroy_context)
+		pcrypt->destroy_context(chan);
 
 	pfifo->reassign(dev, true);
 
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+	/* aside from its resources, the channel should now be dead,
+	 * remove it from the channel list
+	 */
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	nouveau_channel_ref(NULL, &dev_priv->channels.ptr[chan->id]);
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
 
-	/* Release the channel's resources */
+	/* destroy any resources the channel owned */
 	nouveau_gpuobj_ref(NULL, &chan->pushbuf);
 	if (chan->pushbuf_bo) {
 		nouveau_bo_unmap(chan->pushbuf_bo);
@@ -310,44 +322,80 @@ nouveau_channel_free(struct nouveau_channel *chan)
 	}
 	nouveau_gpuobj_channel_takedown(chan);
 	nouveau_notifier_takedown_channel(chan);
-	if (chan->user)
-		iounmap(chan->user);
 
-	dev_priv->fifos[chan->id] = NULL;
+	nouveau_channel_ref(NULL, pchan);
+}
+
+void
+nouveau_channel_put(struct nouveau_channel **pchan)
+{
+	mutex_unlock(&(*pchan)->mutex);
+	nouveau_channel_put_unlocked(pchan);
+}
+
+static void
+nouveau_channel_del(struct kref *ref)
+{
+	struct nouveau_channel *chan =
+		container_of(ref, struct nouveau_channel, ref);
+
 	kfree(chan);
 }
 
+void
+nouveau_channel_ref(struct nouveau_channel *chan,
+		    struct nouveau_channel **pchan)
+{
+	if (chan)
+		kref_get(&chan->ref);
+
+	if (*pchan)
+		kref_put(&(*pchan)->ref, nouveau_channel_del);
+
+	*pchan = chan;
+}
+
+void
+nouveau_channel_idle(struct nouveau_channel *chan)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_fence *fence = NULL;
+	int ret;
+
+	nouveau_fence_update(chan);
+
+	if (chan->fence.sequence != chan->fence.sequence_ack) {
+		ret = nouveau_fence_new(chan, &fence, true);
+		if (!ret) {
+			ret = nouveau_fence_wait(fence, false, false);
+			nouveau_fence_unref(&fence);
+		}
+
+		if (ret)
+			NV_ERROR(dev, "Failed to idle channel %d.\n", chan->id);
+	}
+}
+
 /* cleans up all the fifos from file_priv */
 void
 nouveau_channel_cleanup(struct drm_device *dev, struct drm_file *file_priv)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_engine *engine = &dev_priv->engine;
+	struct nouveau_channel *chan;
 	int i;
 
 	NV_DEBUG(dev, "clearing FIFO enables from file_priv\n");
 	for (i = 0; i < engine->fifo.channels; i++) {
-		struct nouveau_channel *chan = dev_priv->fifos[i];
+		chan = nouveau_channel_get(dev, file_priv, i);
+		if (IS_ERR(chan))
+			continue;
 
-		if (chan && chan->file_priv == file_priv)
-			nouveau_channel_free(chan);
+		atomic_dec(&chan->users);
+		nouveau_channel_put(&chan);
 	}
 }
 
-int
-nouveau_channel_owner(struct drm_device *dev, struct drm_file *file_priv,
-		      int channel)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
-
-	if (channel >= engine->fifo.channels)
-		return 0;
-	if (dev_priv->fifos[channel] == NULL)
-		return 0;
-
-	return (dev_priv->fifos[channel]->file_priv == file_priv);
-}
 
 /***********************************
  * ioctls wrapping the functions
@@ -395,24 +443,26 @@ nouveau_ioctl_fifo_alloc(struct drm_device *dev, void *data,
 	/* Named memory object area */
 	ret = drm_gem_handle_create(file_priv, chan->notifier_bo->gem,
 				    &init->notifier_handle);
-	if (ret) {
-		nouveau_channel_free(chan);
-		return ret;
-	}
 
-	return 0;
+	if (ret == 0)
+		atomic_inc(&chan->users); /* userspace reference */
+	nouveau_channel_put(&chan);
+	return ret;
 }
 
 static int
 nouveau_ioctl_fifo_free(struct drm_device *dev, void *data,
 			struct drm_file *file_priv)
 {
-	struct drm_nouveau_channel_free *cfree = data;
+	struct drm_nouveau_channel_free *req = data;
 	struct nouveau_channel *chan;
 
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(cfree->channel, file_priv, chan);
+	chan = nouveau_channel_get(dev, file_priv, req->channel);
+	if (IS_ERR(chan))
+		return PTR_ERR(chan);
 
-	nouveau_channel_free(chan);
+	atomic_dec(&chan->users);
+	nouveau_channel_put(&chan);
 	return 0;
 }
 
@@ -421,18 +471,18 @@ nouveau_ioctl_fifo_free(struct drm_device *dev, void *data,
  ***********************************/
 
 struct drm_ioctl_desc nouveau_ioctls[] = {
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GETPARAM, nouveau_ioctl_getparam, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_SETPARAM, nouveau_ioctl_setparam, DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_CHANNEL_ALLOC, nouveau_ioctl_fifo_alloc, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_CHANNEL_FREE, nouveau_ioctl_fifo_free, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GROBJ_ALLOC, nouveau_ioctl_grobj_alloc, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_NOTIFIEROBJ_ALLOC, nouveau_ioctl_notifier_alloc, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GPUOBJ_FREE, nouveau_ioctl_gpuobj_free, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_NEW, nouveau_gem_ioctl_new, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_PUSHBUF, nouveau_gem_ioctl_pushbuf, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_PREP, nouveau_gem_ioctl_cpu_prep, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_FINI, nouveau_gem_ioctl_cpu_fini, DRM_AUTH),
-	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_INFO, nouveau_gem_ioctl_info, DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GETPARAM, nouveau_ioctl_getparam, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_SETPARAM, nouveau_ioctl_setparam, DRM_UNLOCKED|DRM_AUTH|DRM_MASTER|DRM_ROOT_ONLY),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_CHANNEL_ALLOC, nouveau_ioctl_fifo_alloc, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_CHANNEL_FREE, nouveau_ioctl_fifo_free, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GROBJ_ALLOC, nouveau_ioctl_grobj_alloc, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_NOTIFIEROBJ_ALLOC, nouveau_ioctl_notifier_alloc, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GPUOBJ_FREE, nouveau_ioctl_gpuobj_free, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_NEW, nouveau_gem_ioctl_new, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_PUSHBUF, nouveau_gem_ioctl_pushbuf, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_PREP, nouveau_gem_ioctl_cpu_prep, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_CPU_FINI, nouveau_gem_ioctl_cpu_fini, DRM_UNLOCKED|DRM_AUTH),
+	DRM_IOCTL_DEF_DRV(NOUVEAU_GEM_INFO, nouveau_gem_ioctl_info, DRM_UNLOCKED|DRM_AUTH),
 };
 
 int nouveau_max_ioctl = DRM_ARRAY_SIZE(nouveau_ioctls);
diff --git a/drivers/gpu/drm/nouveau/nouveau_connector.c b/drivers/gpu/drm/nouveau/nouveau_connector.c
index 52c356e9a3d1..a21e00076839 100644
--- a/drivers/gpu/drm/nouveau/nouveau_connector.c
+++ b/drivers/gpu/drm/nouveau/nouveau_connector.c
@@ -37,6 +37,8 @@
 #include "nouveau_connector.h"
 #include "nouveau_hw.h"
 
+static void nouveau_connector_hotplug(void *, int);
+
 static struct nouveau_encoder *
 find_encoder_by_type(struct drm_connector *connector, int type)
 {
@@ -94,22 +96,30 @@ nouveau_connector_bpp(struct drm_connector *connector)
 }
 
 static void
-nouveau_connector_destroy(struct drm_connector *drm_connector)
+nouveau_connector_destroy(struct drm_connector *connector)
 {
-	struct nouveau_connector *nv_connector =
-		nouveau_connector(drm_connector);
+	struct nouveau_connector *nv_connector = nouveau_connector(connector);
+	struct drm_nouveau_private *dev_priv;
+	struct nouveau_gpio_engine *pgpio;
 	struct drm_device *dev;
 
 	if (!nv_connector)
 		return;
 
 	dev = nv_connector->base.dev;
+	dev_priv = dev->dev_private;
 	NV_DEBUG_KMS(dev, "\n");
 
+	pgpio = &dev_priv->engine.gpio;
+	if (pgpio->irq_unregister) {
+		pgpio->irq_unregister(dev, nv_connector->dcb->gpio_tag,
+				      nouveau_connector_hotplug, connector);
+	}
+
 	kfree(nv_connector->edid);
-	drm_sysfs_connector_remove(drm_connector);
-	drm_connector_cleanup(drm_connector);
-	kfree(drm_connector);
+	drm_sysfs_connector_remove(connector);
+	drm_connector_cleanup(connector);
+	kfree(connector);
 }
 
 static struct nouveau_i2c_chan *
@@ -760,6 +770,7 @@ nouveau_connector_create(struct drm_device *dev, int index)
 {
 	const struct drm_connector_funcs *funcs = &nouveau_connector_funcs;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
 	struct nouveau_connector *nv_connector = NULL;
 	struct dcb_connector_table_entry *dcb = NULL;
 	struct drm_connector *connector;
@@ -876,6 +887,11 @@ nouveau_connector_create(struct drm_device *dev, int index)
 		break;
 	}
 
+	if (pgpio->irq_register) {
+		pgpio->irq_register(dev, nv_connector->dcb->gpio_tag,
+				    nouveau_connector_hotplug, connector);
+	}
+
 	drm_sysfs_connector_add(connector);
 	dcb->drm = connector;
 	return dcb->drm;
@@ -886,3 +902,29 @@ fail:
 	return ERR_PTR(ret);
 
 }
+
+static void
+nouveau_connector_hotplug(void *data, int plugged)
+{
+	struct drm_connector *connector = data;
+	struct drm_device *dev = connector->dev;
+
+	NV_INFO(dev, "%splugged %s\n", plugged ? "" : "un",
+		drm_get_connector_name(connector));
+
+	if (connector->encoder && connector->encoder->crtc &&
+	    connector->encoder->crtc->enabled) {
+		struct nouveau_encoder *nv_encoder = nouveau_encoder(connector->encoder);
+		struct drm_encoder_helper_funcs *helper =
+			connector->encoder->helper_private;
+
+		if (nv_encoder->dcb->type == OUTPUT_DP) {
+			if (plugged)
+				helper->dpms(connector->encoder, DRM_MODE_DPMS_ON);
+			else
+				helper->dpms(connector->encoder, DRM_MODE_DPMS_OFF);
+		}
+	}
+
+	drm_helper_hpd_irq_event(dev);
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_display.c b/drivers/gpu/drm/nouveau/nouveau_display.c
index 2e11fd65b4dd..505c6bfb4d75 100644
--- a/drivers/gpu/drm/nouveau/nouveau_display.c
+++ b/drivers/gpu/drm/nouveau/nouveau_display.c
@@ -29,6 +29,9 @@
 #include "nouveau_drv.h"
 #include "nouveau_fb.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_hw.h"
+#include "nouveau_crtc.h"
+#include "nouveau_dma.h"
 
 static void
 nouveau_user_framebuffer_destroy(struct drm_framebuffer *drm_fb)
@@ -104,3 +107,207 @@ const struct drm_mode_config_funcs nouveau_mode_config_funcs = {
 	.output_poll_changed = nouveau_fbcon_output_poll_changed,
 };
 
+int
+nouveau_vblank_enable(struct drm_device *dev, int crtc)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->card_type >= NV_50)
+		nv_mask(dev, NV50_PDISPLAY_INTR_EN_1, 0,
+			NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_(crtc));
+	else
+		NVWriteCRTC(dev, crtc, NV_PCRTC_INTR_EN_0,
+			    NV_PCRTC_INTR_0_VBLANK);
+
+	return 0;
+}
+
+void
+nouveau_vblank_disable(struct drm_device *dev, int crtc)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->card_type >= NV_50)
+		nv_mask(dev, NV50_PDISPLAY_INTR_EN_1,
+			NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_(crtc), 0);
+	else
+		NVWriteCRTC(dev, crtc, NV_PCRTC_INTR_EN_0, 0);
+}
+
+static int
+nouveau_page_flip_reserve(struct nouveau_bo *old_bo,
+			  struct nouveau_bo *new_bo)
+{
+	int ret;
+
+	ret = nouveau_bo_pin(new_bo, TTM_PL_FLAG_VRAM);
+	if (ret)
+		return ret;
+
+	ret = ttm_bo_reserve(&new_bo->bo, false, false, false, 0);
+	if (ret)
+		goto fail;
+
+	ret = ttm_bo_reserve(&old_bo->bo, false, false, false, 0);
+	if (ret)
+		goto fail_unreserve;
+
+	return 0;
+
+fail_unreserve:
+	ttm_bo_unreserve(&new_bo->bo);
+fail:
+	nouveau_bo_unpin(new_bo);
+	return ret;
+}
+
+static void
+nouveau_page_flip_unreserve(struct nouveau_bo *old_bo,
+			    struct nouveau_bo *new_bo,
+			    struct nouveau_fence *fence)
+{
+	nouveau_bo_fence(new_bo, fence);
+	ttm_bo_unreserve(&new_bo->bo);
+
+	nouveau_bo_fence(old_bo, fence);
+	ttm_bo_unreserve(&old_bo->bo);
+
+	nouveau_bo_unpin(old_bo);
+}
+
+static int
+nouveau_page_flip_emit(struct nouveau_channel *chan,
+		       struct nouveau_bo *old_bo,
+		       struct nouveau_bo *new_bo,
+		       struct nouveau_page_flip_state *s,
+		       struct nouveau_fence **pfence)
+{
+	struct drm_device *dev = chan->dev;
+	unsigned long flags;
+	int ret;
+
+	/* Queue it to the pending list */
+	spin_lock_irqsave(&dev->event_lock, flags);
+	list_add_tail(&s->head, &chan->nvsw.flip);
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+
+	/* Synchronize with the old framebuffer */
+	ret = nouveau_fence_sync(old_bo->bo.sync_obj, chan);
+	if (ret)
+		goto fail;
+
+	/* Emit the pageflip */
+	ret = RING_SPACE(chan, 2);
+	if (ret)
+		goto fail;
+
+	BEGIN_RING(chan, NvSubSw, NV_SW_PAGE_FLIP, 1);
+	OUT_RING(chan, 0);
+	FIRE_RING(chan);
+
+	ret = nouveau_fence_new(chan, pfence, true);
+	if (ret)
+		goto fail;
+
+	return 0;
+fail:
+	spin_lock_irqsave(&dev->event_lock, flags);
+	list_del(&s->head);
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+	return ret;
+}
+
+int
+nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
+		       struct drm_pending_vblank_event *event)
+{
+	struct drm_device *dev = crtc->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_bo *old_bo = nouveau_framebuffer(crtc->fb)->nvbo;
+	struct nouveau_bo *new_bo = nouveau_framebuffer(fb)->nvbo;
+	struct nouveau_page_flip_state *s;
+	struct nouveau_channel *chan;
+	struct nouveau_fence *fence;
+	int ret;
+
+	if (dev_priv->engine.graph.accel_blocked)
+		return -ENODEV;
+
+	s = kzalloc(sizeof(*s), GFP_KERNEL);
+	if (!s)
+		return -ENOMEM;
+
+	/* Don't let the buffers go away while we flip */
+	ret = nouveau_page_flip_reserve(old_bo, new_bo);
+	if (ret)
+		goto fail_free;
+
+	/* Initialize a page flip struct */
+	*s = (struct nouveau_page_flip_state)
+		{ { }, s->event, nouveau_crtc(crtc)->index,
+		  fb->bits_per_pixel, fb->pitch, crtc->x, crtc->y,
+		  new_bo->bo.offset };
+
+	/* Choose the channel the flip will be handled in */
+	chan = nouveau_fence_channel(new_bo->bo.sync_obj);
+	if (!chan)
+		chan = nouveau_channel_get_unlocked(dev_priv->channel);
+	mutex_lock(&chan->mutex);
+
+	/* Emit a page flip */
+	ret = nouveau_page_flip_emit(chan, old_bo, new_bo, s, &fence);
+	nouveau_channel_put(&chan);
+	if (ret)
+		goto fail_unreserve;
+
+	/* Update the crtc struct and cleanup */
+	crtc->fb = fb;
+
+	nouveau_page_flip_unreserve(old_bo, new_bo, fence);
+	nouveau_fence_unref(&fence);
+	return 0;
+
+fail_unreserve:
+	nouveau_page_flip_unreserve(old_bo, new_bo, NULL);
+fail_free:
+	kfree(s);
+	return ret;
+}
+
+int
+nouveau_finish_page_flip(struct nouveau_channel *chan,
+			 struct nouveau_page_flip_state *ps)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_page_flip_state *s;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->event_lock, flags);
+
+	if (list_empty(&chan->nvsw.flip)) {
+		NV_ERROR(dev, "Unexpected pageflip in channel %d.\n", chan->id);
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+		return -EINVAL;
+	}
+
+	s = list_first_entry(&chan->nvsw.flip,
+			     struct nouveau_page_flip_state, head);
+	if (s->event) {
+		struct drm_pending_vblank_event *e = s->event;
+		struct timeval now;
+
+		do_gettimeofday(&now);
+		e->event.sequence = 0;
+		e->event.tv_sec = now.tv_sec;
+		e->event.tv_usec = now.tv_usec;
+		list_add_tail(&e->base.link, &e->base.file_priv->event_list);
+		wake_up_interruptible(&e->base.file_priv->event_wait);
+	}
+
+	list_del(&s->head);
+	*ps = *s;
+	kfree(s);
+
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_dma.c b/drivers/gpu/drm/nouveau/nouveau_dma.c
index 82581e600dcd..6ff77cedc008 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dma.c
@@ -59,17 +59,11 @@ nouveau_dma_init(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *obj = NULL;
 	int ret, i;
 
 	/* Create NV_MEMORY_TO_MEMORY_FORMAT for buffer moves */
-	ret = nouveau_gpuobj_gr_new(chan, dev_priv->card_type < NV_50 ?
-				    0x0039 : 0x5039, &obj);
-	if (ret)
-		return ret;
-
-	ret = nouveau_ramht_insert(chan, NvM2MF, obj);
-	nouveau_gpuobj_ref(NULL, &obj);
+	ret = nouveau_gpuobj_gr_new(chan, NvM2MF, dev_priv->card_type < NV_50 ?
+				    0x0039 : 0x5039);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_dp.c b/drivers/gpu/drm/nouveau/nouveau_dp.c
index 4562f309ae3d..38d599554bce 100644
--- a/drivers/gpu/drm/nouveau/nouveau_dp.c
+++ b/drivers/gpu/drm/nouveau/nouveau_dp.c
@@ -279,7 +279,7 @@ nouveau_dp_link_train(struct drm_encoder *encoder)
 	struct bit_displayport_encoder_table *dpe;
 	int dpe_headerlen;
 	uint8_t config[4], status[3];
-	bool cr_done, cr_max_vs, eq_done;
+	bool cr_done, cr_max_vs, eq_done, hpd_state;
 	int ret = 0, i, tries, voltage;
 
 	NV_DEBUG_KMS(dev, "link training!!\n");
@@ -297,7 +297,7 @@ nouveau_dp_link_train(struct drm_encoder *encoder)
 	/* disable hotplug detect, this flips around on some panels during
 	 * link training.
 	 */
-	pgpio->irq_enable(dev, nv_connector->dcb->gpio_tag, false);
+	hpd_state = pgpio->irq_enable(dev, nv_connector->dcb->gpio_tag, false);
 
 	if (dpe->script0) {
 		NV_DEBUG_KMS(dev, "SOR-%d: running DP script 0\n", nv_encoder->or);
@@ -439,7 +439,7 @@ stop:
 	}
 
 	/* re-enable hotplug detect */
-	pgpio->irq_enable(dev, nv_connector->dcb->gpio_tag, true);
+	pgpio->irq_enable(dev, nv_connector->dcb->gpio_tag, hpd_state);
 
 	return eq_done;
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.c b/drivers/gpu/drm/nouveau/nouveau_drv.c
index 90875494a65a..bb170570938b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.c
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.c
@@ -115,6 +115,10 @@ MODULE_PARM_DESC(perflvl_wr, "Allow perflvl changes (warning: dangerous!)\n");
 int nouveau_perflvl_wr;
 module_param_named(perflvl_wr, nouveau_perflvl_wr, int, 0400);
 
+MODULE_PARM_DESC(msi, "Enable MSI (default: off)\n");
+int nouveau_msi;
+module_param_named(msi, nouveau_msi, int, 0400);
+
 int nouveau_fbpercrtc;
 #if 0
 module_param_named(fbpercrtc, nouveau_fbpercrtc, int, 0400);
@@ -193,23 +197,10 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 
 	NV_INFO(dev, "Idling channels...\n");
 	for (i = 0; i < pfifo->channels; i++) {
-		struct nouveau_fence *fence = NULL;
-
-		chan = dev_priv->fifos[i];
-		if (!chan || (dev_priv->card_type >= NV_50 &&
-			      chan == dev_priv->fifos[0]))
-			continue;
-
-		ret = nouveau_fence_new(chan, &fence, true);
-		if (ret == 0) {
-			ret = nouveau_fence_wait(fence, NULL, false, false);
-			nouveau_fence_unref((void *)&fence);
-		}
+		chan = dev_priv->channels.ptr[i];
 
-		if (ret) {
-			NV_ERROR(dev, "Failed to idle channel %d for suspend\n",
-				 chan->id);
-		}
+		if (chan && chan->pushbuf_bo)
+			nouveau_channel_idle(chan);
 	}
 
 	pgraph->fifo_access(dev, false);
@@ -219,17 +210,17 @@ nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state)
 	pfifo->unload_context(dev);
 	pgraph->unload_context(dev);
 
-	NV_INFO(dev, "Suspending GPU objects...\n");
-	ret = nouveau_gpuobj_suspend(dev);
+	ret = pinstmem->suspend(dev);
 	if (ret) {
 		NV_ERROR(dev, "... failed: %d\n", ret);
 		goto out_abort;
 	}
 
-	ret = pinstmem->suspend(dev);
+	NV_INFO(dev, "Suspending GPU objects...\n");
+	ret = nouveau_gpuobj_suspend(dev);
 	if (ret) {
 		NV_ERROR(dev, "... failed: %d\n", ret);
-		nouveau_gpuobj_suspend_cleanup(dev);
+		pinstmem->resume(dev);
 		goto out_abort;
 	}
 
@@ -294,17 +285,18 @@ nouveau_pci_resume(struct pci_dev *pdev)
 		}
 	}
 
+	NV_INFO(dev, "Restoring GPU objects...\n");
+	nouveau_gpuobj_resume(dev);
+
 	NV_INFO(dev, "Reinitialising engines...\n");
 	engine->instmem.resume(dev);
 	engine->mc.init(dev);
 	engine->timer.init(dev);
 	engine->fb.init(dev);
 	engine->graph.init(dev);
+	engine->crypt.init(dev);
 	engine->fifo.init(dev);
 
-	NV_INFO(dev, "Restoring GPU objects...\n");
-	nouveau_gpuobj_resume(dev);
-
 	nouveau_irq_postinstall(dev);
 
 	/* Re-write SKIPS, they'll have been lost over the suspend */
@@ -313,7 +305,7 @@ nouveau_pci_resume(struct pci_dev *pdev)
 		int j;
 
 		for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-			chan = dev_priv->fifos[i];
+			chan = dev_priv->channels.ptr[i];
 			if (!chan || !chan->pushbuf_bo)
 				continue;
 
@@ -347,13 +339,11 @@ nouveau_pci_resume(struct pci_dev *pdev)
 
 	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
 		struct nouveau_crtc *nv_crtc = nouveau_crtc(crtc);
+		u32 offset = nv_crtc->cursor.nvbo->bo.mem.start << PAGE_SHIFT;
 
-		nv_crtc->cursor.set_offset(nv_crtc,
-					nv_crtc->cursor.nvbo->bo.offset -
-					dev_priv->vm_vram_base);
-
+		nv_crtc->cursor.set_offset(nv_crtc, offset);
 		nv_crtc->cursor.set_pos(nv_crtc, nv_crtc->cursor_saved_x,
-			nv_crtc->cursor_saved_y);
+						 nv_crtc->cursor_saved_y);
 	}
 
 	/* Force CLUT to get re-loaded during modeset */
@@ -393,6 +383,9 @@ static struct drm_driver driver = {
 	.irq_postinstall = nouveau_irq_postinstall,
 	.irq_uninstall = nouveau_irq_uninstall,
 	.irq_handler = nouveau_irq_handler,
+	.get_vblank_counter = drm_vblank_count,
+	.enable_vblank = nouveau_vblank_enable,
+	.disable_vblank = nouveau_vblank_disable,
 	.reclaim_buffers = drm_core_reclaim_buffers,
 	.ioctls = nouveau_ioctls,
 	.fops = {
@@ -403,6 +396,7 @@ static struct drm_driver driver = {
 		.mmap = nouveau_ttm_mmap,
 		.poll = drm_poll,
 		.fasync = drm_fasync,
+		.read = drm_read,
 #if defined(CONFIG_COMPAT)
 		.compat_ioctl = nouveau_compat_ioctl,
 #endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 1c7db64c03bf..8f13906185b2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -54,22 +54,36 @@ struct nouveau_fpriv {
 #include "nouveau_drm.h"
 #include "nouveau_reg.h"
 #include "nouveau_bios.h"
+#include "nouveau_util.h"
+
 struct nouveau_grctx;
+struct nouveau_vram;
+#include "nouveau_vm.h"
 
 #define MAX_NUM_DCB_ENTRIES 16
 
 #define NOUVEAU_MAX_CHANNEL_NR 128
 #define NOUVEAU_MAX_TILE_NR 15
 
-#define NV50_VM_MAX_VRAM (2*1024*1024*1024ULL)
-#define NV50_VM_BLOCK    (512*1024*1024ULL)
-#define NV50_VM_VRAM_NR  (NV50_VM_MAX_VRAM / NV50_VM_BLOCK)
+struct nouveau_vram {
+	struct drm_device *dev;
+
+	struct nouveau_vma bar_vma;
+
+	struct list_head regions;
+	u32 memtype;
+	u64 offset;
+	u64 size;
+};
 
 struct nouveau_tile_reg {
-	struct nouveau_fence *fence;
-	uint32_t addr;
-	uint32_t size;
 	bool used;
+	uint32_t addr;
+	uint32_t limit;
+	uint32_t pitch;
+	uint32_t zcomp;
+	struct drm_mm_node *tag_mem;
+	struct nouveau_fence *fence;
 };
 
 struct nouveau_bo {
@@ -88,6 +102,7 @@ struct nouveau_bo {
 
 	struct nouveau_channel *channel;
 
+	struct nouveau_vma vma;
 	bool mappable;
 	bool no_vm;
 
@@ -96,7 +111,6 @@ struct nouveau_bo {
 	struct nouveau_tile_reg *tile;
 
 	struct drm_gem_object *gem;
-	struct drm_file *cpu_filp;
 	int pin_refcnt;
 };
 
@@ -133,20 +147,28 @@ enum nouveau_flags {
 
 #define NVOBJ_ENGINE_SW		0
 #define NVOBJ_ENGINE_GR		1
-#define NVOBJ_ENGINE_DISPLAY	2
+#define NVOBJ_ENGINE_PPP	2
+#define NVOBJ_ENGINE_COPY	3
+#define NVOBJ_ENGINE_VP		4
+#define NVOBJ_ENGINE_CRYPT      5
+#define NVOBJ_ENGINE_BSP	6
+#define NVOBJ_ENGINE_DISPLAY	0xcafe0001
 #define NVOBJ_ENGINE_INT	0xdeadbeef
 
+#define NVOBJ_FLAG_DONT_MAP             (1 << 0)
 #define NVOBJ_FLAG_ZERO_ALLOC		(1 << 1)
 #define NVOBJ_FLAG_ZERO_FREE		(1 << 2)
+#define NVOBJ_FLAG_VM			(1 << 3)
+
+#define NVOBJ_CINST_GLOBAL	0xdeadbeef
+
 struct nouveau_gpuobj {
 	struct drm_device *dev;
 	struct kref refcount;
 	struct list_head list;
 
-	struct drm_mm_node *im_pramin;
-	struct nouveau_bo *im_backing;
-	uint32_t *im_backing_suspend;
-	int im_bound;
+	void *node;
+	u32 *suspend;
 
 	uint32_t flags;
 
@@ -162,10 +184,29 @@ struct nouveau_gpuobj {
 	void *priv;
 };
 
+struct nouveau_page_flip_state {
+	struct list_head head;
+	struct drm_pending_vblank_event *event;
+	int crtc, bpp, pitch, x, y;
+	uint64_t offset;
+};
+
+enum nouveau_channel_mutex_class {
+	NOUVEAU_UCHANNEL_MUTEX,
+	NOUVEAU_KCHANNEL_MUTEX
+};
+
 struct nouveau_channel {
 	struct drm_device *dev;
 	int id;
 
+	/* references to the channel data structure */
+	struct kref ref;
+	/* users of the hardware channel resources, the hardware
+	 * context will be kicked off when it reaches zero. */
+	atomic_t users;
+	struct mutex mutex;
+
 	/* owner of this fifo */
 	struct drm_file *file_priv;
 	/* mapping of the fifo itself */
@@ -202,12 +243,12 @@ struct nouveau_channel {
 	/* PGRAPH context */
 	/* XXX may be merge 2 pointers as private data ??? */
 	struct nouveau_gpuobj *ramin_grctx;
+	struct nouveau_gpuobj *crypt_ctx;
 	void *pgraph_ctx;
 
 	/* NV50 VM */
+	struct nouveau_vm     *vm;
 	struct nouveau_gpuobj *vm_pd;
-	struct nouveau_gpuobj *vm_gart_pt;
-	struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR];
 
 	/* Objects */
 	struct nouveau_gpuobj *ramin; /* Private instmem */
@@ -238,9 +279,11 @@ struct nouveau_channel {
 
 	struct {
 		struct nouveau_gpuobj *vblsem;
+		uint32_t vblsem_head;
 		uint32_t vblsem_offset;
 		uint32_t vblsem_rval;
 		struct list_head vbl_wait;
+		struct list_head flip;
 	} nvsw;
 
 	struct {
@@ -258,11 +301,11 @@ struct nouveau_instmem_engine {
 	int	(*suspend)(struct drm_device *dev);
 	void	(*resume)(struct drm_device *dev);
 
-	int	(*populate)(struct drm_device *, struct nouveau_gpuobj *,
-			    uint32_t *size);
-	void	(*clear)(struct drm_device *, struct nouveau_gpuobj *);
-	int	(*bind)(struct drm_device *, struct nouveau_gpuobj *);
-	int	(*unbind)(struct drm_device *, struct nouveau_gpuobj *);
+	int	(*get)(struct nouveau_gpuobj *, u32 size, u32 align);
+	void	(*put)(struct nouveau_gpuobj *);
+	int	(*map)(struct nouveau_gpuobj *);
+	void	(*unmap)(struct nouveau_gpuobj *);
+
 	void	(*flush)(struct drm_device *);
 };
 
@@ -279,12 +322,17 @@ struct nouveau_timer_engine {
 
 struct nouveau_fb_engine {
 	int num_tiles;
+	struct drm_mm tag_heap;
+	void *priv;
 
 	int  (*init)(struct drm_device *dev);
 	void (*takedown)(struct drm_device *dev);
 
-	void (*set_region_tiling)(struct drm_device *dev, int i, uint32_t addr,
-				 uint32_t size, uint32_t pitch);
+	void (*init_tile_region)(struct drm_device *dev, int i,
+				 uint32_t addr, uint32_t size,
+				 uint32_t pitch, uint32_t flags);
+	void (*set_tile_region)(struct drm_device *dev, int i);
+	void (*free_tile_region)(struct drm_device *dev, int i);
 };
 
 struct nouveau_fifo_engine {
@@ -310,21 +358,9 @@ struct nouveau_fifo_engine {
 	void (*tlb_flush)(struct drm_device *dev);
 };
 
-struct nouveau_pgraph_object_method {
-	int id;
-	int (*exec)(struct nouveau_channel *chan, int grclass, int mthd,
-		      uint32_t data);
-};
-
-struct nouveau_pgraph_object_class {
-	int id;
-	bool software;
-	struct nouveau_pgraph_object_method *methods;
-};
-
 struct nouveau_pgraph_engine {
-	struct nouveau_pgraph_object_class *grclass;
 	bool accel_blocked;
+	bool registered;
 	int grctx_size;
 
 	/* NV2x/NV3x context table (0x400780) */
@@ -342,8 +378,7 @@ struct nouveau_pgraph_engine {
 	int  (*unload_context)(struct drm_device *);
 	void (*tlb_flush)(struct drm_device *dev);
 
-	void (*set_region_tiling)(struct drm_device *dev, int i, uint32_t addr,
-				  uint32_t size, uint32_t pitch);
+	void (*set_tile_region)(struct drm_device *dev, int i);
 };
 
 struct nouveau_display_engine {
@@ -355,13 +390,19 @@ struct nouveau_display_engine {
 };
 
 struct nouveau_gpio_engine {
+	void *priv;
+
 	int  (*init)(struct drm_device *);
 	void (*takedown)(struct drm_device *);
 
 	int  (*get)(struct drm_device *, enum dcb_gpio_tag);
 	int  (*set)(struct drm_device *, enum dcb_gpio_tag, int state);
 
-	void (*irq_enable)(struct drm_device *, enum dcb_gpio_tag, bool on);
+	int  (*irq_register)(struct drm_device *, enum dcb_gpio_tag,
+			     void (*)(void *, int), void *);
+	void (*irq_unregister)(struct drm_device *, enum dcb_gpio_tag,
+			       void (*)(void *, int), void *);
+	bool (*irq_enable)(struct drm_device *, enum dcb_gpio_tag, bool on);
 };
 
 struct nouveau_pm_voltage_level {
@@ -437,6 +478,7 @@ struct nouveau_pm_engine {
 	struct nouveau_pm_level *cur;
 
 	struct device *hwmon;
+	struct notifier_block acpi_nb;
 
 	int (*clock_get)(struct drm_device *, u32 id);
 	void *(*clock_pre)(struct drm_device *, struct nouveau_pm_level *,
@@ -449,6 +491,25 @@ struct nouveau_pm_engine {
 	int (*temp_get)(struct drm_device *);
 };
 
+struct nouveau_crypt_engine {
+	bool registered;
+
+	int  (*init)(struct drm_device *);
+	void (*takedown)(struct drm_device *);
+	int  (*create_context)(struct nouveau_channel *);
+	void (*destroy_context)(struct nouveau_channel *);
+	void (*tlb_flush)(struct drm_device *dev);
+};
+
+struct nouveau_vram_engine {
+	int  (*init)(struct drm_device *);
+	int  (*get)(struct drm_device *, u64, u32 align, u32 size_nc,
+		    u32 type, struct nouveau_vram **);
+	void (*put)(struct drm_device *, struct nouveau_vram **);
+
+	bool (*flags_valid)(struct drm_device *, u32 tile_flags);
+};
+
 struct nouveau_engine {
 	struct nouveau_instmem_engine instmem;
 	struct nouveau_mc_engine      mc;
@@ -459,6 +520,8 @@ struct nouveau_engine {
 	struct nouveau_display_engine display;
 	struct nouveau_gpio_engine    gpio;
 	struct nouveau_pm_engine      pm;
+	struct nouveau_crypt_engine   crypt;
+	struct nouveau_vram_engine    vram;
 };
 
 struct nouveau_pll_vals {
@@ -577,18 +640,15 @@ struct drm_nouveau_private {
 	bool ramin_available;
 	struct drm_mm ramin_heap;
 	struct list_head gpuobj_list;
+	struct list_head classes;
 
 	struct nouveau_bo *vga_ram;
 
+	/* interrupt handling */
+	void (*irq_handler[32])(struct drm_device *);
+	bool msi_enabled;
 	struct workqueue_struct *wq;
 	struct work_struct irq_work;
-	struct work_struct hpd_work;
-
-	struct {
-		spinlock_t lock;
-		uint32_t hpd0_bits;
-		uint32_t hpd1_bits;
-	} hpd_state;
 
 	struct list_head vbl_waiting;
 
@@ -605,8 +665,10 @@ struct drm_nouveau_private {
 		struct nouveau_bo *bo;
 	} fence;
 
-	int fifo_alloc_count;
-	struct nouveau_channel *fifos[NOUVEAU_MAX_CHANNEL_NR];
+	struct {
+		spinlock_t lock;
+		struct nouveau_channel *ptr[NOUVEAU_MAX_CHANNEL_NR];
+	} channels;
 
 	struct nouveau_engine engine;
 	struct nouveau_channel *channel;
@@ -632,12 +694,14 @@ struct drm_nouveau_private {
 		uint64_t aper_free;
 
 		struct nouveau_gpuobj *sg_ctxdma;
-		struct page *sg_dummy_page;
-		dma_addr_t sg_dummy_bus;
+		struct nouveau_vma vma;
 	} gart_info;
 
 	/* nv10-nv40 tiling regions */
-	struct nouveau_tile_reg tile[NOUVEAU_MAX_TILE_NR];
+	struct {
+		struct nouveau_tile_reg reg[NOUVEAU_MAX_TILE_NR];
+		spinlock_t lock;
+	} tile;
 
 	/* VRAM/fb configuration */
 	uint64_t vram_size;
@@ -650,14 +714,12 @@ struct drm_nouveau_private {
 	uint64_t fb_aper_free;
 	int fb_mtrr;
 
+	/* BAR control (NV50-) */
+	struct nouveau_vm *bar1_vm;
+	struct nouveau_vm *bar3_vm;
+
 	/* G8x/G9x virtual address space */
-	uint64_t vm_gart_base;
-	uint64_t vm_gart_size;
-	uint64_t vm_vram_base;
-	uint64_t vm_vram_size;
-	uint64_t vm_end;
-	struct nouveau_gpuobj *vm_vram_pt[NV50_VM_VRAM_NR];
-	int vm_vram_pt_nr;
+	struct nouveau_vm *chan_vm;
 
 	struct nvbios vbios;
 
@@ -674,6 +736,7 @@ struct drm_nouveau_private {
 	struct backlight_device *backlight;
 
 	struct nouveau_channel *evo;
+	u32 evo_alloc;
 	struct {
 		struct dcb_entry *dcb;
 		u16 script;
@@ -719,16 +782,6 @@ nouveau_bo_ref(struct nouveau_bo *ref, struct nouveau_bo **pnvbo)
 	return 0;
 }
 
-#define NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(id, cl, ch) do {    \
-	struct drm_nouveau_private *nv = dev->dev_private;       \
-	if (!nouveau_channel_owner(dev, (cl), (id))) {           \
-		NV_ERROR(dev, "pid %d doesn't own channel %d\n", \
-			 DRM_CURRENTPID, (id));                  \
-		return -EPERM;                                   \
-	}                                                        \
-	(ch) = nv->fifos[(id)];                                  \
-} while (0)
-
 /* nouveau_drv.c */
 extern int nouveau_agpmode;
 extern int nouveau_duallink;
@@ -748,6 +801,7 @@ extern int nouveau_force_post;
 extern int nouveau_override_conntype;
 extern char *nouveau_perflvl;
 extern int nouveau_perflvl_wr;
+extern int nouveau_msi;
 
 extern int nouveau_pci_suspend(struct pci_dev *pdev, pm_message_t pm_state);
 extern int nouveau_pci_resume(struct pci_dev *pdev);
@@ -762,8 +816,10 @@ extern int  nouveau_ioctl_getparam(struct drm_device *, void *data,
 				   struct drm_file *);
 extern int  nouveau_ioctl_setparam(struct drm_device *, void *data,
 				   struct drm_file *);
-extern bool nouveau_wait_until(struct drm_device *, uint64_t timeout,
-			       uint32_t reg, uint32_t mask, uint32_t val);
+extern bool nouveau_wait_eq(struct drm_device *, uint64_t timeout,
+			    uint32_t reg, uint32_t mask, uint32_t val);
+extern bool nouveau_wait_ne(struct drm_device *, uint64_t timeout,
+			    uint32_t reg, uint32_t mask, uint32_t val);
 extern bool nouveau_wait_for_idle(struct drm_device *);
 extern int  nouveau_card_init(struct drm_device *);
 
@@ -775,18 +831,15 @@ extern void nouveau_mem_gart_fini(struct drm_device *);
 extern int  nouveau_mem_init_agp(struct drm_device *);
 extern int  nouveau_mem_reset_agp(struct drm_device *);
 extern void nouveau_mem_close(struct drm_device *);
-extern struct nouveau_tile_reg *nv10_mem_set_tiling(struct drm_device *dev,
-						    uint32_t addr,
-						    uint32_t size,
-						    uint32_t pitch);
-extern void nv10_mem_expire_tiling(struct drm_device *dev,
-				   struct nouveau_tile_reg *tile,
-				   struct nouveau_fence *fence);
-extern int  nv50_mem_vm_bind_linear(struct drm_device *, uint64_t virt,
-				    uint32_t size, uint32_t flags,
-				    uint64_t phys);
-extern void nv50_mem_vm_unbind(struct drm_device *, uint64_t virt,
-			       uint32_t size);
+extern int  nouveau_mem_detect(struct drm_device *);
+extern bool nouveau_mem_flags_valid(struct drm_device *, u32 tile_flags);
+extern struct nouveau_tile_reg *nv10_mem_set_tiling(
+	struct drm_device *dev, uint32_t addr, uint32_t size,
+	uint32_t pitch, uint32_t flags);
+extern void nv10_mem_put_tile_region(struct drm_device *dev,
+				     struct nouveau_tile_reg *tile,
+				     struct nouveau_fence *fence);
+extern const struct ttm_mem_type_manager_func nouveau_vram_manager;
 
 /* nouveau_notifier.c */
 extern int  nouveau_notifier_init_channel(struct nouveau_channel *);
@@ -803,21 +856,44 @@ extern int  nouveau_ioctl_notifier_free(struct drm_device *, void *data,
 extern struct drm_ioctl_desc nouveau_ioctls[];
 extern int nouveau_max_ioctl;
 extern void nouveau_channel_cleanup(struct drm_device *, struct drm_file *);
-extern int  nouveau_channel_owner(struct drm_device *, struct drm_file *,
-				  int channel);
 extern int  nouveau_channel_alloc(struct drm_device *dev,
 				  struct nouveau_channel **chan,
 				  struct drm_file *file_priv,
 				  uint32_t fb_ctxdma, uint32_t tt_ctxdma);
-extern void nouveau_channel_free(struct nouveau_channel *);
+extern struct nouveau_channel *
+nouveau_channel_get_unlocked(struct nouveau_channel *);
+extern struct nouveau_channel *
+nouveau_channel_get(struct drm_device *, struct drm_file *, int id);
+extern void nouveau_channel_put_unlocked(struct nouveau_channel **);
+extern void nouveau_channel_put(struct nouveau_channel **);
+extern void nouveau_channel_ref(struct nouveau_channel *chan,
+				struct nouveau_channel **pchan);
+extern void nouveau_channel_idle(struct nouveau_channel *chan);
 
 /* nouveau_object.c */
+#define NVOBJ_CLASS(d,c,e) do {                                                \
+	int ret = nouveau_gpuobj_class_new((d), (c), NVOBJ_ENGINE_##e);        \
+	if (ret)                                                               \
+		return ret;                                                    \
+} while(0)
+
+#define NVOBJ_MTHD(d,c,m,e) do {                                               \
+	int ret = nouveau_gpuobj_mthd_new((d), (c), (m), (e));                 \
+	if (ret)                                                               \
+		return ret;                                                    \
+} while(0)
+
 extern int  nouveau_gpuobj_early_init(struct drm_device *);
 extern int  nouveau_gpuobj_init(struct drm_device *);
 extern void nouveau_gpuobj_takedown(struct drm_device *);
 extern int  nouveau_gpuobj_suspend(struct drm_device *dev);
-extern void nouveau_gpuobj_suspend_cleanup(struct drm_device *dev);
 extern void nouveau_gpuobj_resume(struct drm_device *dev);
+extern int  nouveau_gpuobj_class_new(struct drm_device *, u32 class, u32 eng);
+extern int  nouveau_gpuobj_mthd_new(struct drm_device *, u32 class, u32 mthd,
+				    int (*exec)(struct nouveau_channel *,
+					        u32 class, u32 mthd, u32 data));
+extern int  nouveau_gpuobj_mthd_call(struct nouveau_channel *, u32, u32, u32);
+extern int  nouveau_gpuobj_mthd_call2(struct drm_device *, int, u32, u32, u32);
 extern int nouveau_gpuobj_channel_init(struct nouveau_channel *,
 				       uint32_t vram_h, uint32_t tt_h);
 extern void nouveau_gpuobj_channel_takedown(struct nouveau_channel *);
@@ -832,21 +908,25 @@ extern int nouveau_gpuobj_new_fake(struct drm_device *, u32 pinst, u64 vinst,
 extern int nouveau_gpuobj_dma_new(struct nouveau_channel *, int class,
 				  uint64_t offset, uint64_t size, int access,
 				  int target, struct nouveau_gpuobj **);
-extern int nouveau_gpuobj_gart_dma_new(struct nouveau_channel *,
-				       uint64_t offset, uint64_t size,
-				       int access, struct nouveau_gpuobj **,
-				       uint32_t *o_ret);
-extern int nouveau_gpuobj_gr_new(struct nouveau_channel *, int class,
-				 struct nouveau_gpuobj **);
-extern int nouveau_gpuobj_sw_new(struct nouveau_channel *, int class,
-				 struct nouveau_gpuobj **);
+extern int nouveau_gpuobj_gr_new(struct nouveau_channel *, u32 handle, int class);
+extern int nv50_gpuobj_dma_new(struct nouveau_channel *, int class, u64 base,
+			       u64 size, int target, int access, u32 type,
+			       u32 comp, struct nouveau_gpuobj **pobj);
+extern void nv50_gpuobj_dma_init(struct nouveau_gpuobj *, u32 offset,
+				 int class, u64 base, u64 size, int target,
+				 int access, u32 type, u32 comp);
 extern int nouveau_ioctl_grobj_alloc(struct drm_device *, void *data,
 				     struct drm_file *);
 extern int nouveau_ioctl_gpuobj_free(struct drm_device *, void *data,
 				     struct drm_file *);
 
 /* nouveau_irq.c */
+extern int         nouveau_irq_init(struct drm_device *);
+extern void        nouveau_irq_fini(struct drm_device *);
 extern irqreturn_t nouveau_irq_handler(DRM_IRQ_ARGS);
+extern void        nouveau_irq_register(struct drm_device *, int status_bit,
+					void (*)(struct drm_device *));
+extern void        nouveau_irq_unregister(struct drm_device *, int status_bit);
 extern void        nouveau_irq_preinstall(struct drm_device *);
 extern int         nouveau_irq_postinstall(struct drm_device *);
 extern void        nouveau_irq_uninstall(struct drm_device *);
@@ -854,8 +934,8 @@ extern void        nouveau_irq_uninstall(struct drm_device *);
 /* nouveau_sgdma.c */
 extern int nouveau_sgdma_init(struct drm_device *);
 extern void nouveau_sgdma_takedown(struct drm_device *);
-extern int nouveau_sgdma_get_page(struct drm_device *, uint32_t offset,
-				  uint32_t *page);
+extern uint32_t nouveau_sgdma_get_physical(struct drm_device *,
+					   uint32_t offset);
 extern struct ttm_backend *nouveau_sgdma_init_ttm(struct drm_device *);
 
 /* nouveau_debugfs.c */
@@ -966,18 +1046,25 @@ extern void nv04_fb_takedown(struct drm_device *);
 /* nv10_fb.c */
 extern int  nv10_fb_init(struct drm_device *);
 extern void nv10_fb_takedown(struct drm_device *);
-extern void nv10_fb_set_region_tiling(struct drm_device *, int, uint32_t,
-				      uint32_t, uint32_t);
+extern void nv10_fb_init_tile_region(struct drm_device *dev, int i,
+				     uint32_t addr, uint32_t size,
+				     uint32_t pitch, uint32_t flags);
+extern void nv10_fb_set_tile_region(struct drm_device *dev, int i);
+extern void nv10_fb_free_tile_region(struct drm_device *dev, int i);
 
 /* nv30_fb.c */
 extern int  nv30_fb_init(struct drm_device *);
 extern void nv30_fb_takedown(struct drm_device *);
+extern void nv30_fb_init_tile_region(struct drm_device *dev, int i,
+				     uint32_t addr, uint32_t size,
+				     uint32_t pitch, uint32_t flags);
+extern void nv30_fb_free_tile_region(struct drm_device *dev, int i);
 
 /* nv40_fb.c */
 extern int  nv40_fb_init(struct drm_device *);
 extern void nv40_fb_takedown(struct drm_device *);
-extern void nv40_fb_set_region_tiling(struct drm_device *, int, uint32_t,
-				      uint32_t, uint32_t);
+extern void nv40_fb_set_tile_region(struct drm_device *dev, int i);
+
 /* nv50_fb.c */
 extern int  nv50_fb_init(struct drm_device *);
 extern void nv50_fb_takedown(struct drm_device *);
@@ -989,6 +1076,7 @@ extern void nvc0_fb_takedown(struct drm_device *);
 
 /* nv04_fifo.c */
 extern int  nv04_fifo_init(struct drm_device *);
+extern void nv04_fifo_fini(struct drm_device *);
 extern void nv04_fifo_disable(struct drm_device *);
 extern void nv04_fifo_enable(struct drm_device *);
 extern bool nv04_fifo_reassign(struct drm_device *, bool);
@@ -998,19 +1086,18 @@ extern int  nv04_fifo_create_context(struct nouveau_channel *);
 extern void nv04_fifo_destroy_context(struct nouveau_channel *);
 extern int  nv04_fifo_load_context(struct nouveau_channel *);
 extern int  nv04_fifo_unload_context(struct drm_device *);
+extern void nv04_fifo_isr(struct drm_device *);
 
 /* nv10_fifo.c */
 extern int  nv10_fifo_init(struct drm_device *);
 extern int  nv10_fifo_channel_id(struct drm_device *);
 extern int  nv10_fifo_create_context(struct nouveau_channel *);
-extern void nv10_fifo_destroy_context(struct nouveau_channel *);
 extern int  nv10_fifo_load_context(struct nouveau_channel *);
 extern int  nv10_fifo_unload_context(struct drm_device *);
 
 /* nv40_fifo.c */
 extern int  nv40_fifo_init(struct drm_device *);
 extern int  nv40_fifo_create_context(struct nouveau_channel *);
-extern void nv40_fifo_destroy_context(struct nouveau_channel *);
 extern int  nv40_fifo_load_context(struct nouveau_channel *);
 extern int  nv40_fifo_unload_context(struct drm_device *);
 
@@ -1038,7 +1125,6 @@ extern int  nvc0_fifo_load_context(struct nouveau_channel *);
 extern int  nvc0_fifo_unload_context(struct drm_device *);
 
 /* nv04_graph.c */
-extern struct nouveau_pgraph_object_class nv04_graph_grclass[];
 extern int  nv04_graph_init(struct drm_device *);
 extern void nv04_graph_takedown(struct drm_device *);
 extern void nv04_graph_fifo_access(struct drm_device *, bool);
@@ -1047,10 +1133,11 @@ extern int  nv04_graph_create_context(struct nouveau_channel *);
 extern void nv04_graph_destroy_context(struct nouveau_channel *);
 extern int  nv04_graph_load_context(struct nouveau_channel *);
 extern int  nv04_graph_unload_context(struct drm_device *);
-extern void nv04_graph_context_switch(struct drm_device *);
+extern int  nv04_graph_mthd_page_flip(struct nouveau_channel *chan,
+				      u32 class, u32 mthd, u32 data);
+extern struct nouveau_bitfield nv04_graph_nsource[];
 
 /* nv10_graph.c */
-extern struct nouveau_pgraph_object_class nv10_graph_grclass[];
 extern int  nv10_graph_init(struct drm_device *);
 extern void nv10_graph_takedown(struct drm_device *);
 extern struct nouveau_channel *nv10_graph_channel(struct drm_device *);
@@ -1058,13 +1145,11 @@ extern int  nv10_graph_create_context(struct nouveau_channel *);
 extern void nv10_graph_destroy_context(struct nouveau_channel *);
 extern int  nv10_graph_load_context(struct nouveau_channel *);
 extern int  nv10_graph_unload_context(struct drm_device *);
-extern void nv10_graph_context_switch(struct drm_device *);
-extern void nv10_graph_set_region_tiling(struct drm_device *, int, uint32_t,
-					 uint32_t, uint32_t);
+extern void nv10_graph_set_tile_region(struct drm_device *dev, int i);
+extern struct nouveau_bitfield nv10_graph_intr[];
+extern struct nouveau_bitfield nv10_graph_nstatus[];
 
 /* nv20_graph.c */
-extern struct nouveau_pgraph_object_class nv20_graph_grclass[];
-extern struct nouveau_pgraph_object_class nv30_graph_grclass[];
 extern int  nv20_graph_create_context(struct nouveau_channel *);
 extern void nv20_graph_destroy_context(struct nouveau_channel *);
 extern int  nv20_graph_load_context(struct nouveau_channel *);
@@ -1072,11 +1157,9 @@ extern int  nv20_graph_unload_context(struct drm_device *);
 extern int  nv20_graph_init(struct drm_device *);
 extern void nv20_graph_takedown(struct drm_device *);
 extern int  nv30_graph_init(struct drm_device *);
-extern void nv20_graph_set_region_tiling(struct drm_device *, int, uint32_t,
-					 uint32_t, uint32_t);
+extern void nv20_graph_set_tile_region(struct drm_device *dev, int i);
 
 /* nv40_graph.c */
-extern struct nouveau_pgraph_object_class nv40_graph_grclass[];
 extern int  nv40_graph_init(struct drm_device *);
 extern void nv40_graph_takedown(struct drm_device *);
 extern struct nouveau_channel *nv40_graph_channel(struct drm_device *);
@@ -1085,11 +1168,9 @@ extern void nv40_graph_destroy_context(struct nouveau_channel *);
 extern int  nv40_graph_load_context(struct nouveau_channel *);
 extern int  nv40_graph_unload_context(struct drm_device *);
 extern void nv40_grctx_init(struct nouveau_grctx *);
-extern void nv40_graph_set_region_tiling(struct drm_device *, int, uint32_t,
-					 uint32_t, uint32_t);
+extern void nv40_graph_set_tile_region(struct drm_device *dev, int i);
 
 /* nv50_graph.c */
-extern struct nouveau_pgraph_object_class nv50_graph_grclass[];
 extern int  nv50_graph_init(struct drm_device *);
 extern void nv50_graph_takedown(struct drm_device *);
 extern void nv50_graph_fifo_access(struct drm_device *, bool);
@@ -1098,7 +1179,6 @@ extern int  nv50_graph_create_context(struct nouveau_channel *);
 extern void nv50_graph_destroy_context(struct nouveau_channel *);
 extern int  nv50_graph_load_context(struct nouveau_channel *);
 extern int  nv50_graph_unload_context(struct drm_device *);
-extern void nv50_graph_context_switch(struct drm_device *);
 extern int  nv50_grctx_init(struct nouveau_grctx *);
 extern void nv50_graph_tlb_flush(struct drm_device *dev);
 extern void nv86_graph_tlb_flush(struct drm_device *dev);
@@ -1113,16 +1193,22 @@ extern void nvc0_graph_destroy_context(struct nouveau_channel *);
 extern int  nvc0_graph_load_context(struct nouveau_channel *);
 extern int  nvc0_graph_unload_context(struct drm_device *);
 
+/* nv84_crypt.c */
+extern int  nv84_crypt_init(struct drm_device *dev);
+extern void nv84_crypt_fini(struct drm_device *dev);
+extern int  nv84_crypt_create_context(struct nouveau_channel *);
+extern void nv84_crypt_destroy_context(struct nouveau_channel *);
+extern void nv84_crypt_tlb_flush(struct drm_device *dev);
+
 /* nv04_instmem.c */
 extern int  nv04_instmem_init(struct drm_device *);
 extern void nv04_instmem_takedown(struct drm_device *);
 extern int  nv04_instmem_suspend(struct drm_device *);
 extern void nv04_instmem_resume(struct drm_device *);
-extern int  nv04_instmem_populate(struct drm_device *, struct nouveau_gpuobj *,
-				  uint32_t *size);
-extern void nv04_instmem_clear(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nv04_instmem_bind(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nv04_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
+extern int  nv04_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
+extern void nv04_instmem_put(struct nouveau_gpuobj *);
+extern int  nv04_instmem_map(struct nouveau_gpuobj *);
+extern void nv04_instmem_unmap(struct nouveau_gpuobj *);
 extern void nv04_instmem_flush(struct drm_device *);
 
 /* nv50_instmem.c */
@@ -1130,25 +1216,22 @@ extern int  nv50_instmem_init(struct drm_device *);
 extern void nv50_instmem_takedown(struct drm_device *);
 extern int  nv50_instmem_suspend(struct drm_device *);
 extern void nv50_instmem_resume(struct drm_device *);
-extern int  nv50_instmem_populate(struct drm_device *, struct nouveau_gpuobj *,
-				  uint32_t *size);
-extern void nv50_instmem_clear(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nv50_instmem_bind(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nv50_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
+extern int  nv50_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
+extern void nv50_instmem_put(struct nouveau_gpuobj *);
+extern int  nv50_instmem_map(struct nouveau_gpuobj *);
+extern void nv50_instmem_unmap(struct nouveau_gpuobj *);
 extern void nv50_instmem_flush(struct drm_device *);
 extern void nv84_instmem_flush(struct drm_device *);
-extern void nv50_vm_flush(struct drm_device *, int engine);
 
 /* nvc0_instmem.c */
 extern int  nvc0_instmem_init(struct drm_device *);
 extern void nvc0_instmem_takedown(struct drm_device *);
 extern int  nvc0_instmem_suspend(struct drm_device *);
 extern void nvc0_instmem_resume(struct drm_device *);
-extern int  nvc0_instmem_populate(struct drm_device *, struct nouveau_gpuobj *,
-				  uint32_t *size);
-extern void nvc0_instmem_clear(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nvc0_instmem_bind(struct drm_device *, struct nouveau_gpuobj *);
-extern int  nvc0_instmem_unbind(struct drm_device *, struct nouveau_gpuobj *);
+extern int  nvc0_instmem_get(struct nouveau_gpuobj *, u32 size, u32 align);
+extern void nvc0_instmem_put(struct nouveau_gpuobj *);
+extern int  nvc0_instmem_map(struct nouveau_gpuobj *);
+extern void nvc0_instmem_unmap(struct nouveau_gpuobj *);
 extern void nvc0_instmem_flush(struct drm_device *);
 
 /* nv04_mc.c */
@@ -1219,6 +1302,9 @@ extern u16 nouveau_bo_rd16(struct nouveau_bo *nvbo, unsigned index);
 extern void nouveau_bo_wr16(struct nouveau_bo *nvbo, unsigned index, u16 val);
 extern u32 nouveau_bo_rd32(struct nouveau_bo *nvbo, unsigned index);
 extern void nouveau_bo_wr32(struct nouveau_bo *nvbo, unsigned index, u32 val);
+extern void nouveau_bo_fence(struct nouveau_bo *, struct nouveau_fence *);
+extern int nouveau_bo_validate(struct nouveau_bo *, bool interruptible,
+			       bool no_wait_reserve, bool no_wait_gpu);
 
 /* nouveau_fence.c */
 struct nouveau_fence;
@@ -1234,12 +1320,35 @@ extern void nouveau_fence_work(struct nouveau_fence *fence,
 			       void (*work)(void *priv, bool signalled),
 			       void *priv);
 struct nouveau_channel *nouveau_fence_channel(struct nouveau_fence *);
-extern bool nouveau_fence_signalled(void *obj, void *arg);
-extern int nouveau_fence_wait(void *obj, void *arg, bool lazy, bool intr);
+
+extern bool __nouveau_fence_signalled(void *obj, void *arg);
+extern int __nouveau_fence_wait(void *obj, void *arg, bool lazy, bool intr);
+extern int __nouveau_fence_flush(void *obj, void *arg);
+extern void __nouveau_fence_unref(void **obj);
+extern void *__nouveau_fence_ref(void *obj);
+
+static inline bool nouveau_fence_signalled(struct nouveau_fence *obj)
+{
+	return __nouveau_fence_signalled(obj, NULL);
+}
+static inline int
+nouveau_fence_wait(struct nouveau_fence *obj, bool lazy, bool intr)
+{
+	return __nouveau_fence_wait(obj, NULL, lazy, intr);
+}
 extern int nouveau_fence_sync(struct nouveau_fence *, struct nouveau_channel *);
-extern int nouveau_fence_flush(void *obj, void *arg);
-extern void nouveau_fence_unref(void **obj);
-extern void *nouveau_fence_ref(void *obj);
+static inline int nouveau_fence_flush(struct nouveau_fence *obj)
+{
+	return __nouveau_fence_flush(obj, NULL);
+}
+static inline void nouveau_fence_unref(struct nouveau_fence **obj)
+{
+	__nouveau_fence_unref((void **)obj);
+}
+static inline struct nouveau_fence *nouveau_fence_ref(struct nouveau_fence *obj)
+{
+	return __nouveau_fence_ref(obj);
+}
 
 /* nouveau_gem.c */
 extern int nouveau_gem_new(struct drm_device *, struct nouveau_channel *,
@@ -1259,15 +1368,28 @@ extern int nouveau_gem_ioctl_cpu_fini(struct drm_device *, void *,
 extern int nouveau_gem_ioctl_info(struct drm_device *, void *,
 				  struct drm_file *);
 
+/* nouveau_display.c */
+int nouveau_vblank_enable(struct drm_device *dev, int crtc);
+void nouveau_vblank_disable(struct drm_device *dev, int crtc);
+int nouveau_crtc_page_flip(struct drm_crtc *crtc, struct drm_framebuffer *fb,
+			   struct drm_pending_vblank_event *event);
+int nouveau_finish_page_flip(struct nouveau_channel *,
+			     struct nouveau_page_flip_state *);
+
 /* nv10_gpio.c */
 int nv10_gpio_get(struct drm_device *dev, enum dcb_gpio_tag tag);
 int nv10_gpio_set(struct drm_device *dev, enum dcb_gpio_tag tag, int state);
 
 /* nv50_gpio.c */
 int nv50_gpio_init(struct drm_device *dev);
+void nv50_gpio_fini(struct drm_device *dev);
 int nv50_gpio_get(struct drm_device *dev, enum dcb_gpio_tag tag);
 int nv50_gpio_set(struct drm_device *dev, enum dcb_gpio_tag tag, int state);
-void nv50_gpio_irq_enable(struct drm_device *, enum dcb_gpio_tag, bool on);
+int  nv50_gpio_irq_register(struct drm_device *, enum dcb_gpio_tag,
+			    void (*)(void *, int), void *);
+void nv50_gpio_irq_unregister(struct drm_device *, enum dcb_gpio_tag,
+			      void (*)(void *, int), void *);
+bool nv50_gpio_irq_enable(struct drm_device *, enum dcb_gpio_tag, bool on);
 
 /* nv50_calc. */
 int nv50_calc_pll(struct drm_device *, struct pll_lims *, int clk,
@@ -1334,7 +1456,9 @@ static inline void nv_wr08(struct drm_device *dev, unsigned reg, u8 val)
 }
 
 #define nv_wait(dev, reg, mask, val) \
-	nouveau_wait_until(dev, 2000000000ULL, (reg), (mask), (val))
+	nouveau_wait_eq(dev, 2000000000ULL, (reg), (mask), (val))
+#define nv_wait_ne(dev, reg, mask, val) \
+	nouveau_wait_ne(dev, 2000000000ULL, (reg), (mask), (val))
 
 /* PRAMIN access */
 static inline u32 nv_ri32(struct drm_device *dev, unsigned offset)
@@ -1447,6 +1571,23 @@ nv_match_device(struct drm_device *dev, unsigned device,
 		dev->pdev->subsystem_device == sub_device;
 }
 
+/* memory type/access flags, do not match hardware values */
+#define NV_MEM_ACCESS_RO  1
+#define NV_MEM_ACCESS_WO  2
+#define NV_MEM_ACCESS_RW (NV_MEM_ACCESS_RO | NV_MEM_ACCESS_WO)
+#define NV_MEM_ACCESS_SYS 4
+#define NV_MEM_ACCESS_VM  8
+
+#define NV_MEM_TARGET_VRAM        0
+#define NV_MEM_TARGET_PCI         1
+#define NV_MEM_TARGET_PCI_NOSNOOP 2
+#define NV_MEM_TARGET_VM          3
+#define NV_MEM_TARGET_GART        4
+
+#define NV_MEM_TYPE_VM 0x7f
+#define NV_MEM_COMP_VM 0x03
+
+/* NV_SW object class */
 #define NV_SW                                                        0x0000506e
 #define NV_SW_DMA_SEMAPHORE                                          0x00000060
 #define NV_SW_SEMAPHORE_OFFSET                                       0x00000064
@@ -1457,5 +1598,6 @@ nv_match_device(struct drm_device *dev, unsigned device,
 #define NV_SW_VBLSEM_OFFSET                                          0x00000400
 #define NV_SW_VBLSEM_RELEASE_VALUE                                   0x00000404
 #define NV_SW_VBLSEM_RELEASE                                         0x00000408
+#define NV_SW_PAGE_FLIP                                              0x00000500
 
 #endif /* __NOUVEAU_DRV_H__ */
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 02a4d1fd4845..ea861c915149 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -49,6 +49,96 @@
 #include "nouveau_fbcon.h"
 #include "nouveau_dma.h"
 
+static void
+nouveau_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
+{
+	struct nouveau_fbdev *nfbdev = info->par;
+	struct drm_device *dev = nfbdev->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int ret;
+
+	if (info->state != FBINFO_STATE_RUNNING)
+		return;
+
+	ret = -ENODEV;
+	if (!in_interrupt() && !(info->flags & FBINFO_HWACCEL_DISABLED) &&
+	    mutex_trylock(&dev_priv->channel->mutex)) {
+		if (dev_priv->card_type < NV_50)
+			ret = nv04_fbcon_fillrect(info, rect);
+		else
+		if (dev_priv->card_type < NV_C0)
+			ret = nv50_fbcon_fillrect(info, rect);
+		mutex_unlock(&dev_priv->channel->mutex);
+	}
+
+	if (ret == 0)
+		return;
+
+	if (ret != -ENODEV)
+		nouveau_fbcon_gpu_lockup(info);
+	cfb_fillrect(info, rect);
+}
+
+static void
+nouveau_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *image)
+{
+	struct nouveau_fbdev *nfbdev = info->par;
+	struct drm_device *dev = nfbdev->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int ret;
+
+	if (info->state != FBINFO_STATE_RUNNING)
+		return;
+
+	ret = -ENODEV;
+	if (!in_interrupt() && !(info->flags & FBINFO_HWACCEL_DISABLED) &&
+	    mutex_trylock(&dev_priv->channel->mutex)) {
+		if (dev_priv->card_type < NV_50)
+			ret = nv04_fbcon_copyarea(info, image);
+		else
+		if (dev_priv->card_type < NV_C0)
+			ret = nv50_fbcon_copyarea(info, image);
+		mutex_unlock(&dev_priv->channel->mutex);
+	}
+
+	if (ret == 0)
+		return;
+
+	if (ret != -ENODEV)
+		nouveau_fbcon_gpu_lockup(info);
+	cfb_copyarea(info, image);
+}
+
+static void
+nouveau_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
+{
+	struct nouveau_fbdev *nfbdev = info->par;
+	struct drm_device *dev = nfbdev->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int ret;
+
+	if (info->state != FBINFO_STATE_RUNNING)
+		return;
+
+	ret = -ENODEV;
+	if (!in_interrupt() && !(info->flags & FBINFO_HWACCEL_DISABLED) &&
+	    mutex_trylock(&dev_priv->channel->mutex)) {
+		if (dev_priv->card_type < NV_50)
+			ret = nv04_fbcon_imageblit(info, image);
+		else
+		if (dev_priv->card_type < NV_C0)
+			ret = nv50_fbcon_imageblit(info, image);
+		mutex_unlock(&dev_priv->channel->mutex);
+	}
+
+	if (ret == 0)
+		return;
+
+	if (ret != -ENODEV)
+		nouveau_fbcon_gpu_lockup(info);
+	cfb_imageblit(info, image);
+}
+
 static int
 nouveau_fbcon_sync(struct fb_info *info)
 {
@@ -58,12 +148,17 @@ nouveau_fbcon_sync(struct fb_info *info)
 	struct nouveau_channel *chan = dev_priv->channel;
 	int ret, i;
 
-	if (!chan || !chan->accel_done ||
+	if (!chan || !chan->accel_done || in_interrupt() ||
 	    info->state != FBINFO_STATE_RUNNING ||
 	    info->flags & FBINFO_HWACCEL_DISABLED)
 		return 0;
 
-	if (RING_SPACE(chan, 4)) {
+	if (!mutex_trylock(&chan->mutex))
+		return 0;
+
+	ret = RING_SPACE(chan, 4);
+	if (ret) {
+		mutex_unlock(&chan->mutex);
 		nouveau_fbcon_gpu_lockup(info);
 		return 0;
 	}
@@ -74,6 +169,7 @@ nouveau_fbcon_sync(struct fb_info *info)
 	OUT_RING(chan, 0);
 	nouveau_bo_wr32(chan->notifier_bo, chan->m2mf_ntfy + 3, 0xffffffff);
 	FIRE_RING(chan);
+	mutex_unlock(&chan->mutex);
 
 	ret = -EBUSY;
 	for (i = 0; i < 100000; i++) {
@@ -97,9 +193,9 @@ static struct fb_ops nouveau_fbcon_ops = {
 	.owner = THIS_MODULE,
 	.fb_check_var = drm_fb_helper_check_var,
 	.fb_set_par = drm_fb_helper_set_par,
-	.fb_fillrect = cfb_fillrect,
-	.fb_copyarea = cfb_copyarea,
-	.fb_imageblit = cfb_imageblit,
+	.fb_fillrect = nouveau_fbcon_fillrect,
+	.fb_copyarea = nouveau_fbcon_copyarea,
+	.fb_imageblit = nouveau_fbcon_imageblit,
 	.fb_sync = nouveau_fbcon_sync,
 	.fb_pan_display = drm_fb_helper_pan_display,
 	.fb_blank = drm_fb_helper_blank,
@@ -108,29 +204,13 @@ static struct fb_ops nouveau_fbcon_ops = {
 	.fb_debug_leave = drm_fb_helper_debug_leave,
 };
 
-static struct fb_ops nv04_fbcon_ops = {
+static struct fb_ops nouveau_fbcon_sw_ops = {
 	.owner = THIS_MODULE,
 	.fb_check_var = drm_fb_helper_check_var,
 	.fb_set_par = drm_fb_helper_set_par,
-	.fb_fillrect = nv04_fbcon_fillrect,
-	.fb_copyarea = nv04_fbcon_copyarea,
-	.fb_imageblit = nv04_fbcon_imageblit,
-	.fb_sync = nouveau_fbcon_sync,
-	.fb_pan_display = drm_fb_helper_pan_display,
-	.fb_blank = drm_fb_helper_blank,
-	.fb_setcmap = drm_fb_helper_setcmap,
-	.fb_debug_enter = drm_fb_helper_debug_enter,
-	.fb_debug_leave = drm_fb_helper_debug_leave,
-};
-
-static struct fb_ops nv50_fbcon_ops = {
-	.owner = THIS_MODULE,
-	.fb_check_var = drm_fb_helper_check_var,
-	.fb_set_par = drm_fb_helper_set_par,
-	.fb_fillrect = nv50_fbcon_fillrect,
-	.fb_copyarea = nv50_fbcon_copyarea,
-	.fb_imageblit = nv50_fbcon_imageblit,
-	.fb_sync = nouveau_fbcon_sync,
+	.fb_fillrect = cfb_fillrect,
+	.fb_copyarea = cfb_copyarea,
+	.fb_imageblit = cfb_imageblit,
 	.fb_pan_display = drm_fb_helper_pan_display,
 	.fb_blank = drm_fb_helper_blank,
 	.fb_setcmap = drm_fb_helper_setcmap,
@@ -257,9 +337,9 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 			      FBINFO_HWACCEL_FILLRECT |
 			      FBINFO_HWACCEL_IMAGEBLIT;
 	info->flags |= FBINFO_CAN_FORCE_OUTPUT;
-	info->fbops = &nouveau_fbcon_ops;
-	info->fix.smem_start = dev->mode_config.fb_base + nvbo->bo.offset -
-			       dev_priv->vm_vram_base;
+	info->fbops = &nouveau_fbcon_sw_ops;
+	info->fix.smem_start = dev->mode_config.fb_base +
+			       (nvbo->bo.mem.start << PAGE_SHIFT);
 	info->fix.smem_len = size;
 
 	info->screen_base = nvbo_kmap_obj_iovirtual(nouveau_fb->nvbo);
@@ -285,19 +365,18 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 	info->pixmap.flags = FB_PIXMAP_SYSTEM;
 	info->pixmap.scan_align = 1;
 
+	mutex_unlock(&dev->struct_mutex);
+
 	if (dev_priv->channel && !nouveau_nofbaccel) {
-		switch (dev_priv->card_type) {
-		case NV_C0:
-			break;
-		case NV_50:
-			nv50_fbcon_accel_init(info);
-			info->fbops = &nv50_fbcon_ops;
-			break;
-		default:
-			nv04_fbcon_accel_init(info);
-			info->fbops = &nv04_fbcon_ops;
-			break;
-		};
+		ret = -ENODEV;
+		if (dev_priv->card_type < NV_50)
+			ret = nv04_fbcon_accel_init(info);
+		else
+		if (dev_priv->card_type < NV_C0)
+			ret = nv50_fbcon_accel_init(info);
+
+		if (ret == 0)
+			info->fbops = &nouveau_fbcon_ops;
 	}
 
 	nouveau_fbcon_zfill(dev, nfbdev);
@@ -308,7 +387,6 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 						nouveau_fb->base.height,
 						nvbo->bo.offset, nvbo);
 
-	mutex_unlock(&dev->struct_mutex);
 	vga_switcheroo_client_fb_set(dev->pdev, info);
 	return 0;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.h b/drivers/gpu/drm/nouveau/nouveau_fbcon.h
index e7e12684c37e..6b933f2c3a5b 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.h
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.h
@@ -40,13 +40,13 @@ struct nouveau_fbdev {
 
 void nouveau_fbcon_restore(void);
 
-void nv04_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
-void nv04_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
-void nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
+int nv04_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
+int nv04_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
+int nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
 int nv04_fbcon_accel_init(struct fb_info *info);
-void nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
-void nv50_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
-void nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
+int nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
+int nv50_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region);
+int nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image);
 int nv50_fbcon_accel_init(struct fb_info *info);
 
 void nouveau_fbcon_gpu_lockup(struct fb_info *info);
diff --git a/drivers/gpu/drm/nouveau/nouveau_fence.c b/drivers/gpu/drm/nouveau/nouveau_fence.c
index ab1bbfbf266e..374a9793b85f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fence.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fence.c
@@ -64,6 +64,7 @@ nouveau_fence_del(struct kref *ref)
 	struct nouveau_fence *fence =
 		container_of(ref, struct nouveau_fence, refcount);
 
+	nouveau_channel_ref(NULL, &fence->channel);
 	kfree(fence);
 }
 
@@ -76,14 +77,17 @@ nouveau_fence_update(struct nouveau_channel *chan)
 
 	spin_lock(&chan->fence.lock);
 
-	if (USE_REFCNT(dev))
-		sequence = nvchan_rd32(chan, 0x48);
-	else
-		sequence = atomic_read(&chan->fence.last_sequence_irq);
+	/* Fetch the last sequence if the channel is still up and running */
+	if (likely(!list_empty(&chan->fence.pending))) {
+		if (USE_REFCNT(dev))
+			sequence = nvchan_rd32(chan, 0x48);
+		else
+			sequence = atomic_read(&chan->fence.last_sequence_irq);
 
-	if (chan->fence.sequence_ack == sequence)
-		goto out;
-	chan->fence.sequence_ack = sequence;
+		if (chan->fence.sequence_ack == sequence)
+			goto out;
+		chan->fence.sequence_ack = sequence;
+	}
 
 	list_for_each_entry_safe(fence, tmp, &chan->fence.pending, entry) {
 		sequence = fence->sequence;
@@ -113,13 +117,13 @@ nouveau_fence_new(struct nouveau_channel *chan, struct nouveau_fence **pfence,
 	if (!fence)
 		return -ENOMEM;
 	kref_init(&fence->refcount);
-	fence->channel = chan;
+	nouveau_channel_ref(chan, &fence->channel);
 
 	if (emit)
 		ret = nouveau_fence_emit(fence);
 
 	if (ret)
-		nouveau_fence_unref((void *)&fence);
+		nouveau_fence_unref(&fence);
 	*pfence = fence;
 	return ret;
 }
@@ -127,7 +131,7 @@ nouveau_fence_new(struct nouveau_channel *chan, struct nouveau_fence **pfence,
 struct nouveau_channel *
 nouveau_fence_channel(struct nouveau_fence *fence)
 {
-	return fence ? fence->channel : NULL;
+	return fence ? nouveau_channel_get_unlocked(fence->channel) : NULL;
 }
 
 int
@@ -182,7 +186,7 @@ nouveau_fence_work(struct nouveau_fence *fence,
 }
 
 void
-nouveau_fence_unref(void **sync_obj)
+__nouveau_fence_unref(void **sync_obj)
 {
 	struct nouveau_fence *fence = nouveau_fence(*sync_obj);
 
@@ -192,7 +196,7 @@ nouveau_fence_unref(void **sync_obj)
 }
 
 void *
-nouveau_fence_ref(void *sync_obj)
+__nouveau_fence_ref(void *sync_obj)
 {
 	struct nouveau_fence *fence = nouveau_fence(sync_obj);
 
@@ -201,7 +205,7 @@ nouveau_fence_ref(void *sync_obj)
 }
 
 bool
-nouveau_fence_signalled(void *sync_obj, void *sync_arg)
+__nouveau_fence_signalled(void *sync_obj, void *sync_arg)
 {
 	struct nouveau_fence *fence = nouveau_fence(sync_obj);
 	struct nouveau_channel *chan = fence->channel;
@@ -214,13 +218,14 @@ nouveau_fence_signalled(void *sync_obj, void *sync_arg)
 }
 
 int
-nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
+__nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
 {
 	unsigned long timeout = jiffies + (3 * DRM_HZ);
+	unsigned long sleep_time = jiffies + 1;
 	int ret = 0;
 
 	while (1) {
-		if (nouveau_fence_signalled(sync_obj, sync_arg))
+		if (__nouveau_fence_signalled(sync_obj, sync_arg))
 			break;
 
 		if (time_after_eq(jiffies, timeout)) {
@@ -230,7 +235,7 @@ nouveau_fence_wait(void *sync_obj, void *sync_arg, bool lazy, bool intr)
 
 		__set_current_state(intr ? TASK_INTERRUPTIBLE
 			: TASK_UNINTERRUPTIBLE);
-		if (lazy)
+		if (lazy && time_after_eq(jiffies, sleep_time))
 			schedule_timeout(1);
 
 		if (intr && signal_pending(current)) {
@@ -368,7 +373,7 @@ emit_semaphore(struct nouveau_channel *chan, int method,
 
 	kref_get(&sema->ref);
 	nouveau_fence_work(fence, semaphore_work, sema);
-	nouveau_fence_unref((void *)&fence);
+	nouveau_fence_unref(&fence);
 
 	return 0;
 }
@@ -380,33 +385,49 @@ nouveau_fence_sync(struct nouveau_fence *fence,
 	struct nouveau_channel *chan = nouveau_fence_channel(fence);
 	struct drm_device *dev = wchan->dev;
 	struct nouveau_semaphore *sema;
-	int ret;
+	int ret = 0;
 
-	if (likely(!fence || chan == wchan ||
-		   nouveau_fence_signalled(fence, NULL)))
-		return 0;
+	if (likely(!chan || chan == wchan ||
+		   nouveau_fence_signalled(fence)))
+		goto out;
 
 	sema = alloc_semaphore(dev);
 	if (!sema) {
 		/* Early card or broken userspace, fall back to
 		 * software sync. */
-		return nouveau_fence_wait(fence, NULL, false, false);
+		ret = nouveau_fence_wait(fence, true, false);
+		goto out;
+	}
+
+	/* try to take chan's mutex, if we can't take it right away
+	 * we have to fallback to software sync to prevent locking
+	 * order issues
+	 */
+	if (!mutex_trylock(&chan->mutex)) {
+		ret = nouveau_fence_wait(fence, true, false);
+		goto out_unref;
 	}
 
 	/* Make wchan wait until it gets signalled */
 	ret = emit_semaphore(wchan, NV_SW_SEMAPHORE_ACQUIRE, sema);
 	if (ret)
-		goto out;
+		goto out_unlock;
 
 	/* Signal the semaphore from chan */
 	ret = emit_semaphore(chan, NV_SW_SEMAPHORE_RELEASE, sema);
-out:
+
+out_unlock:
+	mutex_unlock(&chan->mutex);
+out_unref:
 	kref_put(&sema->ref, free_semaphore);
+out:
+	if (chan)
+		nouveau_channel_put_unlocked(&chan);
 	return ret;
 }
 
 int
-nouveau_fence_flush(void *sync_obj, void *sync_arg)
+__nouveau_fence_flush(void *sync_obj, void *sync_arg)
 {
 	return 0;
 }
@@ -420,12 +441,7 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 	int ret;
 
 	/* Create an NV_SW object for various sync purposes */
-	ret = nouveau_gpuobj_sw_new(chan, NV_SW, &obj);
-	if (ret)
-		return ret;
-
-	ret = nouveau_ramht_insert(chan, NvSw, obj);
-	nouveau_gpuobj_ref(NULL, &obj);
+	ret = nouveau_gpuobj_gr_new(chan, NvSw, NV_SW);
 	if (ret)
 		return ret;
 
@@ -437,13 +453,12 @@ nouveau_fence_channel_init(struct nouveau_channel *chan)
 
 	/* Create a DMA object for the shared cross-channel sync area. */
 	if (USE_SEMA(dev)) {
-		struct drm_mm_node *mem = dev_priv->fence.bo->bo.mem.mm_node;
+		struct ttm_mem_reg *mem = &dev_priv->fence.bo->bo.mem;
 
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
 					     mem->start << PAGE_SHIFT,
-					     mem->size << PAGE_SHIFT,
-					     NV_DMA_ACCESS_RW,
-					     NV_DMA_TARGET_VIDMEM, &obj);
+					     mem->size, NV_MEM_ACCESS_RW,
+					     NV_MEM_TARGET_VRAM, &obj);
 		if (ret)
 			return ret;
 
@@ -473,6 +488,8 @@ nouveau_fence_channel_fini(struct nouveau_channel *chan)
 {
 	struct nouveau_fence *tmp, *fence;
 
+	spin_lock(&chan->fence.lock);
+
 	list_for_each_entry_safe(fence, tmp, &chan->fence.pending, entry) {
 		fence->signalled = true;
 		list_del(&fence->entry);
@@ -482,6 +499,8 @@ nouveau_fence_channel_fini(struct nouveau_channel *chan)
 
 		kref_put(&fence->refcount, nouveau_fence_del);
 	}
+
+	spin_unlock(&chan->fence.lock);
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nouveau_gem.c b/drivers/gpu/drm/nouveau/nouveau_gem.c
index 9a1fdcf400c2..506c508b7eda 100644
--- a/drivers/gpu/drm/nouveau/nouveau_gem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_gem.c
@@ -48,9 +48,6 @@ nouveau_gem_object_del(struct drm_gem_object *gem)
 		return;
 	nvbo->gem = NULL;
 
-	if (unlikely(nvbo->cpu_filp))
-		ttm_bo_synccpu_write_release(bo);
-
 	if (unlikely(nvbo->pin_refcnt)) {
 		nvbo->pin_refcnt = 1;
 		nouveau_bo_unpin(nvbo);
@@ -106,32 +103,6 @@ nouveau_gem_info(struct drm_gem_object *gem, struct drm_nouveau_gem_info *rep)
 	return 0;
 }
 
-static bool
-nouveau_gem_tile_flags_valid(struct drm_device *dev, uint32_t tile_flags)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (dev_priv->card_type >= NV_50) {
-		switch (tile_flags & NOUVEAU_GEM_TILE_LAYOUT_MASK) {
-		case 0x0000:
-		case 0x1800:
-		case 0x2800:
-		case 0x4800:
-		case 0x7000:
-		case 0x7400:
-		case 0x7a00:
-		case 0xe000:
-			return true;
-		}
-	} else {
-		if (!(tile_flags & NOUVEAU_GEM_TILE_LAYOUT_MASK))
-			return true;
-	}
-
-	NV_ERROR(dev, "bad page flags: 0x%08x\n", tile_flags);
-	return false;
-}
-
 int
 nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 		      struct drm_file *file_priv)
@@ -146,11 +117,6 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 	if (unlikely(dev_priv->ttm.bdev.dev_mapping == NULL))
 		dev_priv->ttm.bdev.dev_mapping = dev_priv->dev->dev_mapping;
 
-	if (req->channel_hint) {
-		NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(req->channel_hint,
-						     file_priv, chan);
-	}
-
 	if (req->info.domain & NOUVEAU_GEM_DOMAIN_VRAM)
 		flags |= TTM_PL_FLAG_VRAM;
 	if (req->info.domain & NOUVEAU_GEM_DOMAIN_GART)
@@ -158,13 +124,23 @@ nouveau_gem_ioctl_new(struct drm_device *dev, void *data,
 	if (!flags || req->info.domain & NOUVEAU_GEM_DOMAIN_CPU)
 		flags |= TTM_PL_FLAG_SYSTEM;
 
-	if (!nouveau_gem_tile_flags_valid(dev, req->info.tile_flags))
+	if (!dev_priv->engine.vram.flags_valid(dev, req->info.tile_flags)) {
+		NV_ERROR(dev, "bad page flags: 0x%08x\n", req->info.tile_flags);
 		return -EINVAL;
+	}
+
+	if (req->channel_hint) {
+		chan = nouveau_channel_get(dev, file_priv, req->channel_hint);
+		if (IS_ERR(chan))
+			return PTR_ERR(chan);
+	}
 
 	ret = nouveau_gem_new(dev, chan, req->info.size, req->align, flags,
 			      req->info.tile_mode, req->info.tile_flags, false,
 			      (req->info.domain & NOUVEAU_GEM_DOMAIN_MAPPABLE),
 			      &nvbo);
+	if (chan)
+		nouveau_channel_put(&chan);
 	if (ret)
 		return ret;
 
@@ -231,15 +207,8 @@ validate_fini_list(struct list_head *list, struct nouveau_fence *fence)
 
 	list_for_each_safe(entry, tmp, list) {
 		nvbo = list_entry(entry, struct nouveau_bo, entry);
-		if (likely(fence)) {
-			struct nouveau_fence *prev_fence;
-
-			spin_lock(&nvbo->bo.lock);
-			prev_fence = nvbo->bo.sync_obj;
-			nvbo->bo.sync_obj = nouveau_fence_ref(fence);
-			spin_unlock(&nvbo->bo.lock);
-			nouveau_fence_unref((void *)&prev_fence);
-		}
+
+		nouveau_bo_fence(nvbo, fence);
 
 		if (unlikely(nvbo->validate_mapped)) {
 			ttm_bo_kunmap(&nvbo->kmap);
@@ -299,14 +268,15 @@ retry:
 			return -EINVAL;
 		}
 
-		ret = ttm_bo_reserve(&nvbo->bo, false, false, true, sequence);
+		ret = ttm_bo_reserve(&nvbo->bo, true, false, true, sequence);
 		if (ret) {
 			validate_fini(op, NULL);
-			if (ret == -EAGAIN)
-				ret = ttm_bo_wait_unreserved(&nvbo->bo, false);
+			if (unlikely(ret == -EAGAIN))
+				ret = ttm_bo_wait_unreserved(&nvbo->bo, true);
 			drm_gem_object_unreference_unlocked(gem);
-			if (ret) {
-				NV_ERROR(dev, "fail reserve\n");
+			if (unlikely(ret)) {
+				if (ret != -ERESTARTSYS)
+					NV_ERROR(dev, "fail reserve\n");
 				return ret;
 			}
 			goto retry;
@@ -331,25 +301,6 @@ retry:
 			validate_fini(op, NULL);
 			return -EINVAL;
 		}
-
-		if (unlikely(atomic_read(&nvbo->bo.cpu_writers) > 0)) {
-			validate_fini(op, NULL);
-
-			if (nvbo->cpu_filp == file_priv) {
-				NV_ERROR(dev, "bo %p mapped by process trying "
-					      "to validate it!\n", nvbo);
-				return -EINVAL;
-			}
-
-			mutex_unlock(&drm_global_mutex);
-			ret = ttm_bo_wait_cpu(&nvbo->bo, false);
-			mutex_lock(&drm_global_mutex);
-			if (ret) {
-				NV_ERROR(dev, "fail wait_cpu\n");
-				return ret;
-			}
-			goto retry;
-		}
 	}
 
 	return 0;
@@ -383,11 +334,11 @@ validate_list(struct nouveau_channel *chan, struct list_head *list,
 		}
 
 		nvbo->channel = (b->read_domains & (1 << 31)) ? NULL : chan;
-		ret = ttm_bo_validate(&nvbo->bo, &nvbo->placement,
-				      false, false, false);
+		ret = nouveau_bo_validate(nvbo, true, false, false);
 		nvbo->channel = NULL;
 		if (unlikely(ret)) {
-			NV_ERROR(dev, "fail ttm_validate\n");
+			if (ret != -ERESTARTSYS)
+				NV_ERROR(dev, "fail ttm_validate\n");
 			return ret;
 		}
 
@@ -439,13 +390,15 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 
 	ret = validate_init(chan, file_priv, pbbo, nr_buffers, op);
 	if (unlikely(ret)) {
-		NV_ERROR(dev, "validate_init\n");
+		if (ret != -ERESTARTSYS)
+			NV_ERROR(dev, "validate_init\n");
 		return ret;
 	}
 
 	ret = validate_list(chan, &op->vram_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
-		NV_ERROR(dev, "validate vram_list\n");
+		if (ret != -ERESTARTSYS)
+			NV_ERROR(dev, "validate vram_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -453,7 +406,8 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 
 	ret = validate_list(chan, &op->gart_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
-		NV_ERROR(dev, "validate gart_list\n");
+		if (ret != -ERESTARTSYS)
+			NV_ERROR(dev, "validate gart_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -461,7 +415,8 @@ nouveau_gem_pushbuf_validate(struct nouveau_channel *chan,
 
 	ret = validate_list(chan, &op->both_list, pbbo, user_buffers);
 	if (unlikely(ret < 0)) {
-		NV_ERROR(dev, "validate both_list\n");
+		if (ret != -ERESTARTSYS)
+			NV_ERROR(dev, "validate both_list\n");
 		validate_fini(op, NULL);
 		return ret;
 	}
@@ -557,9 +512,9 @@ nouveau_gem_pushbuf_reloc_apply(struct drm_device *dev,
 				data |= r->vor;
 		}
 
-		spin_lock(&nvbo->bo.lock);
+		spin_lock(&nvbo->bo.bdev->fence_lock);
 		ret = ttm_bo_wait(&nvbo->bo, false, false, false);
-		spin_unlock(&nvbo->bo.lock);
+		spin_unlock(&nvbo->bo.bdev->fence_lock);
 		if (ret) {
 			NV_ERROR(dev, "reloc wait_idle failed: %d\n", ret);
 			break;
@@ -585,7 +540,9 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	struct nouveau_fence *fence = NULL;
 	int i, j, ret = 0, do_reloc = 0;
 
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(req->channel, file_priv, chan);
+	chan = nouveau_channel_get(dev, file_priv, req->channel);
+	if (IS_ERR(chan))
+		return PTR_ERR(chan);
 
 	req->vram_available = dev_priv->fb_aper_free;
 	req->gart_available = dev_priv->gart_info.aper_free;
@@ -595,28 +552,34 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	if (unlikely(req->nr_push > NOUVEAU_GEM_MAX_PUSH)) {
 		NV_ERROR(dev, "pushbuf push count exceeds limit: %d max %d\n",
 			 req->nr_push, NOUVEAU_GEM_MAX_PUSH);
+		nouveau_channel_put(&chan);
 		return -EINVAL;
 	}
 
 	if (unlikely(req->nr_buffers > NOUVEAU_GEM_MAX_BUFFERS)) {
 		NV_ERROR(dev, "pushbuf bo count exceeds limit: %d max %d\n",
 			 req->nr_buffers, NOUVEAU_GEM_MAX_BUFFERS);
+		nouveau_channel_put(&chan);
 		return -EINVAL;
 	}
 
 	if (unlikely(req->nr_relocs > NOUVEAU_GEM_MAX_RELOCS)) {
 		NV_ERROR(dev, "pushbuf reloc count exceeds limit: %d max %d\n",
 			 req->nr_relocs, NOUVEAU_GEM_MAX_RELOCS);
+		nouveau_channel_put(&chan);
 		return -EINVAL;
 	}
 
 	push = u_memcpya(req->push, req->nr_push, sizeof(*push));
-	if (IS_ERR(push))
+	if (IS_ERR(push)) {
+		nouveau_channel_put(&chan);
 		return PTR_ERR(push);
+	}
 
 	bo = u_memcpya(req->buffers, req->nr_buffers, sizeof(*bo));
 	if (IS_ERR(bo)) {
 		kfree(push);
+		nouveau_channel_put(&chan);
 		return PTR_ERR(bo);
 	}
 
@@ -639,7 +602,8 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 	ret = nouveau_gem_pushbuf_validate(chan, file_priv, bo, req->buffers,
 					   req->nr_buffers, &op, &do_reloc);
 	if (ret) {
-		NV_ERROR(dev, "validate: %d\n", ret);
+		if (ret != -ERESTARTSYS)
+			NV_ERROR(dev, "validate: %d\n", ret);
 		goto out;
 	}
 
@@ -732,7 +696,7 @@ nouveau_gem_ioctl_pushbuf(struct drm_device *dev, void *data,
 
 out:
 	validate_fini(&op, fence);
-	nouveau_fence_unref((void**)&fence);
+	nouveau_fence_unref(&fence);
 	kfree(bo);
 	kfree(push);
 
@@ -750,6 +714,7 @@ out_next:
 		req->suffix1 = 0x00000000;
 	}
 
+	nouveau_channel_put(&chan);
 	return ret;
 }
 
@@ -781,26 +746,9 @@ nouveau_gem_ioctl_cpu_prep(struct drm_device *dev, void *data,
 		return -ENOENT;
 	nvbo = nouveau_gem_object(gem);
 
-	if (nvbo->cpu_filp) {
-		if (nvbo->cpu_filp == file_priv)
-			goto out;
-
-		ret = ttm_bo_wait_cpu(&nvbo->bo, no_wait);
-		if (ret)
-			goto out;
-	}
-
-	if (req->flags & NOUVEAU_GEM_CPU_PREP_NOBLOCK) {
-		spin_lock(&nvbo->bo.lock);
-		ret = ttm_bo_wait(&nvbo->bo, false, false, no_wait);
-		spin_unlock(&nvbo->bo.lock);
-	} else {
-		ret = ttm_bo_synccpu_write_grab(&nvbo->bo, no_wait);
-		if (ret == 0)
-			nvbo->cpu_filp = file_priv;
-	}
-
-out:
+	spin_lock(&nvbo->bo.bdev->fence_lock);
+	ret = ttm_bo_wait(&nvbo->bo, true, true, no_wait);
+	spin_unlock(&nvbo->bo.bdev->fence_lock);
 	drm_gem_object_unreference_unlocked(gem);
 	return ret;
 }
@@ -809,26 +757,7 @@ int
 nouveau_gem_ioctl_cpu_fini(struct drm_device *dev, void *data,
 			   struct drm_file *file_priv)
 {
-	struct drm_nouveau_gem_cpu_prep *req = data;
-	struct drm_gem_object *gem;
-	struct nouveau_bo *nvbo;
-	int ret = -EINVAL;
-
-	gem = drm_gem_object_lookup(dev, file_priv, req->handle);
-	if (!gem)
-		return -ENOENT;
-	nvbo = nouveau_gem_object(gem);
-
-	if (nvbo->cpu_filp != file_priv)
-		goto out;
-	nvbo->cpu_filp = NULL;
-
-	ttm_bo_synccpu_write_release(&nvbo->bo);
-	ret = 0;
-
-out:
-	drm_gem_object_unreference_unlocked(gem);
-	return ret;
+	return 0;
 }
 
 int
diff --git a/drivers/gpu/drm/nouveau/nouveau_hw.c b/drivers/gpu/drm/nouveau/nouveau_hw.c
index b9672a05c411..053edf9d2f67 100644
--- a/drivers/gpu/drm/nouveau/nouveau_hw.c
+++ b/drivers/gpu/drm/nouveau/nouveau_hw.c
@@ -953,7 +953,7 @@ nv_load_state_ext(struct drm_device *dev, int head,
 			NVWriteCRTC(dev, head, NV_PCRTC_850, regp->crtc_850);
 
 			reg900 = NVReadRAMDAC(dev, head, NV_PRAMDAC_900);
-			if (regp->crtc_cfg == NV_PCRTC_CONFIG_START_ADDRESS_HSYNC)
+			if (regp->crtc_cfg == NV10_PCRTC_CONFIG_START_ADDRESS_HSYNC)
 				NVWriteRAMDAC(dev, head, NV_PRAMDAC_900, reg900 | 0x10000);
 			else
 				NVWriteRAMDAC(dev, head, NV_PRAMDAC_900, reg900 & ~0x10000);
@@ -999,8 +999,8 @@ nv_load_state_ext(struct drm_device *dev, int head,
 		if (dev_priv->card_type == NV_10) {
 			/* Not waiting for vertical retrace before modifying
 			   CRE_53/CRE_54 causes lockups. */
-			nouveau_wait_until(dev, 650000000, NV_PRMCIO_INP0__COLOR, 0x8, 0x8);
-			nouveau_wait_until(dev, 650000000, NV_PRMCIO_INP0__COLOR, 0x8, 0x0);
+			nouveau_wait_eq(dev, 650000000, NV_PRMCIO_INP0__COLOR, 0x8, 0x8);
+			nouveau_wait_eq(dev, 650000000, NV_PRMCIO_INP0__COLOR, 0x8, 0x0);
 		}
 
 		wr_cio_state(dev, head, regp, NV_CIO_CRE_53);
@@ -1017,8 +1017,9 @@ nv_load_state_ext(struct drm_device *dev, int head,
 
 	NVWriteCRTC(dev, head, NV_PCRTC_START, regp->fb_start);
 
-	/* Setting 1 on this value gives you interrupts for every vblank period. */
-	NVWriteCRTC(dev, head, NV_PCRTC_INTR_EN_0, 0);
+	/* Enable vblank interrupts. */
+	NVWriteCRTC(dev, head, NV_PCRTC_INTR_EN_0,
+		    (dev->vblank_enabled[head] ? 1 : 0));
 	NVWriteCRTC(dev, head, NV_PCRTC_INTR_0, NV_PCRTC_INTR_0_VBLANK);
 }
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_irq.c b/drivers/gpu/drm/nouveau/nouveau_irq.c
index 7bfd9e6c9d67..2ba7265bc967 100644
--- a/drivers/gpu/drm/nouveau/nouveau_irq.c
+++ b/drivers/gpu/drm/nouveau/nouveau_irq.c
@@ -36,18 +36,7 @@
 #include "nouveau_drv.h"
 #include "nouveau_reg.h"
 #include "nouveau_ramht.h"
-#include <linux/ratelimit.h>
-
-/* needed for hotplug irq */
-#include "nouveau_connector.h"
-#include "nv50_display.h"
-
-static DEFINE_RATELIMIT_STATE(nouveau_ratelimit_state, 3 * HZ, 20);
-
-static int nouveau_ratelimit(void)
-{
-	return __ratelimit(&nouveau_ratelimit_state);
-}
+#include "nouveau_util.h"
 
 void
 nouveau_irq_preinstall(struct drm_device *dev)
@@ -57,19 +46,19 @@ nouveau_irq_preinstall(struct drm_device *dev)
 	/* Master disable */
 	nv_wr32(dev, NV03_PMC_INTR_EN_0, 0);
 
-	if (dev_priv->card_type >= NV_50) {
-		INIT_WORK(&dev_priv->irq_work, nv50_display_irq_handler_bh);
-		INIT_WORK(&dev_priv->hpd_work, nv50_display_irq_hotplug_bh);
-		spin_lock_init(&dev_priv->hpd_state.lock);
-		INIT_LIST_HEAD(&dev_priv->vbl_waiting);
-	}
+	INIT_LIST_HEAD(&dev_priv->vbl_waiting);
 }
 
 int
 nouveau_irq_postinstall(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
 	/* Master enable */
 	nv_wr32(dev, NV03_PMC_INTR_EN_0, NV_PMC_INTR_EN_0_MASTER_ENABLE);
+	if (dev_priv->msi_enabled)
+		nv_wr08(dev, 0x00088068, 0xff);
+
 	return 0;
 }
 
@@ -80,1178 +69,83 @@ nouveau_irq_uninstall(struct drm_device *dev)
 	nv_wr32(dev, NV03_PMC_INTR_EN_0, 0);
 }
 
-static int
-nouveau_call_method(struct nouveau_channel *chan, int class, int mthd, int data)
-{
-	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
-	struct nouveau_pgraph_object_method *grm;
-	struct nouveau_pgraph_object_class *grc;
-
-	grc = dev_priv->engine.graph.grclass;
-	while (grc->id) {
-		if (grc->id == class)
-			break;
-		grc++;
-	}
-
-	if (grc->id != class || !grc->methods)
-		return -ENOENT;
-
-	grm = grc->methods;
-	while (grm->id) {
-		if (grm->id == mthd)
-			return grm->exec(chan, class, mthd, data);
-		grm++;
-	}
-
-	return -ENOENT;
-}
-
-static bool
-nouveau_fifo_swmthd(struct nouveau_channel *chan, uint32_t addr, uint32_t data)
-{
-	struct drm_device *dev = chan->dev;
-	const int subc = (addr >> 13) & 0x7;
-	const int mthd = addr & 0x1ffc;
-
-	if (mthd == 0x0000) {
-		struct nouveau_gpuobj *gpuobj;
-
-		gpuobj = nouveau_ramht_find(chan, data);
-		if (!gpuobj)
-			return false;
-
-		if (gpuobj->engine != NVOBJ_ENGINE_SW)
-			return false;
-
-		chan->sw_subchannel[subc] = gpuobj->class;
-		nv_wr32(dev, NV04_PFIFO_CACHE1_ENGINE, nv_rd32(dev,
-			NV04_PFIFO_CACHE1_ENGINE) & ~(0xf << subc * 4));
-		return true;
-	}
-
-	/* hw object */
-	if (nv_rd32(dev, NV04_PFIFO_CACHE1_ENGINE) & (1 << (subc*4)))
-		return false;
-
-	if (nouveau_call_method(chan, chan->sw_subchannel[subc], mthd, data))
-		return false;
-
-	return true;
-}
-
-static void
-nouveau_fifo_irq_handler(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
-	uint32_t status, reassign;
-	int cnt = 0;
-
-	reassign = nv_rd32(dev, NV03_PFIFO_CACHES) & 1;
-	while ((status = nv_rd32(dev, NV03_PFIFO_INTR_0)) && (cnt++ < 100)) {
-		struct nouveau_channel *chan = NULL;
-		uint32_t chid, get;
-
-		nv_wr32(dev, NV03_PFIFO_CACHES, 0);
-
-		chid = engine->fifo.channel_id(dev);
-		if (chid >= 0 && chid < engine->fifo.channels)
-			chan = dev_priv->fifos[chid];
-		get  = nv_rd32(dev, NV03_PFIFO_CACHE1_GET);
-
-		if (status & NV_PFIFO_INTR_CACHE_ERROR) {
-			uint32_t mthd, data;
-			int ptr;
-
-			/* NV_PFIFO_CACHE1_GET actually goes to 0xffc before
-			 * wrapping on my G80 chips, but CACHE1 isn't big
-			 * enough for this much data.. Tests show that it
-			 * wraps around to the start at GET=0x800.. No clue
-			 * as to why..
-			 */
-			ptr = (get & 0x7ff) >> 2;
-
-			if (dev_priv->card_type < NV_40) {
-				mthd = nv_rd32(dev,
-					NV04_PFIFO_CACHE1_METHOD(ptr));
-				data = nv_rd32(dev,
-					NV04_PFIFO_CACHE1_DATA(ptr));
-			} else {
-				mthd = nv_rd32(dev,
-					NV40_PFIFO_CACHE1_METHOD(ptr));
-				data = nv_rd32(dev,
-					NV40_PFIFO_CACHE1_DATA(ptr));
-			}
-
-			if (!chan || !nouveau_fifo_swmthd(chan, mthd, data)) {
-				NV_INFO(dev, "PFIFO_CACHE_ERROR - Ch %d/%d "
-					     "Mthd 0x%04x Data 0x%08x\n",
-					chid, (mthd >> 13) & 7, mthd & 0x1ffc,
-					data);
-			}
-
-			nv_wr32(dev, NV04_PFIFO_CACHE1_DMA_PUSH, 0);
-			nv_wr32(dev, NV03_PFIFO_INTR_0,
-						NV_PFIFO_INTR_CACHE_ERROR);
-
-			nv_wr32(dev, NV03_PFIFO_CACHE1_PUSH0,
-				nv_rd32(dev, NV03_PFIFO_CACHE1_PUSH0) & ~1);
-			nv_wr32(dev, NV03_PFIFO_CACHE1_GET, get + 4);
-			nv_wr32(dev, NV03_PFIFO_CACHE1_PUSH0,
-				nv_rd32(dev, NV03_PFIFO_CACHE1_PUSH0) | 1);
-			nv_wr32(dev, NV04_PFIFO_CACHE1_HASH, 0);
-
-			nv_wr32(dev, NV04_PFIFO_CACHE1_DMA_PUSH,
-				nv_rd32(dev, NV04_PFIFO_CACHE1_DMA_PUSH) | 1);
-			nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, 1);
-
-			status &= ~NV_PFIFO_INTR_CACHE_ERROR;
-		}
-
-		if (status & NV_PFIFO_INTR_DMA_PUSHER) {
-			u32 dma_get = nv_rd32(dev, 0x003244);
-			u32 dma_put = nv_rd32(dev, 0x003240);
-			u32 push = nv_rd32(dev, 0x003220);
-			u32 state = nv_rd32(dev, 0x003228);
-
-			if (dev_priv->card_type == NV_50) {
-				u32 ho_get = nv_rd32(dev, 0x003328);
-				u32 ho_put = nv_rd32(dev, 0x003320);
-				u32 ib_get = nv_rd32(dev, 0x003334);
-				u32 ib_put = nv_rd32(dev, 0x003330);
-
-				if (nouveau_ratelimit())
-					NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%02x%08x "
-					     "Put 0x%02x%08x IbGet 0x%08x IbPut 0x%08x "
-					     "State 0x%08x Push 0x%08x\n",
-						chid, ho_get, dma_get, ho_put,
-						dma_put, ib_get, ib_put, state,
-						push);
-
-				/* METHOD_COUNT, in DMA_STATE on earlier chipsets */
-				nv_wr32(dev, 0x003364, 0x00000000);
-				if (dma_get != dma_put || ho_get != ho_put) {
-					nv_wr32(dev, 0x003244, dma_put);
-					nv_wr32(dev, 0x003328, ho_put);
-				} else
-				if (ib_get != ib_put) {
-					nv_wr32(dev, 0x003334, ib_put);
-				}
-			} else {
-				NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%08x "
-					     "Put 0x%08x State 0x%08x Push 0x%08x\n",
-					chid, dma_get, dma_put, state, push);
-
-				if (dma_get != dma_put)
-					nv_wr32(dev, 0x003244, dma_put);
-			}
-
-			nv_wr32(dev, 0x003228, 0x00000000);
-			nv_wr32(dev, 0x003220, 0x00000001);
-			nv_wr32(dev, 0x002100, NV_PFIFO_INTR_DMA_PUSHER);
-			status &= ~NV_PFIFO_INTR_DMA_PUSHER;
-		}
-
-		if (status & NV_PFIFO_INTR_SEMAPHORE) {
-			uint32_t sem;
-
-			status &= ~NV_PFIFO_INTR_SEMAPHORE;
-			nv_wr32(dev, NV03_PFIFO_INTR_0,
-				NV_PFIFO_INTR_SEMAPHORE);
-
-			sem = nv_rd32(dev, NV10_PFIFO_CACHE1_SEMAPHORE);
-			nv_wr32(dev, NV10_PFIFO_CACHE1_SEMAPHORE, sem | 0x1);
-
-			nv_wr32(dev, NV03_PFIFO_CACHE1_GET, get + 4);
-			nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, 1);
-		}
-
-		if (dev_priv->card_type == NV_50) {
-			if (status & 0x00000010) {
-				nv50_fb_vm_trap(dev, 1, "PFIFO_BAR_FAULT");
-				status &= ~0x00000010;
-				nv_wr32(dev, 0x002100, 0x00000010);
-			}
-		}
-
-		if (status) {
-			if (nouveau_ratelimit())
-				NV_INFO(dev, "PFIFO_INTR 0x%08x - Ch %d\n",
-					status, chid);
-			nv_wr32(dev, NV03_PFIFO_INTR_0, status);
-			status = 0;
-		}
-
-		nv_wr32(dev, NV03_PFIFO_CACHES, reassign);
-	}
-
-	if (status) {
-		NV_INFO(dev, "PFIFO still angry after %d spins, halt\n", cnt);
-		nv_wr32(dev, 0x2140, 0);
-		nv_wr32(dev, 0x140, 0);
-	}
-
-	nv_wr32(dev, NV03_PMC_INTR_0, NV_PMC_INTR_0_PFIFO_PENDING);
-}
-
-struct nouveau_bitfield_names {
-	uint32_t mask;
-	const char *name;
-};
-
-static struct nouveau_bitfield_names nstatus_names[] =
-{
-	{ NV04_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
-	{ NV04_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
-	{ NV04_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
-	{ NV04_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" }
-};
-
-static struct nouveau_bitfield_names nstatus_names_nv10[] =
-{
-	{ NV10_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
-	{ NV10_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
-	{ NV10_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
-	{ NV10_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" }
-};
-
-static struct nouveau_bitfield_names nsource_names[] =
-{
-	{ NV03_PGRAPH_NSOURCE_NOTIFICATION,       "NOTIFICATION" },
-	{ NV03_PGRAPH_NSOURCE_DATA_ERROR,         "DATA_ERROR" },
-	{ NV03_PGRAPH_NSOURCE_PROTECTION_ERROR,   "PROTECTION_ERROR" },
-	{ NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION,    "RANGE_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_LIMIT_COLOR,        "LIMIT_COLOR" },
-	{ NV03_PGRAPH_NSOURCE_LIMIT_ZETA,         "LIMIT_ZETA" },
-	{ NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD,       "ILLEGAL_MTHD" },
-	{ NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION,   "DMA_R_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION,   "DMA_W_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION,   "FORMAT_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION,    "PATCH_EXCEPTION" },
-	{ NV03_PGRAPH_NSOURCE_STATE_INVALID,      "STATE_INVALID" },
-	{ NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY,      "DOUBLE_NOTIFY" },
-	{ NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE,      "NOTIFY_IN_USE" },
-	{ NV03_PGRAPH_NSOURCE_METHOD_CNT,         "METHOD_CNT" },
-	{ NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION,   "BFR_NOTIFICATION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION, "DMA_VTX_PROTECTION" },
-	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_A,        "DMA_WIDTH_A" },
-	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_B,        "DMA_WIDTH_B" },
-};
-
-static void
-nouveau_print_bitfield_names_(uint32_t value,
-				const struct nouveau_bitfield_names *namelist,
-				const int namelist_len)
-{
-	/*
-	 * Caller must have already printed the KERN_* log level for us.
-	 * Also the caller is responsible for adding the newline.
-	 */
-	int i;
-	for (i = 0; i < namelist_len; ++i) {
-		uint32_t mask = namelist[i].mask;
-		if (value & mask) {
-			printk(" %s", namelist[i].name);
-			value &= ~mask;
-		}
-	}
-	if (value)
-		printk(" (unknown bits 0x%08x)", value);
-}
-#define nouveau_print_bitfield_names(val, namelist) \
-	nouveau_print_bitfield_names_((val), (namelist), ARRAY_SIZE(namelist))
-
-struct nouveau_enum_names {
-	uint32_t value;
-	const char *name;
-};
-
-static void
-nouveau_print_enum_names_(uint32_t value,
-				const struct nouveau_enum_names *namelist,
-				const int namelist_len)
-{
-	/*
-	 * Caller must have already printed the KERN_* log level for us.
-	 * Also the caller is responsible for adding the newline.
-	 */
-	int i;
-	for (i = 0; i < namelist_len; ++i) {
-		if (value == namelist[i].value) {
-			printk("%s", namelist[i].name);
-			return;
-		}
-	}
-	printk("unknown value 0x%08x", value);
-}
-#define nouveau_print_enum_names(val, namelist) \
-	nouveau_print_enum_names_((val), (namelist), ARRAY_SIZE(namelist))
-
-static int
-nouveau_graph_chid_from_grctx(struct drm_device *dev)
+irqreturn_t
+nouveau_irq_handler(DRM_IRQ_ARGS)
 {
+	struct drm_device *dev = (struct drm_device *)arg;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t inst;
+	unsigned long flags;
+	u32 stat;
 	int i;
 
-	if (dev_priv->card_type < NV_40)
-		return dev_priv->engine.fifo.channels;
-	else
-	if (dev_priv->card_type < NV_50) {
-		inst = (nv_rd32(dev, 0x40032c) & 0xfffff) << 4;
-
-		for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-			struct nouveau_channel *chan = dev_priv->fifos[i];
-
-			if (!chan || !chan->ramin_grctx)
-				continue;
-
-			if (inst == chan->ramin_grctx->pinst)
-				break;
-		}
-	} else {
-		inst = (nv_rd32(dev, 0x40032c) & 0xfffff) << 12;
-
-		for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-			struct nouveau_channel *chan = dev_priv->fifos[i];
-
-			if (!chan || !chan->ramin)
-				continue;
-
-			if (inst == chan->ramin->vinst)
-				break;
-		}
-	}
-
-
-	return i;
-}
-
-static int
-nouveau_graph_trapped_channel(struct drm_device *dev, int *channel_ret)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
-	int channel;
-
-	if (dev_priv->card_type < NV_10)
-		channel = (nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR) >> 24) & 0xf;
-	else
-	if (dev_priv->card_type < NV_40)
-		channel = (nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR) >> 20) & 0x1f;
-	else
-		channel = nouveau_graph_chid_from_grctx(dev);
-
-	if (channel >= engine->fifo.channels || !dev_priv->fifos[channel]) {
-		NV_ERROR(dev, "AIII, invalid/inactive channel id %d\n", channel);
-		return -EINVAL;
-	}
-
-	*channel_ret = channel;
-	return 0;
-}
-
-struct nouveau_pgraph_trap {
-	int channel;
-	int class;
-	int subc, mthd, size;
-	uint32_t data, data2;
-	uint32_t nsource, nstatus;
-};
-
-static void
-nouveau_graph_trap_info(struct drm_device *dev,
-			struct nouveau_pgraph_trap *trap)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t address;
-
-	trap->nsource = trap->nstatus = 0;
-	if (dev_priv->card_type < NV_50) {
-		trap->nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
-		trap->nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
-	}
-
-	if (nouveau_graph_trapped_channel(dev, &trap->channel))
-		trap->channel = -1;
-	address = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
-
-	trap->mthd = address & 0x1FFC;
-	trap->data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
-	if (dev_priv->card_type < NV_10) {
-		trap->subc  = (address >> 13) & 0x7;
-	} else {
-		trap->subc  = (address >> 16) & 0x7;
-		trap->data2 = nv_rd32(dev, NV10_PGRAPH_TRAPPED_DATA_HIGH);
-	}
-
-	if (dev_priv->card_type < NV_10)
-		trap->class = nv_rd32(dev, 0x400180 + trap->subc*4) & 0xFF;
-	else if (dev_priv->card_type < NV_40)
-		trap->class = nv_rd32(dev, 0x400160 + trap->subc*4) & 0xFFF;
-	else if (dev_priv->card_type < NV_50)
-		trap->class = nv_rd32(dev, 0x400160 + trap->subc*4) & 0xFFFF;
-	else
-		trap->class = nv_rd32(dev, 0x400814);
-}
-
-static void
-nouveau_graph_dump_trap_info(struct drm_device *dev, const char *id,
-			     struct nouveau_pgraph_trap *trap)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t nsource = trap->nsource, nstatus = trap->nstatus;
-
-	if (dev_priv->card_type < NV_50) {
-		NV_INFO(dev, "%s - nSource:", id);
-		nouveau_print_bitfield_names(nsource, nsource_names);
-		printk(", nStatus:");
-		if (dev_priv->card_type < NV_10)
-			nouveau_print_bitfield_names(nstatus, nstatus_names);
-		else
-			nouveau_print_bitfield_names(nstatus, nstatus_names_nv10);
-		printk("\n");
-	}
-
-	NV_INFO(dev, "%s - Ch %d/%d Class 0x%04x Mthd 0x%04x "
-					"Data 0x%08x:0x%08x\n",
-					id, trap->channel, trap->subc,
-					trap->class, trap->mthd,
-					trap->data2, trap->data);
-}
-
-static int
-nouveau_pgraph_intr_swmthd(struct drm_device *dev,
-			   struct nouveau_pgraph_trap *trap)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if (trap->channel < 0 ||
-	    trap->channel >= dev_priv->engine.fifo.channels ||
-	    !dev_priv->fifos[trap->channel])
-		return -ENODEV;
-
-	return nouveau_call_method(dev_priv->fifos[trap->channel],
-				   trap->class, trap->mthd, trap->data);
-}
-
-static inline void
-nouveau_pgraph_intr_notify(struct drm_device *dev, uint32_t nsource)
-{
-	struct nouveau_pgraph_trap trap;
-	int unhandled = 0;
+	stat = nv_rd32(dev, NV03_PMC_INTR_0);
+	if (!stat)
+		return IRQ_NONE;
 
-	nouveau_graph_trap_info(dev, &trap);
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	for (i = 0; i < 32 && stat; i++) {
+		if (!(stat & (1 << i)) || !dev_priv->irq_handler[i])
+			continue;
 
-	if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
-		if (nouveau_pgraph_intr_swmthd(dev, &trap))
-			unhandled = 1;
-	} else {
-		unhandled = 1;
+		dev_priv->irq_handler[i](dev);
+		stat &= ~(1 << i);
 	}
 
-	if (unhandled)
-		nouveau_graph_dump_trap_info(dev, "PGRAPH_NOTIFY", &trap);
-}
-
-
-static inline void
-nouveau_pgraph_intr_error(struct drm_device *dev, uint32_t nsource)
-{
-	struct nouveau_pgraph_trap trap;
-	int unhandled = 0;
-
-	nouveau_graph_trap_info(dev, &trap);
-	trap.nsource = nsource;
-
-	if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
-		if (nouveau_pgraph_intr_swmthd(dev, &trap))
-			unhandled = 1;
-	} else if (nsource & NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION) {
-		uint32_t v = nv_rd32(dev, 0x402000);
-		nv_wr32(dev, 0x402000, v);
-
-		/* dump the error anyway for now: it's useful for
-		   Gallium development */
-		unhandled = 1;
-	} else {
-		unhandled = 1;
-	}
+	if (dev_priv->msi_enabled)
+		nv_wr08(dev, 0x00088068, 0xff);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 
-	if (unhandled && nouveau_ratelimit())
-		nouveau_graph_dump_trap_info(dev, "PGRAPH_ERROR", &trap);
+	if (stat && nouveau_ratelimit())
+		NV_ERROR(dev, "PMC - unhandled INTR 0x%08x\n", stat);
+	return IRQ_HANDLED;
 }
 
-static inline void
-nouveau_pgraph_intr_context_switch(struct drm_device *dev)
+int
+nouveau_irq_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
-	uint32_t chid;
-
-	chid = engine->fifo.channel_id(dev);
-	NV_DEBUG(dev, "PGRAPH context switch interrupt channel %x\n", chid);
-
-	switch (dev_priv->card_type) {
-	case NV_04:
-		nv04_graph_context_switch(dev);
-		break;
-	case NV_10:
-		nv10_graph_context_switch(dev);
-		break;
-	default:
-		NV_ERROR(dev, "Context switch not implemented\n");
-		break;
-	}
-}
-
-static void
-nouveau_pgraph_irq_handler(struct drm_device *dev)
-{
-	uint32_t status;
-
-	while ((status = nv_rd32(dev, NV03_PGRAPH_INTR))) {
-		uint32_t nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
-
-		if (status & NV_PGRAPH_INTR_NOTIFY) {
-			nouveau_pgraph_intr_notify(dev, nsource);
-
-			status &= ~NV_PGRAPH_INTR_NOTIFY;
-			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_NOTIFY);
-		}
-
-		if (status & NV_PGRAPH_INTR_ERROR) {
-			nouveau_pgraph_intr_error(dev, nsource);
+	int ret;
 
-			status &= ~NV_PGRAPH_INTR_ERROR;
-			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_ERROR);
+	if (nouveau_msi != 0 && dev_priv->card_type >= NV_50) {
+		ret = pci_enable_msi(dev->pdev);
+		if (ret == 0) {
+			NV_INFO(dev, "enabled MSI\n");
+			dev_priv->msi_enabled = true;
 		}
-
-		if (status & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
-			status &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
-			nv_wr32(dev, NV03_PGRAPH_INTR,
-				 NV_PGRAPH_INTR_CONTEXT_SWITCH);
-
-			nouveau_pgraph_intr_context_switch(dev);
-		}
-
-		if (status) {
-			NV_INFO(dev, "Unhandled PGRAPH_INTR - 0x%08x\n", status);
-			nv_wr32(dev, NV03_PGRAPH_INTR, status);
-		}
-
-		if ((nv_rd32(dev, NV04_PGRAPH_FIFO) & (1 << 0)) == 0)
-			nv_wr32(dev, NV04_PGRAPH_FIFO, 1);
 	}
 
-	nv_wr32(dev, NV03_PMC_INTR_0, NV_PMC_INTR_0_PGRAPH_PENDING);
-}
-
-static struct nouveau_enum_names nv50_mp_exec_error_names[] =
-{
-	{ 3, "STACK_UNDERFLOW" },
-	{ 4, "QUADON_ACTIVE" },
-	{ 8, "TIMEOUT" },
-	{ 0x10, "INVALID_OPCODE" },
-	{ 0x40, "BREAKPOINT" },
-};
-
-static void
-nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t units = nv_rd32(dev, 0x1540);
-	uint32_t addr, mp10, status, pc, oplow, ophigh;
-	int i;
-	int mps = 0;
-	for (i = 0; i < 4; i++) {
-		if (!(units & 1 << (i+24)))
-			continue;
-		if (dev_priv->chipset < 0xa0)
-			addr = 0x408200 + (tpid << 12) + (i << 7);
-		else
-			addr = 0x408100 + (tpid << 11) + (i << 7);
-		mp10 = nv_rd32(dev, addr + 0x10);
-		status = nv_rd32(dev, addr + 0x14);
-		if (!status)
-			continue;
-		if (display) {
-			nv_rd32(dev, addr + 0x20);
-			pc = nv_rd32(dev, addr + 0x24);
-			oplow = nv_rd32(dev, addr + 0x70);
-			ophigh= nv_rd32(dev, addr + 0x74);
-			NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - "
-					"TP %d MP %d: ", tpid, i);
-			nouveau_print_enum_names(status,
-					nv50_mp_exec_error_names);
-			printk(" at %06x warp %d, opcode %08x %08x\n",
-					pc&0xffffff, pc >> 24,
-					oplow, ophigh);
-		}
-		nv_wr32(dev, addr + 0x10, mp10);
-		nv_wr32(dev, addr + 0x14, 0);
-		mps++;
-	}
-	if (!mps && display)
-		NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - TP %d: "
-				"No MPs claiming errors?\n", tpid);
+	return drm_irq_install(dev);
 }
 
-static void
-nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
-		uint32_t ustatus_new, int display, const char *name)
+void
+nouveau_irq_fini(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	int tps = 0;
-	uint32_t units = nv_rd32(dev, 0x1540);
-	int i, r;
-	uint32_t ustatus_addr, ustatus;
-	for (i = 0; i < 16; i++) {
-		if (!(units & (1 << i)))
-			continue;
-		if (dev_priv->chipset < 0xa0)
-			ustatus_addr = ustatus_old + (i << 12);
-		else
-			ustatus_addr = ustatus_new + (i << 11);
-		ustatus = nv_rd32(dev, ustatus_addr) & 0x7fffffff;
-		if (!ustatus)
-			continue;
-		tps++;
-		switch (type) {
-		case 6: /* texture error... unknown for now */
-			nv50_fb_vm_trap(dev, display, name);
-			if (display) {
-				NV_ERROR(dev, "magic set %d:\n", i);
-				for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
-					NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
-						nv_rd32(dev, r));
-			}
-			break;
-		case 7: /* MP error */
-			if (ustatus & 0x00010000) {
-				nv50_pgraph_mp_trap(dev, i, display);
-				ustatus &= ~0x00010000;
-			}
-			break;
-		case 8: /* TPDMA error */
-			{
-			uint32_t e0c = nv_rd32(dev, ustatus_addr + 4);
-			uint32_t e10 = nv_rd32(dev, ustatus_addr + 8);
-			uint32_t e14 = nv_rd32(dev, ustatus_addr + 0xc);
-			uint32_t e18 = nv_rd32(dev, ustatus_addr + 0x10);
-			uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14);
-			uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18);
-			uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c);
-			nv50_fb_vm_trap(dev, display, name);
-			/* 2d engine destination */
-			if (ustatus & 0x00000010) {
-				if (display) {
-					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n",
-							i, e14, e10);
-					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
-							i, e0c, e18, e1c, e20, e24);
-				}
-				ustatus &= ~0x00000010;
-			}
-			/* Render target */
-			if (ustatus & 0x00000040) {
-				if (display) {
-					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n",
-							i, e14, e10);
-					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
-							i, e0c, e18, e1c, e20, e24);
-				}
-				ustatus &= ~0x00000040;
-			}
-			/* CUDA memory: l[], g[] or stack. */
-			if (ustatus & 0x00000080) {
-				if (display) {
-					if (e18 & 0x80000000) {
-						/* g[] read fault? */
-						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n",
-								i, e14, e10 | ((e18 >> 24) & 0x1f));
-						e18 &= ~0x1f000000;
-					} else if (e18 & 0xc) {
-						/* g[] write fault? */
-						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n",
-								i, e14, e10 | ((e18 >> 7) & 0x1f));
-						e18 &= ~0x00000f80;
-					} else {
-						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n",
-								i, e14, e10);
-					}
-					NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
-							i, e0c, e18, e1c, e20, e24);
-				}
-				ustatus &= ~0x00000080;
-			}
-			}
-			break;
-		}
-		if (ustatus) {
-			if (display)
-				NV_INFO(dev, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
-		}
-		nv_wr32(dev, ustatus_addr, 0xc0000000);
-	}
-
-	if (!tps && display)
-		NV_INFO(dev, "%s - No TPs claiming errors?\n", name);
-}
-
-static void
-nv50_pgraph_trap_handler(struct drm_device *dev)
-{
-	struct nouveau_pgraph_trap trap;
-	uint32_t status = nv_rd32(dev, 0x400108);
-	uint32_t ustatus;
-	int display = nouveau_ratelimit();
-
-
-	if (!status && display) {
-		nouveau_graph_trap_info(dev, &trap);
-		nouveau_graph_dump_trap_info(dev, "PGRAPH_TRAP", &trap);
-		NV_INFO(dev, "PGRAPH_TRAP - no units reporting traps?\n");
-	}
-
-	/* DISPATCH: Relays commands to other units and handles NOTIFY,
-	 * COND, QUERY. If you get a trap from it, the command is still stuck
-	 * in DISPATCH and you need to do something about it. */
-	if (status & 0x001) {
-		ustatus = nv_rd32(dev, 0x400804) & 0x7fffffff;
-		if (!ustatus && display) {
-			NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - no ustatus?\n");
-		}
-
-		/* Known to be triggered by screwed up NOTIFY and COND... */
-		if (ustatus & 0x00000001) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_DISPATCH_FAULT");
-			nv_wr32(dev, 0x400500, 0);
-			if (nv_rd32(dev, 0x400808) & 0x80000000) {
-				if (display) {
-					if (nouveau_graph_trapped_channel(dev, &trap.channel))
-						trap.channel = -1;
-					trap.class = nv_rd32(dev, 0x400814);
-					trap.mthd = nv_rd32(dev, 0x400808) & 0x1ffc;
-					trap.subc = (nv_rd32(dev, 0x400808) >> 16) & 0x7;
-					trap.data = nv_rd32(dev, 0x40080c);
-					trap.data2 = nv_rd32(dev, 0x400810);
-					nouveau_graph_dump_trap_info(dev,
-							"PGRAPH_TRAP_DISPATCH_FAULT", &trap);
-					NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - 400808: %08x\n", nv_rd32(dev, 0x400808));
-					NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - 400848: %08x\n", nv_rd32(dev, 0x400848));
-				}
-				nv_wr32(dev, 0x400808, 0);
-			} else if (display) {
-				NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_FAULT - No stuck command?\n");
-			}
-			nv_wr32(dev, 0x4008e8, nv_rd32(dev, 0x4008e8) & 3);
-			nv_wr32(dev, 0x400848, 0);
-			ustatus &= ~0x00000001;
-		}
-		if (ustatus & 0x00000002) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_DISPATCH_QUERY");
-			nv_wr32(dev, 0x400500, 0);
-			if (nv_rd32(dev, 0x40084c) & 0x80000000) {
-				if (display) {
-					if (nouveau_graph_trapped_channel(dev, &trap.channel))
-						trap.channel = -1;
-					trap.class = nv_rd32(dev, 0x400814);
-					trap.mthd = nv_rd32(dev, 0x40084c) & 0x1ffc;
-					trap.subc = (nv_rd32(dev, 0x40084c) >> 16) & 0x7;
-					trap.data = nv_rd32(dev, 0x40085c);
-					trap.data2 = 0;
-					nouveau_graph_dump_trap_info(dev,
-							"PGRAPH_TRAP_DISPATCH_QUERY", &trap);
-					NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_QUERY - 40084c: %08x\n", nv_rd32(dev, 0x40084c));
-				}
-				nv_wr32(dev, 0x40084c, 0);
-			} else if (display) {
-				NV_INFO(dev, "PGRAPH_TRAP_DISPATCH_QUERY - No stuck command?\n");
-			}
-			ustatus &= ~0x00000002;
-		}
-		if (ustatus && display)
-			NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - Unhandled ustatus 0x%08x\n", ustatus);
-		nv_wr32(dev, 0x400804, 0xc0000000);
-		nv_wr32(dev, 0x400108, 0x001);
-		status &= ~0x001;
-	}
-
-	/* TRAPs other than dispatch use the "normal" trap regs. */
-	if (status && display) {
-		nouveau_graph_trap_info(dev, &trap);
-		nouveau_graph_dump_trap_info(dev,
-				"PGRAPH_TRAP", &trap);
-	}
-
-	/* M2MF: Memory to memory copy engine. */
-	if (status & 0x002) {
-		ustatus = nv_rd32(dev, 0x406800) & 0x7fffffff;
-		if (!ustatus && display) {
-			NV_INFO(dev, "PGRAPH_TRAP_M2MF - no ustatus?\n");
-		}
-		if (ustatus & 0x00000001) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_NOTIFY");
-			ustatus &= ~0x00000001;
-		}
-		if (ustatus & 0x00000002) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_IN");
-			ustatus &= ~0x00000002;
-		}
-		if (ustatus & 0x00000004) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_M2MF_OUT");
-			ustatus &= ~0x00000004;
-		}
-		NV_INFO (dev, "PGRAPH_TRAP_M2MF - %08x %08x %08x %08x\n",
-				nv_rd32(dev, 0x406804),
-				nv_rd32(dev, 0x406808),
-				nv_rd32(dev, 0x40680c),
-				nv_rd32(dev, 0x406810));
-		if (ustatus && display)
-			NV_INFO(dev, "PGRAPH_TRAP_M2MF - Unhandled ustatus 0x%08x\n", ustatus);
-		/* No sane way found yet -- just reset the bugger. */
-		nv_wr32(dev, 0x400040, 2);
-		nv_wr32(dev, 0x400040, 0);
-		nv_wr32(dev, 0x406800, 0xc0000000);
-		nv_wr32(dev, 0x400108, 0x002);
-		status &= ~0x002;
-	}
-
-	/* VFETCH: Fetches data from vertex buffers. */
-	if (status & 0x004) {
-		ustatus = nv_rd32(dev, 0x400c04) & 0x7fffffff;
-		if (!ustatus && display) {
-			NV_INFO(dev, "PGRAPH_TRAP_VFETCH - no ustatus?\n");
-		}
-		if (ustatus & 0x00000001) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_VFETCH_FAULT");
-			NV_INFO (dev, "PGRAPH_TRAP_VFETCH_FAULT - %08x %08x %08x %08x\n",
-					nv_rd32(dev, 0x400c00),
-					nv_rd32(dev, 0x400c08),
-					nv_rd32(dev, 0x400c0c),
-					nv_rd32(dev, 0x400c10));
-			ustatus &= ~0x00000001;
-		}
-		if (ustatus && display)
-			NV_INFO(dev, "PGRAPH_TRAP_VFETCH - Unhandled ustatus 0x%08x\n", ustatus);
-		nv_wr32(dev, 0x400c04, 0xc0000000);
-		nv_wr32(dev, 0x400108, 0x004);
-		status &= ~0x004;
-	}
-
-	/* STRMOUT: DirectX streamout / OpenGL transform feedback. */
-	if (status & 0x008) {
-		ustatus = nv_rd32(dev, 0x401800) & 0x7fffffff;
-		if (!ustatus && display) {
-			NV_INFO(dev, "PGRAPH_TRAP_STRMOUT - no ustatus?\n");
-		}
-		if (ustatus & 0x00000001) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_STRMOUT_FAULT");
-			NV_INFO (dev, "PGRAPH_TRAP_STRMOUT_FAULT - %08x %08x %08x %08x\n",
-					nv_rd32(dev, 0x401804),
-					nv_rd32(dev, 0x401808),
-					nv_rd32(dev, 0x40180c),
-					nv_rd32(dev, 0x401810));
-			ustatus &= ~0x00000001;
-		}
-		if (ustatus && display)
-			NV_INFO(dev, "PGRAPH_TRAP_STRMOUT - Unhandled ustatus 0x%08x\n", ustatus);
-		/* No sane way found yet -- just reset the bugger. */
-		nv_wr32(dev, 0x400040, 0x80);
-		nv_wr32(dev, 0x400040, 0);
-		nv_wr32(dev, 0x401800, 0xc0000000);
-		nv_wr32(dev, 0x400108, 0x008);
-		status &= ~0x008;
-	}
-
-	/* CCACHE: Handles code and c[] caches and fills them. */
-	if (status & 0x010) {
-		ustatus = nv_rd32(dev, 0x405018) & 0x7fffffff;
-		if (!ustatus && display) {
-			NV_INFO(dev, "PGRAPH_TRAP_CCACHE - no ustatus?\n");
-		}
-		if (ustatus & 0x00000001) {
-			nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_CCACHE_FAULT");
-			NV_INFO (dev, "PGRAPH_TRAP_CCACHE_FAULT - %08x %08x %08x %08x %08x %08x %08x\n",
-					nv_rd32(dev, 0x405800),
-					nv_rd32(dev, 0x405804),
-					nv_rd32(dev, 0x405808),
-					nv_rd32(dev, 0x40580c),
-					nv_rd32(dev, 0x405810),
-					nv_rd32(dev, 0x405814),
-					nv_rd32(dev, 0x40581c));
-			ustatus &= ~0x00000001;
-		}
-		if (ustatus && display)
-			NV_INFO(dev, "PGRAPH_TRAP_CCACHE - Unhandled ustatus 0x%08x\n", ustatus);
-		nv_wr32(dev, 0x405018, 0xc0000000);
-		nv_wr32(dev, 0x400108, 0x010);
-		status &= ~0x010;
-	}
-
-	/* Unknown, not seen yet... 0x402000 is the only trap status reg
-	 * remaining, so try to handle it anyway. Perhaps related to that
-	 * unknown DMA slot on tesla? */
-	if (status & 0x20) {
-		nv50_fb_vm_trap(dev, display, "PGRAPH_TRAP_UNKC04");
-		ustatus = nv_rd32(dev, 0x402000) & 0x7fffffff;
-		if (display)
-			NV_INFO(dev, "PGRAPH_TRAP_UNKC04 - Unhandled ustatus 0x%08x\n", ustatus);
-		nv_wr32(dev, 0x402000, 0xc0000000);
-		/* no status modifiction on purpose */
-	}
-
-	/* TEXTURE: CUDA texturing units */
-	if (status & 0x040) {
-		nv50_pgraph_tp_trap (dev, 6, 0x408900, 0x408600, display,
-				"PGRAPH_TRAP_TEXTURE");
-		nv_wr32(dev, 0x400108, 0x040);
-		status &= ~0x040;
-	}
-
-	/* MP: CUDA execution engines. */
-	if (status & 0x080) {
-		nv50_pgraph_tp_trap (dev, 7, 0x408314, 0x40831c, display,
-				"PGRAPH_TRAP_MP");
-		nv_wr32(dev, 0x400108, 0x080);
-		status &= ~0x080;
-	}
-
-	/* TPDMA:  Handles TP-initiated uncached memory accesses:
-	 * l[], g[], stack, 2d surfaces, render targets. */
-	if (status & 0x100) {
-		nv50_pgraph_tp_trap (dev, 8, 0x408e08, 0x408708, display,
-				"PGRAPH_TRAP_TPDMA");
-		nv_wr32(dev, 0x400108, 0x100);
-		status &= ~0x100;
-	}
-
-	if (status) {
-		if (display)
-			NV_INFO(dev, "PGRAPH_TRAP - Unknown trap 0x%08x\n",
-				status);
-		nv_wr32(dev, 0x400108, status);
-	}
-}
-
-/* There must be a *lot* of these. Will take some time to gather them up. */
-static struct nouveau_enum_names nv50_data_error_names[] =
-{
-	{ 4,	"INVALID_VALUE" },
-	{ 5,	"INVALID_ENUM" },
-	{ 8,	"INVALID_OBJECT" },
-	{ 0xc,	"INVALID_BITFIELD" },
-	{ 0x28,	"MP_NO_REG_SPACE" },
-	{ 0x2b,	"MP_BLOCK_SIZE_MISMATCH" },
-};
-
-static void
-nv50_pgraph_irq_handler(struct drm_device *dev)
-{
-	struct nouveau_pgraph_trap trap;
-	int unhandled = 0;
-	uint32_t status;
-
-	while ((status = nv_rd32(dev, NV03_PGRAPH_INTR))) {
-		/* NOTIFY: You've set a NOTIFY an a command and it's done. */
-		if (status & 0x00000001) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_NOTIFY", &trap);
-			status &= ~0x00000001;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000001);
-		}
-
-		/* COMPUTE_QUERY: Purpose and exact cause unknown, happens
-		 * when you write 0x200 to 0x50c0 method 0x31c. */
-		if (status & 0x00000002) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_COMPUTE_QUERY", &trap);
-			status &= ~0x00000002;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000002);
-		}
-
-		/* Unknown, never seen: 0x4 */
-
-		/* ILLEGAL_MTHD: You used a wrong method for this class. */
-		if (status & 0x00000010) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_pgraph_intr_swmthd(dev, &trap))
-				unhandled = 1;
-			if (unhandled && nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_ILLEGAL_MTHD", &trap);
-			status &= ~0x00000010;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000010);
-		}
-
-		/* ILLEGAL_CLASS: You used a wrong class. */
-		if (status & 0x00000020) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_ILLEGAL_CLASS", &trap);
-			status &= ~0x00000020;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000020);
-		}
-
-		/* DOUBLE_NOTIFY: You tried to set a NOTIFY on another NOTIFY. */
-		if (status & 0x00000040) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_DOUBLE_NOTIFY", &trap);
-			status &= ~0x00000040;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00000040);
-		}
-
-		/* CONTEXT_SWITCH: PGRAPH needs us to load a new context */
-		if (status & 0x00001000) {
-			nv_wr32(dev, 0x400500, 0x00000000);
-			nv_wr32(dev, NV03_PGRAPH_INTR,
-				NV_PGRAPH_INTR_CONTEXT_SWITCH);
-			nv_wr32(dev, NV40_PGRAPH_INTR_EN, nv_rd32(dev,
-				NV40_PGRAPH_INTR_EN) &
-				~NV_PGRAPH_INTR_CONTEXT_SWITCH);
-			nv_wr32(dev, 0x400500, 0x00010001);
-
-			nv50_graph_context_switch(dev);
-
-			status &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
-		}
-
-		/* BUFFER_NOTIFY: Your m2mf transfer finished */
-		if (status & 0x00010000) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_BUFFER_NOTIFY", &trap);
-			status &= ~0x00010000;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00010000);
-		}
-
-		/* DATA_ERROR: Invalid value for this method, or invalid
-		 * state in current PGRAPH context for this operation */
-		if (status & 0x00100000) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit()) {
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_DATA_ERROR", &trap);
-				NV_INFO (dev, "PGRAPH_DATA_ERROR - ");
-				nouveau_print_enum_names(nv_rd32(dev, 0x400110),
-						nv50_data_error_names);
-				printk("\n");
-			}
-			status &= ~0x00100000;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00100000);
-		}
 
-		/* TRAP: Something bad happened in the middle of command
-		 * execution.  Has a billion types, subtypes, and even
-		 * subsubtypes. */
-		if (status & 0x00200000) {
-			nv50_pgraph_trap_handler(dev);
-			status &= ~0x00200000;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x00200000);
-		}
-
-		/* Unknown, never seen: 0x00400000 */
-
-		/* SINGLE_STEP: Happens on every method if you turned on
-		 * single stepping in 40008c */
-		if (status & 0x01000000) {
-			nouveau_graph_trap_info(dev, &trap);
-			if (nouveau_ratelimit())
-				nouveau_graph_dump_trap_info(dev,
-						"PGRAPH_SINGLE_STEP", &trap);
-			status &= ~0x01000000;
-			nv_wr32(dev, NV03_PGRAPH_INTR, 0x01000000);
-		}
-
-		/* 0x02000000 happens when you pause a ctxprog...
-		 * but the only way this can happen that I know is by
-		 * poking the relevant MMIO register, and we don't
-		 * do that. */
-
-		if (status) {
-			NV_INFO(dev, "Unhandled PGRAPH_INTR - 0x%08x\n",
-				status);
-			nv_wr32(dev, NV03_PGRAPH_INTR, status);
-		}
-
-		{
-			const int isb = (1 << 16) | (1 << 0);
-
-			if ((nv_rd32(dev, 0x400500) & isb) != isb)
-				nv_wr32(dev, 0x400500,
-					nv_rd32(dev, 0x400500) | isb);
-		}
-	}
-
-	nv_wr32(dev, NV03_PMC_INTR_0, NV_PMC_INTR_0_PGRAPH_PENDING);
-	if (nv_rd32(dev, 0x400824) & (1 << 31))
-		nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
+	drm_irq_uninstall(dev);
+	if (dev_priv->msi_enabled)
+		pci_disable_msi(dev->pdev);
 }
 
-static void
-nouveau_crtc_irq_handler(struct drm_device *dev, int crtc)
+void
+nouveau_irq_register(struct drm_device *dev, int status_bit,
+		     void (*handler)(struct drm_device *))
 {
-	if (crtc & 1)
-		nv_wr32(dev, NV_CRTC0_INTSTAT, NV_CRTC_INTR_VBLANK);
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	unsigned long flags;
 
-	if (crtc & 2)
-		nv_wr32(dev, NV_CRTC1_INTSTAT, NV_CRTC_INTR_VBLANK);
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	dev_priv->irq_handler[status_bit] = handler;
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
 
-irqreturn_t
-nouveau_irq_handler(DRM_IRQ_ARGS)
+void
+nouveau_irq_unregister(struct drm_device *dev, int status_bit)
 {
-	struct drm_device *dev = (struct drm_device *)arg;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t status;
 	unsigned long flags;
 
-	status = nv_rd32(dev, NV03_PMC_INTR_0);
-	if (!status)
-		return IRQ_NONE;
-
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-
-	if (status & NV_PMC_INTR_0_PFIFO_PENDING) {
-		nouveau_fifo_irq_handler(dev);
-		status &= ~NV_PMC_INTR_0_PFIFO_PENDING;
-	}
-
-	if (status & NV_PMC_INTR_0_PGRAPH_PENDING) {
-		if (dev_priv->card_type >= NV_50)
-			nv50_pgraph_irq_handler(dev);
-		else
-			nouveau_pgraph_irq_handler(dev);
-
-		status &= ~NV_PMC_INTR_0_PGRAPH_PENDING;
-	}
-
-	if (status & NV_PMC_INTR_0_CRTCn_PENDING) {
-		nouveau_crtc_irq_handler(dev, (status>>24)&3);
-		status &= ~NV_PMC_INTR_0_CRTCn_PENDING;
-	}
-
-	if (status & (NV_PMC_INTR_0_NV50_DISPLAY_PENDING |
-		      NV_PMC_INTR_0_NV50_I2C_PENDING)) {
-		nv50_display_irq_handler(dev);
-		status &= ~(NV_PMC_INTR_0_NV50_DISPLAY_PENDING |
-			    NV_PMC_INTR_0_NV50_I2C_PENDING);
-	}
-
-	if (status)
-		NV_ERROR(dev, "Unhandled PMC INTR status bits 0x%08x\n", status);
-
+	dev_priv->irq_handler[status_bit] = NULL;
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
-
-	return IRQ_HANDLED;
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index fe4a30dc4b42..224181193a1f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -36,183 +36,112 @@
 
 #include "nouveau_drv.h"
 #include "nouveau_pm.h"
+#include "nouveau_mm.h"
+#include "nouveau_vm.h"
 
 /*
  * NV10-NV40 tiling helpers
  */
 
 static void
-nv10_mem_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			   uint32_t size, uint32_t pitch)
+nv10_mem_update_tile_region(struct drm_device *dev,
+			    struct nouveau_tile_reg *tile, uint32_t addr,
+			    uint32_t size, uint32_t pitch, uint32_t flags)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_tile_reg *tile = &dev_priv->tile[i];
+	int i = tile - dev_priv->tile.reg;
+	unsigned long save;
 
-	tile->addr = addr;
-	tile->size = size;
-	tile->used = !!pitch;
-	nouveau_fence_unref((void **)&tile->fence);
+	nouveau_fence_unref(&tile->fence);
 
+	if (tile->pitch)
+		pfb->free_tile_region(dev, i);
+
+	if (pitch)
+		pfb->init_tile_region(dev, i, addr, size, pitch, flags);
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, save);
 	pfifo->reassign(dev, false);
 	pfifo->cache_pull(dev, false);
 
 	nouveau_wait_for_idle(dev);
 
-	pgraph->set_region_tiling(dev, i, addr, size, pitch);
-	pfb->set_region_tiling(dev, i, addr, size, pitch);
+	pfb->set_tile_region(dev, i);
+	pgraph->set_tile_region(dev, i);
 
 	pfifo->cache_pull(dev, true);
 	pfifo->reassign(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, save);
 }
 
-struct nouveau_tile_reg *
-nv10_mem_set_tiling(struct drm_device *dev, uint32_t addr, uint32_t size,
-		    uint32_t pitch)
+static struct nouveau_tile_reg *
+nv10_mem_get_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
-	struct nouveau_tile_reg *found = NULL;
-	unsigned long i, flags;
-
-	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
-
-	for (i = 0; i < pfb->num_tiles; i++) {
-		struct nouveau_tile_reg *tile = &dev_priv->tile[i];
-
-		if (tile->used)
-			/* Tile region in use. */
-			continue;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
-		if (tile->fence &&
-		    !nouveau_fence_signalled(tile->fence, NULL))
-			/* Pending tile region. */
-			continue;
-
-		if (max(tile->addr, addr) <
-		    min(tile->addr + tile->size, addr + size))
-			/* Kill an intersecting tile region. */
-			nv10_mem_set_region_tiling(dev, i, 0, 0, 0);
-
-		if (pitch && !found) {
-			/* Free tile region. */
-			nv10_mem_set_region_tiling(dev, i, addr, size, pitch);
-			found = tile;
-		}
-	}
+	spin_lock(&dev_priv->tile.lock);
 
-	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+	if (!tile->used &&
+	    (!tile->fence || nouveau_fence_signalled(tile->fence)))
+		tile->used = true;
+	else
+		tile = NULL;
 
-	return found;
+	spin_unlock(&dev_priv->tile.lock);
+	return tile;
 }
 
 void
-nv10_mem_expire_tiling(struct drm_device *dev, struct nouveau_tile_reg *tile,
-		       struct nouveau_fence *fence)
-{
-	if (fence) {
-		/* Mark it as pending. */
-		tile->fence = fence;
-		nouveau_fence_ref(fence);
-	}
-
-	tile->used = false;
-}
-
-/*
- * NV50 VM helpers
- */
-int
-nv50_mem_vm_bind_linear(struct drm_device *dev, uint64_t virt, uint32_t size,
-			uint32_t flags, uint64_t phys)
+nv10_mem_put_tile_region(struct drm_device *dev, struct nouveau_tile_reg *tile,
+			 struct nouveau_fence *fence)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *pgt;
-	unsigned block;
-	int i;
 
-	virt = ((virt - dev_priv->vm_vram_base) >> 16) << 1;
-	size = (size >> 16) << 1;
-
-	phys |= ((uint64_t)flags << 32);
-	phys |= 1;
-	if (dev_priv->vram_sys_base) {
-		phys += dev_priv->vram_sys_base;
-		phys |= 0x30;
-	}
-
-	while (size) {
-		unsigned offset_h = upper_32_bits(phys);
-		unsigned offset_l = lower_32_bits(phys);
-		unsigned pte, end;
-
-		for (i = 7; i >= 0; i--) {
-			block = 1 << (i + 1);
-			if (size >= block && !(virt & (block - 1)))
-				break;
+	if (tile) {
+		spin_lock(&dev_priv->tile.lock);
+		if (fence) {
+			/* Mark it as pending. */
+			tile->fence = fence;
+			nouveau_fence_ref(fence);
 		}
-		offset_l |= (i << 7);
-
-		phys += block << 15;
-		size -= block;
-
-		while (block) {
-			pgt = dev_priv->vm_vram_pt[virt >> 14];
-			pte = virt & 0x3ffe;
-
-			end = pte + block;
-			if (end > 16384)
-				end = 16384;
-			block -= (end - pte);
-			virt  += (end - pte);
-
-			while (pte < end) {
-				nv_wo32(pgt, (pte * 4) + 0, offset_l);
-				nv_wo32(pgt, (pte * 4) + 4, offset_h);
-				pte += 2;
-			}
-		}
-	}
 
-	dev_priv->engine.instmem.flush(dev);
-	dev_priv->engine.fifo.tlb_flush(dev);
-	dev_priv->engine.graph.tlb_flush(dev);
-	nv50_vm_flush(dev, 6);
-	return 0;
+		tile->used = false;
+		spin_unlock(&dev_priv->tile.lock);
+	}
 }
 
-void
-nv50_mem_vm_unbind(struct drm_device *dev, uint64_t virt, uint32_t size)
+struct nouveau_tile_reg *
+nv10_mem_set_tiling(struct drm_device *dev, uint32_t addr, uint32_t size,
+		    uint32_t pitch, uint32_t flags)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *pgt;
-	unsigned pages, pte, end;
-
-	virt -= dev_priv->vm_vram_base;
-	pages = (size >> 16) << 1;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	struct nouveau_tile_reg *tile, *found = NULL;
+	int i;
 
-	while (pages) {
-		pgt = dev_priv->vm_vram_pt[virt >> 29];
-		pte = (virt & 0x1ffe0000ULL) >> 15;
+	for (i = 0; i < pfb->num_tiles; i++) {
+		tile = nv10_mem_get_tile_region(dev, i);
 
-		end = pte + pages;
-		if (end > 16384)
-			end = 16384;
-		pages -= (end - pte);
-		virt  += (end - pte) << 15;
+		if (pitch && !found) {
+			found = tile;
+			continue;
 
-		while (pte < end) {
-			nv_wo32(pgt, (pte * 4), 0);
-			pte++;
+		} else if (tile && tile->pitch) {
+			/* Kill an unused tile region. */
+			nv10_mem_update_tile_region(dev, tile, 0, 0, 0, 0);
 		}
+
+		nv10_mem_put_tile_region(dev, tile, NULL);
 	}
 
-	dev_priv->engine.instmem.flush(dev);
-	dev_priv->engine.fifo.tlb_flush(dev);
-	dev_priv->engine.graph.tlb_flush(dev);
-	nv50_vm_flush(dev, 6);
+	if (found)
+		nv10_mem_update_tile_region(dev, found, addr, size,
+					    pitch, flags);
+	return found;
 }
 
 /*
@@ -312,62 +241,7 @@ nouveau_mem_detect_nforce(struct drm_device *dev)
 	return 0;
 }
 
-static void
-nv50_vram_preinit(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	int i, parts, colbits, rowbitsa, rowbitsb, banks;
-	u64 rowsize, predicted;
-	u32 r0, r4, rt, ru;
-
-	r0 = nv_rd32(dev, 0x100200);
-	r4 = nv_rd32(dev, 0x100204);
-	rt = nv_rd32(dev, 0x100250);
-	ru = nv_rd32(dev, 0x001540);
-	NV_DEBUG(dev, "memcfg 0x%08x 0x%08x 0x%08x 0x%08x\n", r0, r4, rt, ru);
-
-	for (i = 0, parts = 0; i < 8; i++) {
-		if (ru & (0x00010000 << i))
-			parts++;
-	}
-
-	colbits  =  (r4 & 0x0000f000) >> 12;
-	rowbitsa = ((r4 & 0x000f0000) >> 16) + 8;
-	rowbitsb = ((r4 & 0x00f00000) >> 20) + 8;
-	banks    = ((r4 & 0x01000000) ? 8 : 4);
-
-	rowsize = parts * banks * (1 << colbits) * 8;
-	predicted = rowsize << rowbitsa;
-	if (r0 & 0x00000004)
-		predicted += rowsize << rowbitsb;
-
-	if (predicted != dev_priv->vram_size) {
-		NV_WARN(dev, "memory controller reports %dMiB VRAM\n",
-			(u32)(dev_priv->vram_size >> 20));
-		NV_WARN(dev, "we calculated %dMiB VRAM\n",
-			(u32)(predicted >> 20));
-	}
-
-	dev_priv->vram_rblock_size = rowsize >> 12;
-	if (rt & 1)
-		dev_priv->vram_rblock_size *= 3;
-
-	NV_DEBUG(dev, "rblock %lld bytes\n",
-		 (u64)dev_priv->vram_rblock_size << 12);
-}
-
-static void
-nvaa_vram_preinit(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	/* To our knowledge, there's no large scale reordering of pages
-	 * that occurs on IGP chipsets.
-	 */
-	dev_priv->vram_rblock_size = 1;
-}
-
-static int
+int
 nouveau_mem_detect(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -381,40 +255,25 @@ nouveau_mem_detect(struct drm_device *dev)
 	if (dev_priv->card_type < NV_50) {
 		dev_priv->vram_size  = nv_rd32(dev, NV04_PFB_FIFO_DATA);
 		dev_priv->vram_size &= NV10_PFB_FIFO_DATA_RAM_AMOUNT_MB_MASK;
-	} else
-	if (dev_priv->card_type < NV_C0) {
-		dev_priv->vram_size = nv_rd32(dev, NV04_PFB_FIFO_DATA);
-		dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32;
-		dev_priv->vram_size &= 0xffffffff00ll;
-
-		switch (dev_priv->chipset) {
-		case 0xaa:
-		case 0xac:
-		case 0xaf:
-			dev_priv->vram_sys_base = nv_rd32(dev, 0x100e10);
-			dev_priv->vram_sys_base <<= 12;
-			nvaa_vram_preinit(dev);
-			break;
-		default:
-			nv50_vram_preinit(dev);
-			break;
-		}
 	} else {
 		dev_priv->vram_size  = nv_rd32(dev, 0x10f20c) << 20;
 		dev_priv->vram_size *= nv_rd32(dev, 0x121c74);
 	}
 
-	NV_INFO(dev, "Detected %dMiB VRAM\n", (int)(dev_priv->vram_size >> 20));
-	if (dev_priv->vram_sys_base) {
-		NV_INFO(dev, "Stolen system memory at: 0x%010llx\n",
-			dev_priv->vram_sys_base);
-	}
-
 	if (dev_priv->vram_size)
 		return 0;
 	return -ENOMEM;
 }
 
+bool
+nouveau_mem_flags_valid(struct drm_device *dev, u32 tile_flags)
+{
+	if (!(tile_flags & NOUVEAU_GEM_TILE_LAYOUT_MASK))
+		return true;
+
+	return false;
+}
+
 #if __OS_HAS_AGP
 static unsigned long
 get_agp_mode(struct drm_device *dev, unsigned long mode)
@@ -547,10 +406,6 @@ nouveau_mem_vram_init(struct drm_device *dev)
 	if (ret)
 		return ret;
 
-	ret = nouveau_mem_detect(dev);
-	if (ret)
-		return ret;
-
 	dev_priv->fb_phys = pci_resource_start(dev->pdev, 1);
 
 	ret = nouveau_ttm_global_init(dev_priv);
@@ -566,13 +421,6 @@ nouveau_mem_vram_init(struct drm_device *dev)
 		return ret;
 	}
 
-	dev_priv->fb_available_size = dev_priv->vram_size;
-	dev_priv->fb_mappable_pages = dev_priv->fb_available_size;
-	if (dev_priv->fb_mappable_pages > pci_resource_len(dev->pdev, 1))
-		dev_priv->fb_mappable_pages =
-			pci_resource_len(dev->pdev, 1);
-	dev_priv->fb_mappable_pages >>= PAGE_SHIFT;
-
 	/* reserve space at end of VRAM for PRAMIN */
 	if (dev_priv->chipset == 0x40 || dev_priv->chipset == 0x47 ||
 	    dev_priv->chipset == 0x49 || dev_priv->chipset == 0x4b)
@@ -583,6 +431,22 @@ nouveau_mem_vram_init(struct drm_device *dev)
 	else
 		dev_priv->ramin_rsvd_vram = (512 * 1024);
 
+	ret = dev_priv->engine.vram.init(dev);
+	if (ret)
+		return ret;
+
+	NV_INFO(dev, "Detected %dMiB VRAM\n", (int)(dev_priv->vram_size >> 20));
+	if (dev_priv->vram_sys_base) {
+		NV_INFO(dev, "Stolen system memory at: 0x%010llx\n",
+			dev_priv->vram_sys_base);
+	}
+
+	dev_priv->fb_available_size = dev_priv->vram_size;
+	dev_priv->fb_mappable_pages = dev_priv->fb_available_size;
+	if (dev_priv->fb_mappable_pages > pci_resource_len(dev->pdev, 1))
+		dev_priv->fb_mappable_pages = pci_resource_len(dev->pdev, 1);
+	dev_priv->fb_mappable_pages >>= PAGE_SHIFT;
+
 	dev_priv->fb_available_size -= dev_priv->ramin_rsvd_vram;
 	dev_priv->fb_aper_free = dev_priv->fb_available_size;
 
@@ -799,3 +663,114 @@ nouveau_mem_timing_fini(struct drm_device *dev)
 
 	kfree(mem->timing);
 }
+
+static int
+nouveau_vram_manager_init(struct ttm_mem_type_manager *man, unsigned long p_size)
+{
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev);
+	struct nouveau_mm *mm;
+	u32 b_size;
+	int ret;
+
+	p_size = (p_size << PAGE_SHIFT) >> 12;
+	b_size = dev_priv->vram_rblock_size >> 12;
+
+	ret = nouveau_mm_init(&mm, 0, p_size, b_size);
+	if (ret)
+		return ret;
+
+	man->priv = mm;
+	return 0;
+}
+
+static int
+nouveau_vram_manager_fini(struct ttm_mem_type_manager *man)
+{
+	struct nouveau_mm *mm = man->priv;
+	int ret;
+
+	ret = nouveau_mm_fini(&mm);
+	if (ret)
+		return ret;
+
+	man->priv = NULL;
+	return 0;
+}
+
+static void
+nouveau_vram_manager_del(struct ttm_mem_type_manager *man,
+			 struct ttm_mem_reg *mem)
+{
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev);
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	struct drm_device *dev = dev_priv->dev;
+
+	vram->put(dev, (struct nouveau_vram **)&mem->mm_node);
+}
+
+static int
+nouveau_vram_manager_new(struct ttm_mem_type_manager *man,
+			 struct ttm_buffer_object *bo,
+			 struct ttm_placement *placement,
+			 struct ttm_mem_reg *mem)
+{
+	struct drm_nouveau_private *dev_priv = nouveau_bdev(man->bdev);
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	struct drm_device *dev = dev_priv->dev;
+	struct nouveau_bo *nvbo = nouveau_bo(bo);
+	struct nouveau_vram *node;
+	u32 size_nc = 0;
+	int ret;
+
+	if (nvbo->tile_flags & NOUVEAU_GEM_TILE_NONCONTIG)
+		size_nc = 1 << nvbo->vma.node->type;
+
+	ret = vram->get(dev, mem->num_pages << PAGE_SHIFT,
+			mem->page_alignment << PAGE_SHIFT, size_nc,
+			(nvbo->tile_flags >> 8) & 0xff, &node);
+	if (ret)
+		return ret;
+
+	mem->mm_node = node;
+	mem->start   = node->offset >> PAGE_SHIFT;
+	return 0;
+}
+
+void
+nouveau_vram_manager_debug(struct ttm_mem_type_manager *man, const char *prefix)
+{
+	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm_node *r;
+	u64 total = 0, ttotal[3] = {}, tused[3] = {}, tfree[3] = {};
+	int i;
+
+	mutex_lock(&mm->mutex);
+	list_for_each_entry(r, &mm->nodes, nl_entry) {
+		printk(KERN_DEBUG "%s %s-%d: 0x%010llx 0x%010llx\n",
+		       prefix, r->free ? "free" : "used", r->type,
+		       ((u64)r->offset << 12),
+		       (((u64)r->offset + r->length) << 12));
+		total += r->length;
+		ttotal[r->type] += r->length;
+		if (r->free)
+			tfree[r->type] += r->length;
+		else
+			tused[r->type] += r->length;
+	}
+	mutex_unlock(&mm->mutex);
+
+	printk(KERN_DEBUG "%s  total: 0x%010llx\n", prefix, total << 12);
+	for (i = 0; i < 3; i++) {
+		printk(KERN_DEBUG "%s type %d: 0x%010llx, "
+				  "used 0x%010llx, free 0x%010llx\n", prefix,
+		       i, ttotal[i] << 12, tused[i] << 12, tfree[i] << 12);
+	}
+}
+
+const struct ttm_mem_type_manager_func nouveau_vram_manager = {
+	nouveau_vram_manager_init,
+	nouveau_vram_manager_fini,
+	nouveau_vram_manager_new,
+	nouveau_vram_manager_del,
+	nouveau_vram_manager_debug
+};
diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.c b/drivers/gpu/drm/nouveau/nouveau_mm.c
new file mode 100644
index 000000000000..cdbb11eb701b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.c
@@ -0,0 +1,271 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_mm.h"
+
+static inline void
+region_put(struct nouveau_mm *rmm, struct nouveau_mm_node *a)
+{
+	list_del(&a->nl_entry);
+	list_del(&a->fl_entry);
+	kfree(a);
+}
+
+static struct nouveau_mm_node *
+region_split(struct nouveau_mm *rmm, struct nouveau_mm_node *a, u32 size)
+{
+	struct nouveau_mm_node *b;
+
+	if (a->length == size)
+		return a;
+
+	b = kmalloc(sizeof(*b), GFP_KERNEL);
+	if (unlikely(b == NULL))
+		return NULL;
+
+	b->offset = a->offset;
+	b->length = size;
+	b->free   = a->free;
+	b->type   = a->type;
+	a->offset += size;
+	a->length -= size;
+	list_add_tail(&b->nl_entry, &a->nl_entry);
+	if (b->free)
+		list_add_tail(&b->fl_entry, &a->fl_entry);
+	return b;
+}
+
+static struct nouveau_mm_node *
+nouveau_mm_merge(struct nouveau_mm *rmm, struct nouveau_mm_node *this)
+{
+	struct nouveau_mm_node *prev, *next;
+
+	/* try to merge with free adjacent entries of same type */
+	prev = list_entry(this->nl_entry.prev, struct nouveau_mm_node, nl_entry);
+	if (this->nl_entry.prev != &rmm->nodes) {
+		if (prev->free && prev->type == this->type) {
+			prev->length += this->length;
+			region_put(rmm, this);
+			this = prev;
+		}
+	}
+
+	next = list_entry(this->nl_entry.next, struct nouveau_mm_node, nl_entry);
+	if (this->nl_entry.next != &rmm->nodes) {
+		if (next->free && next->type == this->type) {
+			next->offset  = this->offset;
+			next->length += this->length;
+			region_put(rmm, this);
+			this = next;
+		}
+	}
+
+	return this;
+}
+
+void
+nouveau_mm_put(struct nouveau_mm *rmm, struct nouveau_mm_node *this)
+{
+	u32 block_s, block_l;
+
+	this->free = true;
+	list_add(&this->fl_entry, &rmm->free);
+	this = nouveau_mm_merge(rmm, this);
+
+	/* any entirely free blocks now?  we'll want to remove typing
+	 * on them now so they can be use for any memory allocation
+	 */
+	block_s = roundup(this->offset, rmm->block_size);
+	if (block_s + rmm->block_size > this->offset + this->length)
+		return;
+
+	/* split off any still-typed region at the start */
+	if (block_s != this->offset) {
+		if (!region_split(rmm, this, block_s - this->offset))
+			return;
+	}
+
+	/* split off the soon-to-be-untyped block(s) */
+	block_l = rounddown(this->length, rmm->block_size);
+	if (block_l != this->length) {
+		this = region_split(rmm, this, block_l);
+		if (!this)
+			return;
+	}
+
+	/* mark as having no type, and retry merge with any adjacent
+	 * untyped blocks
+	 */
+	this->type = 0;
+	nouveau_mm_merge(rmm, this);
+}
+
+int
+nouveau_mm_get(struct nouveau_mm *rmm, int type, u32 size, u32 size_nc,
+	       u32 align, struct nouveau_mm_node **pnode)
+{
+	struct nouveau_mm_node *this, *tmp, *next;
+	u32 splitoff, avail, alloc;
+
+	list_for_each_entry_safe(this, tmp, &rmm->free, fl_entry) {
+		next = list_entry(this->nl_entry.next, struct nouveau_mm_node, nl_entry);
+		if (this->nl_entry.next == &rmm->nodes)
+			next = NULL;
+
+		/* skip wrongly typed blocks */
+		if (this->type && this->type != type)
+			continue;
+
+		/* account for alignment */
+		splitoff = this->offset & (align - 1);
+		if (splitoff)
+			splitoff = align - splitoff;
+
+		if (this->length <= splitoff)
+			continue;
+
+		/* determine total memory available from this, and
+		 * the next block (if appropriate)
+		 */
+		avail = this->length;
+		if (next && next->free && (!next->type || next->type == type))
+			avail += next->length;
+
+		avail -= splitoff;
+
+		/* determine allocation size */
+		if (size_nc) {
+			alloc = min(avail, size);
+			alloc = rounddown(alloc, size_nc);
+			if (alloc == 0)
+				continue;
+		} else {
+			alloc = size;
+			if (avail < alloc)
+				continue;
+		}
+
+		/* untyped block, split off a chunk that's a multiple
+		 * of block_size and type it
+		 */
+		if (!this->type) {
+			u32 block = roundup(alloc + splitoff, rmm->block_size);
+			if (this->length < block)
+				continue;
+
+			this = region_split(rmm, this, block);
+			if (!this)
+				return -ENOMEM;
+
+			this->type = type;
+		}
+
+		/* stealing memory from adjacent block */
+		if (alloc > this->length) {
+			u32 amount = alloc - (this->length - splitoff);
+
+			if (!next->type) {
+				amount = roundup(amount, rmm->block_size);
+
+				next = region_split(rmm, next, amount);
+				if (!next)
+					return -ENOMEM;
+
+				next->type = type;
+			}
+
+			this->length += amount;
+			next->offset += amount;
+			next->length -= amount;
+			if (!next->length) {
+				list_del(&next->nl_entry);
+				list_del(&next->fl_entry);
+				kfree(next);
+			}
+		}
+
+		if (splitoff) {
+			if (!region_split(rmm, this, splitoff))
+				return -ENOMEM;
+		}
+
+		this = region_split(rmm, this, alloc);
+		if (this == NULL)
+			return -ENOMEM;
+
+		this->free = false;
+		list_del(&this->fl_entry);
+		*pnode = this;
+		return 0;
+	}
+
+	return -ENOMEM;
+}
+
+int
+nouveau_mm_init(struct nouveau_mm **prmm, u32 offset, u32 length, u32 block)
+{
+	struct nouveau_mm *rmm;
+	struct nouveau_mm_node *heap;
+
+	heap = kzalloc(sizeof(*heap), GFP_KERNEL);
+	if (!heap)
+		return -ENOMEM;
+	heap->free = true;
+	heap->offset = roundup(offset, block);
+	heap->length = rounddown(offset + length, block) - heap->offset;
+
+	rmm = kzalloc(sizeof(*rmm), GFP_KERNEL);
+	if (!rmm) {
+		kfree(heap);
+		return -ENOMEM;
+	}
+	rmm->block_size = block;
+	mutex_init(&rmm->mutex);
+	INIT_LIST_HEAD(&rmm->nodes);
+	INIT_LIST_HEAD(&rmm->free);
+	list_add(&heap->nl_entry, &rmm->nodes);
+	list_add(&heap->fl_entry, &rmm->free);
+
+	*prmm = rmm;
+	return 0;
+}
+
+int
+nouveau_mm_fini(struct nouveau_mm **prmm)
+{
+	struct nouveau_mm *rmm = *prmm;
+	struct nouveau_mm_node *heap =
+		list_first_entry(&rmm->nodes, struct nouveau_mm_node, nl_entry);
+
+	if (!list_is_singular(&rmm->nodes))
+		return -EBUSY;
+
+	kfree(heap);
+	kfree(rmm);
+	*prmm = NULL;
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.h b/drivers/gpu/drm/nouveau/nouveau_mm.h
new file mode 100644
index 000000000000..250e642de0a7
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.h
@@ -0,0 +1,62 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#ifndef __NOUVEAU_REGION_H__
+#define __NOUVEAU_REGION_H__
+
+struct nouveau_mm_node {
+	struct list_head nl_entry;
+	struct list_head fl_entry;
+	struct list_head rl_entry;
+
+	bool free;
+	int  type;
+
+	u32 offset;
+	u32 length;
+};
+
+struct nouveau_mm {
+	struct list_head nodes;
+	struct list_head free;
+
+	struct mutex mutex;
+
+	u32 block_size;
+};
+
+int  nouveau_mm_init(struct nouveau_mm **, u32 offset, u32 length, u32 block);
+int  nouveau_mm_fini(struct nouveau_mm **);
+int  nouveau_mm_pre(struct nouveau_mm *);
+int  nouveau_mm_get(struct nouveau_mm *, int type, u32 size, u32 size_nc,
+		    u32 align, struct nouveau_mm_node **);
+void nouveau_mm_put(struct nouveau_mm *, struct nouveau_mm_node *);
+
+int  nv50_vram_init(struct drm_device *);
+int  nv50_vram_new(struct drm_device *, u64 size, u32 align, u32 size_nc,
+		    u32 memtype, struct nouveau_vram **);
+void nv50_vram_del(struct drm_device *, struct nouveau_vram **);
+bool nv50_vram_flags_valid(struct drm_device *, u32 tile_flags);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_notifier.c b/drivers/gpu/drm/nouveau/nouveau_notifier.c
index 2cc59f8c658b..a050b7b69782 100644
--- a/drivers/gpu/drm/nouveau/nouveau_notifier.c
+++ b/drivers/gpu/drm/nouveau/nouveau_notifier.c
@@ -99,7 +99,6 @@ nouveau_notifier_alloc(struct nouveau_channel *chan, uint32_t handle,
 		       int size, uint32_t *b_offset)
 {
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *nobj = NULL;
 	struct drm_mm_node *mem;
 	uint32_t offset;
@@ -113,31 +112,15 @@ nouveau_notifier_alloc(struct nouveau_channel *chan, uint32_t handle,
 		return -ENOMEM;
 	}
 
-	offset = chan->notifier_bo->bo.mem.start << PAGE_SHIFT;
-	if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_VRAM) {
-		target = NV_DMA_TARGET_VIDMEM;
-	} else
-	if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_TT) {
-		if (dev_priv->gart_info.type == NOUVEAU_GART_SGDMA &&
-		    dev_priv->card_type < NV_50) {
-			ret = nouveau_sgdma_get_page(dev, offset, &offset);
-			if (ret)
-				return ret;
-			target = NV_DMA_TARGET_PCI;
-		} else {
-			target = NV_DMA_TARGET_AGP;
-			if (dev_priv->card_type >= NV_50)
-				offset += dev_priv->vm_gart_base;
-		}
-	} else {
-		NV_ERROR(dev, "Bad DMA target, mem_type %d!\n",
-			 chan->notifier_bo->bo.mem.mem_type);
-		return -EINVAL;
-	}
+	if (chan->notifier_bo->bo.mem.mem_type == TTM_PL_VRAM)
+		target = NV_MEM_TARGET_VRAM;
+	else
+		target = NV_MEM_TARGET_GART;
+	offset  = chan->notifier_bo->bo.mem.start << PAGE_SHIFT;
 	offset += mem->start;
 
 	ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY, offset,
-				     mem->size, NV_DMA_ACCESS_RW, target,
+				     mem->size, NV_MEM_ACCESS_RW, target,
 				     &nobj);
 	if (ret) {
 		drm_mm_put_block(mem);
@@ -185,11 +168,11 @@ nouveau_ioctl_notifier_alloc(struct drm_device *dev, void *data,
 	struct nouveau_channel *chan;
 	int ret;
 
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(na->channel, file_priv, chan);
+	chan = nouveau_channel_get(dev, file_priv, na->channel);
+	if (IS_ERR(chan))
+		return PTR_ERR(chan);
 
 	ret = nouveau_notifier_alloc(chan, na->handle, na->size, &na->offset);
-	if (ret)
-		return ret;
-
-	return 0;
+	nouveau_channel_put(&chan);
+	return ret;
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_object.c b/drivers/gpu/drm/nouveau/nouveau_object.c
index dd572adca02a..55c9fdcfa67f 100644
--- a/drivers/gpu/drm/nouveau/nouveau_object.c
+++ b/drivers/gpu/drm/nouveau/nouveau_object.c
@@ -35,6 +35,102 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 #include "nouveau_ramht.h"
+#include "nouveau_vm.h"
+
+struct nouveau_gpuobj_method {
+	struct list_head head;
+	u32 mthd;
+	int (*exec)(struct nouveau_channel *, u32 class, u32 mthd, u32 data);
+};
+
+struct nouveau_gpuobj_class {
+	struct list_head head;
+	struct list_head methods;
+	u32 id;
+	u32 engine;
+};
+
+int
+nouveau_gpuobj_class_new(struct drm_device *dev, u32 class, u32 engine)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj_class *oc;
+
+	oc = kzalloc(sizeof(*oc), GFP_KERNEL);
+	if (!oc)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&oc->methods);
+	oc->id = class;
+	oc->engine = engine;
+	list_add(&oc->head, &dev_priv->classes);
+	return 0;
+}
+
+int
+nouveau_gpuobj_mthd_new(struct drm_device *dev, u32 class, u32 mthd,
+			int (*exec)(struct nouveau_channel *, u32, u32, u32))
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj_method *om;
+	struct nouveau_gpuobj_class *oc;
+
+	list_for_each_entry(oc, &dev_priv->classes, head) {
+		if (oc->id == class)
+			goto found;
+	}
+
+	return -EINVAL;
+
+found:
+	om = kzalloc(sizeof(*om), GFP_KERNEL);
+	if (!om)
+		return -ENOMEM;
+
+	om->mthd = mthd;
+	om->exec = exec;
+	list_add(&om->head, &oc->methods);
+	return 0;
+}
+
+int
+nouveau_gpuobj_mthd_call(struct nouveau_channel *chan,
+			 u32 class, u32 mthd, u32 data)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_gpuobj_method *om;
+	struct nouveau_gpuobj_class *oc;
+
+	list_for_each_entry(oc, &dev_priv->classes, head) {
+		if (oc->id != class)
+			continue;
+
+		list_for_each_entry(om, &oc->methods, head) {
+			if (om->mthd == mthd)
+				return om->exec(chan, class, mthd, data);
+		}
+	}
+
+	return -ENOENT;
+}
+
+int
+nouveau_gpuobj_mthd_call2(struct drm_device *dev, int chid,
+			  u32 class, u32 mthd, u32 data)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan = NULL;
+	unsigned long flags;
+	int ret = -EINVAL;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	if (chid > 0 && chid < dev_priv->engine.fifo.channels)
+		chan = dev_priv->channels.ptr[chid];
+	if (chan)
+		ret = nouveau_gpuobj_mthd_call(chan, class, mthd, data);
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+	return ret;
+}
 
 /* NVidia uses context objects to drive drawing operations.
 
@@ -73,17 +169,14 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 		   struct nouveau_gpuobj **gpuobj_ret)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
+	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
 	struct nouveau_gpuobj *gpuobj;
 	struct drm_mm_node *ramin = NULL;
-	int ret;
+	int ret, i;
 
 	NV_DEBUG(dev, "ch%d size=%u align=%d flags=0x%08x\n",
 		 chan ? chan->id : -1, size, align, flags);
 
-	if (!dev_priv || !gpuobj_ret || *gpuobj_ret != NULL)
-		return -EINVAL;
-
 	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
 	if (!gpuobj)
 		return -ENOMEM;
@@ -98,88 +191,41 @@ nouveau_gpuobj_new(struct drm_device *dev, struct nouveau_channel *chan,
 	spin_unlock(&dev_priv->ramin_lock);
 
 	if (chan) {
-		NV_DEBUG(dev, "channel heap\n");
-
 		ramin = drm_mm_search_free(&chan->ramin_heap, size, align, 0);
 		if (ramin)
 			ramin = drm_mm_get_block(ramin, size, align);
-
 		if (!ramin) {
 			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return -ENOMEM;
 		}
-	} else {
-		NV_DEBUG(dev, "global heap\n");
-
-		/* allocate backing pages, sets vinst */
-		ret = engine->instmem.populate(dev, gpuobj, &size);
-		if (ret) {
-			nouveau_gpuobj_ref(NULL, &gpuobj);
-			return ret;
-		}
-
-		/* try and get aperture space */
-		do {
-			if (drm_mm_pre_get(&dev_priv->ramin_heap))
-				return -ENOMEM;
 
-			spin_lock(&dev_priv->ramin_lock);
-			ramin = drm_mm_search_free(&dev_priv->ramin_heap, size,
-						   align, 0);
-			if (ramin == NULL) {
-				spin_unlock(&dev_priv->ramin_lock);
-				nouveau_gpuobj_ref(NULL, &gpuobj);
-				return -ENOMEM;
-			}
-
-			ramin = drm_mm_get_block_atomic(ramin, size, align);
-			spin_unlock(&dev_priv->ramin_lock);
-		} while (ramin == NULL);
-
-		/* on nv50 it's ok to fail, we have a fallback path */
-		if (!ramin && dev_priv->card_type < NV_50) {
-			nouveau_gpuobj_ref(NULL, &gpuobj);
-			return -ENOMEM;
-		}
-	}
+		gpuobj->pinst = chan->ramin->pinst;
+		if (gpuobj->pinst != ~0)
+			gpuobj->pinst += ramin->start;
 
-	/* if we got a chunk of the aperture, map pages into it */
-	gpuobj->im_pramin = ramin;
-	if (!chan && gpuobj->im_pramin && dev_priv->ramin_available) {
-		ret = engine->instmem.bind(dev, gpuobj);
+		gpuobj->cinst = ramin->start;
+		gpuobj->vinst = ramin->start + chan->ramin->vinst;
+		gpuobj->node  = ramin;
+	} else {
+		ret = instmem->get(gpuobj, size, align);
 		if (ret) {
 			nouveau_gpuobj_ref(NULL, &gpuobj);
 			return ret;
 		}
-	}
 
-	/* calculate the various different addresses for the object */
-	if (chan) {
-		gpuobj->pinst = chan->ramin->pinst;
-		if (gpuobj->pinst != ~0)
-			gpuobj->pinst += gpuobj->im_pramin->start;
-
-		if (dev_priv->card_type < NV_50) {
-			gpuobj->cinst = gpuobj->pinst;
-		} else {
-			gpuobj->cinst = gpuobj->im_pramin->start;
-			gpuobj->vinst = gpuobj->im_pramin->start +
-					chan->ramin->vinst;
-		}
-	} else {
-		if (gpuobj->im_pramin)
-			gpuobj->pinst = gpuobj->im_pramin->start;
-		else
+		ret = -ENOSYS;
+		if (!(flags & NVOBJ_FLAG_DONT_MAP))
+			ret = instmem->map(gpuobj);
+		if (ret)
 			gpuobj->pinst = ~0;
-		gpuobj->cinst = 0xdeadbeef;
+
+		gpuobj->cinst = NVOBJ_CINST_GLOBAL;
 	}
 
 	if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) {
-		int i;
-
 		for (i = 0; i < gpuobj->size; i += 4)
 			nv_wo32(gpuobj, i, 0);
-		engine->instmem.flush(dev);
+		instmem->flush(dev);
 	}
 
 
@@ -195,6 +241,7 @@ nouveau_gpuobj_init(struct drm_device *dev)
 	NV_DEBUG(dev, "\n");
 
 	INIT_LIST_HEAD(&dev_priv->gpuobj_list);
+	INIT_LIST_HEAD(&dev_priv->classes);
 	spin_lock_init(&dev_priv->ramin_lock);
 	dev_priv->ramin_base = ~0;
 
@@ -205,9 +252,20 @@ void
 nouveau_gpuobj_takedown(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj_method *om, *tm;
+	struct nouveau_gpuobj_class *oc, *tc;
 
 	NV_DEBUG(dev, "\n");
 
+	list_for_each_entry_safe(oc, tc, &dev_priv->classes, head) {
+		list_for_each_entry_safe(om, tm, &oc->methods, head) {
+			list_del(&om->head);
+			kfree(om);
+		}
+		list_del(&oc->head);
+		kfree(oc);
+	}
+
 	BUG_ON(!list_empty(&dev_priv->gpuobj_list));
 }
 
@@ -219,26 +277,34 @@ nouveau_gpuobj_del(struct kref *ref)
 		container_of(ref, struct nouveau_gpuobj, refcount);
 	struct drm_device *dev = gpuobj->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_engine *engine = &dev_priv->engine;
+	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
 	int i;
 
 	NV_DEBUG(dev, "gpuobj %p\n", gpuobj);
 
-	if (gpuobj->im_pramin && (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE)) {
+	if (gpuobj->node && (gpuobj->flags & NVOBJ_FLAG_ZERO_FREE)) {
 		for (i = 0; i < gpuobj->size; i += 4)
 			nv_wo32(gpuobj, i, 0);
-		engine->instmem.flush(dev);
+		instmem->flush(dev);
 	}
 
 	if (gpuobj->dtor)
 		gpuobj->dtor(dev, gpuobj);
 
-	if (gpuobj->im_backing)
-		engine->instmem.clear(dev, gpuobj);
+	if (gpuobj->cinst == NVOBJ_CINST_GLOBAL) {
+		if (gpuobj->node) {
+			instmem->unmap(gpuobj);
+			instmem->put(gpuobj);
+		}
+	} else {
+		if (gpuobj->node) {
+			spin_lock(&dev_priv->ramin_lock);
+			drm_mm_put_block(gpuobj->node);
+			spin_unlock(&dev_priv->ramin_lock);
+		}
+	}
 
 	spin_lock(&dev_priv->ramin_lock);
-	if (gpuobj->im_pramin)
-		drm_mm_put_block(gpuobj->im_pramin);
 	list_del(&gpuobj->list);
 	spin_unlock(&dev_priv->ramin_lock);
 
@@ -278,7 +344,7 @@ nouveau_gpuobj_new_fake(struct drm_device *dev, u32 pinst, u64 vinst,
 	kref_init(&gpuobj->refcount);
 	gpuobj->size  = size;
 	gpuobj->pinst = pinst;
-	gpuobj->cinst = 0xdeadbeef;
+	gpuobj->cinst = NVOBJ_CINST_GLOBAL;
 	gpuobj->vinst = vinst;
 
 	if (gpuobj->flags & NVOBJ_FLAG_ZERO_ALLOC) {
@@ -335,113 +401,150 @@ nouveau_gpuobj_class_instmem_size(struct drm_device *dev, int class)
    The method below creates a DMA object in instance RAM and returns a handle
    to it that can be used to set up context objects.
 */
-int
-nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class,
-		       uint64_t offset, uint64_t size, int access,
-		       int target, struct nouveau_gpuobj **gpuobj)
+
+void
+nv50_gpuobj_dma_init(struct nouveau_gpuobj *obj, u32 offset, int class,
+		     u64 base, u64 size, int target, int access,
+		     u32 type, u32 comp)
 {
-	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
-	int ret;
+	struct drm_nouveau_private *dev_priv = obj->dev->dev_private;
+	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
+	u32 flags0;
 
-	NV_DEBUG(dev, "ch%d class=0x%04x offset=0x%llx size=0x%llx\n",
-		 chan->id, class, offset, size);
-	NV_DEBUG(dev, "access=%d target=%d\n", access, target);
+	flags0  = (comp << 29) | (type << 22) | class;
+	flags0 |= 0x00100000;
+
+	switch (access) {
+	case NV_MEM_ACCESS_RO: flags0 |= 0x00040000; break;
+	case NV_MEM_ACCESS_RW:
+	case NV_MEM_ACCESS_WO: flags0 |= 0x00080000; break;
+	default:
+		break;
+	}
 
 	switch (target) {
-	case NV_DMA_TARGET_AGP:
-		offset += dev_priv->gart_info.aper_base;
+	case NV_MEM_TARGET_VRAM:
+		flags0 |= 0x00010000;
+		break;
+	case NV_MEM_TARGET_PCI:
+		flags0 |= 0x00020000;
+		break;
+	case NV_MEM_TARGET_PCI_NOSNOOP:
+		flags0 |= 0x00030000;
 		break;
+	case NV_MEM_TARGET_GART:
+		base += dev_priv->gart_info.aper_base;
 	default:
+		flags0 &= ~0x00100000;
 		break;
 	}
 
-	ret = nouveau_gpuobj_new(dev, chan,
-				 nouveau_gpuobj_class_instmem_size(dev, class),
-				 16, NVOBJ_FLAG_ZERO_ALLOC |
-				 NVOBJ_FLAG_ZERO_FREE, gpuobj);
-	if (ret) {
-		NV_ERROR(dev, "Error creating gpuobj: %d\n", ret);
-		return ret;
-	}
+	/* convert to base + limit */
+	size = (base + size) - 1;
 
-	if (dev_priv->card_type < NV_50) {
-		uint32_t frame, adjust, pte_flags = 0;
-
-		if (access != NV_DMA_ACCESS_RO)
-			pte_flags |= (1<<1);
-		adjust = offset &  0x00000fff;
-		frame  = offset & ~0x00000fff;
-
-		nv_wo32(*gpuobj,  0, ((1<<12) | (1<<13) | (adjust << 20) |
-				      (access << 14) | (target << 16) |
-				      class));
-		nv_wo32(*gpuobj,  4, size - 1);
-		nv_wo32(*gpuobj,  8, frame | pte_flags);
-		nv_wo32(*gpuobj, 12, frame | pte_flags);
-	} else {
-		uint64_t limit = offset + size - 1;
-		uint32_t flags0, flags5;
+	nv_wo32(obj, offset + 0x00, flags0);
+	nv_wo32(obj, offset + 0x04, lower_32_bits(size));
+	nv_wo32(obj, offset + 0x08, lower_32_bits(base));
+	nv_wo32(obj, offset + 0x0c, upper_32_bits(size) << 24 |
+				    upper_32_bits(base));
+	nv_wo32(obj, offset + 0x10, 0x00000000);
+	nv_wo32(obj, offset + 0x14, 0x00000000);
 
-		if (target == NV_DMA_TARGET_VIDMEM) {
-			flags0 = 0x00190000;
-			flags5 = 0x00010000;
-		} else {
-			flags0 = 0x7fc00000;
-			flags5 = 0x00080000;
-		}
+	pinstmem->flush(obj->dev);
+}
 
-		nv_wo32(*gpuobj,  0, flags0 | class);
-		nv_wo32(*gpuobj,  4, lower_32_bits(limit));
-		nv_wo32(*gpuobj,  8, lower_32_bits(offset));
-		nv_wo32(*gpuobj, 12, ((upper_32_bits(limit) & 0xff) << 24) |
-				      (upper_32_bits(offset) & 0xff));
-		nv_wo32(*gpuobj, 20, flags5);
-	}
+int
+nv50_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base, u64 size,
+		    int target, int access, u32 type, u32 comp,
+		    struct nouveau_gpuobj **pobj)
+{
+	struct drm_device *dev = chan->dev;
+	int ret;
 
-	instmem->flush(dev);
+	ret = nouveau_gpuobj_new(dev, chan, 24, 16, NVOBJ_FLAG_ZERO_FREE, pobj);
+	if (ret)
+		return ret;
 
-	(*gpuobj)->engine = NVOBJ_ENGINE_SW;
-	(*gpuobj)->class  = class;
+	nv50_gpuobj_dma_init(*pobj, 0, class, base, size, target,
+			     access, type, comp);
 	return 0;
 }
 
 int
-nouveau_gpuobj_gart_dma_new(struct nouveau_channel *chan,
-			    uint64_t offset, uint64_t size, int access,
-			    struct nouveau_gpuobj **gpuobj,
-			    uint32_t *o_ret)
+nouveau_gpuobj_dma_new(struct nouveau_channel *chan, int class, u64 base,
+		       u64 size, int access, int target,
+		       struct nouveau_gpuobj **pobj)
 {
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *obj;
+	u32 flags0, flags2;
 	int ret;
 
-	if (dev_priv->gart_info.type == NOUVEAU_GART_AGP ||
-	    (dev_priv->card_type >= NV_50 &&
-	     dev_priv->gart_info.type == NOUVEAU_GART_SGDMA)) {
-		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
-					     offset + dev_priv->vm_gart_base,
-					     size, access, NV_DMA_TARGET_AGP,
-					     gpuobj);
-		if (o_ret)
-			*o_ret = 0;
-	} else
-	if (dev_priv->gart_info.type == NOUVEAU_GART_SGDMA) {
-		nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, gpuobj);
-		if (offset & ~0xffffffffULL) {
-			NV_ERROR(dev, "obj offset exceeds 32-bits\n");
-			return -EINVAL;
+	if (dev_priv->card_type >= NV_50) {
+		u32 comp = (target == NV_MEM_TARGET_VM) ? NV_MEM_COMP_VM : 0;
+		u32 type = (target == NV_MEM_TARGET_VM) ? NV_MEM_TYPE_VM : 0;
+
+		return nv50_gpuobj_dma_new(chan, class, base, size,
+					   target, access, type, comp, pobj);
+	}
+
+	if (target == NV_MEM_TARGET_GART) {
+		if (dev_priv->gart_info.type == NOUVEAU_GART_AGP) {
+			target = NV_MEM_TARGET_PCI_NOSNOOP;
+			base  += dev_priv->gart_info.aper_base;
+		} else
+		if (base != 0) {
+			base = nouveau_sgdma_get_physical(dev, base);
+			target = NV_MEM_TARGET_PCI;
+		} else {
+			nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma, pobj);
+			return 0;
 		}
-		if (o_ret)
-			*o_ret = (uint32_t)offset;
-		ret = (*gpuobj != NULL) ? 0 : -EINVAL;
-	} else {
-		NV_ERROR(dev, "Invalid GART type %d\n", dev_priv->gart_info.type);
-		return -EINVAL;
 	}
 
-	return ret;
+	flags0  = class;
+	flags0 |= 0x00003000; /* PT present, PT linear */
+	flags2  = 0;
+
+	switch (target) {
+	case NV_MEM_TARGET_PCI:
+		flags0 |= 0x00020000;
+		break;
+	case NV_MEM_TARGET_PCI_NOSNOOP:
+		flags0 |= 0x00030000;
+		break;
+	default:
+		break;
+	}
+
+	switch (access) {
+	case NV_MEM_ACCESS_RO:
+		flags0 |= 0x00004000;
+		break;
+	case NV_MEM_ACCESS_WO:
+		flags0 |= 0x00008000;
+	default:
+		flags2 |= 0x00000002;
+		break;
+	}
+
+	flags0 |= (base & 0x00000fff) << 20;
+	flags2 |= (base & 0xfffff000);
+
+	ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj);
+	if (ret)
+		return ret;
+
+	nv_wo32(obj, 0x00, flags0);
+	nv_wo32(obj, 0x04, size - 1);
+	nv_wo32(obj, 0x08, flags2);
+	nv_wo32(obj, 0x0c, flags2);
+
+	obj->engine = NVOBJ_ENGINE_SW;
+	obj->class  = class;
+	*pobj = obj;
+	return 0;
 }
 
 /* Context objects in the instance RAM have the following structure.
@@ -495,82 +598,122 @@ nouveau_gpuobj_gart_dma_new(struct nouveau_channel *chan,
    entry[5]:
    set to 0?
 */
+static int
+nouveau_gpuobj_sw_new(struct nouveau_channel *chan, int class,
+		      struct nouveau_gpuobj **gpuobj_ret)
+{
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
+	struct nouveau_gpuobj *gpuobj;
+
+	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
+	if (!gpuobj)
+		return -ENOMEM;
+	gpuobj->dev = chan->dev;
+	gpuobj->engine = NVOBJ_ENGINE_SW;
+	gpuobj->class = class;
+	kref_init(&gpuobj->refcount);
+	gpuobj->cinst = 0x40;
+
+	spin_lock(&dev_priv->ramin_lock);
+	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
+	spin_unlock(&dev_priv->ramin_lock);
+	*gpuobj_ret = gpuobj;
+	return 0;
+}
+
 int
-nouveau_gpuobj_gr_new(struct nouveau_channel *chan, int class,
-		      struct nouveau_gpuobj **gpuobj)
+nouveau_gpuobj_gr_new(struct nouveau_channel *chan, u32 handle, int class)
 {
+	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj_class *oc;
+	struct nouveau_gpuobj *gpuobj;
 	int ret;
 
 	NV_DEBUG(dev, "ch%d class=0x%04x\n", chan->id, class);
 
+	list_for_each_entry(oc, &dev_priv->classes, head) {
+		if (oc->id == class)
+			goto found;
+	}
+
+	NV_ERROR(dev, "illegal object class: 0x%x\n", class);
+	return -EINVAL;
+
+found:
+	switch (oc->engine) {
+	case NVOBJ_ENGINE_SW:
+		ret = nouveau_gpuobj_sw_new(chan, class, &gpuobj);
+		if (ret)
+			return ret;
+		goto insert;
+	case NVOBJ_ENGINE_GR:
+		if (dev_priv->card_type >= NV_50 && !chan->ramin_grctx) {
+			struct nouveau_pgraph_engine *pgraph =
+				&dev_priv->engine.graph;
+
+			ret = pgraph->create_context(chan);
+			if (ret)
+				return ret;
+		}
+		break;
+	case NVOBJ_ENGINE_CRYPT:
+		if (!chan->crypt_ctx) {
+			struct nouveau_crypt_engine *pcrypt =
+				&dev_priv->engine.crypt;
+
+			ret = pcrypt->create_context(chan);
+			if (ret)
+				return ret;
+		}
+		break;
+	}
+
 	ret = nouveau_gpuobj_new(dev, chan,
 				 nouveau_gpuobj_class_instmem_size(dev, class),
 				 16,
 				 NVOBJ_FLAG_ZERO_ALLOC | NVOBJ_FLAG_ZERO_FREE,
-				 gpuobj);
+				 &gpuobj);
 	if (ret) {
-		NV_ERROR(dev, "Error creating gpuobj: %d\n", ret);
+		NV_ERROR(dev, "error creating gpuobj: %d\n", ret);
 		return ret;
 	}
 
 	if (dev_priv->card_type >= NV_50) {
-		nv_wo32(*gpuobj,  0, class);
-		nv_wo32(*gpuobj, 20, 0x00010000);
+		nv_wo32(gpuobj,  0, class);
+		nv_wo32(gpuobj, 20, 0x00010000);
 	} else {
 		switch (class) {
 		case NV_CLASS_NULL:
-			nv_wo32(*gpuobj, 0, 0x00001030);
-			nv_wo32(*gpuobj, 4, 0xFFFFFFFF);
+			nv_wo32(gpuobj, 0, 0x00001030);
+			nv_wo32(gpuobj, 4, 0xFFFFFFFF);
 			break;
 		default:
 			if (dev_priv->card_type >= NV_40) {
-				nv_wo32(*gpuobj, 0, class);
+				nv_wo32(gpuobj, 0, class);
 #ifdef __BIG_ENDIAN
-				nv_wo32(*gpuobj, 8, 0x01000000);
+				nv_wo32(gpuobj, 8, 0x01000000);
 #endif
 			} else {
 #ifdef __BIG_ENDIAN
-				nv_wo32(*gpuobj, 0, class | 0x00080000);
+				nv_wo32(gpuobj, 0, class | 0x00080000);
 #else
-				nv_wo32(*gpuobj, 0, class);
+				nv_wo32(gpuobj, 0, class);
 #endif
 			}
 		}
 	}
 	dev_priv->engine.instmem.flush(dev);
 
-	(*gpuobj)->engine = NVOBJ_ENGINE_GR;
-	(*gpuobj)->class  = class;
-	return 0;
-}
-
-int
-nouveau_gpuobj_sw_new(struct nouveau_channel *chan, int class,
-		      struct nouveau_gpuobj **gpuobj_ret)
-{
-	struct drm_nouveau_private *dev_priv;
-	struct nouveau_gpuobj *gpuobj;
-
-	if (!chan || !gpuobj_ret || *gpuobj_ret != NULL)
-		return -EINVAL;
-	dev_priv = chan->dev->dev_private;
-
-	gpuobj = kzalloc(sizeof(*gpuobj), GFP_KERNEL);
-	if (!gpuobj)
-		return -ENOMEM;
-	gpuobj->dev = chan->dev;
-	gpuobj->engine = NVOBJ_ENGINE_SW;
-	gpuobj->class = class;
-	kref_init(&gpuobj->refcount);
-	gpuobj->cinst = 0x40;
+	gpuobj->engine = oc->engine;
+	gpuobj->class  = oc->id;
 
-	spin_lock(&dev_priv->ramin_lock);
-	list_add_tail(&gpuobj->list, &dev_priv->gpuobj_list);
-	spin_unlock(&dev_priv->ramin_lock);
-	*gpuobj_ret = gpuobj;
-	return 0;
+insert:
+	ret = nouveau_ramht_insert(chan, handle, gpuobj);
+	if (ret)
+		NV_ERROR(dev, "error adding gpuobj to RAMHT: %d\n", ret);
+	nouveau_gpuobj_ref(NULL, &gpuobj);
+	return ret;
 }
 
 static int
@@ -585,7 +728,7 @@ nouveau_gpuobj_channel_init_pramin(struct nouveau_channel *chan)
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
 	/* Base amount for object storage (4KiB enough?) */
-	size = 0x1000;
+	size = 0x2000;
 	base = 0;
 
 	/* PGRAPH context */
@@ -624,9 +767,8 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_instmem_engine *instmem = &dev_priv->engine.instmem;
 	struct nouveau_gpuobj *vram = NULL, *tt = NULL;
-	int ret, i;
+	int ret;
 
 	NV_DEBUG(dev, "ch%d vram=0x%08x tt=0x%08x\n", chan->id, vram_h, tt_h);
 
@@ -637,16 +779,14 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 		return ret;
 	}
 
-	/* NV50 VM
+	/* NV50/NVC0 VM
 	 *  - Allocate per-channel page-directory
-	 *  - Map GART and VRAM into the channel's address space at the
-	 *    locations determined during init.
+	 *  - Link with shared channel VM
 	 */
-	if (dev_priv->card_type >= NV_50) {
+	if (dev_priv->chan_vm) {
 		u32 pgd_offs = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200;
 		u64 vm_vinst = chan->ramin->vinst + pgd_offs;
 		u32 vm_pinst = chan->ramin->pinst;
-		u32 pde;
 
 		if (vm_pinst != ~0)
 			vm_pinst += pgd_offs;
@@ -655,29 +795,8 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 					      0, &chan->vm_pd);
 		if (ret)
 			return ret;
-		for (i = 0; i < 0x4000; i += 8) {
-			nv_wo32(chan->vm_pd, i + 0, 0x00000000);
-			nv_wo32(chan->vm_pd, i + 4, 0xdeadcafe);
-		}
-
-		nouveau_gpuobj_ref(dev_priv->gart_info.sg_ctxdma,
-				   &chan->vm_gart_pt);
-		pde = (dev_priv->vm_gart_base / (512*1024*1024)) * 8;
-		nv_wo32(chan->vm_pd, pde + 0, chan->vm_gart_pt->vinst | 3);
-		nv_wo32(chan->vm_pd, pde + 4, 0x00000000);
-
-		pde = (dev_priv->vm_vram_base / (512*1024*1024)) * 8;
-		for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
-			nouveau_gpuobj_ref(dev_priv->vm_vram_pt[i],
-					   &chan->vm_vram_pt[i]);
-
-			nv_wo32(chan->vm_pd, pde + 0,
-				chan->vm_vram_pt[i]->vinst | 0x61);
-			nv_wo32(chan->vm_pd, pde + 4, 0x00000000);
-			pde += 8;
-		}
 
-		instmem->flush(dev);
+		nouveau_vm_ref(dev_priv->chan_vm, &chan->vm, chan->vm_pd);
 	}
 
 	/* RAMHT */
@@ -700,9 +819,8 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 	/* VRAM ctxdma */
 	if (dev_priv->card_type >= NV_50) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
-					     0, dev_priv->vm_end,
-					     NV_DMA_ACCESS_RW,
-					     NV_DMA_TARGET_AGP, &vram);
+					     0, (1ULL << 40), NV_MEM_ACCESS_RW,
+					     NV_MEM_TARGET_VM, &vram);
 		if (ret) {
 			NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
 			return ret;
@@ -710,8 +828,8 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 	} else {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
 					     0, dev_priv->fb_available_size,
-					     NV_DMA_ACCESS_RW,
-					     NV_DMA_TARGET_VIDMEM, &vram);
+					     NV_MEM_ACCESS_RW,
+					     NV_MEM_TARGET_VRAM, &vram);
 		if (ret) {
 			NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
 			return ret;
@@ -728,21 +846,13 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 	/* TT memory ctxdma */
 	if (dev_priv->card_type >= NV_50) {
 		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
-					     0, dev_priv->vm_end,
-					     NV_DMA_ACCESS_RW,
-					     NV_DMA_TARGET_AGP, &tt);
-		if (ret) {
-			NV_ERROR(dev, "Error creating VRAM ctxdma: %d\n", ret);
-			return ret;
-		}
-	} else
-	if (dev_priv->gart_info.type != NOUVEAU_GART_NONE) {
-		ret = nouveau_gpuobj_gart_dma_new(chan, 0,
-						  dev_priv->gart_info.aper_size,
-						  NV_DMA_ACCESS_RW, &tt, NULL);
+					     0, (1ULL << 40), NV_MEM_ACCESS_RW,
+					     NV_MEM_TARGET_VM, &tt);
 	} else {
-		NV_ERROR(dev, "Invalid GART type %d\n", dev_priv->gart_info.type);
-		ret = -EINVAL;
+		ret = nouveau_gpuobj_dma_new(chan, NV_CLASS_DMA_IN_MEMORY,
+					     0, dev_priv->gart_info.aper_size,
+					     NV_MEM_ACCESS_RW,
+					     NV_MEM_TARGET_GART, &tt);
 	}
 
 	if (ret) {
@@ -763,9 +873,7 @@ nouveau_gpuobj_channel_init(struct nouveau_channel *chan,
 void
 nouveau_gpuobj_channel_takedown(struct nouveau_channel *chan)
 {
-	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
 	struct drm_device *dev = chan->dev;
-	int i;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
@@ -774,10 +882,8 @@ nouveau_gpuobj_channel_takedown(struct nouveau_channel *chan)
 
 	nouveau_ramht_ref(NULL, &chan->ramht, chan);
 
+	nouveau_vm_ref(NULL, &chan->vm, chan->vm_pd);
 	nouveau_gpuobj_ref(NULL, &chan->vm_pd);
-	nouveau_gpuobj_ref(NULL, &chan->vm_gart_pt);
-	for (i = 0; i < dev_priv->vm_vram_pt_nr; i++)
-		nouveau_gpuobj_ref(NULL, &chan->vm_vram_pt[i]);
 
 	if (chan->ramin_heap.free_stack.next)
 		drm_mm_takedown(&chan->ramin_heap);
@@ -791,147 +897,91 @@ nouveau_gpuobj_suspend(struct drm_device *dev)
 	struct nouveau_gpuobj *gpuobj;
 	int i;
 
-	if (dev_priv->card_type < NV_50) {
-		dev_priv->susres.ramin_copy = vmalloc(dev_priv->ramin_rsvd_vram);
-		if (!dev_priv->susres.ramin_copy)
-			return -ENOMEM;
-
-		for (i = 0; i < dev_priv->ramin_rsvd_vram; i += 4)
-			dev_priv->susres.ramin_copy[i/4] = nv_ri32(dev, i);
-		return 0;
-	}
-
 	list_for_each_entry(gpuobj, &dev_priv->gpuobj_list, list) {
-		if (!gpuobj->im_backing)
+		if (gpuobj->cinst != NVOBJ_CINST_GLOBAL)
 			continue;
 
-		gpuobj->im_backing_suspend = vmalloc(gpuobj->size);
-		if (!gpuobj->im_backing_suspend) {
+		gpuobj->suspend = vmalloc(gpuobj->size);
+		if (!gpuobj->suspend) {
 			nouveau_gpuobj_resume(dev);
 			return -ENOMEM;
 		}
 
 		for (i = 0; i < gpuobj->size; i += 4)
-			gpuobj->im_backing_suspend[i/4] = nv_ro32(gpuobj, i);
+			gpuobj->suspend[i/4] = nv_ro32(gpuobj, i);
 	}
 
 	return 0;
 }
 
 void
-nouveau_gpuobj_suspend_cleanup(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *gpuobj;
-
-	if (dev_priv->card_type < NV_50) {
-		vfree(dev_priv->susres.ramin_copy);
-		dev_priv->susres.ramin_copy = NULL;
-		return;
-	}
-
-	list_for_each_entry(gpuobj, &dev_priv->gpuobj_list, list) {
-		if (!gpuobj->im_backing_suspend)
-			continue;
-
-		vfree(gpuobj->im_backing_suspend);
-		gpuobj->im_backing_suspend = NULL;
-	}
-}
-
-void
 nouveau_gpuobj_resume(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *gpuobj;
 	int i;
 
-	if (dev_priv->card_type < NV_50) {
-		for (i = 0; i < dev_priv->ramin_rsvd_vram; i += 4)
-			nv_wi32(dev, i, dev_priv->susres.ramin_copy[i/4]);
-		nouveau_gpuobj_suspend_cleanup(dev);
-		return;
-	}
-
 	list_for_each_entry(gpuobj, &dev_priv->gpuobj_list, list) {
-		if (!gpuobj->im_backing_suspend)
+		if (!gpuobj->suspend)
 			continue;
 
 		for (i = 0; i < gpuobj->size; i += 4)
-			nv_wo32(gpuobj, i, gpuobj->im_backing_suspend[i/4]);
-		dev_priv->engine.instmem.flush(dev);
+			nv_wo32(gpuobj, i, gpuobj->suspend[i/4]);
+
+		vfree(gpuobj->suspend);
+		gpuobj->suspend = NULL;
 	}
 
-	nouveau_gpuobj_suspend_cleanup(dev);
+	dev_priv->engine.instmem.flush(dev);
 }
 
 int nouveau_ioctl_grobj_alloc(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct drm_nouveau_grobj_alloc *init = data;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
-	struct nouveau_pgraph_object_class *grc;
-	struct nouveau_gpuobj *gr = NULL;
 	struct nouveau_channel *chan;
 	int ret;
 
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(init->channel, file_priv, chan);
-
 	if (init->handle == ~0)
 		return -EINVAL;
 
-	grc = pgraph->grclass;
-	while (grc->id) {
-		if (grc->id == init->class)
-			break;
-		grc++;
-	}
+	chan = nouveau_channel_get(dev, file_priv, init->channel);
+	if (IS_ERR(chan))
+		return PTR_ERR(chan);
 
-	if (!grc->id) {
-		NV_ERROR(dev, "Illegal object class: 0x%x\n", init->class);
-		return -EPERM;
+	if (nouveau_ramht_find(chan, init->handle)) {
+		ret = -EEXIST;
+		goto out;
 	}
 
-	if (nouveau_ramht_find(chan, init->handle))
-		return -EEXIST;
-
-	if (!grc->software)
-		ret = nouveau_gpuobj_gr_new(chan, grc->id, &gr);
-	else
-		ret = nouveau_gpuobj_sw_new(chan, grc->id, &gr);
+	ret = nouveau_gpuobj_gr_new(chan, init->handle, init->class);
 	if (ret) {
 		NV_ERROR(dev, "Error creating object: %d (%d/0x%08x)\n",
 			 ret, init->channel, init->handle);
-		return ret;
-	}
-
-	ret = nouveau_ramht_insert(chan, init->handle, gr);
-	nouveau_gpuobj_ref(NULL, &gr);
-	if (ret) {
-		NV_ERROR(dev, "Error referencing object: %d (%d/0x%08x)\n",
-			 ret, init->channel, init->handle);
-		return ret;
 	}
 
-	return 0;
+out:
+	nouveau_channel_put(&chan);
+	return ret;
 }
 
 int nouveau_ioctl_gpuobj_free(struct drm_device *dev, void *data,
 			      struct drm_file *file_priv)
 {
 	struct drm_nouveau_gpuobj_free *objfree = data;
-	struct nouveau_gpuobj *gpuobj;
 	struct nouveau_channel *chan;
+	int ret;
 
-	NOUVEAU_GET_USER_CHANNEL_WITH_RETURN(objfree->channel, file_priv, chan);
+	chan = nouveau_channel_get(dev, file_priv, objfree->channel);
+	if (IS_ERR(chan))
+		return PTR_ERR(chan);
 
-	gpuobj = nouveau_ramht_find(chan, objfree->handle);
-	if (!gpuobj)
-		return -ENOENT;
+	/* Synchronize with the user channel */
+	nouveau_channel_idle(chan);
 
-	nouveau_ramht_remove(chan, objfree->handle);
-	return 0;
+	ret = nouveau_ramht_remove(chan, objfree->handle);
+	nouveau_channel_put(&chan);
+	return ret;
 }
 
 u32
diff --git a/drivers/gpu/drm/nouveau/nouveau_pm.c b/drivers/gpu/drm/nouveau/nouveau_pm.c
index 9f7b158f5825..d93814160bcf 100644
--- a/drivers/gpu/drm/nouveau/nouveau_pm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_pm.c
@@ -27,6 +27,10 @@
 #include "nouveau_drv.h"
 #include "nouveau_pm.h"
 
+#ifdef CONFIG_ACPI
+#include <linux/acpi.h>
+#endif
+#include <linux/power_supply.h>
 #include <linux/hwmon.h>
 #include <linux/hwmon-sysfs.h>
 
@@ -446,6 +450,25 @@ nouveau_hwmon_fini(struct drm_device *dev)
 #endif
 }
 
+#ifdef CONFIG_ACPI
+static int
+nouveau_pm_acpi_event(struct notifier_block *nb, unsigned long val, void *data)
+{
+	struct drm_nouveau_private *dev_priv =
+		container_of(nb, struct drm_nouveau_private, engine.pm.acpi_nb);
+	struct drm_device *dev = dev_priv->dev;
+	struct acpi_bus_event *entry = (struct acpi_bus_event *)data;
+
+	if (strcmp(entry->device_class, "ac_adapter") == 0) {
+		bool ac = power_supply_is_system_supplied();
+
+		NV_DEBUG(dev, "power supply changed: %s\n", ac ? "AC" : "DC");
+	}
+
+	return NOTIFY_OK;
+}
+#endif
+
 int
 nouveau_pm_init(struct drm_device *dev)
 {
@@ -485,6 +508,10 @@ nouveau_pm_init(struct drm_device *dev)
 
 	nouveau_sysfs_init(dev);
 	nouveau_hwmon_init(dev);
+#ifdef CONFIG_ACPI
+	pm->acpi_nb.notifier_call = nouveau_pm_acpi_event;
+	register_acpi_notifier(&pm->acpi_nb);
+#endif
 
 	return 0;
 }
@@ -503,6 +530,9 @@ nouveau_pm_fini(struct drm_device *dev)
 	nouveau_perf_fini(dev);
 	nouveau_volt_fini(dev);
 
+#ifdef CONFIG_ACPI
+	unregister_acpi_notifier(&pm->acpi_nb);
+#endif
 	nouveau_hwmon_fini(dev);
 	nouveau_sysfs_fini(dev);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_ramht.c b/drivers/gpu/drm/nouveau/nouveau_ramht.c
index 2d8580927ca4..bef3e6910418 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ramht.c
+++ b/drivers/gpu/drm/nouveau/nouveau_ramht.c
@@ -104,17 +104,17 @@ nouveau_ramht_insert(struct nouveau_channel *chan, u32 handle,
 	nouveau_gpuobj_ref(gpuobj, &entry->gpuobj);
 
 	if (dev_priv->card_type < NV_40) {
-		ctx = NV_RAMHT_CONTEXT_VALID | (gpuobj->cinst >> 4) |
+		ctx = NV_RAMHT_CONTEXT_VALID | (gpuobj->pinst >> 4) |
 		      (chan->id << NV_RAMHT_CONTEXT_CHANNEL_SHIFT) |
 		      (gpuobj->engine << NV_RAMHT_CONTEXT_ENGINE_SHIFT);
 	} else
 	if (dev_priv->card_type < NV_50) {
-		ctx = (gpuobj->cinst >> 4) |
+		ctx = (gpuobj->pinst >> 4) |
 		      (chan->id << NV40_RAMHT_CONTEXT_CHANNEL_SHIFT) |
 		      (gpuobj->engine << NV40_RAMHT_CONTEXT_ENGINE_SHIFT);
 	} else {
 		if (gpuobj->engine == NVOBJ_ENGINE_DISPLAY) {
-			ctx = (gpuobj->cinst << 10) | 2;
+			ctx = (gpuobj->cinst << 10) | chan->id;
 		} else {
 			ctx = (gpuobj->cinst >> 4) |
 			      ((gpuobj->engine <<
@@ -214,18 +214,19 @@ out:
 	spin_unlock_irqrestore(&chan->ramht->lock, flags);
 }
 
-void
+int
 nouveau_ramht_remove(struct nouveau_channel *chan, u32 handle)
 {
 	struct nouveau_ramht_entry *entry;
 
 	entry = nouveau_ramht_remove_entry(chan, handle);
 	if (!entry)
-		return;
+		return -ENOENT;
 
 	nouveau_ramht_remove_hash(chan, entry->handle);
 	nouveau_gpuobj_ref(NULL, &entry->gpuobj);
 	kfree(entry);
+	return 0;
 }
 
 struct nouveau_gpuobj *
diff --git a/drivers/gpu/drm/nouveau/nouveau_ramht.h b/drivers/gpu/drm/nouveau/nouveau_ramht.h
index b79cb5e1a8f1..c82de98fee0e 100644
--- a/drivers/gpu/drm/nouveau/nouveau_ramht.h
+++ b/drivers/gpu/drm/nouveau/nouveau_ramht.h
@@ -48,7 +48,7 @@ extern void nouveau_ramht_ref(struct nouveau_ramht *, struct nouveau_ramht **,
 
 extern int  nouveau_ramht_insert(struct nouveau_channel *, u32 handle,
 				 struct nouveau_gpuobj *);
-extern void nouveau_ramht_remove(struct nouveau_channel *, u32 handle);
+extern int  nouveau_ramht_remove(struct nouveau_channel *, u32 handle);
 extern struct nouveau_gpuobj *
 nouveau_ramht_find(struct nouveau_channel *chan, u32 handle);
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_reg.h b/drivers/gpu/drm/nouveau/nouveau_reg.h
index 1b42541ca9e5..04e8fb795269 100644
--- a/drivers/gpu/drm/nouveau/nouveau_reg.h
+++ b/drivers/gpu/drm/nouveau/nouveau_reg.h
@@ -45,6 +45,11 @@
 #	define NV04_PFB_REF_CMD_REFRESH				(1 << 0)
 #define NV04_PFB_PRE						0x001002d4
 #	define NV04_PFB_PRE_CMD_PRECHARGE			(1 << 0)
+#define NV20_PFB_ZCOMP(i)                              (0x00100300 + 4*(i))
+#	define NV20_PFB_ZCOMP_MODE_32				(4 << 24)
+#	define NV20_PFB_ZCOMP_EN				(1 << 31)
+#	define NV25_PFB_ZCOMP_MODE_16				(1 << 20)
+#	define NV25_PFB_ZCOMP_MODE_32				(2 << 20)
 #define NV10_PFB_CLOSE_PAGE2					0x0010033c
 #define NV04_PFB_SCRAMBLE(i)                         (0x00100400 + 4 * (i))
 #define NV40_PFB_TILE(i)                              (0x00100600 + (i*16))
@@ -74,17 +79,6 @@
 #    define NV40_RAMHT_CONTEXT_ENGINE_SHIFT                20
 #    define NV40_RAMHT_CONTEXT_INSTANCE_SHIFT              0
 
-/* DMA object defines */
-#define NV_DMA_ACCESS_RW 0
-#define NV_DMA_ACCESS_RO 1
-#define NV_DMA_ACCESS_WO 2
-#define NV_DMA_TARGET_VIDMEM 0
-#define NV_DMA_TARGET_PCI    2
-#define NV_DMA_TARGET_AGP    3
-/* The following is not a real value used by the card, it's changed by
- * nouveau_object_dma_create */
-#define NV_DMA_TARGET_PCI_NONLINEAR 8
-
 /* Some object classes we care about in the drm */
 #define NV_CLASS_DMA_FROM_MEMORY                           0x00000002
 #define NV_CLASS_DMA_TO_MEMORY                             0x00000003
@@ -332,6 +326,7 @@
 #define NV04_PGRAPH_BSWIZZLE5                              0x004006A0
 #define NV03_PGRAPH_STATUS                                 0x004006B0
 #define NV04_PGRAPH_STATUS                                 0x00400700
+#    define NV40_PGRAPH_STATUS_SYNC_STALL                  0x00004000
 #define NV04_PGRAPH_TRAPPED_ADDR                           0x00400704
 #define NV04_PGRAPH_TRAPPED_DATA                           0x00400708
 #define NV04_PGRAPH_SURFACE                                0x0040070C
@@ -378,6 +373,7 @@
 #define NV20_PGRAPH_TLIMIT(i)                              (0x00400904 + (i*16))
 #define NV20_PGRAPH_TSIZE(i)                               (0x00400908 + (i*16))
 #define NV20_PGRAPH_TSTATUS(i)                             (0x0040090C + (i*16))
+#define NV20_PGRAPH_ZCOMP(i)                               (0x00400980 + 4*(i))
 #define NV10_PGRAPH_TILE(i)                                (0x00400B00 + (i*16))
 #define NV10_PGRAPH_TLIMIT(i)                              (0x00400B04 + (i*16))
 #define NV10_PGRAPH_TSIZE(i)                               (0x00400B08 + (i*16))
@@ -714,31 +710,32 @@
 #define NV50_PDISPLAY_INTR_1_CLK_UNK10                               0x00000010
 #define NV50_PDISPLAY_INTR_1_CLK_UNK20                               0x00000020
 #define NV50_PDISPLAY_INTR_1_CLK_UNK40                               0x00000040
-#define NV50_PDISPLAY_INTR_EN                                        0x0061002c
-#define NV50_PDISPLAY_INTR_EN_VBLANK_CRTC                            0x0000000c
-#define NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_(n)                   (1 << ((n) + 2))
-#define NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_0                          0x00000004
-#define NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_1                          0x00000008
-#define NV50_PDISPLAY_INTR_EN_CLK_UNK10                              0x00000010
-#define NV50_PDISPLAY_INTR_EN_CLK_UNK20                              0x00000020
-#define NV50_PDISPLAY_INTR_EN_CLK_UNK40                              0x00000040
+#define NV50_PDISPLAY_INTR_EN_0                                      0x00610028
+#define NV50_PDISPLAY_INTR_EN_1                                      0x0061002c
+#define NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC                          0x0000000c
+#define NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_(n)                 (1 << ((n) + 2))
+#define NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_0                        0x00000004
+#define NV50_PDISPLAY_INTR_EN_1_VBLANK_CRTC_1                        0x00000008
+#define NV50_PDISPLAY_INTR_EN_1_CLK_UNK10                            0x00000010
+#define NV50_PDISPLAY_INTR_EN_1_CLK_UNK20                            0x00000020
+#define NV50_PDISPLAY_INTR_EN_1_CLK_UNK40                            0x00000040
 #define NV50_PDISPLAY_UNK30_CTRL                                     0x00610030
 #define NV50_PDISPLAY_UNK30_CTRL_UPDATE_VCLK0                        0x00000200
 #define NV50_PDISPLAY_UNK30_CTRL_UPDATE_VCLK1                        0x00000400
 #define NV50_PDISPLAY_UNK30_CTRL_PENDING                             0x80000000
-#define NV50_PDISPLAY_TRAPPED_ADDR                                   0x00610080
-#define NV50_PDISPLAY_TRAPPED_DATA                                   0x00610084
-#define NV50_PDISPLAY_CHANNEL_STAT(i)                  ((i) * 0x10 + 0x00610200)
-#define NV50_PDISPLAY_CHANNEL_STAT_DMA                               0x00000010
-#define NV50_PDISPLAY_CHANNEL_STAT_DMA_DISABLED                      0x00000000
-#define NV50_PDISPLAY_CHANNEL_STAT_DMA_ENABLED                       0x00000010
-#define NV50_PDISPLAY_CHANNEL_DMA_CB(i)                ((i) * 0x10 + 0x00610204)
-#define NV50_PDISPLAY_CHANNEL_DMA_CB_LOCATION                        0x00000002
-#define NV50_PDISPLAY_CHANNEL_DMA_CB_LOCATION_VRAM                   0x00000000
-#define NV50_PDISPLAY_CHANNEL_DMA_CB_LOCATION_SYSTEM                 0x00000002
-#define NV50_PDISPLAY_CHANNEL_DMA_CB_VALID                           0x00000001
-#define NV50_PDISPLAY_CHANNEL_UNK2(i)                  ((i) * 0x10 + 0x00610208)
-#define NV50_PDISPLAY_CHANNEL_UNK3(i)                  ((i) * 0x10 + 0x0061020c)
+#define NV50_PDISPLAY_TRAPPED_ADDR(i)                  ((i) * 0x08 + 0x00610080)
+#define NV50_PDISPLAY_TRAPPED_DATA(i)                  ((i) * 0x08 + 0x00610084)
+#define NV50_PDISPLAY_EVO_CTRL(i)                      ((i) * 0x10 + 0x00610200)
+#define NV50_PDISPLAY_EVO_CTRL_DMA                                   0x00000010
+#define NV50_PDISPLAY_EVO_CTRL_DMA_DISABLED                          0x00000000
+#define NV50_PDISPLAY_EVO_CTRL_DMA_ENABLED                           0x00000010
+#define NV50_PDISPLAY_EVO_DMA_CB(i)                    ((i) * 0x10 + 0x00610204)
+#define NV50_PDISPLAY_EVO_DMA_CB_LOCATION                            0x00000002
+#define NV50_PDISPLAY_EVO_DMA_CB_LOCATION_VRAM                       0x00000000
+#define NV50_PDISPLAY_EVO_DMA_CB_LOCATION_SYSTEM                     0x00000002
+#define NV50_PDISPLAY_EVO_DMA_CB_VALID                               0x00000001
+#define NV50_PDISPLAY_EVO_UNK2(i)                      ((i) * 0x10 + 0x00610208)
+#define NV50_PDISPLAY_EVO_HASH_TAG(i)                  ((i) * 0x10 + 0x0061020c)
 
 #define NV50_PDISPLAY_CURSOR                                         0x00610270
 #define NV50_PDISPLAY_CURSOR_CURSOR_CTRL2(i)           ((i) * 0x10 + 0x00610270)
@@ -746,15 +743,11 @@
 #define NV50_PDISPLAY_CURSOR_CURSOR_CTRL2_STATUS                     0x00030000
 #define NV50_PDISPLAY_CURSOR_CURSOR_CTRL2_STATUS_ACTIVE              0x00010000
 
-#define NV50_PDISPLAY_CTRL_STATE                                     0x00610300
-#define NV50_PDISPLAY_CTRL_STATE_PENDING                             0x80000000
-#define NV50_PDISPLAY_CTRL_STATE_METHOD                              0x00001ffc
-#define NV50_PDISPLAY_CTRL_STATE_ENABLE                              0x00000001
-#define NV50_PDISPLAY_CTRL_VAL                                       0x00610304
-#define NV50_PDISPLAY_UNK_380                                        0x00610380
-#define NV50_PDISPLAY_RAM_AMOUNT                                     0x00610384
-#define NV50_PDISPLAY_UNK_388                                        0x00610388
-#define NV50_PDISPLAY_UNK_38C                                        0x0061038c
+#define NV50_PDISPLAY_PIO_CTRL                                       0x00610300
+#define NV50_PDISPLAY_PIO_CTRL_PENDING                               0x80000000
+#define NV50_PDISPLAY_PIO_CTRL_MTHD                                  0x00001ffc
+#define NV50_PDISPLAY_PIO_CTRL_ENABLED                               0x00000001
+#define NV50_PDISPLAY_PIO_DATA                                       0x00610304
 
 #define NV50_PDISPLAY_CRTC_P(i, r)        ((i) * 0x540 + NV50_PDISPLAY_CRTC_##r)
 #define NV50_PDISPLAY_CRTC_C(i, r)    (4 + (i) * 0x540 + NV50_PDISPLAY_CRTC_##r)
diff --git a/drivers/gpu/drm/nouveau/nouveau_sgdma.c b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
index d4ac97007038..9a250eb53098 100644
--- a/drivers/gpu/drm/nouveau/nouveau_sgdma.c
+++ b/drivers/gpu/drm/nouveau/nouveau_sgdma.c
@@ -14,7 +14,7 @@ struct nouveau_sgdma_be {
 	dma_addr_t *pages;
 	unsigned nr_pages;
 
-	unsigned pte_start;
+	u64 offset;
 	bool bound;
 };
 
@@ -74,18 +74,6 @@ nouveau_sgdma_clear(struct ttm_backend *be)
 	}
 }
 
-static inline unsigned
-nouveau_sgdma_pte(struct drm_device *dev, uint64_t offset)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	unsigned pte = (offset >> NV_CTXDMA_PAGE_SHIFT);
-
-	if (dev_priv->card_type < NV_50)
-		return pte + 2;
-
-	return pte << 1;
-}
-
 static int
 nouveau_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 {
@@ -97,32 +85,17 @@ nouveau_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
 
 	NV_DEBUG(dev, "pg=0x%lx\n", mem->start);
 
-	pte = nouveau_sgdma_pte(nvbe->dev, mem->start << PAGE_SHIFT);
-	nvbe->pte_start = pte;
+	nvbe->offset = mem->start << PAGE_SHIFT;
+	pte = (nvbe->offset >> NV_CTXDMA_PAGE_SHIFT) + 2;
 	for (i = 0; i < nvbe->nr_pages; i++) {
 		dma_addr_t dma_offset = nvbe->pages[i];
 		uint32_t offset_l = lower_32_bits(dma_offset);
-		uint32_t offset_h = upper_32_bits(dma_offset);
-
-		for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++) {
-			if (dev_priv->card_type < NV_50) {
-				nv_wo32(gpuobj, (pte * 4) + 0, offset_l | 3);
-				pte += 1;
-			} else {
-				nv_wo32(gpuobj, (pte * 4) + 0, offset_l | 0x21);
-				nv_wo32(gpuobj, (pte * 4) + 4, offset_h & 0xff);
-				pte += 2;
-			}
 
+		for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++, pte++) {
+			nv_wo32(gpuobj, (pte * 4) + 0, offset_l | 3);
 			dma_offset += NV_CTXDMA_PAGE_SIZE;
 		}
 	}
-	dev_priv->engine.instmem.flush(nvbe->dev);
-
-	if (dev_priv->card_type == NV_50) {
-		dev_priv->engine.fifo.tlb_flush(dev);
-		dev_priv->engine.graph.tlb_flush(dev);
-	}
 
 	nvbe->bound = true;
 	return 0;
@@ -142,28 +115,10 @@ nouveau_sgdma_unbind(struct ttm_backend *be)
 	if (!nvbe->bound)
 		return 0;
 
-	pte = nvbe->pte_start;
+	pte = (nvbe->offset >> NV_CTXDMA_PAGE_SHIFT) + 2;
 	for (i = 0; i < nvbe->nr_pages; i++) {
-		dma_addr_t dma_offset = dev_priv->gart_info.sg_dummy_bus;
-
-		for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++) {
-			if (dev_priv->card_type < NV_50) {
-				nv_wo32(gpuobj, (pte * 4) + 0, dma_offset | 3);
-				pte += 1;
-			} else {
-				nv_wo32(gpuobj, (pte * 4) + 0, 0x00000000);
-				nv_wo32(gpuobj, (pte * 4) + 4, 0x00000000);
-				pte += 2;
-			}
-
-			dma_offset += NV_CTXDMA_PAGE_SIZE;
-		}
-	}
-	dev_priv->engine.instmem.flush(nvbe->dev);
-
-	if (dev_priv->card_type == NV_50) {
-		dev_priv->engine.fifo.tlb_flush(dev);
-		dev_priv->engine.graph.tlb_flush(dev);
+		for (j = 0; j < PAGE_SIZE / NV_CTXDMA_PAGE_SIZE; j++, pte++)
+			nv_wo32(gpuobj, (pte * 4) + 0, 0x00000000);
 	}
 
 	nvbe->bound = false;
@@ -186,6 +141,35 @@ nouveau_sgdma_destroy(struct ttm_backend *be)
 	}
 }
 
+static int
+nv50_sgdma_bind(struct ttm_backend *be, struct ttm_mem_reg *mem)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+
+	nvbe->offset = mem->start << PAGE_SHIFT;
+
+	nouveau_vm_map_sg(&dev_priv->gart_info.vma, nvbe->offset,
+			  nvbe->nr_pages << PAGE_SHIFT, nvbe->pages);
+	nvbe->bound = true;
+	return 0;
+}
+
+static int
+nv50_sgdma_unbind(struct ttm_backend *be)
+{
+	struct nouveau_sgdma_be *nvbe = (struct nouveau_sgdma_be *)be;
+	struct drm_nouveau_private *dev_priv = nvbe->dev->dev_private;
+
+	if (!nvbe->bound)
+		return 0;
+
+	nouveau_vm_unmap_at(&dev_priv->gart_info.vma, nvbe->offset,
+			    nvbe->nr_pages << PAGE_SHIFT);
+	nvbe->bound = false;
+	return 0;
+}
+
 static struct ttm_backend_func nouveau_sgdma_backend = {
 	.populate		= nouveau_sgdma_populate,
 	.clear			= nouveau_sgdma_clear,
@@ -194,23 +178,30 @@ static struct ttm_backend_func nouveau_sgdma_backend = {
 	.destroy		= nouveau_sgdma_destroy
 };
 
+static struct ttm_backend_func nv50_sgdma_backend = {
+	.populate		= nouveau_sgdma_populate,
+	.clear			= nouveau_sgdma_clear,
+	.bind			= nv50_sgdma_bind,
+	.unbind			= nv50_sgdma_unbind,
+	.destroy		= nouveau_sgdma_destroy
+};
+
 struct ttm_backend *
 nouveau_sgdma_init_ttm(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_sgdma_be *nvbe;
 
-	if (!dev_priv->gart_info.sg_ctxdma)
-		return NULL;
-
 	nvbe = kzalloc(sizeof(*nvbe), GFP_KERNEL);
 	if (!nvbe)
 		return NULL;
 
 	nvbe->dev = dev;
 
-	nvbe->backend.func	= &nouveau_sgdma_backend;
-
+	if (dev_priv->card_type < NV_50)
+		nvbe->backend.func = &nouveau_sgdma_backend;
+	else
+		nvbe->backend.func = &nv50_sgdma_backend;
 	return &nvbe->backend;
 }
 
@@ -218,7 +209,6 @@ int
 nouveau_sgdma_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct pci_dev *pdev = dev->pdev;
 	struct nouveau_gpuobj *gpuobj = NULL;
 	uint32_t aper_size, obj_size;
 	int i, ret;
@@ -231,68 +221,40 @@ nouveau_sgdma_init(struct drm_device *dev)
 
 		obj_size  = (aper_size >> NV_CTXDMA_PAGE_SHIFT) * 4;
 		obj_size += 8; /* ctxdma header */
-	} else {
-		/* 1 entire VM page table */
-		aper_size = (512 * 1024 * 1024);
-		obj_size  = (aper_size >> NV_CTXDMA_PAGE_SHIFT) * 8;
-	}
-
-	ret = nouveau_gpuobj_new(dev, NULL, obj_size, 16,
-				      NVOBJ_FLAG_ZERO_ALLOC |
-				      NVOBJ_FLAG_ZERO_FREE, &gpuobj);
-	if (ret) {
-		NV_ERROR(dev, "Error creating sgdma object: %d\n", ret);
-		return ret;
-	}
-
-	dev_priv->gart_info.sg_dummy_page =
-		alloc_page(GFP_KERNEL|__GFP_DMA32|__GFP_ZERO);
-	if (!dev_priv->gart_info.sg_dummy_page) {
-		nouveau_gpuobj_ref(NULL, &gpuobj);
-		return -ENOMEM;
-	}
 
-	set_bit(PG_locked, &dev_priv->gart_info.sg_dummy_page->flags);
-	dev_priv->gart_info.sg_dummy_bus =
-		pci_map_page(pdev, dev_priv->gart_info.sg_dummy_page, 0,
-			     PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	if (pci_dma_mapping_error(pdev, dev_priv->gart_info.sg_dummy_bus)) {
-		nouveau_gpuobj_ref(NULL, &gpuobj);
-		return -EFAULT;
-	}
+		ret = nouveau_gpuobj_new(dev, NULL, obj_size, 16,
+					      NVOBJ_FLAG_ZERO_ALLOC |
+					      NVOBJ_FLAG_ZERO_FREE, &gpuobj);
+		if (ret) {
+			NV_ERROR(dev, "Error creating sgdma object: %d\n", ret);
+			return ret;
+		}
 
-	if (dev_priv->card_type < NV_50) {
-		/* special case, allocated from global instmem heap so
-		 * cinst is invalid, we use it on all channels though so
-		 * cinst needs to be valid, set it the same as pinst
-		 */
-		gpuobj->cinst = gpuobj->pinst;
-
-		/* Maybe use NV_DMA_TARGET_AGP for PCIE? NVIDIA do this, and
-		 * confirmed to work on c51.  Perhaps means NV_DMA_TARGET_PCIE
-		 * on those cards? */
 		nv_wo32(gpuobj, 0, NV_CLASS_DMA_IN_MEMORY |
 				   (1 << 12) /* PT present */ |
 				   (0 << 13) /* PT *not* linear */ |
-				   (NV_DMA_ACCESS_RW  << 14) |
-				   (NV_DMA_TARGET_PCI << 16));
+				   (0 << 14) /* RW */ |
+				   (2 << 16) /* PCI */);
 		nv_wo32(gpuobj, 4, aper_size - 1);
-		for (i = 2; i < 2 + (aper_size >> 12); i++) {
-			nv_wo32(gpuobj, i * 4,
-				dev_priv->gart_info.sg_dummy_bus | 3);
-		}
-	} else {
-		for (i = 0; i < obj_size; i += 8) {
-			nv_wo32(gpuobj, i + 0, 0x00000000);
-			nv_wo32(gpuobj, i + 4, 0x00000000);
-		}
+		for (i = 2; i < 2 + (aper_size >> 12); i++)
+			nv_wo32(gpuobj, i * 4, 0x00000000);
+
+		dev_priv->gart_info.sg_ctxdma = gpuobj;
+		dev_priv->gart_info.aper_base = 0;
+		dev_priv->gart_info.aper_size = aper_size;
+	} else
+	if (dev_priv->chan_vm) {
+		ret = nouveau_vm_get(dev_priv->chan_vm, 512 * 1024 * 1024,
+				     12, NV_MEM_ACCESS_RW,
+				     &dev_priv->gart_info.vma);
+		if (ret)
+			return ret;
+
+		dev_priv->gart_info.aper_base = dev_priv->gart_info.vma.offset;
+		dev_priv->gart_info.aper_size = 512 * 1024 * 1024;
 	}
-	dev_priv->engine.instmem.flush(dev);
 
 	dev_priv->gart_info.type      = NOUVEAU_GART_SGDMA;
-	dev_priv->gart_info.aper_base = 0;
-	dev_priv->gart_info.aper_size = aper_size;
-	dev_priv->gart_info.sg_ctxdma = gpuobj;
 	return 0;
 }
 
@@ -301,31 +263,19 @@ nouveau_sgdma_takedown(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
-	if (dev_priv->gart_info.sg_dummy_page) {
-		pci_unmap_page(dev->pdev, dev_priv->gart_info.sg_dummy_bus,
-			       NV_CTXDMA_PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-		unlock_page(dev_priv->gart_info.sg_dummy_page);
-		__free_page(dev_priv->gart_info.sg_dummy_page);
-		dev_priv->gart_info.sg_dummy_page = NULL;
-		dev_priv->gart_info.sg_dummy_bus = 0;
-	}
-
 	nouveau_gpuobj_ref(NULL, &dev_priv->gart_info.sg_ctxdma);
+	nouveau_vm_put(&dev_priv->gart_info.vma);
 }
 
-int
-nouveau_sgdma_get_page(struct drm_device *dev, uint32_t offset, uint32_t *page)
+uint32_t
+nouveau_sgdma_get_physical(struct drm_device *dev, uint32_t offset)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_gpuobj *gpuobj = dev_priv->gart_info.sg_ctxdma;
-	int pte;
+	int pte = (offset >> NV_CTXDMA_PAGE_SHIFT) + 2;
 
-	pte = (offset >> NV_CTXDMA_PAGE_SHIFT) << 2;
-	if (dev_priv->card_type < NV_50) {
-		*page = nv_ro32(gpuobj, (pte + 8)) & ~NV_CTXDMA_PAGE_MASK;
-		return 0;
-	}
+	BUG_ON(dev_priv->card_type >= NV_50);
 
-	NV_ERROR(dev, "Unimplemented on NV50\n");
-	return -EINVAL;
+	return (nv_ro32(gpuobj, 4 * pte) & ~NV_CTXDMA_PAGE_MASK) |
+		(offset & NV_CTXDMA_PAGE_MASK);
 }
diff --git a/drivers/gpu/drm/nouveau/nouveau_state.c b/drivers/gpu/drm/nouveau/nouveau_state.c
index 049f755567e5..8eac943e8fd2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_state.c
+++ b/drivers/gpu/drm/nouveau/nouveau_state.c
@@ -53,10 +53,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->instmem.takedown	= nv04_instmem_takedown;
 		engine->instmem.suspend		= nv04_instmem_suspend;
 		engine->instmem.resume		= nv04_instmem_resume;
-		engine->instmem.populate	= nv04_instmem_populate;
-		engine->instmem.clear		= nv04_instmem_clear;
-		engine->instmem.bind		= nv04_instmem_bind;
-		engine->instmem.unbind		= nv04_instmem_unbind;
+		engine->instmem.get		= nv04_instmem_get;
+		engine->instmem.put		= nv04_instmem_put;
+		engine->instmem.map		= nv04_instmem_map;
+		engine->instmem.unmap		= nv04_instmem_unmap;
 		engine->instmem.flush		= nv04_instmem_flush;
 		engine->mc.init			= nv04_mc_init;
 		engine->mc.takedown		= nv04_mc_takedown;
@@ -65,7 +65,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv04_fb_init;
 		engine->fb.takedown		= nv04_fb_takedown;
-		engine->graph.grclass		= nv04_graph_grclass;
 		engine->graph.init		= nv04_graph_init;
 		engine->graph.takedown		= nv04_graph_takedown;
 		engine->graph.fifo_access	= nv04_graph_fifo_access;
@@ -76,7 +75,7 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->graph.unload_context	= nv04_graph_unload_context;
 		engine->fifo.channels		= 16;
 		engine->fifo.init		= nv04_fifo_init;
-		engine->fifo.takedown		= nouveau_stub_takedown;
+		engine->fifo.takedown		= nv04_fifo_fini;
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
@@ -99,16 +98,20 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x10:
 		engine->instmem.init		= nv04_instmem_init;
 		engine->instmem.takedown	= nv04_instmem_takedown;
 		engine->instmem.suspend		= nv04_instmem_suspend;
 		engine->instmem.resume		= nv04_instmem_resume;
-		engine->instmem.populate	= nv04_instmem_populate;
-		engine->instmem.clear		= nv04_instmem_clear;
-		engine->instmem.bind		= nv04_instmem_bind;
-		engine->instmem.unbind		= nv04_instmem_unbind;
+		engine->instmem.get		= nv04_instmem_get;
+		engine->instmem.put		= nv04_instmem_put;
+		engine->instmem.map		= nv04_instmem_map;
+		engine->instmem.unmap		= nv04_instmem_unmap;
 		engine->instmem.flush		= nv04_instmem_flush;
 		engine->mc.init			= nv04_mc_init;
 		engine->mc.takedown		= nv04_mc_takedown;
@@ -117,8 +120,9 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv10_fb_init;
 		engine->fb.takedown		= nv10_fb_takedown;
-		engine->fb.set_region_tiling	= nv10_fb_set_region_tiling;
-		engine->graph.grclass		= nv10_graph_grclass;
+		engine->fb.init_tile_region	= nv10_fb_init_tile_region;
+		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
+		engine->fb.free_tile_region	= nv10_fb_free_tile_region;
 		engine->graph.init		= nv10_graph_init;
 		engine->graph.takedown		= nv10_graph_takedown;
 		engine->graph.channel		= nv10_graph_channel;
@@ -127,17 +131,17 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->graph.fifo_access	= nv04_graph_fifo_access;
 		engine->graph.load_context	= nv10_graph_load_context;
 		engine->graph.unload_context	= nv10_graph_unload_context;
-		engine->graph.set_region_tiling	= nv10_graph_set_region_tiling;
+		engine->graph.set_tile_region	= nv10_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
-		engine->fifo.takedown		= nouveau_stub_takedown;
+		engine->fifo.takedown		= nv04_fifo_fini;
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
-		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.destroy_context	= nv04_fifo_destroy_context;
 		engine->fifo.load_context	= nv10_fifo_load_context;
 		engine->fifo.unload_context	= nv10_fifo_unload_context;
 		engine->display.early_init	= nv04_display_early_init;
@@ -153,16 +157,20 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x20:
 		engine->instmem.init		= nv04_instmem_init;
 		engine->instmem.takedown	= nv04_instmem_takedown;
 		engine->instmem.suspend		= nv04_instmem_suspend;
 		engine->instmem.resume		= nv04_instmem_resume;
-		engine->instmem.populate	= nv04_instmem_populate;
-		engine->instmem.clear		= nv04_instmem_clear;
-		engine->instmem.bind		= nv04_instmem_bind;
-		engine->instmem.unbind		= nv04_instmem_unbind;
+		engine->instmem.get		= nv04_instmem_get;
+		engine->instmem.put		= nv04_instmem_put;
+		engine->instmem.map		= nv04_instmem_map;
+		engine->instmem.unmap		= nv04_instmem_unmap;
 		engine->instmem.flush		= nv04_instmem_flush;
 		engine->mc.init			= nv04_mc_init;
 		engine->mc.takedown		= nv04_mc_takedown;
@@ -171,8 +179,9 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv10_fb_init;
 		engine->fb.takedown		= nv10_fb_takedown;
-		engine->fb.set_region_tiling	= nv10_fb_set_region_tiling;
-		engine->graph.grclass		= nv20_graph_grclass;
+		engine->fb.init_tile_region	= nv10_fb_init_tile_region;
+		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
+		engine->fb.free_tile_region	= nv10_fb_free_tile_region;
 		engine->graph.init		= nv20_graph_init;
 		engine->graph.takedown		= nv20_graph_takedown;
 		engine->graph.channel		= nv10_graph_channel;
@@ -181,17 +190,17 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->graph.fifo_access	= nv04_graph_fifo_access;
 		engine->graph.load_context	= nv20_graph_load_context;
 		engine->graph.unload_context	= nv20_graph_unload_context;
-		engine->graph.set_region_tiling	= nv20_graph_set_region_tiling;
+		engine->graph.set_tile_region	= nv20_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
-		engine->fifo.takedown		= nouveau_stub_takedown;
+		engine->fifo.takedown		= nv04_fifo_fini;
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
-		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.destroy_context	= nv04_fifo_destroy_context;
 		engine->fifo.load_context	= nv10_fifo_load_context;
 		engine->fifo.unload_context	= nv10_fifo_unload_context;
 		engine->display.early_init	= nv04_display_early_init;
@@ -207,16 +216,20 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_get		= nv04_pm_clock_get;
 		engine->pm.clock_pre		= nv04_pm_clock_pre;
 		engine->pm.clock_set		= nv04_pm_clock_set;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x30:
 		engine->instmem.init		= nv04_instmem_init;
 		engine->instmem.takedown	= nv04_instmem_takedown;
 		engine->instmem.suspend		= nv04_instmem_suspend;
 		engine->instmem.resume		= nv04_instmem_resume;
-		engine->instmem.populate	= nv04_instmem_populate;
-		engine->instmem.clear		= nv04_instmem_clear;
-		engine->instmem.bind		= nv04_instmem_bind;
-		engine->instmem.unbind		= nv04_instmem_unbind;
+		engine->instmem.get		= nv04_instmem_get;
+		engine->instmem.put		= nv04_instmem_put;
+		engine->instmem.map		= nv04_instmem_map;
+		engine->instmem.unmap		= nv04_instmem_unmap;
 		engine->instmem.flush		= nv04_instmem_flush;
 		engine->mc.init			= nv04_mc_init;
 		engine->mc.takedown		= nv04_mc_takedown;
@@ -225,8 +238,9 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv30_fb_init;
 		engine->fb.takedown		= nv30_fb_takedown;
-		engine->fb.set_region_tiling	= nv10_fb_set_region_tiling;
-		engine->graph.grclass		= nv30_graph_grclass;
+		engine->fb.init_tile_region	= nv30_fb_init_tile_region;
+		engine->fb.set_tile_region	= nv10_fb_set_tile_region;
+		engine->fb.free_tile_region	= nv30_fb_free_tile_region;
 		engine->graph.init		= nv30_graph_init;
 		engine->graph.takedown		= nv20_graph_takedown;
 		engine->graph.fifo_access	= nv04_graph_fifo_access;
@@ -235,17 +249,17 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->graph.destroy_context	= nv20_graph_destroy_context;
 		engine->graph.load_context	= nv20_graph_load_context;
 		engine->graph.unload_context	= nv20_graph_unload_context;
-		engine->graph.set_region_tiling	= nv20_graph_set_region_tiling;
+		engine->graph.set_tile_region	= nv20_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv10_fifo_init;
-		engine->fifo.takedown		= nouveau_stub_takedown;
+		engine->fifo.takedown		= nv04_fifo_fini;
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv10_fifo_create_context;
-		engine->fifo.destroy_context	= nv10_fifo_destroy_context;
+		engine->fifo.destroy_context	= nv04_fifo_destroy_context;
 		engine->fifo.load_context	= nv10_fifo_load_context;
 		engine->fifo.unload_context	= nv10_fifo_unload_context;
 		engine->display.early_init	= nv04_display_early_init;
@@ -263,6 +277,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.clock_set		= nv04_pm_clock_set;
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x40:
 	case 0x60:
@@ -270,10 +288,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->instmem.takedown	= nv04_instmem_takedown;
 		engine->instmem.suspend		= nv04_instmem_suspend;
 		engine->instmem.resume		= nv04_instmem_resume;
-		engine->instmem.populate	= nv04_instmem_populate;
-		engine->instmem.clear		= nv04_instmem_clear;
-		engine->instmem.bind		= nv04_instmem_bind;
-		engine->instmem.unbind		= nv04_instmem_unbind;
+		engine->instmem.get		= nv04_instmem_get;
+		engine->instmem.put		= nv04_instmem_put;
+		engine->instmem.map		= nv04_instmem_map;
+		engine->instmem.unmap		= nv04_instmem_unmap;
 		engine->instmem.flush		= nv04_instmem_flush;
 		engine->mc.init			= nv40_mc_init;
 		engine->mc.takedown		= nv40_mc_takedown;
@@ -282,8 +300,9 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv40_fb_init;
 		engine->fb.takedown		= nv40_fb_takedown;
-		engine->fb.set_region_tiling	= nv40_fb_set_region_tiling;
-		engine->graph.grclass		= nv40_graph_grclass;
+		engine->fb.init_tile_region	= nv30_fb_init_tile_region;
+		engine->fb.set_tile_region	= nv40_fb_set_tile_region;
+		engine->fb.free_tile_region	= nv30_fb_free_tile_region;
 		engine->graph.init		= nv40_graph_init;
 		engine->graph.takedown		= nv40_graph_takedown;
 		engine->graph.fifo_access	= nv04_graph_fifo_access;
@@ -292,17 +311,17 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->graph.destroy_context	= nv40_graph_destroy_context;
 		engine->graph.load_context	= nv40_graph_load_context;
 		engine->graph.unload_context	= nv40_graph_unload_context;
-		engine->graph.set_region_tiling	= nv40_graph_set_region_tiling;
+		engine->graph.set_tile_region	= nv40_graph_set_tile_region;
 		engine->fifo.channels		= 32;
 		engine->fifo.init		= nv40_fifo_init;
-		engine->fifo.takedown		= nouveau_stub_takedown;
+		engine->fifo.takedown		= nv04_fifo_fini;
 		engine->fifo.disable		= nv04_fifo_disable;
 		engine->fifo.enable		= nv04_fifo_enable;
 		engine->fifo.reassign		= nv04_fifo_reassign;
 		engine->fifo.cache_pull		= nv04_fifo_cache_pull;
 		engine->fifo.channel_id		= nv10_fifo_channel_id;
 		engine->fifo.create_context	= nv40_fifo_create_context;
-		engine->fifo.destroy_context	= nv40_fifo_destroy_context;
+		engine->fifo.destroy_context	= nv04_fifo_destroy_context;
 		engine->fifo.load_context	= nv40_fifo_load_context;
 		engine->fifo.unload_context	= nv40_fifo_unload_context;
 		engine->display.early_init	= nv04_display_early_init;
@@ -321,6 +340,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
 		engine->pm.temp_get		= nv40_temp_get;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	case 0x50:
 	case 0x80: /* gotta love NVIDIA's consistency.. */
@@ -330,10 +353,10 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->instmem.takedown	= nv50_instmem_takedown;
 		engine->instmem.suspend		= nv50_instmem_suspend;
 		engine->instmem.resume		= nv50_instmem_resume;
-		engine->instmem.populate	= nv50_instmem_populate;
-		engine->instmem.clear		= nv50_instmem_clear;
-		engine->instmem.bind		= nv50_instmem_bind;
-		engine->instmem.unbind		= nv50_instmem_unbind;
+		engine->instmem.get		= nv50_instmem_get;
+		engine->instmem.put		= nv50_instmem_put;
+		engine->instmem.map		= nv50_instmem_map;
+		engine->instmem.unmap		= nv50_instmem_unmap;
 		if (dev_priv->chipset == 0x50)
 			engine->instmem.flush	= nv50_instmem_flush;
 		else
@@ -345,7 +368,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nv50_fb_init;
 		engine->fb.takedown		= nv50_fb_takedown;
-		engine->graph.grclass		= nv50_graph_grclass;
 		engine->graph.init		= nv50_graph_init;
 		engine->graph.takedown		= nv50_graph_takedown;
 		engine->graph.fifo_access	= nv50_graph_fifo_access;
@@ -381,24 +403,32 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->display.init		= nv50_display_init;
 		engine->display.destroy		= nv50_display_destroy;
 		engine->gpio.init		= nv50_gpio_init;
-		engine->gpio.takedown		= nouveau_stub_takedown;
+		engine->gpio.takedown		= nv50_gpio_fini;
 		engine->gpio.get		= nv50_gpio_get;
 		engine->gpio.set		= nv50_gpio_set;
+		engine->gpio.irq_register	= nv50_gpio_irq_register;
+		engine->gpio.irq_unregister	= nv50_gpio_irq_unregister;
 		engine->gpio.irq_enable		= nv50_gpio_irq_enable;
 		switch (dev_priv->chipset) {
-		case 0xa3:
-		case 0xa5:
-		case 0xa8:
-		case 0xaf:
-			engine->pm.clock_get	= nva3_pm_clock_get;
-			engine->pm.clock_pre	= nva3_pm_clock_pre;
-			engine->pm.clock_set	= nva3_pm_clock_set;
-			break;
-		default:
+		case 0x84:
+		case 0x86:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+		case 0x98:
+		case 0xa0:
+		case 0xaa:
+		case 0xac:
+		case 0x50:
 			engine->pm.clock_get	= nv50_pm_clock_get;
 			engine->pm.clock_pre	= nv50_pm_clock_pre;
 			engine->pm.clock_set	= nv50_pm_clock_set;
 			break;
+		default:
+			engine->pm.clock_get	= nva3_pm_clock_get;
+			engine->pm.clock_pre	= nva3_pm_clock_pre;
+			engine->pm.clock_set	= nva3_pm_clock_set;
+			break;
 		}
 		engine->pm.voltage_get		= nouveau_voltage_gpio_get;
 		engine->pm.voltage_set		= nouveau_voltage_gpio_set;
@@ -406,16 +436,38 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 			engine->pm.temp_get	= nv84_temp_get;
 		else
 			engine->pm.temp_get	= nv40_temp_get;
+		switch (dev_priv->chipset) {
+		case 0x84:
+		case 0x86:
+		case 0x92:
+		case 0x94:
+		case 0x96:
+		case 0xa0:
+			engine->crypt.init	= nv84_crypt_init;
+			engine->crypt.takedown	= nv84_crypt_fini;
+			engine->crypt.create_context = nv84_crypt_create_context;
+			engine->crypt.destroy_context = nv84_crypt_destroy_context;
+			engine->crypt.tlb_flush	= nv84_crypt_tlb_flush;
+			break;
+		default:
+			engine->crypt.init	= nouveau_stub_init;
+			engine->crypt.takedown	= nouveau_stub_takedown;
+			break;
+		}
+		engine->vram.init		= nv50_vram_init;
+		engine->vram.get		= nv50_vram_new;
+		engine->vram.put		= nv50_vram_del;
+		engine->vram.flags_valid	= nv50_vram_flags_valid;
 		break;
 	case 0xC0:
 		engine->instmem.init		= nvc0_instmem_init;
 		engine->instmem.takedown	= nvc0_instmem_takedown;
 		engine->instmem.suspend		= nvc0_instmem_suspend;
 		engine->instmem.resume		= nvc0_instmem_resume;
-		engine->instmem.populate	= nvc0_instmem_populate;
-		engine->instmem.clear		= nvc0_instmem_clear;
-		engine->instmem.bind		= nvc0_instmem_bind;
-		engine->instmem.unbind		= nvc0_instmem_unbind;
+		engine->instmem.get		= nvc0_instmem_get;
+		engine->instmem.put		= nvc0_instmem_put;
+		engine->instmem.map		= nvc0_instmem_map;
+		engine->instmem.unmap		= nvc0_instmem_unmap;
 		engine->instmem.flush		= nvc0_instmem_flush;
 		engine->mc.init			= nv50_mc_init;
 		engine->mc.takedown		= nv50_mc_takedown;
@@ -424,7 +476,6 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->timer.takedown		= nv04_timer_takedown;
 		engine->fb.init			= nvc0_fb_init;
 		engine->fb.takedown		= nvc0_fb_takedown;
-		engine->graph.grclass		= NULL;  //nvc0_graph_grclass;
 		engine->graph.init		= nvc0_graph_init;
 		engine->graph.takedown		= nvc0_graph_takedown;
 		engine->graph.fifo_access	= nvc0_graph_fifo_access;
@@ -453,7 +504,13 @@ static int nouveau_init_engine_ptrs(struct drm_device *dev)
 		engine->gpio.takedown		= nouveau_stub_takedown;
 		engine->gpio.get		= nv50_gpio_get;
 		engine->gpio.set		= nv50_gpio_set;
+		engine->gpio.irq_register	= nv50_gpio_irq_register;
+		engine->gpio.irq_unregister	= nv50_gpio_irq_unregister;
 		engine->gpio.irq_enable		= nv50_gpio_irq_enable;
+		engine->crypt.init		= nouveau_stub_init;
+		engine->crypt.takedown		= nouveau_stub_takedown;
+		engine->vram.init		= nouveau_mem_detect;
+		engine->vram.flags_valid	= nouveau_mem_flags_valid;
 		break;
 	default:
 		NV_ERROR(dev, "NV%02x unsupported\n", dev_priv->chipset);
@@ -495,7 +552,7 @@ nouveau_card_init_channel(struct drm_device *dev)
 
 	ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
 				     0, dev_priv->vram_size,
-				     NV_DMA_ACCESS_RW, NV_DMA_TARGET_VIDMEM,
+				     NV_MEM_ACCESS_RW, NV_MEM_TARGET_VRAM,
 				     &gpuobj);
 	if (ret)
 		goto out_err;
@@ -505,9 +562,10 @@ nouveau_card_init_channel(struct drm_device *dev)
 	if (ret)
 		goto out_err;
 
-	ret = nouveau_gpuobj_gart_dma_new(dev_priv->channel, 0,
-					  dev_priv->gart_info.aper_size,
-					  NV_DMA_ACCESS_RW, &gpuobj, NULL);
+	ret = nouveau_gpuobj_dma_new(dev_priv->channel, NV_CLASS_DMA_IN_MEMORY,
+				     0, dev_priv->gart_info.aper_size,
+				     NV_MEM_ACCESS_RW, NV_MEM_TARGET_GART,
+				     &gpuobj);
 	if (ret)
 		goto out_err;
 
@@ -516,11 +574,11 @@ nouveau_card_init_channel(struct drm_device *dev)
 	if (ret)
 		goto out_err;
 
+	mutex_unlock(&dev_priv->channel->mutex);
 	return 0;
 
 out_err:
-	nouveau_channel_free(dev_priv->channel);
-	dev_priv->channel = NULL;
+	nouveau_channel_put(&dev_priv->channel);
 	return ret;
 }
 
@@ -567,6 +625,8 @@ nouveau_card_init(struct drm_device *dev)
 	if (ret)
 		goto out;
 	engine = &dev_priv->engine;
+	spin_lock_init(&dev_priv->channels.lock);
+	spin_lock_init(&dev_priv->tile.lock);
 	spin_lock_init(&dev_priv->context_switch_lock);
 
 	/* Make the CRTCs and I2C buses accessible */
@@ -625,26 +685,28 @@ nouveau_card_init(struct drm_device *dev)
 		if (ret)
 			goto out_fb;
 
+		/* PCRYPT */
+		ret = engine->crypt.init(dev);
+		if (ret)
+			goto out_graph;
+
 		/* PFIFO */
 		ret = engine->fifo.init(dev);
 		if (ret)
-			goto out_graph;
+			goto out_crypt;
 	}
 
 	ret = engine->display.create(dev);
 	if (ret)
 		goto out_fifo;
 
-	/* this call irq_preinstall, register irq handler and
-	 * call irq_postinstall
-	 */
-	ret = drm_irq_install(dev);
+	ret = drm_vblank_init(dev, nv_two_heads(dev) ? 2 : 1);
 	if (ret)
-		goto out_display;
+		goto out_vblank;
 
-	ret = drm_vblank_init(dev, 0);
+	ret = nouveau_irq_init(dev);
 	if (ret)
-		goto out_irq;
+		goto out_vblank;
 
 	/* what about PVIDEO/PCRTC/PRAMDAC etc? */
 
@@ -669,12 +731,16 @@ nouveau_card_init(struct drm_device *dev)
 out_fence:
 	nouveau_fence_fini(dev);
 out_irq:
-	drm_irq_uninstall(dev);
-out_display:
+	nouveau_irq_fini(dev);
+out_vblank:
+	drm_vblank_cleanup(dev);
 	engine->display.destroy(dev);
 out_fifo:
 	if (!nouveau_noaccel)
 		engine->fifo.takedown(dev);
+out_crypt:
+	if (!nouveau_noaccel)
+		engine->crypt.takedown(dev);
 out_graph:
 	if (!nouveau_noaccel)
 		engine->graph.takedown(dev);
@@ -713,12 +779,12 @@ static void nouveau_card_takedown(struct drm_device *dev)
 
 	if (!engine->graph.accel_blocked) {
 		nouveau_fence_fini(dev);
-		nouveau_channel_free(dev_priv->channel);
-		dev_priv->channel = NULL;
+		nouveau_channel_put_unlocked(&dev_priv->channel);
 	}
 
 	if (!nouveau_noaccel) {
 		engine->fifo.takedown(dev);
+		engine->crypt.takedown(dev);
 		engine->graph.takedown(dev);
 	}
 	engine->fb.takedown(dev);
@@ -737,7 +803,8 @@ static void nouveau_card_takedown(struct drm_device *dev)
 	nouveau_gpuobj_takedown(dev);
 	nouveau_mem_vram_fini(dev);
 
-	drm_irq_uninstall(dev);
+	nouveau_irq_fini(dev);
+	drm_vblank_cleanup(dev);
 
 	nouveau_pm_fini(dev);
 	nouveau_bios_takedown(dev);
@@ -1024,21 +1091,6 @@ int nouveau_ioctl_getparam(struct drm_device *dev, void *data,
 		else
 			getparam->value = NV_PCI;
 		break;
-	case NOUVEAU_GETPARAM_FB_PHYSICAL:
-		getparam->value = dev_priv->fb_phys;
-		break;
-	case NOUVEAU_GETPARAM_AGP_PHYSICAL:
-		getparam->value = dev_priv->gart_info.aper_base;
-		break;
-	case NOUVEAU_GETPARAM_PCI_PHYSICAL:
-		if (dev->sg) {
-			getparam->value = (unsigned long)dev->sg->virtual;
-		} else {
-			NV_ERROR(dev, "Requested PCIGART address, "
-					"while no PCIGART was created\n");
-			return -EINVAL;
-		}
-		break;
 	case NOUVEAU_GETPARAM_FB_SIZE:
 		getparam->value = dev_priv->fb_available_size;
 		break;
@@ -1046,7 +1098,7 @@ int nouveau_ioctl_getparam(struct drm_device *dev, void *data,
 		getparam->value = dev_priv->gart_info.aper_size;
 		break;
 	case NOUVEAU_GETPARAM_VM_VRAM_BASE:
-		getparam->value = dev_priv->vm_vram_base;
+		getparam->value = 0; /* deprecated */
 		break;
 	case NOUVEAU_GETPARAM_PTIMER_TIME:
 		getparam->value = dev_priv->engine.timer.read(dev);
@@ -1054,6 +1106,9 @@ int nouveau_ioctl_getparam(struct drm_device *dev, void *data,
 	case NOUVEAU_GETPARAM_HAS_BO_USAGE:
 		getparam->value = 1;
 		break;
+	case NOUVEAU_GETPARAM_HAS_PAGEFLIP:
+		getparam->value = (dev_priv->card_type < NV_50);
+		break;
 	case NOUVEAU_GETPARAM_GRAPH_UNITS:
 		/* NV40 and NV50 versions are quite different, but register
 		 * address is the same. User is supposed to know the card
@@ -1087,8 +1142,9 @@ nouveau_ioctl_setparam(struct drm_device *dev, void *data,
 }
 
 /* Wait until (value(reg) & mask) == val, up until timeout has hit */
-bool nouveau_wait_until(struct drm_device *dev, uint64_t timeout,
-			uint32_t reg, uint32_t mask, uint32_t val)
+bool
+nouveau_wait_eq(struct drm_device *dev, uint64_t timeout,
+		uint32_t reg, uint32_t mask, uint32_t val)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
@@ -1102,10 +1158,33 @@ bool nouveau_wait_until(struct drm_device *dev, uint64_t timeout,
 	return false;
 }
 
+/* Wait until (value(reg) & mask) != val, up until timeout has hit */
+bool
+nouveau_wait_ne(struct drm_device *dev, uint64_t timeout,
+		uint32_t reg, uint32_t mask, uint32_t val)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
+	uint64_t start = ptimer->read(dev);
+
+	do {
+		if ((nv_rd32(dev, reg) & mask) != val)
+			return true;
+	} while (ptimer->read(dev) - start < timeout);
+
+	return false;
+}
+
 /* Waits for PGRAPH to go completely idle */
 bool nouveau_wait_for_idle(struct drm_device *dev)
 {
-	if (!nv_wait(dev, NV04_PGRAPH_STATUS, 0xffffffff, 0x00000000)) {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	uint32_t mask = ~0;
+
+	if (dev_priv->card_type == NV_40)
+		mask &= ~NV40_PGRAPH_STATUS_SYNC_STALL;
+
+	if (!nv_wait(dev, NV04_PGRAPH_STATUS, mask, 0)) {
 		NV_ERROR(dev, "PGRAPH idle timed out with status 0x%08x\n",
 			 nv_rd32(dev, NV04_PGRAPH_STATUS));
 		return false;
diff --git a/drivers/gpu/drm/nouveau/nouveau_util.c b/drivers/gpu/drm/nouveau/nouveau_util.c
new file mode 100644
index 000000000000..fbe0fb13bc1e
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_util.c
@@ -0,0 +1,69 @@
+/*
+ * Copyright (C) 2010 Nouveau Project
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#include <linux/ratelimit.h>
+
+#include "nouveau_util.h"
+
+static DEFINE_RATELIMIT_STATE(nouveau_ratelimit_state, 3 * HZ, 20);
+
+void
+nouveau_bitfield_print(const struct nouveau_bitfield *bf, u32 value)
+{
+	while (bf->name) {
+		if (value & bf->mask) {
+			printk(" %s", bf->name);
+			value &= ~bf->mask;
+		}
+
+		bf++;
+	}
+
+	if (value)
+		printk(" (unknown bits 0x%08x)", value);
+}
+
+void
+nouveau_enum_print(const struct nouveau_enum *en, u32 value)
+{
+	while (en->name) {
+		if (value == en->value) {
+			printk("%s", en->name);
+			return;
+		}
+
+		en++;
+	}
+
+	printk("(unknown enum 0x%08x)", value);
+}
+
+int
+nouveau_ratelimit(void)
+{
+	return __ratelimit(&nouveau_ratelimit_state);
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_util.h b/drivers/gpu/drm/nouveau/nouveau_util.h
new file mode 100644
index 000000000000..d9ceaea26f4b
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_util.h
@@ -0,0 +1,45 @@
+/*
+ * Copyright (C) 2010 Nouveau Project
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __NOUVEAU_UTIL_H__
+#define __NOUVEAU_UTIL_H__
+
+struct nouveau_bitfield {
+	u32 mask;
+	const char *name;
+};
+
+struct nouveau_enum {
+	u32 value;
+	const char *name;
+};
+
+void nouveau_bitfield_print(const struct nouveau_bitfield *, u32 value);
+void nouveau_enum_print(const struct nouveau_enum *, u32 value);
+int nouveau_ratelimit(void);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.c b/drivers/gpu/drm/nouveau/nouveau_vm.c
new file mode 100644
index 000000000000..07ab1749cf7d
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.c
@@ -0,0 +1,421 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_mm.h"
+#include "nouveau_vm.h"
+
+void
+nouveau_vm_map_at(struct nouveau_vma *vma, u64 delta, struct nouveau_vram *vram)
+{
+	struct nouveau_vm *vm = vma->vm;
+	struct nouveau_mm_node *r;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vm->pgt_bits - bits);
+	u32 end, len;
+
+	list_for_each_entry(r, &vram->regions, rl_entry) {
+		u64 phys = (u64)r->offset << 12;
+		u32 num  = r->length >> bits;
+
+		while (num) {
+			struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+
+			end = (pte + num);
+			if (unlikely(end >= max))
+				end = max;
+			len = end - pte;
+
+			vm->map(vma, pgt, vram, pte, len, phys);
+
+			num -= len;
+			pte += len;
+			if (unlikely(end >= max)) {
+				pde++;
+				pte = 0;
+			}
+		}
+	}
+
+	vm->flush(vm);
+}
+
+void
+nouveau_vm_map(struct nouveau_vma *vma, struct nouveau_vram *vram)
+{
+	nouveau_vm_map_at(vma, 0, vram);
+}
+
+void
+nouveau_vm_map_sg(struct nouveau_vma *vma, u64 delta, u64 length,
+		  dma_addr_t *list)
+{
+	struct nouveau_vm *vm = vma->vm;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vm->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		vm->map_sg(vma, pgt, pte, list, len);
+
+		num  -= len;
+		pte  += len;
+		list += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	vm->flush(vm);
+}
+
+void
+nouveau_vm_unmap_at(struct nouveau_vma *vma, u64 delta, u64 length)
+{
+	struct nouveau_vm *vm = vma->vm;
+	u32 offset = vma->node->offset + (delta >> 12);
+	u32 bits = vma->node->type - 12;
+	u32 num  = length >> vma->node->type;
+	u32 pde  = (offset >> vm->pgt_bits) - vm->fpde;
+	u32 pte  = (offset & ((1 << vm->pgt_bits) - 1)) >> bits;
+	u32 max  = 1 << (vm->pgt_bits - bits);
+	u32 end, len;
+
+	while (num) {
+		struct nouveau_gpuobj *pgt = vm->pgt[pde].obj;
+
+		end = (pte + num);
+		if (unlikely(end >= max))
+			end = max;
+		len = end - pte;
+
+		vm->unmap(pgt, pte, len);
+
+		num -= len;
+		pte += len;
+		if (unlikely(end >= max)) {
+			pde++;
+			pte = 0;
+		}
+	}
+
+	vm->flush(vm);
+}
+
+void
+nouveau_vm_unmap(struct nouveau_vma *vma)
+{
+	nouveau_vm_unmap_at(vma, 0, (u64)vma->node->length << 12);
+}
+
+static void
+nouveau_vm_unmap_pgt(struct nouveau_vm *vm, u32 fpde, u32 lpde)
+{
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_vm_pgt *vpgt;
+	struct nouveau_gpuobj *pgt;
+	u32 pde;
+
+	for (pde = fpde; pde <= lpde; pde++) {
+		vpgt = &vm->pgt[pde - vm->fpde];
+		if (--vpgt->refcount)
+			continue;
+
+		list_for_each_entry(vpgd, &vm->pgd_list, head) {
+			vm->unmap_pgt(vpgd->obj, pde);
+		}
+
+		pgt = vpgt->obj;
+		vpgt->obj = NULL;
+
+		mutex_unlock(&vm->mm->mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&vm->mm->mutex);
+	}
+}
+
+static int
+nouveau_vm_map_pgt(struct nouveau_vm *vm, u32 pde, u32 type)
+{
+	struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+	struct nouveau_vm_pgd *vpgd;
+	struct nouveau_gpuobj *pgt;
+	u32 pgt_size;
+	int ret;
+
+	pgt_size  = (1 << (vm->pgt_bits + 12)) >> type;
+	pgt_size *= 8;
+
+	mutex_unlock(&vm->mm->mutex);
+	ret = nouveau_gpuobj_new(vm->dev, NULL, pgt_size, 0x1000,
+				 NVOBJ_FLAG_ZERO_ALLOC, &pgt);
+	mutex_lock(&vm->mm->mutex);
+	if (unlikely(ret))
+		return ret;
+
+	/* someone beat us to filling the PDE while we didn't have the lock */
+	if (unlikely(vpgt->refcount++)) {
+		mutex_unlock(&vm->mm->mutex);
+		nouveau_gpuobj_ref(NULL, &pgt);
+		mutex_lock(&vm->mm->mutex);
+		return 0;
+	}
+
+	list_for_each_entry(vpgd, &vm->pgd_list, head) {
+		vm->map_pgt(vpgd->obj, type, pde, pgt);
+	}
+
+	vpgt->page_shift = type;
+	vpgt->obj = pgt;
+	return 0;
+}
+
+int
+nouveau_vm_get(struct nouveau_vm *vm, u64 size, u32 page_shift,
+	       u32 access, struct nouveau_vma *vma)
+{
+	u32 align = (1 << page_shift) >> 12;
+	u32 msize = size >> 12;
+	u32 fpde, lpde, pde;
+	int ret;
+
+	mutex_lock(&vm->mm->mutex);
+	ret = nouveau_mm_get(vm->mm, page_shift, msize, 0, align, &vma->node);
+	if (unlikely(ret != 0)) {
+		mutex_unlock(&vm->mm->mutex);
+		return ret;
+	}
+
+	fpde = (vma->node->offset >> vm->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
+	for (pde = fpde; pde <= lpde; pde++) {
+		struct nouveau_vm_pgt *vpgt = &vm->pgt[pde - vm->fpde];
+
+		if (likely(vpgt->refcount)) {
+			vpgt->refcount++;
+			continue;
+		}
+
+		ret = nouveau_vm_map_pgt(vm, pde, vma->node->type);
+		if (ret) {
+			if (pde != fpde)
+				nouveau_vm_unmap_pgt(vm, fpde, pde - 1);
+			nouveau_mm_put(vm->mm, vma->node);
+			mutex_unlock(&vm->mm->mutex);
+			vma->node = NULL;
+			return ret;
+		}
+	}
+	mutex_unlock(&vm->mm->mutex);
+
+	vma->vm     = vm;
+	vma->offset = (u64)vma->node->offset << 12;
+	vma->access = access;
+	return 0;
+}
+
+void
+nouveau_vm_put(struct nouveau_vma *vma)
+{
+	struct nouveau_vm *vm = vma->vm;
+	u32 fpde, lpde;
+
+	if (unlikely(vma->node == NULL))
+		return;
+	fpde = (vma->node->offset >> vm->pgt_bits);
+	lpde = (vma->node->offset + vma->node->length - 1) >> vm->pgt_bits;
+
+	mutex_lock(&vm->mm->mutex);
+	nouveau_mm_put(vm->mm, vma->node);
+	vma->node = NULL;
+	nouveau_vm_unmap_pgt(vm, fpde, lpde);
+	mutex_unlock(&vm->mm->mutex);
+}
+
+int
+nouveau_vm_new(struct drm_device *dev, u64 offset, u64 length, u64 mm_offset,
+	       u8 pgt_bits, u8 spg_shift, u8 lpg_shift,
+	       struct nouveau_vm **pvm)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vm *vm;
+	u64 mm_length = (offset + length) - mm_offset;
+	u32 block;
+	int ret;
+
+	vm = kzalloc(sizeof(*vm), GFP_KERNEL);
+	if (!vm)
+		return -ENOMEM;
+
+	if (dev_priv->card_type == NV_50) {
+		vm->map_pgt = nv50_vm_map_pgt;
+		vm->unmap_pgt = nv50_vm_unmap_pgt;
+		vm->map = nv50_vm_map;
+		vm->map_sg = nv50_vm_map_sg;
+		vm->unmap = nv50_vm_unmap;
+		vm->flush = nv50_vm_flush;
+	} else {
+		kfree(vm);
+		return -ENOSYS;
+	}
+
+	vm->fpde   = offset >> pgt_bits;
+	vm->lpde   = (offset + length - 1) >> pgt_bits;
+	vm->pgt = kcalloc(vm->lpde - vm->fpde + 1, sizeof(*vm->pgt), GFP_KERNEL);
+	if (!vm->pgt) {
+		kfree(vm);
+		return -ENOMEM;
+	}
+
+	INIT_LIST_HEAD(&vm->pgd_list);
+	vm->dev = dev;
+	vm->refcount = 1;
+	vm->pgt_bits = pgt_bits - 12;
+	vm->spg_shift = spg_shift;
+	vm->lpg_shift = lpg_shift;
+
+	block = (1 << pgt_bits);
+	if (length < block)
+		block = length;
+
+	ret = nouveau_mm_init(&vm->mm, mm_offset >> 12, mm_length >> 12,
+			      block >> 12);
+	if (ret) {
+		kfree(vm);
+		return ret;
+	}
+
+	*pvm = vm;
+	return 0;
+}
+
+static int
+nouveau_vm_link(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_vm_pgd *vpgd;
+	int i;
+
+	if (!pgd)
+		return 0;
+
+	vpgd = kzalloc(sizeof(*vpgd), GFP_KERNEL);
+	if (!vpgd)
+		return -ENOMEM;
+
+	nouveau_gpuobj_ref(pgd, &vpgd->obj);
+
+	mutex_lock(&vm->mm->mutex);
+	for (i = vm->fpde; i <= vm->lpde; i++) {
+		struct nouveau_vm_pgt *vpgt = &vm->pgt[i - vm->fpde];
+
+		if (!vpgt->obj) {
+			vm->unmap_pgt(pgd, i);
+			continue;
+		}
+
+		vm->map_pgt(pgd, vpgt->page_shift, i, vpgt->obj);
+	}
+	list_add(&vpgd->head, &vm->pgd_list);
+	mutex_unlock(&vm->mm->mutex);
+	return 0;
+}
+
+static void
+nouveau_vm_unlink(struct nouveau_vm *vm, struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_vm_pgd *vpgd, *tmp;
+
+	if (!pgd)
+		return;
+
+	mutex_lock(&vm->mm->mutex);
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		if (vpgd->obj != pgd)
+			continue;
+
+		list_del(&vpgd->head);
+		nouveau_gpuobj_ref(NULL, &vpgd->obj);
+		kfree(vpgd);
+	}
+	mutex_unlock(&vm->mm->mutex);
+}
+
+static void
+nouveau_vm_del(struct nouveau_vm *vm)
+{
+	struct nouveau_vm_pgd *vpgd, *tmp;
+
+	list_for_each_entry_safe(vpgd, tmp, &vm->pgd_list, head) {
+		nouveau_vm_unlink(vm, vpgd->obj);
+	}
+	WARN_ON(nouveau_mm_fini(&vm->mm) != 0);
+
+	kfree(vm->pgt);
+	kfree(vm);
+}
+
+int
+nouveau_vm_ref(struct nouveau_vm *ref, struct nouveau_vm **ptr,
+	       struct nouveau_gpuobj *pgd)
+{
+	struct nouveau_vm *vm;
+	int ret;
+
+	vm = ref;
+	if (vm) {
+		ret = nouveau_vm_link(vm, pgd);
+		if (ret)
+			return ret;
+
+		vm->refcount++;
+	}
+
+	vm = *ptr;
+	*ptr = ref;
+
+	if (vm) {
+		nouveau_vm_unlink(vm, pgd);
+
+		if (--vm->refcount == 0)
+			nouveau_vm_del(vm);
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nouveau_vm.h b/drivers/gpu/drm/nouveau/nouveau_vm.h
new file mode 100644
index 000000000000..b6755cfa7b71
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nouveau_vm.h
@@ -0,0 +1,107 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#ifndef __NOUVEAU_VM_H__
+#define __NOUVEAU_VM_H__
+
+#include "drmP.h"
+
+#include "nouveau_drv.h"
+#include "nouveau_mm.h"
+
+struct nouveau_vm_pgt {
+	struct nouveau_gpuobj *obj;
+	u32 page_shift;
+	u32 refcount;
+};
+
+struct nouveau_vm_pgd {
+	struct list_head head;
+	struct nouveau_gpuobj *obj;
+};
+
+struct nouveau_vma {
+	struct nouveau_vm *vm;
+	struct nouveau_mm_node *node;
+	u64 offset;
+	u32 access;
+};
+
+struct nouveau_vm {
+	struct drm_device *dev;
+	struct nouveau_mm *mm;
+	int refcount;
+
+	struct list_head pgd_list;
+	atomic_t pgraph_refs;
+	atomic_t pcrypt_refs;
+
+	struct nouveau_vm_pgt *pgt;
+	u32 fpde;
+	u32 lpde;
+
+	u32 pgt_bits;
+	u8  spg_shift;
+	u8  lpg_shift;
+
+	void (*map_pgt)(struct nouveau_gpuobj *pgd, u32 type, u32 pde,
+			struct nouveau_gpuobj *pgt);
+	void (*unmap_pgt)(struct nouveau_gpuobj *pgd, u32 pde);
+	void (*map)(struct nouveau_vma *, struct nouveau_gpuobj *,
+		    struct nouveau_vram *, u32 pte, u32 cnt, u64 phys);
+	void (*map_sg)(struct nouveau_vma *, struct nouveau_gpuobj *,
+		       u32 pte, dma_addr_t *, u32 cnt);
+	void (*unmap)(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt);
+	void (*flush)(struct nouveau_vm *);
+};
+
+/* nouveau_vm.c */
+int  nouveau_vm_new(struct drm_device *, u64 offset, u64 length, u64 mm_offset,
+		    u8 pgt_bits, u8 spg_shift, u8 lpg_shift,
+		    struct nouveau_vm **);
+int  nouveau_vm_ref(struct nouveau_vm *, struct nouveau_vm **,
+		    struct nouveau_gpuobj *pgd);
+int  nouveau_vm_get(struct nouveau_vm *, u64 size, u32 page_shift,
+		    u32 access, struct nouveau_vma *);
+void nouveau_vm_put(struct nouveau_vma *);
+void nouveau_vm_map(struct nouveau_vma *, struct nouveau_vram *);
+void nouveau_vm_map_at(struct nouveau_vma *, u64 offset, struct nouveau_vram *);
+void nouveau_vm_unmap(struct nouveau_vma *);
+void nouveau_vm_unmap_at(struct nouveau_vma *, u64 offset, u64 length);
+void nouveau_vm_map_sg(struct nouveau_vma *, u64 offset, u64 length,
+		       dma_addr_t *);
+
+/* nv50_vm.c */
+void nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 type, u32 pde,
+		     struct nouveau_gpuobj *pgt);
+void nv50_vm_unmap_pgt(struct nouveau_gpuobj *pgd, u32 pde);
+void nv50_vm_map(struct nouveau_vma *, struct nouveau_gpuobj *,
+		 struct nouveau_vram *, u32 pte, u32 cnt, u64 phys);
+void nv50_vm_map_sg(struct nouveau_vma *, struct nouveau_gpuobj *,
+		    u32 pte, dma_addr_t *, u32 cnt);
+void nv50_vm_unmap(struct nouveau_gpuobj *, u32 pte, u32 cnt);
+void nv50_vm_flush(struct nouveau_vm *);
+void nv50_vm_flush_engine(struct drm_device *, int engine);
+
+#endif
diff --git a/drivers/gpu/drm/nouveau/nv04_crtc.c b/drivers/gpu/drm/nouveau/nv04_crtc.c
index 40e180741629..297505eb98d5 100644
--- a/drivers/gpu/drm/nouveau/nv04_crtc.c
+++ b/drivers/gpu/drm/nouveau/nv04_crtc.c
@@ -551,7 +551,10 @@ nv_crtc_mode_set_regs(struct drm_crtc *crtc, struct drm_display_mode * mode)
 	if (dev_priv->card_type >= NV_30)
 		regp->gpio_ext = NVReadCRTC(dev, 0, NV_PCRTC_GPIO_EXT);
 
-	regp->crtc_cfg = NV_PCRTC_CONFIG_START_ADDRESS_HSYNC;
+	if (dev_priv->card_type >= NV_10)
+		regp->crtc_cfg = NV10_PCRTC_CONFIG_START_ADDRESS_HSYNC;
+	else
+		regp->crtc_cfg = NV04_PCRTC_CONFIG_START_ADDRESS_HSYNC;
 
 	/* Some misc regs */
 	if (dev_priv->card_type == NV_40) {
@@ -669,6 +672,7 @@ static void nv_crtc_prepare(struct drm_crtc *crtc)
 	if (nv_two_heads(dev))
 		NVSetOwner(dev, nv_crtc->index);
 
+	drm_vblank_pre_modeset(dev, nv_crtc->index);
 	funcs->dpms(crtc, DRM_MODE_DPMS_OFF);
 
 	NVBlankScreen(dev, nv_crtc->index, true);
@@ -701,6 +705,7 @@ static void nv_crtc_commit(struct drm_crtc *crtc)
 #endif
 
 	funcs->dpms(crtc, DRM_MODE_DPMS_ON);
+	drm_vblank_post_modeset(dev, nv_crtc->index);
 }
 
 static void nv_crtc_destroy(struct drm_crtc *crtc)
@@ -986,6 +991,7 @@ static const struct drm_crtc_funcs nv04_crtc_funcs = {
 	.cursor_move = nv04_crtc_cursor_move,
 	.gamma_set = nv_crtc_gamma_set,
 	.set_config = drm_crtc_helper_set_config,
+	.page_flip = nouveau_crtc_page_flip,
 	.destroy = nv_crtc_destroy,
 };
 
diff --git a/drivers/gpu/drm/nouveau/nv04_dac.c b/drivers/gpu/drm/nouveau/nv04_dac.c
index ba6423f2ffcc..e000455e06d0 100644
--- a/drivers/gpu/drm/nouveau/nv04_dac.c
+++ b/drivers/gpu/drm/nouveau/nv04_dac.c
@@ -74,14 +74,14 @@ static int sample_load_twice(struct drm_device *dev, bool sense[2])
 		 * use a 10ms timeout (guards against crtc being inactive, in
 		 * which case blank state would never change)
 		 */
-		if (!nouveau_wait_until(dev, 10000000, NV_PRMCIO_INP0__COLOR,
-					0x00000001, 0x00000000))
+		if (!nouveau_wait_eq(dev, 10000000, NV_PRMCIO_INP0__COLOR,
+				     0x00000001, 0x00000000))
 			return -EBUSY;
-		if (!nouveau_wait_until(dev, 10000000, NV_PRMCIO_INP0__COLOR,
-					0x00000001, 0x00000001))
+		if (!nouveau_wait_eq(dev, 10000000, NV_PRMCIO_INP0__COLOR,
+				     0x00000001, 0x00000001))
 			return -EBUSY;
-		if (!nouveau_wait_until(dev, 10000000, NV_PRMCIO_INP0__COLOR,
-					0x00000001, 0x00000000))
+		if (!nouveau_wait_eq(dev, 10000000, NV_PRMCIO_INP0__COLOR,
+				     0x00000001, 0x00000000))
 			return -EBUSY;
 
 		udelay(100);
diff --git a/drivers/gpu/drm/nouveau/nv04_display.c b/drivers/gpu/drm/nouveau/nv04_display.c
index 9e28cf772e3c..1715e1464b7d 100644
--- a/drivers/gpu/drm/nouveau/nv04_display.c
+++ b/drivers/gpu/drm/nouveau/nv04_display.c
@@ -32,6 +32,9 @@
 #include "nouveau_encoder.h"
 #include "nouveau_connector.h"
 
+static void nv04_vblank_crtc0_isr(struct drm_device *);
+static void nv04_vblank_crtc1_isr(struct drm_device *);
+
 static void
 nv04_display_store_initial_head_owner(struct drm_device *dev)
 {
@@ -197,6 +200,8 @@ nv04_display_create(struct drm_device *dev)
 		func->save(encoder);
 	}
 
+	nouveau_irq_register(dev, 24, nv04_vblank_crtc0_isr);
+	nouveau_irq_register(dev, 25, nv04_vblank_crtc1_isr);
 	return 0;
 }
 
@@ -208,6 +213,9 @@ nv04_display_destroy(struct drm_device *dev)
 
 	NV_DEBUG_KMS(dev, "\n");
 
+	nouveau_irq_unregister(dev, 24);
+	nouveau_irq_unregister(dev, 25);
+
 	/* Turn every CRTC off. */
 	list_for_each_entry(crtc, &dev->mode_config.crtc_list, head) {
 		struct drm_mode_set modeset = {
@@ -258,3 +266,16 @@ nv04_display_init(struct drm_device *dev)
 	return 0;
 }
 
+static void
+nv04_vblank_crtc0_isr(struct drm_device *dev)
+{
+	nv_wr32(dev, NV_CRTC0_INTSTAT, NV_CRTC_INTR_VBLANK);
+	drm_handle_vblank(dev, 0);
+}
+
+static void
+nv04_vblank_crtc1_isr(struct drm_device *dev)
+{
+	nv_wr32(dev, NV_CRTC1_INTSTAT, NV_CRTC_INTR_VBLANK);
+	drm_handle_vblank(dev, 1);
+}
diff --git a/drivers/gpu/drm/nouveau/nv04_fbcon.c b/drivers/gpu/drm/nouveau/nv04_fbcon.c
index 33e4c9388bc1..7a1189371096 100644
--- a/drivers/gpu/drm/nouveau/nv04_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv04_fbcon.c
@@ -28,52 +28,39 @@
 #include "nouveau_ramht.h"
 #include "nouveau_fbcon.h"
 
-void
+int
 nv04_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) && RING_SPACE(chan, 4)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
-
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_copyarea(info, region);
-		return;
-	}
+	ret = RING_SPACE(chan, 4);
+	if (ret)
+		return ret;
 
 	BEGIN_RING(chan, NvSubImageBlit, 0x0300, 3);
 	OUT_RING(chan, (region->sy << 16) | region->sx);
 	OUT_RING(chan, (region->dy << 16) | region->dx);
 	OUT_RING(chan, (region->height << 16) | region->width);
 	FIRE_RING(chan);
+	return 0;
 }
 
-void
+int
 nv04_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) && RING_SPACE(chan, 7)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
-
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_fillrect(info, rect);
-		return;
-	}
+	ret = RING_SPACE(chan, 7);
+	if (ret)
+		return ret;
 
 	BEGIN_RING(chan, NvSubGdiRect, 0x02fc, 1);
 	OUT_RING(chan, (rect->rop != ROP_COPY) ? 1 : 3);
@@ -87,9 +74,10 @@ nv04_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 	OUT_RING(chan, (rect->dx << 16) | rect->dy);
 	OUT_RING(chan, (rect->width << 16) | rect->height);
 	FIRE_RING(chan);
+	return 0;
 }
 
-void
+int
 nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
@@ -101,23 +89,14 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	uint32_t dsize;
 	uint32_t width;
 	uint32_t *data = (uint32_t *)image->data;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (image->depth != 1) {
-		cfb_imageblit(info, image);
-		return;
-	}
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) && RING_SPACE(chan, 8)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
+	if (image->depth != 1)
+		return -ENODEV;
 
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_imageblit(info, image);
-		return;
-	}
+	ret = RING_SPACE(chan, 8);
+	if (ret)
+		return ret;
 
 	width = ALIGN(image->width, 8);
 	dsize = ALIGN(width * image->height, 32) >> 5;
@@ -144,11 +123,9 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	while (dsize) {
 		int iter_len = dsize > 128 ? 128 : dsize;
 
-		if (RING_SPACE(chan, iter_len + 1)) {
-			nouveau_fbcon_gpu_lockup(info);
-			cfb_imageblit(info, image);
-			return;
-		}
+		ret = RING_SPACE(chan, iter_len + 1);
+		if (ret)
+			return ret;
 
 		BEGIN_RING(chan, NvSubGdiRect, 0x0c00, iter_len);
 		OUT_RINGp(chan, data, iter_len);
@@ -157,22 +134,7 @@ nv04_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	}
 
 	FIRE_RING(chan);
-}
-
-static int
-nv04_fbcon_grobj_new(struct drm_device *dev, int class, uint32_t handle)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *obj = NULL;
-	int ret;
-
-	ret = nouveau_gpuobj_gr_new(dev_priv->channel, class, &obj);
-	if (ret)
-		return ret;
-
-	ret = nouveau_ramht_insert(dev_priv->channel, handle, obj);
-	nouveau_gpuobj_ref(NULL, &obj);
-	return ret;
+	return 0;
 }
 
 int
@@ -214,29 +176,31 @@ nv04_fbcon_accel_init(struct fb_info *info)
 		return -EINVAL;
 	}
 
-	ret = nv04_fbcon_grobj_new(dev, dev_priv->card_type >= NV_10 ?
-				   0x0062 : 0x0042, NvCtxSurf2D);
+	ret = nouveau_gpuobj_gr_new(chan, NvCtxSurf2D,
+				    dev_priv->card_type >= NV_10 ?
+				    0x0062 : 0x0042);
 	if (ret)
 		return ret;
 
-	ret = nv04_fbcon_grobj_new(dev, 0x0019, NvClipRect);
+	ret = nouveau_gpuobj_gr_new(chan, NvClipRect, 0x0019);
 	if (ret)
 		return ret;
 
-	ret = nv04_fbcon_grobj_new(dev, 0x0043, NvRop);
+	ret = nouveau_gpuobj_gr_new(chan, NvRop, 0x0043);
 	if (ret)
 		return ret;
 
-	ret = nv04_fbcon_grobj_new(dev, 0x0044, NvImagePatt);
+	ret = nouveau_gpuobj_gr_new(chan, NvImagePatt, 0x0044);
 	if (ret)
 		return ret;
 
-	ret = nv04_fbcon_grobj_new(dev, 0x004a, NvGdiRect);
+	ret = nouveau_gpuobj_gr_new(chan, NvGdiRect, 0x004a);
 	if (ret)
 		return ret;
 
-	ret = nv04_fbcon_grobj_new(dev, dev_priv->chipset >= 0x11 ?
-				   0x009f : 0x005f, NvImageBlit);
+	ret = nouveau_gpuobj_gr_new(chan, NvImageBlit,
+				    dev_priv->chipset >= 0x11 ?
+				    0x009f : 0x005f);
 	if (ret)
 		return ret;
 
diff --git a/drivers/gpu/drm/nouveau/nv04_fifo.c b/drivers/gpu/drm/nouveau/nv04_fifo.c
index 708293b7ddcd..f89d104698df 100644
--- a/drivers/gpu/drm/nouveau/nv04_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv04_fifo.c
@@ -28,6 +28,7 @@
 #include "drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_ramht.h"
+#include "nouveau_util.h"
 
 #define NV04_RAMFC(c) (dev_priv->ramfc->pinst + ((c) * NV04_RAMFC__SIZE))
 #define NV04_RAMFC__SIZE 32
@@ -128,6 +129,11 @@ nv04_fifo_create_context(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
+	chan->user = ioremap(pci_resource_start(dev->pdev, 0) +
+			     NV03_USER(chan->id), PAGE_SIZE);
+	if (!chan->user)
+		return -ENOMEM;
+
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 
 	/* Setup initial state */
@@ -151,10 +157,31 @@ void
 nv04_fifo_destroy_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	unsigned long flags;
 
-	nv_wr32(dev, NV04_PFIFO_MODE,
-		nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pfifo->reassign(dev, false);
 
+	/* Unload the context if it's the currently active one */
+	if (pfifo->channel_id(dev) == chan->id) {
+		pfifo->disable(dev);
+		pfifo->unload_context(dev);
+		pfifo->enable(dev);
+	}
+
+	/* Keep it from being rescheduled */
+	nv_mask(dev, NV04_PFIFO_MODE, 1 << chan->id, 0);
+
+	pfifo->reassign(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the channel resources */
+	if (chan->user) {
+		iounmap(chan->user);
+		chan->user = NULL;
+	}
 	nouveau_gpuobj_ref(NULL, &chan->ramfc);
 }
 
@@ -208,7 +235,7 @@ nv04_fifo_unload_context(struct drm_device *dev)
 	if (chid < 0 || chid >= dev_priv->engine.fifo.channels)
 		return 0;
 
-	chan = dev_priv->fifos[chid];
+	chan = dev_priv->channels.ptr[chid];
 	if (!chan) {
 		NV_ERROR(dev, "Inactive channel on PFIFO: %d\n", chid);
 		return -EINVAL;
@@ -267,6 +294,7 @@ nv04_fifo_init_ramxx(struct drm_device *dev)
 static void
 nv04_fifo_init_intr(struct drm_device *dev)
 {
+	nouveau_irq_register(dev, 8, nv04_fifo_isr);
 	nv_wr32(dev, 0x002100, 0xffffffff);
 	nv_wr32(dev, 0x002140, 0xffffffff);
 }
@@ -289,7 +317,7 @@ nv04_fifo_init(struct drm_device *dev)
 	pfifo->reassign(dev, true);
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		if (dev_priv->fifos[i]) {
+		if (dev_priv->channels.ptr[i]) {
 			uint32_t mode = nv_rd32(dev, NV04_PFIFO_MODE);
 			nv_wr32(dev, NV04_PFIFO_MODE, mode | (1 << i));
 		}
@@ -298,3 +326,207 @@ nv04_fifo_init(struct drm_device *dev)
 	return 0;
 }
 
+void
+nv04_fifo_fini(struct drm_device *dev)
+{
+	nv_wr32(dev, 0x2140, 0x00000000);
+	nouveau_irq_unregister(dev, 8);
+}
+
+static bool
+nouveau_fifo_swmthd(struct drm_device *dev, u32 chid, u32 addr, u32 data)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan = NULL;
+	struct nouveau_gpuobj *obj;
+	unsigned long flags;
+	const int subc = (addr >> 13) & 0x7;
+	const int mthd = addr & 0x1ffc;
+	bool handled = false;
+	u32 engine;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	if (likely(chid >= 0 && chid < dev_priv->engine.fifo.channels))
+		chan = dev_priv->channels.ptr[chid];
+	if (unlikely(!chan))
+		goto out;
+
+	switch (mthd) {
+	case 0x0000: /* bind object to subchannel */
+		obj = nouveau_ramht_find(chan, data);
+		if (unlikely(!obj || obj->engine != NVOBJ_ENGINE_SW))
+			break;
+
+		chan->sw_subchannel[subc] = obj->class;
+		engine = 0x0000000f << (subc * 4);
+
+		nv_mask(dev, NV04_PFIFO_CACHE1_ENGINE, engine, 0x00000000);
+		handled = true;
+		break;
+	default:
+		engine = nv_rd32(dev, NV04_PFIFO_CACHE1_ENGINE);
+		if (unlikely(((engine >> (subc * 4)) & 0xf) != 0))
+			break;
+
+		if (!nouveau_gpuobj_mthd_call(chan, chan->sw_subchannel[subc],
+					      mthd, data))
+			handled = true;
+		break;
+	}
+
+out:
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+	return handled;
+}
+
+void
+nv04_fifo_isr(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_engine *engine = &dev_priv->engine;
+	uint32_t status, reassign;
+	int cnt = 0;
+
+	reassign = nv_rd32(dev, NV03_PFIFO_CACHES) & 1;
+	while ((status = nv_rd32(dev, NV03_PFIFO_INTR_0)) && (cnt++ < 100)) {
+		uint32_t chid, get;
+
+		nv_wr32(dev, NV03_PFIFO_CACHES, 0);
+
+		chid = engine->fifo.channel_id(dev);
+		get  = nv_rd32(dev, NV03_PFIFO_CACHE1_GET);
+
+		if (status & NV_PFIFO_INTR_CACHE_ERROR) {
+			uint32_t mthd, data;
+			int ptr;
+
+			/* NV_PFIFO_CACHE1_GET actually goes to 0xffc before
+			 * wrapping on my G80 chips, but CACHE1 isn't big
+			 * enough for this much data.. Tests show that it
+			 * wraps around to the start at GET=0x800.. No clue
+			 * as to why..
+			 */
+			ptr = (get & 0x7ff) >> 2;
+
+			if (dev_priv->card_type < NV_40) {
+				mthd = nv_rd32(dev,
+					NV04_PFIFO_CACHE1_METHOD(ptr));
+				data = nv_rd32(dev,
+					NV04_PFIFO_CACHE1_DATA(ptr));
+			} else {
+				mthd = nv_rd32(dev,
+					NV40_PFIFO_CACHE1_METHOD(ptr));
+				data = nv_rd32(dev,
+					NV40_PFIFO_CACHE1_DATA(ptr));
+			}
+
+			if (!nouveau_fifo_swmthd(dev, chid, mthd, data)) {
+				NV_INFO(dev, "PFIFO_CACHE_ERROR - Ch %d/%d "
+					     "Mthd 0x%04x Data 0x%08x\n",
+					chid, (mthd >> 13) & 7, mthd & 0x1ffc,
+					data);
+			}
+
+			nv_wr32(dev, NV04_PFIFO_CACHE1_DMA_PUSH, 0);
+			nv_wr32(dev, NV03_PFIFO_INTR_0,
+						NV_PFIFO_INTR_CACHE_ERROR);
+
+			nv_wr32(dev, NV03_PFIFO_CACHE1_PUSH0,
+				nv_rd32(dev, NV03_PFIFO_CACHE1_PUSH0) & ~1);
+			nv_wr32(dev, NV03_PFIFO_CACHE1_GET, get + 4);
+			nv_wr32(dev, NV03_PFIFO_CACHE1_PUSH0,
+				nv_rd32(dev, NV03_PFIFO_CACHE1_PUSH0) | 1);
+			nv_wr32(dev, NV04_PFIFO_CACHE1_HASH, 0);
+
+			nv_wr32(dev, NV04_PFIFO_CACHE1_DMA_PUSH,
+				nv_rd32(dev, NV04_PFIFO_CACHE1_DMA_PUSH) | 1);
+			nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, 1);
+
+			status &= ~NV_PFIFO_INTR_CACHE_ERROR;
+		}
+
+		if (status & NV_PFIFO_INTR_DMA_PUSHER) {
+			u32 dma_get = nv_rd32(dev, 0x003244);
+			u32 dma_put = nv_rd32(dev, 0x003240);
+			u32 push = nv_rd32(dev, 0x003220);
+			u32 state = nv_rd32(dev, 0x003228);
+
+			if (dev_priv->card_type == NV_50) {
+				u32 ho_get = nv_rd32(dev, 0x003328);
+				u32 ho_put = nv_rd32(dev, 0x003320);
+				u32 ib_get = nv_rd32(dev, 0x003334);
+				u32 ib_put = nv_rd32(dev, 0x003330);
+
+				if (nouveau_ratelimit())
+					NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%02x%08x "
+					     "Put 0x%02x%08x IbGet 0x%08x IbPut 0x%08x "
+					     "State 0x%08x Push 0x%08x\n",
+						chid, ho_get, dma_get, ho_put,
+						dma_put, ib_get, ib_put, state,
+						push);
+
+				/* METHOD_COUNT, in DMA_STATE on earlier chipsets */
+				nv_wr32(dev, 0x003364, 0x00000000);
+				if (dma_get != dma_put || ho_get != ho_put) {
+					nv_wr32(dev, 0x003244, dma_put);
+					nv_wr32(dev, 0x003328, ho_put);
+				} else
+				if (ib_get != ib_put) {
+					nv_wr32(dev, 0x003334, ib_put);
+				}
+			} else {
+				NV_INFO(dev, "PFIFO_DMA_PUSHER - Ch %d Get 0x%08x "
+					     "Put 0x%08x State 0x%08x Push 0x%08x\n",
+					chid, dma_get, dma_put, state, push);
+
+				if (dma_get != dma_put)
+					nv_wr32(dev, 0x003244, dma_put);
+			}
+
+			nv_wr32(dev, 0x003228, 0x00000000);
+			nv_wr32(dev, 0x003220, 0x00000001);
+			nv_wr32(dev, 0x002100, NV_PFIFO_INTR_DMA_PUSHER);
+			status &= ~NV_PFIFO_INTR_DMA_PUSHER;
+		}
+
+		if (status & NV_PFIFO_INTR_SEMAPHORE) {
+			uint32_t sem;
+
+			status &= ~NV_PFIFO_INTR_SEMAPHORE;
+			nv_wr32(dev, NV03_PFIFO_INTR_0,
+				NV_PFIFO_INTR_SEMAPHORE);
+
+			sem = nv_rd32(dev, NV10_PFIFO_CACHE1_SEMAPHORE);
+			nv_wr32(dev, NV10_PFIFO_CACHE1_SEMAPHORE, sem | 0x1);
+
+			nv_wr32(dev, NV03_PFIFO_CACHE1_GET, get + 4);
+			nv_wr32(dev, NV04_PFIFO_CACHE1_PULL0, 1);
+		}
+
+		if (dev_priv->card_type == NV_50) {
+			if (status & 0x00000010) {
+				nv50_fb_vm_trap(dev, 1, "PFIFO_BAR_FAULT");
+				status &= ~0x00000010;
+				nv_wr32(dev, 0x002100, 0x00000010);
+			}
+		}
+
+		if (status) {
+			if (nouveau_ratelimit())
+				NV_INFO(dev, "PFIFO_INTR 0x%08x - Ch %d\n",
+					status, chid);
+			nv_wr32(dev, NV03_PFIFO_INTR_0, status);
+			status = 0;
+		}
+
+		nv_wr32(dev, NV03_PFIFO_CACHES, reassign);
+	}
+
+	if (status) {
+		NV_INFO(dev, "PFIFO still angry after %d spins, halt\n", cnt);
+		nv_wr32(dev, 0x2140, 0);
+		nv_wr32(dev, 0x140, 0);
+	}
+
+	nv_wr32(dev, NV03_PMC_INTR_0, NV_PMC_INTR_0_PFIFO_PENDING);
+}
diff --git a/drivers/gpu/drm/nouveau/nv04_graph.c b/drivers/gpu/drm/nouveau/nv04_graph.c
index c8973421b635..af75015068d6 100644
--- a/drivers/gpu/drm/nouveau/nv04_graph.c
+++ b/drivers/gpu/drm/nouveau/nv04_graph.c
@@ -26,6 +26,11 @@
 #include "drm.h"
 #include "nouveau_drm.h"
 #include "nouveau_drv.h"
+#include "nouveau_hw.h"
+#include "nouveau_util.h"
+
+static int  nv04_graph_register(struct drm_device *dev);
+static void nv04_graph_isr(struct drm_device *dev);
 
 static uint32_t nv04_graph_ctx_regs[] = {
 	0x0040053c,
@@ -357,10 +362,10 @@ nv04_graph_channel(struct drm_device *dev)
 	if (chid >= dev_priv->engine.fifo.channels)
 		return NULL;
 
-	return dev_priv->fifos[chid];
+	return dev_priv->channels.ptr[chid];
 }
 
-void
+static void
 nv04_graph_context_switch(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -368,7 +373,6 @@ nv04_graph_context_switch(struct drm_device *dev)
 	struct nouveau_channel *chan = NULL;
 	int chid;
 
-	pgraph->fifo_access(dev, false);
 	nouveau_wait_for_idle(dev);
 
 	/* If previous context is valid, we need to save it */
@@ -376,11 +380,9 @@ nv04_graph_context_switch(struct drm_device *dev)
 
 	/* Load context for next channel */
 	chid = dev_priv->engine.fifo.channel_id(dev);
-	chan = dev_priv->fifos[chid];
+	chan = dev_priv->channels.ptr[chid];
 	if (chan)
 		nv04_graph_load_context(chan);
-
-	pgraph->fifo_access(dev, true);
 }
 
 static uint32_t *ctx_reg(struct graph_state *ctx, uint32_t reg)
@@ -412,10 +414,25 @@ int nv04_graph_create_context(struct nouveau_channel *chan)
 
 void nv04_graph_destroy_context(struct nouveau_channel *chan)
 {
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pgraph->fifo_access(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (pgraph->channel(dev) == chan)
+		pgraph->unload_context(dev);
 
+	/* Free the context resources */
 	kfree(pgraph_ctx);
 	chan->pgraph_ctx = NULL;
+
+	pgraph->fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
 
 int nv04_graph_load_context(struct nouveau_channel *chan)
@@ -468,13 +485,19 @@ int nv04_graph_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	uint32_t tmp;
+	int ret;
 
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |
 			 NV_PMC_ENABLE_PGRAPH);
 
+	ret = nv04_graph_register(dev);
+	if (ret)
+		return ret;
+
 	/* Enable PGRAPH interrupts */
+	nouveau_irq_register(dev, 12, nv04_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR, 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -510,6 +533,8 @@ int nv04_graph_init(struct drm_device *dev)
 
 void nv04_graph_takedown(struct drm_device *dev)
 {
+	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
+	nouveau_irq_unregister(dev, 12);
 }
 
 void
@@ -524,13 +549,27 @@ nv04_graph_fifo_access(struct drm_device *dev, bool enabled)
 }
 
 static int
-nv04_graph_mthd_set_ref(struct nouveau_channel *chan, int grclass,
-			int mthd, uint32_t data)
+nv04_graph_mthd_set_ref(struct nouveau_channel *chan,
+			u32 class, u32 mthd, u32 data)
 {
 	atomic_set(&chan->fence.last_sequence_irq, data);
 	return 0;
 }
 
+int
+nv04_graph_mthd_page_flip(struct nouveau_channel *chan,
+			  u32 class, u32 mthd, u32 data)
+{
+	struct drm_device *dev = chan->dev;
+	struct nouveau_page_flip_state s;
+
+	if (!nouveau_finish_page_flip(chan, &s))
+		nv_set_crtc_base(dev, s.crtc,
+				 s.offset + s.y * s.pitch + s.x * s.bpp / 8);
+
+	return 0;
+}
+
 /*
  * Software methods, why they are needed, and how they all work:
  *
@@ -606,12 +645,12 @@ nv04_graph_mthd_set_ref(struct nouveau_channel *chan, int grclass,
  */
 
 static void
-nv04_graph_set_ctx1(struct nouveau_channel *chan, uint32_t mask, uint32_t value)
+nv04_graph_set_ctx1(struct nouveau_channel *chan, u32 mask, u32 value)
 {
 	struct drm_device *dev = chan->dev;
-	uint32_t instance = (nv_rd32(dev, NV04_PGRAPH_CTX_SWITCH4) & 0xffff) << 4;
+	u32 instance = (nv_rd32(dev, NV04_PGRAPH_CTX_SWITCH4) & 0xffff) << 4;
 	int subc = (nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR) >> 13) & 0x7;
-	uint32_t tmp;
+	u32 tmp;
 
 	tmp  = nv_ri32(dev, instance);
 	tmp &= ~mask;
@@ -623,11 +662,11 @@ nv04_graph_set_ctx1(struct nouveau_channel *chan, uint32_t mask, uint32_t value)
 }
 
 static void
-nv04_graph_set_ctx_val(struct nouveau_channel *chan, uint32_t mask, uint32_t value)
+nv04_graph_set_ctx_val(struct nouveau_channel *chan, u32 mask, u32 value)
 {
 	struct drm_device *dev = chan->dev;
-	uint32_t instance = (nv_rd32(dev, NV04_PGRAPH_CTX_SWITCH4) & 0xffff) << 4;
-	uint32_t tmp, ctx1;
+	u32 instance = (nv_rd32(dev, NV04_PGRAPH_CTX_SWITCH4) & 0xffff) << 4;
+	u32 tmp, ctx1;
 	int class, op, valid = 1;
 
 	ctx1 = nv_ri32(dev, instance);
@@ -672,13 +711,13 @@ nv04_graph_set_ctx_val(struct nouveau_channel *chan, uint32_t mask, uint32_t val
 }
 
 static int
-nv04_graph_mthd_set_operation(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_set_operation(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	if (data > 5)
 		return 1;
 	/* Old versions of the objects only accept first three operations. */
-	if (data > 2 && grclass < 0x40)
+	if (data > 2 && class < 0x40)
 		return 1;
 	nv04_graph_set_ctx1(chan, 0x00038000, data << 15);
 	/* changing operation changes set of objects needed for validation */
@@ -687,8 +726,8 @@ nv04_graph_mthd_set_operation(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_surf3d_clip_h(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_surf3d_clip_h(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	uint32_t min = data & 0xffff, max;
 	uint32_t w = data >> 16;
@@ -706,8 +745,8 @@ nv04_graph_mthd_surf3d_clip_h(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_surf3d_clip_v(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_surf3d_clip_v(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	uint32_t min = data & 0xffff, max;
 	uint32_t w = data >> 16;
@@ -725,8 +764,8 @@ nv04_graph_mthd_surf3d_clip_v(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf2d(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf2d(struct nouveau_channel *chan,
+			    u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -742,8 +781,8 @@ nv04_graph_mthd_bind_surf2d(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf2d_swzsurf(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf2d_swzsurf(struct nouveau_channel *chan,
+				    u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -763,8 +802,8 @@ nv04_graph_mthd_bind_surf2d_swzsurf(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_nv01_patt(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_nv01_patt(struct nouveau_channel *chan,
+			       u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -778,8 +817,8 @@ nv04_graph_mthd_bind_nv01_patt(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_nv04_patt(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_nv04_patt(struct nouveau_channel *chan,
+			       u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -793,8 +832,8 @@ nv04_graph_mthd_bind_nv04_patt(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_rop(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_rop(struct nouveau_channel *chan,
+			 u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -808,8 +847,8 @@ nv04_graph_mthd_bind_rop(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_beta1(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_beta1(struct nouveau_channel *chan,
+			   u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -823,8 +862,8 @@ nv04_graph_mthd_bind_beta1(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_beta4(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_beta4(struct nouveau_channel *chan,
+			   u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -838,8 +877,8 @@ nv04_graph_mthd_bind_beta4(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf_dst(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf_dst(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -853,8 +892,8 @@ nv04_graph_mthd_bind_surf_dst(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf_src(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf_src(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -868,8 +907,8 @@ nv04_graph_mthd_bind_surf_src(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf_color(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf_color(struct nouveau_channel *chan,
+				u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -883,8 +922,8 @@ nv04_graph_mthd_bind_surf_color(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_surf_zeta(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_surf_zeta(struct nouveau_channel *chan,
+			       u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -898,8 +937,8 @@ nv04_graph_mthd_bind_surf_zeta(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_clip(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_clip(struct nouveau_channel *chan,
+			  u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -913,8 +952,8 @@ nv04_graph_mthd_bind_clip(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv04_graph_mthd_bind_chroma(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv04_graph_mthd_bind_chroma(struct nouveau_channel *chan,
+			    u32 class, u32 mthd, u32 data)
 {
 	switch (nv_ri32(chan->dev, data << 4) & 0xff) {
 	case 0x30:
@@ -930,194 +969,346 @@ nv04_graph_mthd_bind_chroma(struct nouveau_channel *chan, int grclass,
 	return 1;
 }
 
-static struct nouveau_pgraph_object_method nv04_graph_mthds_sw[] = {
-	{ 0x0150, nv04_graph_mthd_set_ref },
-	{}
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv03_gdirect[] = {
-	{ 0x0184, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x0188, nv04_graph_mthd_bind_rop },
-	{ 0x018c, nv04_graph_mthd_bind_beta1 },
-	{ 0x0190, nv04_graph_mthd_bind_surf_dst },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_gdirect[] = {
-	{ 0x0188, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_beta4 },
-	{ 0x0198, nv04_graph_mthd_bind_surf2d },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv01_imageblit[] = {
-	{ 0x0184, nv04_graph_mthd_bind_chroma },
-	{ 0x0188, nv04_graph_mthd_bind_clip },
-	{ 0x018c, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x0190, nv04_graph_mthd_bind_rop },
-	{ 0x0194, nv04_graph_mthd_bind_beta1 },
-	{ 0x0198, nv04_graph_mthd_bind_surf_dst },
-	{ 0x019c, nv04_graph_mthd_bind_surf_src },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_imageblit_ifc[] = {
-	{ 0x0184, nv04_graph_mthd_bind_chroma },
-	{ 0x0188, nv04_graph_mthd_bind_clip },
-	{ 0x018c, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x0190, nv04_graph_mthd_bind_rop },
-	{ 0x0194, nv04_graph_mthd_bind_beta1 },
-	{ 0x0198, nv04_graph_mthd_bind_beta4 },
-	{ 0x019c, nv04_graph_mthd_bind_surf2d },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_iifc[] = {
-	{ 0x0188, nv04_graph_mthd_bind_chroma },
-	{ 0x018c, nv04_graph_mthd_bind_clip },
-	{ 0x0190, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x0194, nv04_graph_mthd_bind_rop },
-	{ 0x0198, nv04_graph_mthd_bind_beta1 },
-	{ 0x019c, nv04_graph_mthd_bind_beta4 },
-	{ 0x01a0, nv04_graph_mthd_bind_surf2d_swzsurf },
-	{ 0x03e4, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv01_ifc[] = {
-	{ 0x0184, nv04_graph_mthd_bind_chroma },
-	{ 0x0188, nv04_graph_mthd_bind_clip },
-	{ 0x018c, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x0190, nv04_graph_mthd_bind_rop },
-	{ 0x0194, nv04_graph_mthd_bind_beta1 },
-	{ 0x0198, nv04_graph_mthd_bind_surf_dst },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv03_sifc[] = {
-	{ 0x0184, nv04_graph_mthd_bind_chroma },
-	{ 0x0188, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_surf_dst },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_sifc[] = {
-	{ 0x0184, nv04_graph_mthd_bind_chroma },
-	{ 0x0188, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_beta4 },
-	{ 0x0198, nv04_graph_mthd_bind_surf2d },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv03_sifm[] = {
-	{ 0x0188, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_surf_dst },
-	{ 0x0304, nv04_graph_mthd_set_operation },
-	{},
-};
-
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_sifm[] = {
-	{ 0x0188, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_beta4 },
-	{ 0x0198, nv04_graph_mthd_bind_surf2d_swzsurf },
-	{ 0x0304, nv04_graph_mthd_set_operation },
-	{},
-};
+static int
+nv04_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv01_shape[] = {
-	{ 0x0184, nv04_graph_mthd_bind_clip },
-	{ 0x0188, nv04_graph_mthd_bind_nv01_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_surf_dst },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
-};
+	if (dev_priv->engine.graph.registered)
+		return 0;
 
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv04_shape[] = {
-	{ 0x0184, nv04_graph_mthd_bind_clip },
-	{ 0x0188, nv04_graph_mthd_bind_nv04_patt },
-	{ 0x018c, nv04_graph_mthd_bind_rop },
-	{ 0x0190, nv04_graph_mthd_bind_beta1 },
-	{ 0x0194, nv04_graph_mthd_bind_beta4 },
-	{ 0x0198, nv04_graph_mthd_bind_surf2d },
-	{ 0x02fc, nv04_graph_mthd_set_operation },
-	{},
+	/* dvd subpicture */
+	NVOBJ_CLASS(dev, 0x0038, GR);
+
+	/* m2mf */
+	NVOBJ_CLASS(dev, 0x0039, GR);
+
+	/* nv03 gdirect */
+	NVOBJ_CLASS(dev, 0x004b, GR);
+	NVOBJ_MTHD (dev, 0x004b, 0x0184, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x004b, 0x0188, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x004b, 0x018c, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x004b, 0x0190, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x004b, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 gdirect */
+	NVOBJ_CLASS(dev, 0x004a, GR);
+	NVOBJ_MTHD (dev, 0x004a, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x004a, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x004a, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x004a, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x004a, 0x0198, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x004a, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv01 imageblit */
+	NVOBJ_CLASS(dev, 0x001f, GR);
+	NVOBJ_MTHD (dev, 0x001f, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x001f, 0x0188, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x001f, 0x018c, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x001f, 0x0190, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x001f, 0x0194, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x001f, 0x0198, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x001f, 0x019c, nv04_graph_mthd_bind_surf_src);
+	NVOBJ_MTHD (dev, 0x001f, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 imageblit */
+	NVOBJ_CLASS(dev, 0x005f, GR);
+	NVOBJ_MTHD (dev, 0x005f, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x005f, 0x0188, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x005f, 0x018c, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x005f, 0x0190, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x005f, 0x0194, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x005f, 0x0198, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x005f, 0x019c, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x005f, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 iifc */
+	NVOBJ_CLASS(dev, 0x0060, GR);
+	NVOBJ_MTHD (dev, 0x0060, 0x0188, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x0060, 0x018c, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x0060, 0x0190, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x0060, 0x0194, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0060, 0x0198, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0060, 0x019c, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x0060, 0x01a0, nv04_graph_mthd_bind_surf2d_swzsurf);
+	NVOBJ_MTHD (dev, 0x0060, 0x03e4, nv04_graph_mthd_set_operation);
+
+	/* nv05 iifc */
+	NVOBJ_CLASS(dev, 0x0064, GR);
+
+	/* nv01 ifc */
+	NVOBJ_CLASS(dev, 0x0021, GR);
+	NVOBJ_MTHD (dev, 0x0021, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x0021, 0x0188, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x0021, 0x018c, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x0021, 0x0190, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0021, 0x0194, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0021, 0x0198, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x0021, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 ifc */
+	NVOBJ_CLASS(dev, 0x0061, GR);
+	NVOBJ_MTHD (dev, 0x0061, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x0061, 0x0188, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x0061, 0x018c, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x0061, 0x0190, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0061, 0x0194, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0061, 0x0198, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x0061, 0x019c, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x0061, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv05 ifc */
+	NVOBJ_CLASS(dev, 0x0065, GR);
+
+	/* nv03 sifc */
+	NVOBJ_CLASS(dev, 0x0036, GR);
+	NVOBJ_MTHD (dev, 0x0036, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x0036, 0x0188, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x0036, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0036, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0036, 0x0194, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x0036, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 sifc */
+	NVOBJ_CLASS(dev, 0x0076, GR);
+	NVOBJ_MTHD (dev, 0x0076, 0x0184, nv04_graph_mthd_bind_chroma);
+	NVOBJ_MTHD (dev, 0x0076, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x0076, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0076, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0076, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x0076, 0x0198, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x0076, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv05 sifc */
+	NVOBJ_CLASS(dev, 0x0066, GR);
+
+	/* nv03 sifm */
+	NVOBJ_CLASS(dev, 0x0037, GR);
+	NVOBJ_MTHD (dev, 0x0037, 0x0188, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x0037, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0037, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0037, 0x0194, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x0037, 0x0304, nv04_graph_mthd_set_operation);
+
+	/* nv04 sifm */
+	NVOBJ_CLASS(dev, 0x0077, GR);
+	NVOBJ_MTHD (dev, 0x0077, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x0077, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x0077, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x0077, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x0077, 0x0198, nv04_graph_mthd_bind_surf2d_swzsurf);
+	NVOBJ_MTHD (dev, 0x0077, 0x0304, nv04_graph_mthd_set_operation);
+
+	/* null */
+	NVOBJ_CLASS(dev, 0x0030, GR);
+
+	/* surf2d */
+	NVOBJ_CLASS(dev, 0x0042, GR);
+
+	/* rop */
+	NVOBJ_CLASS(dev, 0x0043, GR);
+
+	/* beta1 */
+	NVOBJ_CLASS(dev, 0x0012, GR);
+
+	/* beta4 */
+	NVOBJ_CLASS(dev, 0x0072, GR);
+
+	/* cliprect */
+	NVOBJ_CLASS(dev, 0x0019, GR);
+
+	/* nv01 pattern */
+	NVOBJ_CLASS(dev, 0x0018, GR);
+
+	/* nv04 pattern */
+	NVOBJ_CLASS(dev, 0x0044, GR);
+
+	/* swzsurf */
+	NVOBJ_CLASS(dev, 0x0052, GR);
+
+	/* surf3d */
+	NVOBJ_CLASS(dev, 0x0053, GR);
+	NVOBJ_MTHD (dev, 0x0053, 0x02f8, nv04_graph_mthd_surf3d_clip_h);
+	NVOBJ_MTHD (dev, 0x0053, 0x02fc, nv04_graph_mthd_surf3d_clip_v);
+
+	/* nv03 tex_tri */
+	NVOBJ_CLASS(dev, 0x0048, GR);
+	NVOBJ_MTHD (dev, 0x0048, 0x0188, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x0048, 0x018c, nv04_graph_mthd_bind_surf_color);
+	NVOBJ_MTHD (dev, 0x0048, 0x0190, nv04_graph_mthd_bind_surf_zeta);
+
+	/* tex_tri */
+	NVOBJ_CLASS(dev, 0x0054, GR);
+
+	/* multitex_tri */
+	NVOBJ_CLASS(dev, 0x0055, GR);
+
+	/* nv01 chroma */
+	NVOBJ_CLASS(dev, 0x0017, GR);
+
+	/* nv04 chroma */
+	NVOBJ_CLASS(dev, 0x0057, GR);
+
+	/* surf_dst */
+	NVOBJ_CLASS(dev, 0x0058, GR);
+
+	/* surf_src */
+	NVOBJ_CLASS(dev, 0x0059, GR);
+
+	/* surf_color */
+	NVOBJ_CLASS(dev, 0x005a, GR);
+
+	/* surf_zeta */
+	NVOBJ_CLASS(dev, 0x005b, GR);
+
+	/* nv01 line */
+	NVOBJ_CLASS(dev, 0x001c, GR);
+	NVOBJ_MTHD (dev, 0x001c, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x001c, 0x0188, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x001c, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x001c, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x001c, 0x0194, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x001c, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 line */
+	NVOBJ_CLASS(dev, 0x005c, GR);
+	NVOBJ_MTHD (dev, 0x005c, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x005c, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x005c, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x005c, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x005c, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x005c, 0x0198, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x005c, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv01 tri */
+	NVOBJ_CLASS(dev, 0x001d, GR);
+	NVOBJ_MTHD (dev, 0x001d, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x001d, 0x0188, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x001d, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x001d, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x001d, 0x0194, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x001d, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 tri */
+	NVOBJ_CLASS(dev, 0x005d, GR);
+	NVOBJ_MTHD (dev, 0x005d, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x005d, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x005d, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x005d, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x005d, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x005d, 0x0198, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x005d, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv01 rect */
+	NVOBJ_CLASS(dev, 0x001e, GR);
+	NVOBJ_MTHD (dev, 0x001e, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x001e, 0x0188, nv04_graph_mthd_bind_nv01_patt);
+	NVOBJ_MTHD (dev, 0x001e, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x001e, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x001e, 0x0194, nv04_graph_mthd_bind_surf_dst);
+	NVOBJ_MTHD (dev, 0x001e, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nv04 rect */
+	NVOBJ_CLASS(dev, 0x005e, GR);
+	NVOBJ_MTHD (dev, 0x005e, 0x0184, nv04_graph_mthd_bind_clip);
+	NVOBJ_MTHD (dev, 0x005e, 0x0188, nv04_graph_mthd_bind_nv04_patt);
+	NVOBJ_MTHD (dev, 0x005e, 0x018c, nv04_graph_mthd_bind_rop);
+	NVOBJ_MTHD (dev, 0x005e, 0x0190, nv04_graph_mthd_bind_beta1);
+	NVOBJ_MTHD (dev, 0x005e, 0x0194, nv04_graph_mthd_bind_beta4);
+	NVOBJ_MTHD (dev, 0x005e, 0x0198, nv04_graph_mthd_bind_surf2d);
+	NVOBJ_MTHD (dev, 0x005e, 0x02fc, nv04_graph_mthd_set_operation);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0150, nv04_graph_mthd_set_ref);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	dev_priv->engine.graph.registered = true;
+	return 0;
 };
 
-static struct nouveau_pgraph_object_method nv04_graph_mthds_nv03_tex_tri[] = {
-	{ 0x0188, nv04_graph_mthd_bind_clip },
-	{ 0x018c, nv04_graph_mthd_bind_surf_color },
-	{ 0x0190, nv04_graph_mthd_bind_surf_zeta },
-	{},
+static struct nouveau_bitfield nv04_graph_intr[] = {
+	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
+	{}
 };
 
-static struct nouveau_pgraph_object_method nv04_graph_mthds_surf3d[] = {
-	{ 0x02f8, nv04_graph_mthd_surf3d_clip_h },
-	{ 0x02fc, nv04_graph_mthd_surf3d_clip_v },
-	{},
+static struct nouveau_bitfield nv04_graph_nstatus[] =
+{
+	{ NV04_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
+	{ NV04_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
+	{ NV04_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
+	{ NV04_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
+	{}
 };
 
-struct nouveau_pgraph_object_class nv04_graph_grclass[] = {
-	{ 0x0038, false, NULL }, /* dvd subpicture */
-	{ 0x0039, false, NULL }, /* m2mf */
-	{ 0x004b, false, nv04_graph_mthds_nv03_gdirect }, /* nv03 gdirect */
-	{ 0x004a, false, nv04_graph_mthds_nv04_gdirect }, /* nv04 gdirect */
-	{ 0x001f, false, nv04_graph_mthds_nv01_imageblit }, /* nv01 imageblit */
-	{ 0x005f, false, nv04_graph_mthds_nv04_imageblit_ifc }, /* nv04 imageblit */
-	{ 0x0060, false, nv04_graph_mthds_nv04_iifc }, /* nv04 iifc */
-	{ 0x0064, false, NULL }, /* nv05 iifc */
-	{ 0x0021, false, nv04_graph_mthds_nv01_ifc }, /* nv01 ifc */
-	{ 0x0061, false, nv04_graph_mthds_nv04_imageblit_ifc }, /* nv04 ifc */
-	{ 0x0065, false, NULL }, /* nv05 ifc */
-	{ 0x0036, false, nv04_graph_mthds_nv03_sifc }, /* nv03 sifc */
-	{ 0x0076, false, nv04_graph_mthds_nv04_sifc }, /* nv04 sifc */
-	{ 0x0066, false, NULL }, /* nv05 sifc */
-	{ 0x0037, false, nv04_graph_mthds_nv03_sifm }, /* nv03 sifm */
-	{ 0x0077, false, nv04_graph_mthds_nv04_sifm }, /* nv04 sifm */
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x0042, false, NULL }, /* surf2d */
-	{ 0x0043, false, NULL }, /* rop */
-	{ 0x0012, false, NULL }, /* beta1 */
-	{ 0x0072, false, NULL }, /* beta4 */
-	{ 0x0019, false, NULL }, /* cliprect */
-	{ 0x0018, false, NULL }, /* nv01 pattern */
-	{ 0x0044, false, NULL }, /* nv04 pattern */
-	{ 0x0052, false, NULL }, /* swzsurf */
-	{ 0x0053, false, nv04_graph_mthds_surf3d }, /* surf3d */
-	{ 0x0048, false, nv04_graph_mthds_nv03_tex_tri }, /* nv03 tex_tri */
-	{ 0x0054, false, NULL }, /* tex_tri */
-	{ 0x0055, false, NULL }, /* multitex_tri */
-	{ 0x0017, false, NULL }, /* nv01 chroma */
-	{ 0x0057, false, NULL }, /* nv04 chroma */
-	{ 0x0058, false, NULL }, /* surf_dst */
-	{ 0x0059, false, NULL }, /* surf_src */
-	{ 0x005a, false, NULL }, /* surf_color */
-	{ 0x005b, false, NULL }, /* surf_zeta */
-	{ 0x001c, false, nv04_graph_mthds_nv01_shape }, /* nv01 line */
-	{ 0x005c, false, nv04_graph_mthds_nv04_shape }, /* nv04 line */
-	{ 0x001d, false, nv04_graph_mthds_nv01_shape }, /* nv01 tri */
-	{ 0x005d, false, nv04_graph_mthds_nv04_shape }, /* nv04 tri */
-	{ 0x001e, false, nv04_graph_mthds_nv01_shape }, /* nv01 rect */
-	{ 0x005e, false, nv04_graph_mthds_nv04_shape }, /* nv04 rect */
-	{ 0x506e, true, nv04_graph_mthds_sw },
+struct nouveau_bitfield nv04_graph_nsource[] =
+{
+	{ NV03_PGRAPH_NSOURCE_NOTIFICATION,       "NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DATA_ERROR,         "DATA_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_PROTECTION_ERROR,   "PROTECTION_ERROR" },
+	{ NV03_PGRAPH_NSOURCE_RANGE_EXCEPTION,    "RANGE_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_COLOR,        "LIMIT_COLOR" },
+	{ NV03_PGRAPH_NSOURCE_LIMIT_ZETA,         "LIMIT_ZETA" },
+	{ NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD,       "ILLEGAL_MTHD" },
+	{ NV03_PGRAPH_NSOURCE_DMA_R_PROTECTION,   "DMA_R_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_W_PROTECTION,   "DMA_W_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_FORMAT_EXCEPTION,   "FORMAT_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_PATCH_EXCEPTION,    "PATCH_EXCEPTION" },
+	{ NV03_PGRAPH_NSOURCE_STATE_INVALID,      "STATE_INVALID" },
+	{ NV03_PGRAPH_NSOURCE_DOUBLE_NOTIFY,      "DOUBLE_NOTIFY" },
+	{ NV03_PGRAPH_NSOURCE_NOTIFY_IN_USE,      "NOTIFY_IN_USE" },
+	{ NV03_PGRAPH_NSOURCE_METHOD_CNT,         "METHOD_CNT" },
+	{ NV03_PGRAPH_NSOURCE_BFR_NOTIFICATION,   "BFR_NOTIFICATION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION, "DMA_VTX_PROTECTION" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_A,        "DMA_WIDTH_A" },
+	{ NV03_PGRAPH_NSOURCE_DMA_WIDTH_B,        "DMA_WIDTH_B" },
 	{}
 };
 
+static void
+nv04_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
+		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
+		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 chid = (addr & 0x0f000000) >> 24;
+		u32 subc = (addr & 0x0000e000) >> 13;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400180 + subc * 4) & 0xff;
+		u32 show = stat;
+
+		if (stat & NV_PGRAPH_INTR_NOTIFY) {
+			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
+					show &= ~NV_PGRAPH_INTR_NOTIFY;
+			}
+		}
+
+		if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
+			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
+			stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			nv04_graph_context_switch(dev);
+		}
+
+		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
+		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
+
+		if (show && nouveau_ratelimit()) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv04_graph_intr, show);
+			printk(" nsource:");
+			nouveau_bitfield_print(nv04_graph_nsource, nsource);
+			printk(" nstatus:");
+			nouveau_bitfield_print(nv04_graph_nstatus, nstatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d/%d class 0x%04x "
+				     "mthd 0x%04x data 0x%08x\n",
+				chid, subc, class, mthd, data);
+		}
+	}
+}
diff --git a/drivers/gpu/drm/nouveau/nv04_instmem.c b/drivers/gpu/drm/nouveau/nv04_instmem.c
index 0b5ae297abde..b8e3edb5c063 100644
--- a/drivers/gpu/drm/nouveau/nv04_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv04_instmem.c
@@ -98,42 +98,66 @@ nv04_instmem_takedown(struct drm_device *dev)
 }
 
 int
-nv04_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj,
-		      uint32_t *sz)
+nv04_instmem_suspend(struct drm_device *dev)
 {
 	return 0;
 }
 
 void
-nv04_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
-{
-}
-
-int
-nv04_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nv04_instmem_resume(struct drm_device *dev)
 {
-	return 0;
 }
 
 int
-nv04_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nv04_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 {
+	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
+	struct drm_mm_node *ramin = NULL;
+
+	do {
+		if (drm_mm_pre_get(&dev_priv->ramin_heap))
+			return -ENOMEM;
+
+		spin_lock(&dev_priv->ramin_lock);
+		ramin = drm_mm_search_free(&dev_priv->ramin_heap, size, align, 0);
+		if (ramin == NULL) {
+			spin_unlock(&dev_priv->ramin_lock);
+			return -ENOMEM;
+		}
+
+		ramin = drm_mm_get_block_atomic(ramin, size, align);
+		spin_unlock(&dev_priv->ramin_lock);
+	} while (ramin == NULL);
+
+	gpuobj->node  = ramin;
+	gpuobj->vinst = ramin->start;
 	return 0;
 }
 
 void
-nv04_instmem_flush(struct drm_device *dev)
+nv04_instmem_put(struct nouveau_gpuobj *gpuobj)
 {
+	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
+
+	spin_lock(&dev_priv->ramin_lock);
+	drm_mm_put_block(gpuobj->node);
+	gpuobj->node = NULL;
+	spin_unlock(&dev_priv->ramin_lock);
 }
 
 int
-nv04_instmem_suspend(struct drm_device *dev)
+nv04_instmem_map(struct nouveau_gpuobj *gpuobj)
 {
+	gpuobj->pinst = gpuobj->vinst;
 	return 0;
 }
 
 void
-nv04_instmem_resume(struct drm_device *dev)
+nv04_instmem_unmap(struct nouveau_gpuobj *gpuobj)
 {
 }
 
+void
+nv04_instmem_flush(struct drm_device *dev)
+{
+}
diff --git a/drivers/gpu/drm/nouveau/nv10_fb.c b/drivers/gpu/drm/nouveau/nv10_fb.c
index cc5cda44e501..f78181a59b4a 100644
--- a/drivers/gpu/drm/nouveau/nv10_fb.c
+++ b/drivers/gpu/drm/nouveau/nv10_fb.c
@@ -3,23 +3,109 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 
+static struct drm_mm_node *
+nv20_fb_alloc_tag(struct drm_device *dev, uint32_t size)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	struct drm_mm_node *mem;
+	int ret;
+
+	ret = drm_mm_pre_get(&pfb->tag_heap);
+	if (ret)
+		return NULL;
+
+	spin_lock(&dev_priv->tile.lock);
+	mem = drm_mm_search_free(&pfb->tag_heap, size, 0, 0);
+	if (mem)
+		mem = drm_mm_get_block_atomic(mem, size, 0);
+	spin_unlock(&dev_priv->tile.lock);
+
+	return mem;
+}
+
+static void
+nv20_fb_free_tag(struct drm_device *dev, struct drm_mm_node *mem)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	spin_lock(&dev_priv->tile.lock);
+	drm_mm_put_block(mem);
+	spin_unlock(&dev_priv->tile.lock);
+}
+
+void
+nv10_fb_init_tile_region(struct drm_device *dev, int i, uint32_t addr,
+			 uint32_t size, uint32_t pitch, uint32_t flags)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+	int bpp = (flags & NOUVEAU_GEM_TILE_32BPP ? 32 : 16);
+
+	tile->addr = addr;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+
+	if (dev_priv->card_type == NV_20) {
+		if (flags & NOUVEAU_GEM_TILE_ZETA) {
+			/*
+			 * Allocate some of the on-die tag memory,
+			 * used to store Z compression meta-data (most
+			 * likely just a bitmap determining if a given
+			 * tile is compressed or not).
+			 */
+			tile->tag_mem = nv20_fb_alloc_tag(dev, size / 256);
+
+			if (tile->tag_mem) {
+				/* Enable Z compression */
+				if (dev_priv->chipset >= 0x25)
+					tile->zcomp = tile->tag_mem->start |
+						(bpp == 16 ?
+						 NV25_PFB_ZCOMP_MODE_16 :
+						 NV25_PFB_ZCOMP_MODE_32);
+				else
+					tile->zcomp = tile->tag_mem->start |
+						NV20_PFB_ZCOMP_EN |
+						(bpp == 16 ? 0 :
+						 NV20_PFB_ZCOMP_MODE_32);
+			}
+
+			tile->addr |= 3;
+		} else {
+			tile->addr |= 1;
+		}
+
+	} else {
+		tile->addr |= 1 << 31;
+	}
+}
+
 void
-nv10_fb_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			  uint32_t size, uint32_t pitch)
+nv10_fb_free_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t limit = max(1u, addr + size) - 1;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
-	if (pitch) {
-		if (dev_priv->card_type >= NV_20)
-			addr |= 1;
-		else
-			addr |= 1 << 31;
+	if (tile->tag_mem) {
+		nv20_fb_free_tag(dev, tile->tag_mem);
+		tile->tag_mem = NULL;
 	}
 
-	nv_wr32(dev, NV10_PFB_TLIMIT(i), limit);
-	nv_wr32(dev, NV10_PFB_TSIZE(i), pitch);
-	nv_wr32(dev, NV10_PFB_TILE(i), addr);
+	tile->addr = tile->limit = tile->pitch = tile->zcomp = 0;
+}
+
+void
+nv10_fb_set_tile_region(struct drm_device *dev, int i)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+
+	nv_wr32(dev, NV10_PFB_TLIMIT(i), tile->limit);
+	nv_wr32(dev, NV10_PFB_TSIZE(i), tile->pitch);
+	nv_wr32(dev, NV10_PFB_TILE(i), tile->addr);
+
+	if (dev_priv->card_type == NV_20)
+		nv_wr32(dev, NV20_PFB_ZCOMP(i), tile->zcomp);
 }
 
 int
@@ -31,9 +117,14 @@ nv10_fb_init(struct drm_device *dev)
 
 	pfb->num_tiles = NV10_PFB_TILE__SIZE;
 
+	if (dev_priv->card_type == NV_20)
+		drm_mm_init(&pfb->tag_heap, 0,
+			    (dev_priv->chipset >= 0x25 ?
+			     64 * 1024 : 32 * 1024));
+
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < pfb->num_tiles; i++)
-		pfb->set_region_tiling(dev, i, 0, 0, 0);
+		pfb->set_tile_region(dev, i);
 
 	return 0;
 }
@@ -41,4 +132,13 @@ nv10_fb_init(struct drm_device *dev)
 void
 nv10_fb_takedown(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
+	int i;
+
+	for (i = 0; i < pfb->num_tiles; i++)
+		pfb->free_tile_region(dev, i);
+
+	if (dev_priv->card_type == NV_20)
+		drm_mm_takedown(&pfb->tag_heap);
 }
diff --git a/drivers/gpu/drm/nouveau/nv10_fifo.c b/drivers/gpu/drm/nouveau/nv10_fifo.c
index f1b03ad58fd5..d2ecbff4bee1 100644
--- a/drivers/gpu/drm/nouveau/nv10_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv10_fifo.c
@@ -53,6 +53,11 @@ nv10_fifo_create_context(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
+	chan->user = ioremap(pci_resource_start(dev->pdev, 0) +
+			     NV03_USER(chan->id), PAGE_SIZE);
+	if (!chan->user)
+		return -ENOMEM;
+
 	/* Fill entries that are seen filled in dumps of nvidia driver just
 	 * after channel's is put into DMA mode
 	 */
@@ -73,17 +78,6 @@ nv10_fifo_create_context(struct nouveau_channel *chan)
 	return 0;
 }
 
-void
-nv10_fifo_destroy_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-
-	nv_wr32(dev, NV04_PFIFO_MODE,
-			nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
-
-	nouveau_gpuobj_ref(NULL, &chan->ramfc);
-}
-
 static void
 nv10_fifo_do_load_context(struct drm_device *dev, int chid)
 {
@@ -219,6 +213,7 @@ nv10_fifo_init_ramxx(struct drm_device *dev)
 static void
 nv10_fifo_init_intr(struct drm_device *dev)
 {
+	nouveau_irq_register(dev, 8, nv04_fifo_isr);
 	nv_wr32(dev, 0x002100, 0xffffffff);
 	nv_wr32(dev, 0x002140, 0xffffffff);
 }
@@ -241,7 +236,7 @@ nv10_fifo_init(struct drm_device *dev)
 	pfifo->reassign(dev, true);
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		if (dev_priv->fifos[i]) {
+		if (dev_priv->channels.ptr[i]) {
 			uint32_t mode = nv_rd32(dev, NV04_PFIFO_MODE);
 			nv_wr32(dev, NV04_PFIFO_MODE, mode | (1 << i));
 		}
diff --git a/drivers/gpu/drm/nouveau/nv10_graph.c b/drivers/gpu/drm/nouveau/nv10_graph.c
index 8e68c9731159..8c92edb7bbcd 100644
--- a/drivers/gpu/drm/nouveau/nv10_graph.c
+++ b/drivers/gpu/drm/nouveau/nv10_graph.c
@@ -26,6 +26,10 @@
 #include "drm.h"
 #include "nouveau_drm.h"
 #include "nouveau_drv.h"
+#include "nouveau_util.h"
+
+static int  nv10_graph_register(struct drm_device *);
+static void nv10_graph_isr(struct drm_device *);
 
 #define NV10_FIFO_NUMBER 32
 
@@ -786,15 +790,13 @@ nv10_graph_unload_context(struct drm_device *dev)
 	return 0;
 }
 
-void
+static void
 nv10_graph_context_switch(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct nouveau_channel *chan = NULL;
 	int chid;
 
-	pgraph->fifo_access(dev, false);
 	nouveau_wait_for_idle(dev);
 
 	/* If previous context is valid, we need to save it */
@@ -802,11 +804,9 @@ nv10_graph_context_switch(struct drm_device *dev)
 
 	/* Load context for next channel */
 	chid = (nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR) >> 20) & 0x1f;
-	chan = dev_priv->fifos[chid];
+	chan = dev_priv->channels.ptr[chid];
 	if (chan && chan->pgraph_ctx)
 		nv10_graph_load_context(chan);
-
-	pgraph->fifo_access(dev, true);
 }
 
 #define NV_WRITE_CTX(reg, val) do { \
@@ -833,7 +833,7 @@ nv10_graph_channel(struct drm_device *dev)
 	if (chid >= dev_priv->engine.fifo.channels)
 		return NULL;
 
-	return dev_priv->fifos[chid];
+	return dev_priv->channels.ptr[chid];
 }
 
 int nv10_graph_create_context(struct nouveau_channel *chan)
@@ -875,37 +875,54 @@ int nv10_graph_create_context(struct nouveau_channel *chan)
 
 void nv10_graph_destroy_context(struct nouveau_channel *chan)
 {
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	struct graph_state *pgraph_ctx = chan->pgraph_ctx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pgraph->fifo_access(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (pgraph->channel(dev) == chan)
+		pgraph->unload_context(dev);
 
+	/* Free the context resources */
 	kfree(pgraph_ctx);
 	chan->pgraph_ctx = NULL;
+
+	pgraph->fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
 
 void
-nv10_graph_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			     uint32_t size, uint32_t pitch)
+nv10_graph_set_tile_region(struct drm_device *dev, int i)
 {
-	uint32_t limit = max(1u, addr + size) - 1;
-
-	if (pitch)
-		addr |= 1 << 31;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
-	nv_wr32(dev, NV10_PGRAPH_TLIMIT(i), limit);
-	nv_wr32(dev, NV10_PGRAPH_TSIZE(i), pitch);
-	nv_wr32(dev, NV10_PGRAPH_TILE(i), addr);
+	nv_wr32(dev, NV10_PGRAPH_TLIMIT(i), tile->limit);
+	nv_wr32(dev, NV10_PGRAPH_TSIZE(i), tile->pitch);
+	nv_wr32(dev, NV10_PGRAPH_TILE(i), tile->addr);
 }
 
 int nv10_graph_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	uint32_t tmp;
-	int i;
+	int ret, i;
 
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) |
 			 NV_PMC_ENABLE_PGRAPH);
 
+	ret = nv10_graph_register(dev);
+	if (ret)
+		return ret;
+
+	nouveau_irq_register(dev, 12, nv10_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -928,7 +945,7 @@ int nv10_graph_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < NV10_PFB_TILE__SIZE; i++)
-		nv10_graph_set_region_tiling(dev, i, 0, 0, 0);
+		nv10_graph_set_tile_region(dev, i);
 
 	nv_wr32(dev, NV10_PGRAPH_CTX_SWITCH(0), 0x00000000);
 	nv_wr32(dev, NV10_PGRAPH_CTX_SWITCH(1), 0x00000000);
@@ -948,17 +965,17 @@ int nv10_graph_init(struct drm_device *dev)
 
 void nv10_graph_takedown(struct drm_device *dev)
 {
+	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
+	nouveau_irq_unregister(dev, 12);
 }
 
 static int
-nv17_graph_mthd_lma_window(struct nouveau_channel *chan, int grclass,
-			   int mthd, uint32_t data)
+nv17_graph_mthd_lma_window(struct nouveau_channel *chan,
+			   u32 class, u32 mthd, u32 data)
 {
 	struct drm_device *dev = chan->dev;
 	struct graph_state *ctx = chan->pgraph_ctx;
 	struct pipe_state *pipe = &ctx->pipe_state;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	uint32_t pipe_0x0040[1], pipe_0x64c0[8], pipe_0x6a80[3], pipe_0x6ab0[3];
 	uint32_t xfmode0, xfmode1;
 	int i;
@@ -1025,18 +1042,14 @@ nv17_graph_mthd_lma_window(struct nouveau_channel *chan, int grclass,
 
 	nouveau_wait_for_idle(dev);
 
-	pgraph->fifo_access(dev, true);
-
 	return 0;
 }
 
 static int
-nv17_graph_mthd_lma_enable(struct nouveau_channel *chan, int grclass,
-			   int mthd, uint32_t data)
+nv17_graph_mthd_lma_enable(struct nouveau_channel *chan,
+			   u32 class, u32 mthd, u32 data)
 {
 	struct drm_device *dev = chan->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 
 	nouveau_wait_for_idle(dev);
 
@@ -1045,40 +1058,118 @@ nv17_graph_mthd_lma_enable(struct nouveau_channel *chan, int grclass,
 	nv_wr32(dev, 0x004006b0,
 		nv_rd32(dev, 0x004006b0) | 0x8 << 24);
 
-	pgraph->fifo_access(dev, true);
+	return 0;
+}
+
+static int
+nv10_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->engine.graph.registered)
+		return 0;
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x005f, GR); /* imageblit */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x0052, GR); /* swzsurf */
+	NVOBJ_CLASS(dev, 0x0093, GR); /* surf3d */
+	NVOBJ_CLASS(dev, 0x0094, GR); /* tex_tri */
+	NVOBJ_CLASS(dev, 0x0095, GR); /* multitex_tri */
+
+	/* celcius */
+	if (dev_priv->chipset <= 0x10) {
+		NVOBJ_CLASS(dev, 0x0056, GR);
+	} else
+	if (dev_priv->chipset < 0x17 || dev_priv->chipset == 0x1a) {
+		NVOBJ_CLASS(dev, 0x0096, GR);
+	} else {
+		NVOBJ_CLASS(dev, 0x0099, GR);
+		NVOBJ_MTHD (dev, 0x0099, 0x1638, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x163c, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1640, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1644, nv17_graph_mthd_lma_window);
+		NVOBJ_MTHD (dev, 0x0099, 0x1658, nv17_graph_mthd_lma_enable);
+	}
 
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	dev_priv->engine.graph.registered = true;
 	return 0;
 }
 
-static struct nouveau_pgraph_object_method nv17_graph_celsius_mthds[] = {
-	{ 0x1638, nv17_graph_mthd_lma_window },
-	{ 0x163c, nv17_graph_mthd_lma_window },
-	{ 0x1640, nv17_graph_mthd_lma_window },
-	{ 0x1644, nv17_graph_mthd_lma_window },
-	{ 0x1658, nv17_graph_mthd_lma_enable },
+struct nouveau_bitfield nv10_graph_intr[] = {
+	{ NV_PGRAPH_INTR_NOTIFY, "NOTIFY" },
+	{ NV_PGRAPH_INTR_ERROR,  "ERROR"  },
 	{}
 };
 
-struct nouveau_pgraph_object_class nv10_graph_grclass[] = {
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x0039, false, NULL }, /* m2mf */
-	{ 0x004a, false, NULL }, /* gdirect */
-	{ 0x005f, false, NULL }, /* imageblit */
-	{ 0x009f, false, NULL }, /* imageblit (nv12) */
-	{ 0x008a, false, NULL }, /* ifc */
-	{ 0x0089, false, NULL }, /* sifm */
-	{ 0x0062, false, NULL }, /* surf2d */
-	{ 0x0043, false, NULL }, /* rop */
-	{ 0x0012, false, NULL }, /* beta1 */
-	{ 0x0072, false, NULL }, /* beta4 */
-	{ 0x0019, false, NULL }, /* cliprect */
-	{ 0x0044, false, NULL }, /* pattern */
-	{ 0x0052, false, NULL }, /* swzsurf */
-	{ 0x0093, false, NULL }, /* surf3d */
-	{ 0x0094, false, NULL }, /* tex_tri */
-	{ 0x0095, false, NULL }, /* multitex_tri */
-	{ 0x0056, false, NULL }, /* celcius (nv10) */
-	{ 0x0096, false, NULL }, /* celcius (nv11) */
-	{ 0x0099, false, nv17_graph_celsius_mthds }, /* celcius (nv17) */
+struct nouveau_bitfield nv10_graph_nstatus[] =
+{
+	{ NV10_PGRAPH_NSTATUS_STATE_IN_USE,       "STATE_IN_USE" },
+	{ NV10_PGRAPH_NSTATUS_INVALID_STATE,      "INVALID_STATE" },
+	{ NV10_PGRAPH_NSTATUS_BAD_ARGUMENT,       "BAD_ARGUMENT" },
+	{ NV10_PGRAPH_NSTATUS_PROTECTION_FAULT,   "PROTECTION_FAULT" },
 	{}
 };
+
+static void
+nv10_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
+		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
+		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 chid = (addr & 0x01f00000) >> 20;
+		u32 subc = (addr & 0x00070000) >> 16;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400160 + subc * 4) & 0xfff;
+		u32 show = stat;
+
+		if (stat & NV_PGRAPH_INTR_ERROR) {
+			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
+					show &= ~NV_PGRAPH_INTR_ERROR;
+			}
+		}
+
+		if (stat & NV_PGRAPH_INTR_CONTEXT_SWITCH) {
+			nv_wr32(dev, NV03_PGRAPH_INTR, NV_PGRAPH_INTR_CONTEXT_SWITCH);
+			stat &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			show &= ~NV_PGRAPH_INTR_CONTEXT_SWITCH;
+			nv10_graph_context_switch(dev);
+		}
+
+		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
+		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
+
+		if (show && nouveau_ratelimit()) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv10_graph_intr, show);
+			printk(" nsource:");
+			nouveau_bitfield_print(nv04_graph_nsource, nsource);
+			printk(" nstatus:");
+			nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d/%d class 0x%04x "
+				     "mthd 0x%04x data 0x%08x\n",
+				chid, subc, class, mthd, data);
+		}
+	}
+}
diff --git a/drivers/gpu/drm/nouveau/nv20_graph.c b/drivers/gpu/drm/nouveau/nv20_graph.c
index 12ab9cd56eca..8464b76798d5 100644
--- a/drivers/gpu/drm/nouveau/nv20_graph.c
+++ b/drivers/gpu/drm/nouveau/nv20_graph.c
@@ -32,6 +32,10 @@
 #define NV34_GRCTX_SIZE    (18140)
 #define NV35_36_GRCTX_SIZE (22396)
 
+static int nv20_graph_register(struct drm_device *);
+static int nv30_graph_register(struct drm_device *);
+static void nv20_graph_isr(struct drm_device *);
+
 static void
 nv20_graph_context_init(struct drm_device *dev, struct nouveau_gpuobj *ctx)
 {
@@ -425,9 +429,21 @@ nv20_graph_destroy_context(struct nouveau_channel *chan)
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	unsigned long flags;
 
-	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pgraph->fifo_access(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (pgraph->channel(dev) == chan)
+		pgraph->unload_context(dev);
+
+	pgraph->fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the context resources */
 	nv_wo32(pgraph->ctx_table, chan->id * 4, 0);
+	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
 }
 
 int
@@ -496,24 +512,27 @@ nv20_graph_rdi(struct drm_device *dev)
 }
 
 void
-nv20_graph_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			     uint32_t size, uint32_t pitch)
+nv20_graph_set_tile_region(struct drm_device *dev, int i)
 {
-	uint32_t limit = max(1u, addr + size) - 1;
-
-	if (pitch)
-		addr |= 1;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
-	nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), limit);
-	nv_wr32(dev, NV20_PGRAPH_TSIZE(i), pitch);
-	nv_wr32(dev, NV20_PGRAPH_TILE(i), addr);
+	nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), tile->limit);
+	nv_wr32(dev, NV20_PGRAPH_TSIZE(i), tile->pitch);
+	nv_wr32(dev, NV20_PGRAPH_TILE(i), tile->addr);
 
 	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0030 + 4 * i);
-	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, limit);
+	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, tile->limit);
 	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0050 + 4 * i);
-	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, pitch);
+	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, tile->pitch);
 	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0010 + 4 * i);
-	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, addr);
+	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, tile->addr);
+
+	if (dev_priv->card_type == NV_20) {
+		nv_wr32(dev, NV20_PGRAPH_ZCOMP(i), tile->zcomp);
+		nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00ea0090 + 4 * i);
+		nv_wr32(dev, NV10_PGRAPH_RDI_DATA, tile->zcomp);
+	}
 }
 
 int
@@ -560,6 +579,13 @@ nv20_graph_init(struct drm_device *dev)
 
 	nv20_graph_rdi(dev);
 
+	ret = nv20_graph_register(dev);
+	if (ret) {
+		nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
+		return ret;
+	}
+
+	nouveau_irq_register(dev, 12, nv20_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -571,16 +597,17 @@ nv20_graph_init(struct drm_device *dev)
 	nv_wr32(dev, 0x40009C           , 0x00000040);
 
 	if (dev_priv->chipset >= 0x25) {
-		nv_wr32(dev, 0x400890, 0x00080000);
+		nv_wr32(dev, 0x400890, 0x00a8cfff);
 		nv_wr32(dev, 0x400610, 0x304B1FB6);
-		nv_wr32(dev, 0x400B80, 0x18B82880);
+		nv_wr32(dev, 0x400B80, 0x1cbd3883);
 		nv_wr32(dev, 0x400B84, 0x44000000);
 		nv_wr32(dev, 0x400098, 0x40000080);
 		nv_wr32(dev, 0x400B88, 0x000000ff);
+
 	} else {
-		nv_wr32(dev, 0x400880, 0x00080000); /* 0x0008c7df */
+		nv_wr32(dev, 0x400880, 0x0008c7df);
 		nv_wr32(dev, 0x400094, 0x00000005);
-		nv_wr32(dev, 0x400B80, 0x45CAA208); /* 0x45eae20e */
+		nv_wr32(dev, 0x400B80, 0x45eae20e);
 		nv_wr32(dev, 0x400B84, 0x24000000);
 		nv_wr32(dev, 0x400098, 0x00000040);
 		nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00E00038);
@@ -591,14 +618,8 @@ nv20_graph_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < NV10_PFB_TILE__SIZE; i++)
-		nv20_graph_set_region_tiling(dev, i, 0, 0, 0);
+		nv20_graph_set_tile_region(dev, i);
 
-	for (i = 0; i < 8; i++) {
-		nv_wr32(dev, 0x400980 + i * 4, nv_rd32(dev, 0x100300 + i * 4));
-		nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA0090 + i * 4);
-		nv_wr32(dev, NV10_PGRAPH_RDI_DATA,
-					nv_rd32(dev, 0x100300 + i * 4));
-	}
 	nv_wr32(dev, 0x4009a0, nv_rd32(dev, 0x100324));
 	nv_wr32(dev, NV10_PGRAPH_RDI_INDEX, 0x00EA000C);
 	nv_wr32(dev, NV10_PGRAPH_RDI_DATA, nv_rd32(dev, 0x100324));
@@ -642,6 +663,9 @@ nv20_graph_takedown(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 
+	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0x00000000);
+	nouveau_irq_unregister(dev, 12);
+
 	nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
 }
 
@@ -684,9 +708,16 @@ nv30_graph_init(struct drm_device *dev)
 			return ret;
 	}
 
+	ret = nv30_graph_register(dev);
+	if (ret) {
+		nouveau_gpuobj_ref(NULL, &pgraph->ctx_table);
+		return ret;
+	}
+
 	nv_wr32(dev, NV20_PGRAPH_CHANNEL_CTX_TABLE,
 		     pgraph->ctx_table->pinst >> 4);
 
+	nouveau_irq_register(dev, 12, nv20_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -724,7 +755,7 @@ nv30_graph_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < NV10_PFB_TILE__SIZE; i++)
-		nv20_graph_set_region_tiling(dev, i, 0, 0, 0);
+		nv20_graph_set_tile_region(dev, i);
 
 	nv_wr32(dev, NV10_PGRAPH_CTX_CONTROL, 0x10000100);
 	nv_wr32(dev, NV10_PGRAPH_STATE      , 0xFFFFFFFF);
@@ -744,46 +775,125 @@ nv30_graph_init(struct drm_device *dev)
 	return 0;
 }
 
-struct nouveau_pgraph_object_class nv20_graph_grclass[] = {
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x0039, false, NULL }, /* m2mf */
-	{ 0x004a, false, NULL }, /* gdirect */
-	{ 0x009f, false, NULL }, /* imageblit (nv12) */
-	{ 0x008a, false, NULL }, /* ifc */
-	{ 0x0089, false, NULL }, /* sifm */
-	{ 0x0062, false, NULL }, /* surf2d */
-	{ 0x0043, false, NULL }, /* rop */
-	{ 0x0012, false, NULL }, /* beta1 */
-	{ 0x0072, false, NULL }, /* beta4 */
-	{ 0x0019, false, NULL }, /* cliprect */
-	{ 0x0044, false, NULL }, /* pattern */
-	{ 0x009e, false, NULL }, /* swzsurf */
-	{ 0x0096, false, NULL }, /* celcius */
-	{ 0x0097, false, NULL }, /* kelvin (nv20) */
-	{ 0x0597, false, NULL }, /* kelvin (nv25) */
-	{}
-};
-
-struct nouveau_pgraph_object_class nv30_graph_grclass[] = {
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x0039, false, NULL }, /* m2mf */
-	{ 0x004a, false, NULL }, /* gdirect */
-	{ 0x009f, false, NULL }, /* imageblit (nv12) */
-	{ 0x008a, false, NULL }, /* ifc */
-	{ 0x038a, false, NULL }, /* ifc (nv30) */
-	{ 0x0089, false, NULL }, /* sifm */
-	{ 0x0389, false, NULL }, /* sifm (nv30) */
-	{ 0x0062, false, NULL }, /* surf2d */
-	{ 0x0362, false, NULL }, /* surf2d (nv30) */
-	{ 0x0043, false, NULL }, /* rop */
-	{ 0x0012, false, NULL }, /* beta1 */
-	{ 0x0072, false, NULL }, /* beta4 */
-	{ 0x0019, false, NULL }, /* cliprect */
-	{ 0x0044, false, NULL }, /* pattern */
-	{ 0x039e, false, NULL }, /* swzsurf */
-	{ 0x0397, false, NULL }, /* rankine (nv30) */
-	{ 0x0497, false, NULL }, /* rankine (nv35) */
-	{ 0x0697, false, NULL }, /* rankine (nv34) */
-	{}
-};
+static int
+nv20_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->engine.graph.registered)
+		return 0;
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x009e, GR); /* swzsurf */
+	NVOBJ_CLASS(dev, 0x0096, GR); /* celcius */
+
+	/* kelvin */
+	if (dev_priv->chipset < 0x25)
+		NVOBJ_CLASS(dev, 0x0097, GR);
+	else
+		NVOBJ_CLASS(dev, 0x0597, GR);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	dev_priv->engine.graph.registered = true;
+	return 0;
+}
+
+static int
+nv30_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
+	if (dev_priv->engine.graph.registered)
+		return 0;
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x038a, GR); /* ifc (nv30) */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x0389, GR); /* sifm (nv30) */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x0362, GR); /* surf2d (nv30) */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x039e, GR); /* swzsurf */
+
+	/* rankine */
+	if (0x00000003 & (1 << (dev_priv->chipset & 0x0f)))
+		NVOBJ_CLASS(dev, 0x0397, GR);
+	else
+	if (0x00000010 & (1 << (dev_priv->chipset & 0x0f)))
+		NVOBJ_CLASS(dev, 0x0697, GR);
+	else
+	if (0x000001e0 & (1 << (dev_priv->chipset & 0x0f)))
+		NVOBJ_CLASS(dev, 0x0497, GR);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	dev_priv->engine.graph.registered = true;
+	return 0;
+}
+
+static void
+nv20_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
+		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
+		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 chid = (addr & 0x01f00000) >> 20;
+		u32 subc = (addr & 0x00070000) >> 16;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400160 + subc * 4) & 0xfff;
+		u32 show = stat;
+
+		if (stat & NV_PGRAPH_INTR_ERROR) {
+			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
+					show &= ~NV_PGRAPH_INTR_ERROR;
+			}
+		}
+
+		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
+		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
+
+		if (show && nouveau_ratelimit()) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv10_graph_intr, show);
+			printk(" nsource:");
+			nouveau_bitfield_print(nv04_graph_nsource, nsource);
+			printk(" nstatus:");
+			nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d/%d class 0x%04x "
+				     "mthd 0x%04x data 0x%08x\n",
+				chid, subc, class, mthd, data);
+		}
+	}
+}
diff --git a/drivers/gpu/drm/nouveau/nv30_fb.c b/drivers/gpu/drm/nouveau/nv30_fb.c
index 4a3f2f095128..e0135f0e2144 100644
--- a/drivers/gpu/drm/nouveau/nv30_fb.c
+++ b/drivers/gpu/drm/nouveau/nv30_fb.c
@@ -29,6 +29,27 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 
+void
+nv30_fb_init_tile_region(struct drm_device *dev, int i, uint32_t addr,
+			 uint32_t size, uint32_t pitch, uint32_t flags)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+
+	tile->addr = addr | 1;
+	tile->limit = max(1u, addr + size) - 1;
+	tile->pitch = pitch;
+}
+
+void
+nv30_fb_free_tile_region(struct drm_device *dev, int i)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
+
+	tile->addr = tile->limit = tile->pitch = 0;
+}
+
 static int
 calc_bias(struct drm_device *dev, int k, int i, int j)
 {
@@ -65,7 +86,7 @@ nv30_fb_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < pfb->num_tiles; i++)
-		pfb->set_region_tiling(dev, i, 0, 0, 0);
+		pfb->set_tile_region(dev, i);
 
 	/* Init the memory timing regs at 0x10037c/0x1003ac */
 	if (dev_priv->chipset == 0x30 ||
diff --git a/drivers/gpu/drm/nouveau/nv40_fb.c b/drivers/gpu/drm/nouveau/nv40_fb.c
index 3cd07d8d5bd7..f3d9c0505f7b 100644
--- a/drivers/gpu/drm/nouveau/nv40_fb.c
+++ b/drivers/gpu/drm/nouveau/nv40_fb.c
@@ -4,26 +4,22 @@
 #include "nouveau_drm.h"
 
 void
-nv40_fb_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			  uint32_t size, uint32_t pitch)
+nv40_fb_set_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t limit = max(1u, addr + size) - 1;
-
-	if (pitch)
-		addr |= 1;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
 	switch (dev_priv->chipset) {
 	case 0x40:
-		nv_wr32(dev, NV10_PFB_TLIMIT(i), limit);
-		nv_wr32(dev, NV10_PFB_TSIZE(i), pitch);
-		nv_wr32(dev, NV10_PFB_TILE(i), addr);
+		nv_wr32(dev, NV10_PFB_TLIMIT(i), tile->limit);
+		nv_wr32(dev, NV10_PFB_TSIZE(i), tile->pitch);
+		nv_wr32(dev, NV10_PFB_TILE(i), tile->addr);
 		break;
 
 	default:
-		nv_wr32(dev, NV40_PFB_TLIMIT(i), limit);
-		nv_wr32(dev, NV40_PFB_TSIZE(i), pitch);
-		nv_wr32(dev, NV40_PFB_TILE(i), addr);
+		nv_wr32(dev, NV40_PFB_TLIMIT(i), tile->limit);
+		nv_wr32(dev, NV40_PFB_TSIZE(i), tile->pitch);
+		nv_wr32(dev, NV40_PFB_TILE(i), tile->addr);
 		break;
 	}
 }
@@ -64,7 +60,7 @@ nv40_fb_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < pfb->num_tiles; i++)
-		pfb->set_region_tiling(dev, i, 0, 0, 0);
+		pfb->set_tile_region(dev, i);
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv40_fifo.c b/drivers/gpu/drm/nouveau/nv40_fifo.c
index d337b8b28cdd..c86e4d4e9b96 100644
--- a/drivers/gpu/drm/nouveau/nv40_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv40_fifo.c
@@ -47,6 +47,11 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
+	chan->user = ioremap(pci_resource_start(dev->pdev, 0) +
+			     NV40_USER(chan->id), PAGE_SIZE);
+	if (!chan->user)
+		return -ENOMEM;
+
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 
 	nv_wi32(dev, fc +  0, chan->pushbuf_base);
@@ -70,17 +75,6 @@ nv40_fifo_create_context(struct nouveau_channel *chan)
 	return 0;
 }
 
-void
-nv40_fifo_destroy_context(struct nouveau_channel *chan)
-{
-	struct drm_device *dev = chan->dev;
-
-	nv_wr32(dev, NV04_PFIFO_MODE,
-		nv_rd32(dev, NV04_PFIFO_MODE) & ~(1 << chan->id));
-
-	nouveau_gpuobj_ref(NULL, &chan->ramfc);
-}
-
 static void
 nv40_fifo_do_load_context(struct drm_device *dev, int chid)
 {
@@ -279,6 +273,7 @@ nv40_fifo_init_ramxx(struct drm_device *dev)
 static void
 nv40_fifo_init_intr(struct drm_device *dev)
 {
+	nouveau_irq_register(dev, 8, nv04_fifo_isr);
 	nv_wr32(dev, 0x002100, 0xffffffff);
 	nv_wr32(dev, 0x002140, 0xffffffff);
 }
@@ -301,7 +296,7 @@ nv40_fifo_init(struct drm_device *dev)
 	pfifo->reassign(dev, true);
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		if (dev_priv->fifos[i]) {
+		if (dev_priv->channels.ptr[i]) {
 			uint32_t mode = nv_rd32(dev, NV04_PFIFO_MODE);
 			nv_wr32(dev, NV04_PFIFO_MODE, mode | (1 << i));
 		}
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 7ee1b91569b8..0618846a97ce 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -29,6 +29,9 @@
 #include "nouveau_drv.h"
 #include "nouveau_grctx.h"
 
+static int nv40_graph_register(struct drm_device *);
+static void nv40_graph_isr(struct drm_device *);
+
 struct nouveau_channel *
 nv40_graph_channel(struct drm_device *dev)
 {
@@ -42,7 +45,7 @@ nv40_graph_channel(struct drm_device *dev)
 	inst = (inst & NV40_PGRAPH_CTXCTL_CUR_INSTANCE) << 4;
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		struct nouveau_channel *chan = dev_priv->fifos[i];
+		struct nouveau_channel *chan = dev_priv->channels.ptr[i];
 
 		if (chan && chan->ramin_grctx &&
 		    chan->ramin_grctx->pinst == inst)
@@ -79,6 +82,22 @@ nv40_graph_create_context(struct nouveau_channel *chan)
 void
 nv40_graph_destroy_context(struct nouveau_channel *chan)
 {
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pgraph->fifo_access(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (pgraph->channel(dev) == chan)
+		pgraph->unload_context(dev);
+
+	pgraph->fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the context resources */
 	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
 }
 
@@ -174,43 +193,39 @@ nv40_graph_unload_context(struct drm_device *dev)
 }
 
 void
-nv40_graph_set_region_tiling(struct drm_device *dev, int i, uint32_t addr,
-			     uint32_t size, uint32_t pitch)
+nv40_graph_set_tile_region(struct drm_device *dev, int i)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t limit = max(1u, addr + size) - 1;
-
-	if (pitch)
-		addr |= 1;
+	struct nouveau_tile_reg *tile = &dev_priv->tile.reg[i];
 
 	switch (dev_priv->chipset) {
 	case 0x44:
 	case 0x4a:
 	case 0x4e:
-		nv_wr32(dev, NV20_PGRAPH_TSIZE(i), pitch);
-		nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), limit);
-		nv_wr32(dev, NV20_PGRAPH_TILE(i), addr);
+		nv_wr32(dev, NV20_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(dev, NV20_PGRAPH_TILE(i), tile->addr);
 		break;
 
 	case 0x46:
 	case 0x47:
 	case 0x49:
 	case 0x4b:
-		nv_wr32(dev, NV47_PGRAPH_TSIZE(i), pitch);
-		nv_wr32(dev, NV47_PGRAPH_TLIMIT(i), limit);
-		nv_wr32(dev, NV47_PGRAPH_TILE(i), addr);
-		nv_wr32(dev, NV40_PGRAPH_TSIZE1(i), pitch);
-		nv_wr32(dev, NV40_PGRAPH_TLIMIT1(i), limit);
-		nv_wr32(dev, NV40_PGRAPH_TILE1(i), addr);
+		nv_wr32(dev, NV47_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(dev, NV47_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(dev, NV47_PGRAPH_TILE(i), tile->addr);
+		nv_wr32(dev, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+		nv_wr32(dev, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+		nv_wr32(dev, NV40_PGRAPH_TILE1(i), tile->addr);
 		break;
 
 	default:
-		nv_wr32(dev, NV20_PGRAPH_TSIZE(i), pitch);
-		nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), limit);
-		nv_wr32(dev, NV20_PGRAPH_TILE(i), addr);
-		nv_wr32(dev, NV40_PGRAPH_TSIZE1(i), pitch);
-		nv_wr32(dev, NV40_PGRAPH_TLIMIT1(i), limit);
-		nv_wr32(dev, NV40_PGRAPH_TILE1(i), addr);
+		nv_wr32(dev, NV20_PGRAPH_TSIZE(i), tile->pitch);
+		nv_wr32(dev, NV20_PGRAPH_TLIMIT(i), tile->limit);
+		nv_wr32(dev, NV20_PGRAPH_TILE(i), tile->addr);
+		nv_wr32(dev, NV40_PGRAPH_TSIZE1(i), tile->pitch);
+		nv_wr32(dev, NV40_PGRAPH_TLIMIT1(i), tile->limit);
+		nv_wr32(dev, NV40_PGRAPH_TILE1(i), tile->addr);
 		break;
 	}
 }
@@ -232,7 +247,7 @@ nv40_graph_init(struct drm_device *dev)
 	struct nouveau_fb_engine *pfb = &dev_priv->engine.fb;
 	struct nouveau_grctx ctx = {};
 	uint32_t vramsz, *cp;
-	int i, j;
+	int ret, i, j;
 
 	nv_wr32(dev, NV03_PMC_ENABLE, nv_rd32(dev, NV03_PMC_ENABLE) &
 			~NV_PMC_ENABLE_PGRAPH);
@@ -256,9 +271,14 @@ nv40_graph_init(struct drm_device *dev)
 
 	kfree(cp);
 
+	ret = nv40_graph_register(dev);
+	if (ret)
+		return ret;
+
 	/* No context present currently */
 	nv_wr32(dev, NV40_PGRAPH_CTXCTL_CUR, 0x00000000);
 
+	nouveau_irq_register(dev, 12, nv40_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR   , 0xFFFFFFFF);
 	nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xFFFFFFFF);
 
@@ -347,7 +367,7 @@ nv40_graph_init(struct drm_device *dev)
 
 	/* Turn all the tiling regions off. */
 	for (i = 0; i < pfb->num_tiles; i++)
-		nv40_graph_set_region_tiling(dev, i, 0, 0, 0);
+		nv40_graph_set_tile_region(dev, i);
 
 	/* begin RAM config */
 	vramsz = pci_resource_len(dev->pdev, 0) - 1;
@@ -390,26 +410,111 @@ nv40_graph_init(struct drm_device *dev)
 
 void nv40_graph_takedown(struct drm_device *dev)
 {
+	nouveau_irq_unregister(dev, 12);
 }
 
-struct nouveau_pgraph_object_class nv40_graph_grclass[] = {
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x0039, false, NULL }, /* m2mf */
-	{ 0x004a, false, NULL }, /* gdirect */
-	{ 0x009f, false, NULL }, /* imageblit (nv12) */
-	{ 0x008a, false, NULL }, /* ifc */
-	{ 0x0089, false, NULL }, /* sifm */
-	{ 0x3089, false, NULL }, /* sifm (nv40) */
-	{ 0x0062, false, NULL }, /* surf2d */
-	{ 0x3062, false, NULL }, /* surf2d (nv40) */
-	{ 0x0043, false, NULL }, /* rop */
-	{ 0x0012, false, NULL }, /* beta1 */
-	{ 0x0072, false, NULL }, /* beta4 */
-	{ 0x0019, false, NULL }, /* cliprect */
-	{ 0x0044, false, NULL }, /* pattern */
-	{ 0x309e, false, NULL }, /* swzsurf */
-	{ 0x4097, false, NULL }, /* curie (nv40) */
-	{ 0x4497, false, NULL }, /* curie (nv44) */
-	{}
-};
+static int
+nv40_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
 
+	if (dev_priv->engine.graph.registered)
+		return 0;
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x0039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x004a, GR); /* gdirect */
+	NVOBJ_CLASS(dev, 0x009f, GR); /* imageblit (nv12) */
+	NVOBJ_CLASS(dev, 0x008a, GR); /* ifc */
+	NVOBJ_CLASS(dev, 0x0089, GR); /* sifm */
+	NVOBJ_CLASS(dev, 0x3089, GR); /* sifm (nv40) */
+	NVOBJ_CLASS(dev, 0x0062, GR); /* surf2d */
+	NVOBJ_CLASS(dev, 0x3062, GR); /* surf2d (nv40) */
+	NVOBJ_CLASS(dev, 0x0043, GR); /* rop */
+	NVOBJ_CLASS(dev, 0x0012, GR); /* beta1 */
+	NVOBJ_CLASS(dev, 0x0072, GR); /* beta4 */
+	NVOBJ_CLASS(dev, 0x0019, GR); /* cliprect */
+	NVOBJ_CLASS(dev, 0x0044, GR); /* pattern */
+	NVOBJ_CLASS(dev, 0x309e, GR); /* swzsurf */
+
+	/* curie */
+	if (dev_priv->chipset >= 0x60 ||
+	    0x00005450 & (1 << (dev_priv->chipset & 0x0f)))
+		NVOBJ_CLASS(dev, 0x4497, GR);
+	else
+		NVOBJ_CLASS(dev, 0x4097, GR);
+
+	/* nvsw */
+	NVOBJ_CLASS(dev, 0x506e, SW);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv04_graph_mthd_page_flip);
+
+	dev_priv->engine.graph.registered = true;
+	return 0;
+}
+
+static int
+nv40_graph_isr_chid(struct drm_device *dev, u32 inst)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
+		chan = dev_priv->channels.ptr[i];
+		if (!chan || !chan->ramin_grctx)
+			continue;
+
+		if (inst == chan->ramin_grctx->pinst)
+			break;
+	}
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+	return i;
+}
+
+static void
+nv40_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, NV03_PGRAPH_INTR))) {
+		u32 nsource = nv_rd32(dev, NV03_PGRAPH_NSOURCE);
+		u32 nstatus = nv_rd32(dev, NV03_PGRAPH_NSTATUS);
+		u32 inst = (nv_rd32(dev, 0x40032c) & 0x000fffff) << 4;
+		u32 chid = nv40_graph_isr_chid(dev, inst);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 subc = (addr & 0x00070000) >> 16;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400160 + subc * 4) & 0xffff;
+		u32 show = stat;
+
+		if (stat & NV_PGRAPH_INTR_ERROR) {
+			if (nsource & NV03_PGRAPH_NSOURCE_ILLEGAL_MTHD) {
+				if (!nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data))
+					show &= ~NV_PGRAPH_INTR_ERROR;
+			} else
+			if (nsource & NV03_PGRAPH_NSOURCE_DMA_VTX_PROTECTION) {
+				nv_mask(dev, 0x402000, 0, 0);
+			}
+		}
+
+		nv_wr32(dev, NV03_PGRAPH_INTR, stat);
+		nv_wr32(dev, NV04_PGRAPH_FIFO, 0x00000001);
+
+		if (show && nouveau_ratelimit()) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv10_graph_intr, show);
+			printk(" nsource:");
+			nouveau_bitfield_print(nv04_graph_nsource, nsource);
+			printk(" nstatus:");
+			nouveau_bitfield_print(nv10_graph_nstatus, nstatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d (0x%08x) subc %d "
+				     "class 0x%04x mthd 0x%04x data 0x%08x\n",
+				chid, inst, subc, class, mthd, data);
+		}
+	}
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_crtc.c b/drivers/gpu/drm/nouveau/nv50_crtc.c
index 56476d0c6de8..2c346f797285 100644
--- a/drivers/gpu/drm/nouveau/nv50_crtc.c
+++ b/drivers/gpu/drm/nouveau/nv50_crtc.c
@@ -345,7 +345,6 @@ nv50_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 		     uint32_t buffer_handle, uint32_t width, uint32_t height)
 {
 	struct drm_device *dev = crtc->dev;
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_crtc *nv_crtc = nouveau_crtc(crtc);
 	struct nouveau_bo *cursor = NULL;
 	struct drm_gem_object *gem;
@@ -374,8 +373,7 @@ nv50_crtc_cursor_set(struct drm_crtc *crtc, struct drm_file *file_priv,
 
 	nouveau_bo_unmap(cursor);
 
-	nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.nvbo->bo.offset -
-					    dev_priv->vm_vram_base);
+	nv_crtc->cursor.set_offset(nv_crtc, nv_crtc->cursor.nvbo->bo.mem.start << PAGE_SHIFT);
 	nv_crtc->cursor.show(nv_crtc, true);
 
 out:
@@ -437,6 +435,7 @@ static const struct drm_crtc_funcs nv50_crtc_funcs = {
 	.cursor_move = nv50_crtc_cursor_move,
 	.gamma_set = nv50_crtc_gamma_set,
 	.set_config = drm_crtc_helper_set_config,
+	.page_flip = nouveau_crtc_page_flip,
 	.destroy = nv50_crtc_destroy,
 };
 
@@ -453,6 +452,7 @@ nv50_crtc_prepare(struct drm_crtc *crtc)
 
 	NV_DEBUG_KMS(dev, "index %d\n", nv_crtc->index);
 
+	drm_vblank_pre_modeset(dev, nv_crtc->index);
 	nv50_crtc_blank(nv_crtc, true);
 }
 
@@ -468,6 +468,7 @@ nv50_crtc_commit(struct drm_crtc *crtc)
 	NV_DEBUG_KMS(dev, "index %d\n", nv_crtc->index);
 
 	nv50_crtc_blank(nv_crtc, false);
+	drm_vblank_post_modeset(dev, nv_crtc->index);
 
 	ret = RING_SPACE(evo, 2);
 	if (ret) {
@@ -545,7 +546,7 @@ nv50_crtc_do_mode_set_base(struct drm_crtc *crtc,
 		 return -EINVAL;
 	}
 
-	nv_crtc->fb.offset = fb->nvbo->bo.offset - dev_priv->vm_vram_base;
+	nv_crtc->fb.offset = fb->nvbo->bo.mem.start << PAGE_SHIFT;
 	nv_crtc->fb.tile_flags = nouveau_bo_tile_layout(fb->nvbo);
 	nv_crtc->fb.cpp = drm_fb->bits_per_pixel / 8;
 	if (!nv_crtc->fb.blanked && dev_priv->chipset != 0x50) {
diff --git a/drivers/gpu/drm/nouveau/nv50_display.c b/drivers/gpu/drm/nouveau/nv50_display.c
index f624c611ddea..7cc94ed9ed95 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.c
+++ b/drivers/gpu/drm/nouveau/nv50_display.c
@@ -33,6 +33,8 @@
 #include "nouveau_ramht.h"
 #include "drm_crtc_helper.h"
 
+static void nv50_display_isr(struct drm_device *);
+
 static inline int
 nv50_sor_nr(struct drm_device *dev)
 {
@@ -46,159 +48,6 @@ nv50_sor_nr(struct drm_device *dev)
 	return 4;
 }
 
-static void
-nv50_evo_channel_del(struct nouveau_channel **pchan)
-{
-	struct nouveau_channel *chan = *pchan;
-
-	if (!chan)
-		return;
-	*pchan = NULL;
-
-	nouveau_gpuobj_channel_takedown(chan);
-	nouveau_bo_unmap(chan->pushbuf_bo);
-	nouveau_bo_ref(NULL, &chan->pushbuf_bo);
-
-	if (chan->user)
-		iounmap(chan->user);
-
-	kfree(chan);
-}
-
-static int
-nv50_evo_dmaobj_new(struct nouveau_channel *evo, uint32_t class, uint32_t name,
-		    uint32_t tile_flags, uint32_t magic_flags,
-		    uint32_t offset, uint32_t limit)
-{
-	struct drm_nouveau_private *dev_priv = evo->dev->dev_private;
-	struct drm_device *dev = evo->dev;
-	struct nouveau_gpuobj *obj = NULL;
-	int ret;
-
-	ret = nouveau_gpuobj_new(dev, evo, 6*4, 32, 0, &obj);
-	if (ret)
-		return ret;
-	obj->engine = NVOBJ_ENGINE_DISPLAY;
-
-	nv_wo32(obj,  0, (tile_flags << 22) | (magic_flags << 16) | class);
-	nv_wo32(obj,  4, limit);
-	nv_wo32(obj,  8, offset);
-	nv_wo32(obj, 12, 0x00000000);
-	nv_wo32(obj, 16, 0x00000000);
-	if (dev_priv->card_type < NV_C0)
-		nv_wo32(obj, 20, 0x00010000);
-	else
-		nv_wo32(obj, 20, 0x00020000);
-	dev_priv->engine.instmem.flush(dev);
-
-	ret = nouveau_ramht_insert(evo, name, obj);
-	nouveau_gpuobj_ref(NULL, &obj);
-	if (ret) {
-		return ret;
-	}
-
-	return 0;
-}
-
-static int
-nv50_evo_channel_new(struct drm_device *dev, struct nouveau_channel **pchan)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_gpuobj *ramht = NULL;
-	struct nouveau_channel *chan;
-	int ret;
-
-	chan = kzalloc(sizeof(struct nouveau_channel), GFP_KERNEL);
-	if (!chan)
-		return -ENOMEM;
-	*pchan = chan;
-
-	chan->id = -1;
-	chan->dev = dev;
-	chan->user_get = 4;
-	chan->user_put = 0;
-
-	ret = nouveau_gpuobj_new(dev, NULL, 32768, 0x1000,
-				 NVOBJ_FLAG_ZERO_ALLOC, &chan->ramin);
-	if (ret) {
-		NV_ERROR(dev, "Error allocating EVO channel memory: %d\n", ret);
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	ret = drm_mm_init(&chan->ramin_heap, 0, 32768);
-	if (ret) {
-		NV_ERROR(dev, "Error initialising EVO PRAMIN heap: %d\n", ret);
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	ret = nouveau_gpuobj_new(dev, chan, 4096, 16, 0, &ramht);
-	if (ret) {
-		NV_ERROR(dev, "Unable to allocate EVO RAMHT: %d\n", ret);
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	ret = nouveau_ramht_new(dev, ramht, &chan->ramht);
-	nouveau_gpuobj_ref(NULL, &ramht);
-	if (ret) {
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	if (dev_priv->chipset != 0x50) {
-		ret = nv50_evo_dmaobj_new(chan, 0x3d, NvEvoFB16, 0x70, 0x19,
-					  0, 0xffffffff);
-		if (ret) {
-			nv50_evo_channel_del(pchan);
-			return ret;
-		}
-
-
-		ret = nv50_evo_dmaobj_new(chan, 0x3d, NvEvoFB32, 0x7a, 0x19,
-					  0, 0xffffffff);
-		if (ret) {
-			nv50_evo_channel_del(pchan);
-			return ret;
-		}
-	}
-
-	ret = nv50_evo_dmaobj_new(chan, 0x3d, NvEvoVRAM, 0, 0x19,
-				  0, dev_priv->vram_size);
-	if (ret) {
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	ret = nouveau_bo_new(dev, NULL, 4096, 0, TTM_PL_FLAG_VRAM, 0, 0,
-			     false, true, &chan->pushbuf_bo);
-	if (ret == 0)
-		ret = nouveau_bo_pin(chan->pushbuf_bo, TTM_PL_FLAG_VRAM);
-	if (ret) {
-		NV_ERROR(dev, "Error creating EVO DMA push buffer: %d\n", ret);
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	ret = nouveau_bo_map(chan->pushbuf_bo);
-	if (ret) {
-		NV_ERROR(dev, "Error mapping EVO DMA push buffer: %d\n", ret);
-		nv50_evo_channel_del(pchan);
-		return ret;
-	}
-
-	chan->user = ioremap(pci_resource_start(dev->pdev, 0) +
-					NV50_PDISPLAY_USER(0), PAGE_SIZE);
-	if (!chan->user) {
-		NV_ERROR(dev, "Error mapping EVO control regs.\n");
-		nv50_evo_channel_del(pchan);
-		return -ENOMEM;
-	}
-
-	return 0;
-}
-
 int
 nv50_display_early_init(struct drm_device *dev)
 {
@@ -214,17 +63,16 @@ int
 nv50_display_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer;
 	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
-	struct nouveau_channel *evo = dev_priv->evo;
 	struct drm_connector *connector;
-	uint32_t val, ram_amount;
-	uint64_t start;
+	struct nouveau_channel *evo;
 	int ret, i;
+	u32 val;
 
 	NV_DEBUG_KMS(dev, "\n");
 
 	nv_wr32(dev, 0x00610184, nv_rd32(dev, 0x00614004));
+
 	/*
 	 * I think the 0x006101XX range is some kind of main control area
 	 * that enables things.
@@ -240,16 +88,19 @@ nv50_display_init(struct drm_device *dev)
 		val = nv_rd32(dev, 0x0061610c + (i * 0x800));
 		nv_wr32(dev, 0x0061019c + (i * 0x10), val);
 	}
+
 	/* DAC */
 	for (i = 0; i < 3; i++) {
 		val = nv_rd32(dev, 0x0061a000 + (i * 0x800));
 		nv_wr32(dev, 0x006101d0 + (i * 0x04), val);
 	}
+
 	/* SOR */
 	for (i = 0; i < nv50_sor_nr(dev); i++) {
 		val = nv_rd32(dev, 0x0061c000 + (i * 0x800));
 		nv_wr32(dev, 0x006101e0 + (i * 0x04), val);
 	}
+
 	/* EXT */
 	for (i = 0; i < 3; i++) {
 		val = nv_rd32(dev, 0x0061e000 + (i * 0x800));
@@ -262,17 +113,6 @@ nv50_display_init(struct drm_device *dev)
 		nv_wr32(dev, NV50_PDISPLAY_DAC_CLK_CTRL1(i), 0x00000001);
 	}
 
-	/* This used to be in crtc unblank, but seems out of place there. */
-	nv_wr32(dev, NV50_PDISPLAY_UNK_380, 0);
-	/* RAM is clamped to 256 MiB. */
-	ram_amount = dev_priv->vram_size;
-	NV_DEBUG_KMS(dev, "ram_amount %d\n", ram_amount);
-	if (ram_amount > 256*1024*1024)
-		ram_amount = 256*1024*1024;
-	nv_wr32(dev, NV50_PDISPLAY_RAM_AMOUNT, ram_amount - 1);
-	nv_wr32(dev, NV50_PDISPLAY_UNK_388, 0x150000);
-	nv_wr32(dev, NV50_PDISPLAY_UNK_38C, 0);
-
 	/* The precise purpose is unknown, i suspect it has something to do
 	 * with text mode.
 	 */
@@ -287,37 +127,6 @@ nv50_display_init(struct drm_device *dev)
 		}
 	}
 
-	/* taken from nv bug #12637, attempts to un-wedge the hw if it's
-	 * stuck in some unspecified state
-	 */
-	start = ptimer->read(dev);
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0), 0x2b00);
-	while ((val = nv_rd32(dev, NV50_PDISPLAY_CHANNEL_STAT(0))) & 0x1e0000) {
-		if ((val & 0x9f0000) == 0x20000)
-			nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0),
-							val | 0x800000);
-
-		if ((val & 0x3f0000) == 0x30000)
-			nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0),
-							val | 0x200000);
-
-		if (ptimer->read(dev) - start > 1000000000ULL) {
-			NV_ERROR(dev, "timeout: (0x610200 & 0x1e0000) != 0\n");
-			NV_ERROR(dev, "0x610200 = 0x%08x\n", val);
-			return -EBUSY;
-		}
-	}
-
-	nv_wr32(dev, NV50_PDISPLAY_CTRL_STATE, NV50_PDISPLAY_CTRL_STATE_ENABLE);
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0), 0x1000b03);
-	if (!nv_wait(dev, NV50_PDISPLAY_CHANNEL_STAT(0),
-		     0x40000000, 0x40000000)) {
-		NV_ERROR(dev, "timeout: (0x610200 & 0x40000000) == 0x40000000\n");
-		NV_ERROR(dev, "0x610200 = 0x%08x\n",
-			  nv_rd32(dev, NV50_PDISPLAY_CHANNEL_STAT(0)));
-		return -EBUSY;
-	}
-
 	for (i = 0; i < 2; i++) {
 		nv_wr32(dev, NV50_PDISPLAY_CURSOR_CURSOR_CTRL2(i), 0x2000);
 		if (!nv_wait(dev, NV50_PDISPLAY_CURSOR_CURSOR_CTRL2(i),
@@ -341,39 +150,31 @@ nv50_display_init(struct drm_device *dev)
 		}
 	}
 
-	nv_wr32(dev, NV50_PDISPLAY_OBJECTS, (evo->ramin->vinst >> 8) | 9);
+	nv_wr32(dev, NV50_PDISPLAY_PIO_CTRL, 0x00000000);
+	nv_mask(dev, NV50_PDISPLAY_INTR_0, 0x00000000, 0x00000000);
+	nv_wr32(dev, NV50_PDISPLAY_INTR_EN_0, 0x00000000);
+	nv_mask(dev, NV50_PDISPLAY_INTR_1, 0x00000000, 0x00000000);
+	nv_wr32(dev, NV50_PDISPLAY_INTR_EN_1,
+		     NV50_PDISPLAY_INTR_EN_1_CLK_UNK10 |
+		     NV50_PDISPLAY_INTR_EN_1_CLK_UNK20 |
+		     NV50_PDISPLAY_INTR_EN_1_CLK_UNK40);
+
+	/* enable hotplug interrupts */
+	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
+		struct nouveau_connector *conn = nouveau_connector(connector);
 
-	/* initialise fifo */
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_DMA_CB(0),
-		((evo->pushbuf_bo->bo.mem.start << PAGE_SHIFT) >> 8) |
-		NV50_PDISPLAY_CHANNEL_DMA_CB_LOCATION_VRAM |
-		NV50_PDISPLAY_CHANNEL_DMA_CB_VALID);
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_UNK2(0), 0x00010000);
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_UNK3(0), 0x00000002);
-	if (!nv_wait(dev, 0x610200, 0x80000000, 0x00000000)) {
-		NV_ERROR(dev, "timeout: (0x610200 & 0x80000000) == 0\n");
-		NV_ERROR(dev, "0x610200 = 0x%08x\n", nv_rd32(dev, 0x610200));
-		return -EBUSY;
+		if (conn->dcb->gpio_tag == 0xff)
+			continue;
+
+		pgpio->irq_enable(dev, conn->dcb->gpio_tag, true);
 	}
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0),
-		(nv_rd32(dev, NV50_PDISPLAY_CHANNEL_STAT(0)) & ~0x00000003) |
-		 NV50_PDISPLAY_CHANNEL_STAT_DMA_ENABLED);
-	nv_wr32(dev, NV50_PDISPLAY_USER_PUT(0), 0);
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0), 0x01000003 |
-		NV50_PDISPLAY_CHANNEL_STAT_DMA_ENABLED);
-	nv_wr32(dev, 0x610300, nv_rd32(dev, 0x610300) & ~1);
-
-	evo->dma.max = (4096/4) - 2;
-	evo->dma.put = 0;
-	evo->dma.cur = evo->dma.put;
-	evo->dma.free = evo->dma.max - evo->dma.cur;
-
-	ret = RING_SPACE(evo, NOUVEAU_DMA_SKIPS);
+
+	ret = nv50_evo_init(dev);
 	if (ret)
 		return ret;
+	evo = dev_priv->evo;
 
-	for (i = 0; i < NOUVEAU_DMA_SKIPS; i++)
-		OUT_RING(evo, 0);
+	nv_wr32(dev, NV50_PDISPLAY_OBJECTS, (evo->ramin->vinst >> 8) | 9);
 
 	ret = RING_SPACE(evo, 11);
 	if (ret)
@@ -393,21 +194,6 @@ nv50_display_init(struct drm_device *dev)
 	if (!nv_wait(dev, 0x640004, 0xffffffff, evo->dma.put << 2))
 		NV_ERROR(dev, "evo pushbuf stalled\n");
 
-	/* enable clock change interrupts. */
-	nv_wr32(dev, 0x610028, 0x00010001);
-	nv_wr32(dev, NV50_PDISPLAY_INTR_EN, (NV50_PDISPLAY_INTR_EN_CLK_UNK10 |
-					     NV50_PDISPLAY_INTR_EN_CLK_UNK20 |
-					     NV50_PDISPLAY_INTR_EN_CLK_UNK40));
-
-	/* enable hotplug interrupts */
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
-		struct nouveau_connector *conn = nouveau_connector(connector);
-
-		if (conn->dcb->gpio_tag == 0xff)
-			continue;
-
-		pgpio->irq_enable(dev, conn->dcb->gpio_tag, true);
-	}
 
 	return 0;
 }
@@ -452,13 +238,7 @@ static int nv50_display_disable(struct drm_device *dev)
 		}
 	}
 
-	nv_wr32(dev, NV50_PDISPLAY_CHANNEL_STAT(0), 0);
-	nv_wr32(dev, NV50_PDISPLAY_CTRL_STATE, 0);
-	if (!nv_wait(dev, NV50_PDISPLAY_CHANNEL_STAT(0), 0x1e0000, 0)) {
-		NV_ERROR(dev, "timeout: (0x610200 & 0x1e0000) == 0\n");
-		NV_ERROR(dev, "0x610200 = 0x%08x\n",
-			  nv_rd32(dev, NV50_PDISPLAY_CHANNEL_STAT(0)));
-	}
+	nv50_evo_fini(dev);
 
 	for (i = 0; i < 3; i++) {
 		if (!nv_wait(dev, NV50_PDISPLAY_SOR_DPMS_STATE(i),
@@ -470,7 +250,7 @@ static int nv50_display_disable(struct drm_device *dev)
 	}
 
 	/* disable interrupts. */
-	nv_wr32(dev, NV50_PDISPLAY_INTR_EN, 0x00000000);
+	nv_wr32(dev, NV50_PDISPLAY_INTR_EN_1, 0x00000000);
 
 	/* disable hotplug interrupts */
 	nv_wr32(dev, 0xe054, 0xffffffff);
@@ -508,13 +288,6 @@ int nv50_display_create(struct drm_device *dev)
 
 	dev->mode_config.fb_base = dev_priv->fb_phys;
 
-	/* Create EVO channel */
-	ret = nv50_evo_channel_new(dev, &dev_priv->evo);
-	if (ret) {
-		NV_ERROR(dev, "Error creating EVO channel: %d\n", ret);
-		return ret;
-	}
-
 	/* Create CRTC objects */
 	for (i = 0; i < 2; i++)
 		nv50_crtc_create(dev, i);
@@ -557,6 +330,9 @@ int nv50_display_create(struct drm_device *dev)
 		}
 	}
 
+	INIT_WORK(&dev_priv->irq_work, nv50_display_irq_handler_bh);
+	nouveau_irq_register(dev, 26, nv50_display_isr);
+
 	ret = nv50_display_init(dev);
 	if (ret) {
 		nv50_display_destroy(dev);
@@ -569,14 +345,12 @@ int nv50_display_create(struct drm_device *dev)
 void
 nv50_display_destroy(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
 	NV_DEBUG_KMS(dev, "\n");
 
 	drm_mode_config_cleanup(dev);
 
 	nv50_display_disable(dev);
-	nv50_evo_channel_del(&dev_priv->evo);
+	nouveau_irq_unregister(dev, 26);
 }
 
 static u16
@@ -660,32 +434,32 @@ static void
 nv50_display_vblank_crtc_handler(struct drm_device *dev, int crtc)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_channel *chan;
-	struct list_head *entry, *tmp;
+	struct nouveau_channel *chan, *tmp;
 
-	list_for_each_safe(entry, tmp, &dev_priv->vbl_waiting) {
-		chan = list_entry(entry, struct nouveau_channel, nvsw.vbl_wait);
+	list_for_each_entry_safe(chan, tmp, &dev_priv->vbl_waiting,
+				 nvsw.vbl_wait) {
+		if (chan->nvsw.vblsem_head != crtc)
+			continue;
 
 		nouveau_bo_wr32(chan->notifier_bo, chan->nvsw.vblsem_offset,
 						chan->nvsw.vblsem_rval);
 		list_del(&chan->nvsw.vbl_wait);
+		drm_vblank_put(dev, crtc);
 	}
+
+	drm_handle_vblank(dev, crtc);
 }
 
 static void
 nv50_display_vblank_handler(struct drm_device *dev, uint32_t intr)
 {
-	intr &= NV50_PDISPLAY_INTR_1_VBLANK_CRTC;
-
 	if (intr & NV50_PDISPLAY_INTR_1_VBLANK_CRTC_0)
 		nv50_display_vblank_crtc_handler(dev, 0);
 
 	if (intr & NV50_PDISPLAY_INTR_1_VBLANK_CRTC_1)
 		nv50_display_vblank_crtc_handler(dev, 1);
 
-	nv_wr32(dev, NV50_PDISPLAY_INTR_EN, nv_rd32(dev,
-		     NV50_PDISPLAY_INTR_EN) & ~intr);
-	nv_wr32(dev, NV50_PDISPLAY_INTR_1, intr);
+	nv_wr32(dev, NV50_PDISPLAY_INTR_1, NV50_PDISPLAY_INTR_1_VBLANK_CRTC);
 }
 
 static void
@@ -1011,108 +785,31 @@ nv50_display_irq_handler_bh(struct work_struct *work)
 static void
 nv50_display_error_handler(struct drm_device *dev)
 {
-	uint32_t addr, data;
-
-	nv_wr32(dev, NV50_PDISPLAY_INTR_0, 0x00010000);
-	addr = nv_rd32(dev, NV50_PDISPLAY_TRAPPED_ADDR);
-	data = nv_rd32(dev, NV50_PDISPLAY_TRAPPED_DATA);
-
-	NV_ERROR(dev, "EvoCh %d Mthd 0x%04x Data 0x%08x (0x%04x 0x%02x)\n",
-		 0, addr & 0xffc, data, addr >> 16, (addr >> 12) & 0xf);
-
-	nv_wr32(dev, NV50_PDISPLAY_TRAPPED_ADDR, 0x90000000);
-}
-
-void
-nv50_display_irq_hotplug_bh(struct work_struct *work)
-{
-	struct drm_nouveau_private *dev_priv =
-		container_of(work, struct drm_nouveau_private, hpd_work);
-	struct drm_device *dev = dev_priv->dev;
-	struct drm_connector *connector;
-	const uint32_t gpio_reg[4] = { 0xe104, 0xe108, 0xe280, 0xe284 };
-	uint32_t unplug_mask, plug_mask, change_mask;
-	uint32_t hpd0, hpd1;
-
-	spin_lock_irq(&dev_priv->hpd_state.lock);
-	hpd0 = dev_priv->hpd_state.hpd0_bits;
-	dev_priv->hpd_state.hpd0_bits = 0;
-	hpd1 = dev_priv->hpd_state.hpd1_bits;
-	dev_priv->hpd_state.hpd1_bits = 0;
-	spin_unlock_irq(&dev_priv->hpd_state.lock);
-
-	hpd0 &= nv_rd32(dev, 0xe050);
-	if (dev_priv->chipset >= 0x90)
-		hpd1 &= nv_rd32(dev, 0xe070);
-
-	plug_mask   = (hpd0 & 0x0000ffff) | (hpd1 << 16);
-	unplug_mask = (hpd0 >> 16) | (hpd1 & 0xffff0000);
-	change_mask = plug_mask | unplug_mask;
-
-	list_for_each_entry(connector, &dev->mode_config.connector_list, head) {
-		struct drm_encoder_helper_funcs *helper;
-		struct nouveau_connector *nv_connector =
-			nouveau_connector(connector);
-		struct nouveau_encoder *nv_encoder;
-		struct dcb_gpio_entry *gpio;
-		uint32_t reg;
-		bool plugged;
-
-		if (!nv_connector->dcb)
-			continue;
+	u32 channels = (nv_rd32(dev, NV50_PDISPLAY_INTR_0) & 0x001f0000) >> 16;
+	u32 addr, data;
+	int chid;
 
-		gpio = nouveau_bios_gpio_entry(dev, nv_connector->dcb->gpio_tag);
-		if (!gpio || !(change_mask & (1 << gpio->line)))
+	for (chid = 0; chid < 5; chid++) {
+		if (!(channels & (1 << chid)))
 			continue;
 
-		reg = nv_rd32(dev, gpio_reg[gpio->line >> 3]);
-		plugged = !!(reg & (4 << ((gpio->line & 7) << 2)));
-		NV_INFO(dev, "%splugged %s\n", plugged ? "" : "un",
-			drm_get_connector_name(connector)) ;
-
-		if (!connector->encoder || !connector->encoder->crtc ||
-		    !connector->encoder->crtc->enabled)
-			continue;
-		nv_encoder = nouveau_encoder(connector->encoder);
-		helper = connector->encoder->helper_private;
-
-		if (nv_encoder->dcb->type != OUTPUT_DP)
-			continue;
+		nv_wr32(dev, NV50_PDISPLAY_INTR_0, 0x00010000 << chid);
+		addr = nv_rd32(dev, NV50_PDISPLAY_TRAPPED_ADDR(chid));
+		data = nv_rd32(dev, NV50_PDISPLAY_TRAPPED_DATA(chid));
+		NV_ERROR(dev, "EvoCh %d Mthd 0x%04x Data 0x%08x "
+			      "(0x%04x 0x%02x)\n", chid,
+			 addr & 0xffc, data, addr >> 16, (addr >> 12) & 0xf);
 
-		if (plugged)
-			helper->dpms(connector->encoder, DRM_MODE_DPMS_ON);
-		else
-			helper->dpms(connector->encoder, DRM_MODE_DPMS_OFF);
+		nv_wr32(dev, NV50_PDISPLAY_TRAPPED_ADDR(chid), 0x90000000);
 	}
-
-	drm_helper_hpd_irq_event(dev);
 }
 
-void
-nv50_display_irq_handler(struct drm_device *dev)
+static void
+nv50_display_isr(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	uint32_t delayed = 0;
 
-	if (nv_rd32(dev, NV50_PMC_INTR_0) & NV50_PMC_INTR_0_HOTPLUG) {
-		uint32_t hpd0_bits, hpd1_bits = 0;
-
-		hpd0_bits = nv_rd32(dev, 0xe054);
-		nv_wr32(dev, 0xe054, hpd0_bits);
-
-		if (dev_priv->chipset >= 0x90) {
-			hpd1_bits = nv_rd32(dev, 0xe074);
-			nv_wr32(dev, 0xe074, hpd1_bits);
-		}
-
-		spin_lock(&dev_priv->hpd_state.lock);
-		dev_priv->hpd_state.hpd0_bits |= hpd0_bits;
-		dev_priv->hpd_state.hpd1_bits |= hpd1_bits;
-		spin_unlock(&dev_priv->hpd_state.lock);
-
-		queue_work(dev_priv->wq, &dev_priv->hpd_work);
-	}
-
 	while (nv_rd32(dev, NV50_PMC_INTR_0) & NV50_PMC_INTR_0_DISPLAY) {
 		uint32_t intr0 = nv_rd32(dev, NV50_PDISPLAY_INTR_0);
 		uint32_t intr1 = nv_rd32(dev, NV50_PDISPLAY_INTR_1);
@@ -1123,9 +820,9 @@ nv50_display_irq_handler(struct drm_device *dev)
 		if (!intr0 && !(intr1 & ~delayed))
 			break;
 
-		if (intr0 & 0x00010000) {
+		if (intr0 & 0x001f0000) {
 			nv50_display_error_handler(dev);
-			intr0 &= ~0x00010000;
+			intr0 &= ~0x001f0000;
 		}
 
 		if (intr1 & NV50_PDISPLAY_INTR_1_VBLANK_CRTC) {
@@ -1156,4 +853,3 @@ nv50_display_irq_handler(struct drm_device *dev)
 		}
 	}
 }
-
diff --git a/drivers/gpu/drm/nouveau/nv50_display.h b/drivers/gpu/drm/nouveau/nv50_display.h
index c551f0b85ee0..f0e30b78ef6b 100644
--- a/drivers/gpu/drm/nouveau/nv50_display.h
+++ b/drivers/gpu/drm/nouveau/nv50_display.h
@@ -35,9 +35,7 @@
 #include "nouveau_crtc.h"
 #include "nv50_evo.h"
 
-void nv50_display_irq_handler(struct drm_device *dev);
 void nv50_display_irq_handler_bh(struct work_struct *work);
-void nv50_display_irq_hotplug_bh(struct work_struct *work);
 int nv50_display_early_init(struct drm_device *dev);
 void nv50_display_late_takedown(struct drm_device *dev);
 int nv50_display_create(struct drm_device *dev);
diff --git a/drivers/gpu/drm/nouveau/nv50_evo.c b/drivers/gpu/drm/nouveau/nv50_evo.c
new file mode 100644
index 000000000000..887b2a97e2a2
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_evo.c
@@ -0,0 +1,318 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+
+#include "nouveau_drv.h"
+#include "nouveau_dma.h"
+#include "nouveau_ramht.h"
+
+static void
+nv50_evo_channel_del(struct nouveau_channel **pevo)
+{
+	struct drm_nouveau_private *dev_priv;
+	struct nouveau_channel *evo = *pevo;
+
+	if (!evo)
+		return;
+	*pevo = NULL;
+
+	dev_priv = evo->dev->dev_private;
+	dev_priv->evo_alloc &= ~(1 << evo->id);
+
+	nouveau_gpuobj_channel_takedown(evo);
+	nouveau_bo_unmap(evo->pushbuf_bo);
+	nouveau_bo_ref(NULL, &evo->pushbuf_bo);
+
+	if (evo->user)
+		iounmap(evo->user);
+
+	kfree(evo);
+}
+
+int
+nv50_evo_dmaobj_new(struct nouveau_channel *evo, u32 class, u32 name,
+		    u32 tile_flags, u32 magic_flags, u32 offset, u32 limit)
+{
+	struct drm_nouveau_private *dev_priv = evo->dev->dev_private;
+	struct drm_device *dev = evo->dev;
+	struct nouveau_gpuobj *obj = NULL;
+	int ret;
+
+	ret = nouveau_gpuobj_new(dev, dev_priv->evo, 6*4, 32, 0, &obj);
+	if (ret)
+		return ret;
+	obj->engine = NVOBJ_ENGINE_DISPLAY;
+
+	nv_wo32(obj,  0, (tile_flags << 22) | (magic_flags << 16) | class);
+	nv_wo32(obj,  4, limit);
+	nv_wo32(obj,  8, offset);
+	nv_wo32(obj, 12, 0x00000000);
+	nv_wo32(obj, 16, 0x00000000);
+	if (dev_priv->card_type < NV_C0)
+		nv_wo32(obj, 20, 0x00010000);
+	else
+		nv_wo32(obj, 20, 0x00020000);
+	dev_priv->engine.instmem.flush(dev);
+
+	ret = nouveau_ramht_insert(evo, name, obj);
+	nouveau_gpuobj_ref(NULL, &obj);
+	if (ret) {
+		return ret;
+	}
+
+	return 0;
+}
+
+static int
+nv50_evo_channel_new(struct drm_device *dev, struct nouveau_channel **pevo)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *evo;
+	int ret;
+
+	evo = kzalloc(sizeof(struct nouveau_channel), GFP_KERNEL);
+	if (!evo)
+		return -ENOMEM;
+	*pevo = evo;
+
+	for (evo->id = 0; evo->id < 5; evo->id++) {
+		if (dev_priv->evo_alloc & (1 << evo->id))
+			continue;
+
+		dev_priv->evo_alloc |= (1 << evo->id);
+		break;
+	}
+
+	if (evo->id == 5) {
+		kfree(evo);
+		return -ENODEV;
+	}
+
+	evo->dev = dev;
+	evo->user_get = 4;
+	evo->user_put = 0;
+
+	ret = nouveau_bo_new(dev, NULL, 4096, 0, TTM_PL_FLAG_VRAM, 0, 0,
+			     false, true, &evo->pushbuf_bo);
+	if (ret == 0)
+		ret = nouveau_bo_pin(evo->pushbuf_bo, TTM_PL_FLAG_VRAM);
+	if (ret) {
+		NV_ERROR(dev, "Error creating EVO DMA push buffer: %d\n", ret);
+		nv50_evo_channel_del(pevo);
+		return ret;
+	}
+
+	ret = nouveau_bo_map(evo->pushbuf_bo);
+	if (ret) {
+		NV_ERROR(dev, "Error mapping EVO DMA push buffer: %d\n", ret);
+		nv50_evo_channel_del(pevo);
+		return ret;
+	}
+
+	evo->user = ioremap(pci_resource_start(dev->pdev, 0) +
+			    NV50_PDISPLAY_USER(evo->id), PAGE_SIZE);
+	if (!evo->user) {
+		NV_ERROR(dev, "Error mapping EVO control regs.\n");
+		nv50_evo_channel_del(pevo);
+		return -ENOMEM;
+	}
+
+	/* bind primary evo channel's ramht to the channel */
+	if (dev_priv->evo && evo != dev_priv->evo)
+		nouveau_ramht_ref(dev_priv->evo->ramht, &evo->ramht, NULL);
+
+	return 0;
+}
+
+static int
+nv50_evo_channel_init(struct nouveau_channel *evo)
+{
+	struct drm_device *dev = evo->dev;
+	int id = evo->id, ret, i;
+	u64 pushbuf = evo->pushbuf_bo->bo.mem.start << PAGE_SHIFT;
+	u32 tmp;
+
+	tmp = nv_rd32(dev, NV50_PDISPLAY_EVO_CTRL(id));
+	if ((tmp & 0x009f0000) == 0x00020000)
+		nv_wr32(dev, NV50_PDISPLAY_EVO_CTRL(id), tmp | 0x00800000);
+
+	tmp = nv_rd32(dev, NV50_PDISPLAY_EVO_CTRL(id));
+	if ((tmp & 0x003f0000) == 0x00030000)
+		nv_wr32(dev, NV50_PDISPLAY_EVO_CTRL(id), tmp | 0x00600000);
+
+	/* initialise fifo */
+	nv_wr32(dev, NV50_PDISPLAY_EVO_DMA_CB(id), pushbuf >> 8 |
+		     NV50_PDISPLAY_EVO_DMA_CB_LOCATION_VRAM |
+		     NV50_PDISPLAY_EVO_DMA_CB_VALID);
+	nv_wr32(dev, NV50_PDISPLAY_EVO_UNK2(id), 0x00010000);
+	nv_wr32(dev, NV50_PDISPLAY_EVO_HASH_TAG(id), id);
+	nv_mask(dev, NV50_PDISPLAY_EVO_CTRL(id), NV50_PDISPLAY_EVO_CTRL_DMA,
+		     NV50_PDISPLAY_EVO_CTRL_DMA_ENABLED);
+
+	nv_wr32(dev, NV50_PDISPLAY_USER_PUT(id), 0x00000000);
+	nv_wr32(dev, NV50_PDISPLAY_EVO_CTRL(id), 0x01000003 |
+		     NV50_PDISPLAY_EVO_CTRL_DMA_ENABLED);
+	if (!nv_wait(dev, NV50_PDISPLAY_EVO_CTRL(id), 0x80000000, 0x00000000)) {
+		NV_ERROR(dev, "EvoCh %d init timeout: 0x%08x\n", id,
+			 nv_rd32(dev, NV50_PDISPLAY_EVO_CTRL(id)));
+		return -EBUSY;
+	}
+
+	/* enable error reporting on the channel */
+	nv_mask(dev, 0x610028, 0x00000000, 0x00010001 << id);
+
+	evo->dma.max = (4096/4) - 2;
+	evo->dma.put = 0;
+	evo->dma.cur = evo->dma.put;
+	evo->dma.free = evo->dma.max - evo->dma.cur;
+
+	ret = RING_SPACE(evo, NOUVEAU_DMA_SKIPS);
+	if (ret)
+		return ret;
+
+	for (i = 0; i < NOUVEAU_DMA_SKIPS; i++)
+		OUT_RING(evo, 0);
+
+	return 0;
+}
+
+static void
+nv50_evo_channel_fini(struct nouveau_channel *evo)
+{
+	struct drm_device *dev = evo->dev;
+	int id = evo->id;
+
+	nv_mask(dev, 0x610028, 0x00010001 << id, 0x00000000);
+	nv_mask(dev, NV50_PDISPLAY_EVO_CTRL(id), 0x00001010, 0x00001000);
+	nv_wr32(dev, NV50_PDISPLAY_INTR_0, (1 << id));
+	nv_mask(dev, NV50_PDISPLAY_EVO_CTRL(id), 0x00000003, 0x00000000);
+	if (!nv_wait(dev, NV50_PDISPLAY_EVO_CTRL(id), 0x001e0000, 0x00000000)) {
+		NV_ERROR(dev, "EvoCh %d takedown timeout: 0x%08x\n", id,
+			 nv_rd32(dev, NV50_PDISPLAY_EVO_CTRL(id)));
+	}
+}
+
+static int
+nv50_evo_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramht = NULL;
+	struct nouveau_channel *evo;
+	int ret;
+
+	/* create primary evo channel, the one we use for modesetting
+	 * purporses
+	 */
+	ret = nv50_evo_channel_new(dev, &dev_priv->evo);
+	if (ret)
+		return ret;
+	evo = dev_priv->evo;
+
+	/* setup object management on it, any other evo channel will
+	 * use this also as there's no per-channel support on the
+	 * hardware
+	 */
+	ret = nouveau_gpuobj_new(dev, NULL, 32768, 65536,
+				 NVOBJ_FLAG_ZERO_ALLOC, &evo->ramin);
+	if (ret) {
+		NV_ERROR(dev, "Error allocating EVO channel memory: %d\n", ret);
+		nv50_evo_channel_del(&dev_priv->evo);
+		return ret;
+	}
+
+	ret = drm_mm_init(&evo->ramin_heap, 0, 32768);
+	if (ret) {
+		NV_ERROR(dev, "Error initialising EVO PRAMIN heap: %d\n", ret);
+		nv50_evo_channel_del(&dev_priv->evo);
+		return ret;
+	}
+
+	ret = nouveau_gpuobj_new(dev, evo, 4096, 16, 0, &ramht);
+	if (ret) {
+		NV_ERROR(dev, "Unable to allocate EVO RAMHT: %d\n", ret);
+		nv50_evo_channel_del(&dev_priv->evo);
+		return ret;
+	}
+
+	ret = nouveau_ramht_new(dev, ramht, &evo->ramht);
+	nouveau_gpuobj_ref(NULL, &ramht);
+	if (ret) {
+		nv50_evo_channel_del(&dev_priv->evo);
+		return ret;
+	}
+
+	/* create some default objects for the scanout memtypes we support */
+	if (dev_priv->chipset != 0x50) {
+		ret = nv50_evo_dmaobj_new(evo, 0x3d, NvEvoFB16, 0x70, 0x19,
+					  0, 0xffffffff);
+		if (ret) {
+			nv50_evo_channel_del(&dev_priv->evo);
+			return ret;
+		}
+
+
+		ret = nv50_evo_dmaobj_new(evo, 0x3d, NvEvoFB32, 0x7a, 0x19,
+					  0, 0xffffffff);
+		if (ret) {
+			nv50_evo_channel_del(&dev_priv->evo);
+			return ret;
+		}
+	}
+
+	ret = nv50_evo_dmaobj_new(evo, 0x3d, NvEvoVRAM, 0, 0x19,
+				  0, dev_priv->vram_size);
+	if (ret) {
+		nv50_evo_channel_del(&dev_priv->evo);
+		return ret;
+	}
+
+	return 0;
+}
+
+int
+nv50_evo_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int ret;
+
+	if (!dev_priv->evo) {
+		ret = nv50_evo_create(dev);
+		if (ret)
+			return ret;
+	}
+
+	return nv50_evo_channel_init(dev_priv->evo);
+}
+
+void
+nv50_evo_fini(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->evo) {
+		nv50_evo_channel_fini(dev_priv->evo);
+		nv50_evo_channel_del(&dev_priv->evo);
+	}
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_evo.h b/drivers/gpu/drm/nouveau/nv50_evo.h
index aae13343bcec..aa4f0d3cea8e 100644
--- a/drivers/gpu/drm/nouveau/nv50_evo.h
+++ b/drivers/gpu/drm/nouveau/nv50_evo.h
@@ -24,6 +24,15 @@
  *
  */
 
+#ifndef __NV50_EVO_H__
+#define __NV50_EVO_H__
+
+int  nv50_evo_init(struct drm_device *dev);
+void nv50_evo_fini(struct drm_device *dev);
+int  nv50_evo_dmaobj_new(struct nouveau_channel *, u32 class, u32 name,
+			 u32 tile_flags, u32 magic_flags,
+			 u32 offset, u32 limit);
+
 #define NV50_EVO_UPDATE                                              0x00000080
 #define NV50_EVO_UNK84                                               0x00000084
 #define NV50_EVO_UNK84_NOTIFY                                        0x40000000
@@ -111,3 +120,4 @@
 #define NV50_EVO_CRTC_SCALE_RES1                                     0x000008d8
 #define NV50_EVO_CRTC_SCALE_RES2                                     0x000008dc
 
+#endif
diff --git a/drivers/gpu/drm/nouveau/nv50_fb.c b/drivers/gpu/drm/nouveau/nv50_fb.c
index cd1988b15d2c..50290dea0ac4 100644
--- a/drivers/gpu/drm/nouveau/nv50_fb.c
+++ b/drivers/gpu/drm/nouveau/nv50_fb.c
@@ -3,30 +3,75 @@
 #include "nouveau_drv.h"
 #include "nouveau_drm.h"
 
+struct nv50_fb_priv {
+	struct page *r100c08_page;
+	dma_addr_t r100c08;
+};
+
+static int
+nv50_fb_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_fb_priv *priv;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	priv->r100c08_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	if (!priv->r100c08_page) {
+		kfree(priv);
+		return -ENOMEM;
+	}
+
+	priv->r100c08 = pci_map_page(dev->pdev, priv->r100c08_page, 0,
+				     PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
+	if (pci_dma_mapping_error(dev->pdev, priv->r100c08)) {
+		__free_page(priv->r100c08_page);
+		kfree(priv);
+		return -EFAULT;
+	}
+
+	dev_priv->engine.fb.priv = priv;
+	return 0;
+}
+
 int
 nv50_fb_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_fb_priv *priv;
+	int ret;
+
+	if (!dev_priv->engine.fb.priv) {
+		ret = nv50_fb_create(dev);
+		if (ret)
+			return ret;
+	}
+	priv = dev_priv->engine.fb.priv;
 
 	/* Not a clue what this is exactly.  Without pointing it at a
 	 * scratch page, VRAM->GART blits with M2MF (as in DDX DFS)
 	 * cause IOMMU "read from address 0" errors (rh#561267)
 	 */
-	nv_wr32(dev, 0x100c08, dev_priv->gart_info.sg_dummy_bus >> 8);
+	nv_wr32(dev, 0x100c08, priv->r100c08 >> 8);
 
 	/* This is needed to get meaningful information from 100c90
 	 * on traps. No idea what these values mean exactly. */
 	switch (dev_priv->chipset) {
 	case 0x50:
-		nv_wr32(dev, 0x100c90, 0x0707ff);
+		nv_wr32(dev, 0x100c90, 0x000707ff);
 		break;
 	case 0xa3:
 	case 0xa5:
 	case 0xa8:
-		nv_wr32(dev, 0x100c90, 0x0d0fff);
+		nv_wr32(dev, 0x100c90, 0x000d0fff);
+		break;
+	case 0xaf:
+		nv_wr32(dev, 0x100c90, 0x089d1fff);
 		break;
 	default:
-		nv_wr32(dev, 0x100c90, 0x1d07ff);
+		nv_wr32(dev, 0x100c90, 0x001d07ff);
 		break;
 	}
 
@@ -36,12 +81,25 @@ nv50_fb_init(struct drm_device *dev)
 void
 nv50_fb_takedown(struct drm_device *dev)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nv50_fb_priv *priv;
+
+	priv = dev_priv->engine.fb.priv;
+	if (!priv)
+		return;
+	dev_priv->engine.fb.priv = NULL;
+
+	pci_unmap_page(dev->pdev, priv->r100c08, PAGE_SIZE,
+		       PCI_DMA_BIDIRECTIONAL);
+	__free_page(priv->r100c08_page);
+	kfree(priv);
 }
 
 void
 nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	unsigned long flags;
 	u32 trap[6], idx, chinst;
 	int i, ch;
 
@@ -60,8 +118,10 @@ nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
 		return;
 
 	chinst = (trap[2] << 16) | trap[1];
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
 	for (ch = 0; ch < dev_priv->engine.fifo.channels; ch++) {
-		struct nouveau_channel *chan = dev_priv->fifos[ch];
+		struct nouveau_channel *chan = dev_priv->channels.ptr[ch];
 
 		if (!chan || !chan->ramin)
 			continue;
@@ -69,6 +129,7 @@ nv50_fb_vm_trap(struct drm_device *dev, int display, const char *name)
 		if (chinst == chan->ramin->vinst >> 12)
 			break;
 	}
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
 
 	NV_INFO(dev, "%s - VM: Trapped %s at %02x%04x%04x status %08x "
 		     "channel %d (0x%08x)\n",
diff --git a/drivers/gpu/drm/nouveau/nv50_fbcon.c b/drivers/gpu/drm/nouveau/nv50_fbcon.c
index 6dcf048eddbc..6d38cb1488ae 100644
--- a/drivers/gpu/drm/nouveau/nv50_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nv50_fbcon.c
@@ -3,27 +3,20 @@
 #include "nouveau_dma.h"
 #include "nouveau_ramht.h"
 #include "nouveau_fbcon.h"
+#include "nouveau_mm.h"
 
-void
+int
 nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) &&
-	     RING_SPACE(chan, rect->rop == ROP_COPY ? 7 : 11)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
-
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_fillrect(info, rect);
-		return;
-	}
+	ret = RING_SPACE(chan, rect->rop == ROP_COPY ? 7 : 11);
+	if (ret)
+		return ret;
 
 	if (rect->rop != ROP_COPY) {
 		BEGIN_RING(chan, NvSub2D, 0x02ac, 1);
@@ -45,27 +38,21 @@ nv50_fbcon_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 		OUT_RING(chan, 3);
 	}
 	FIRE_RING(chan);
+	return 0;
 }
 
-void
+int
 nv50_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) && RING_SPACE(chan, 12)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
-
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_copyarea(info, region);
-		return;
-	}
+	ret = RING_SPACE(chan, 12);
+	if (ret)
+		return ret;
 
 	BEGIN_RING(chan, NvSub2D, 0x0110, 1);
 	OUT_RING(chan, 0);
@@ -80,9 +67,10 @@ nv50_fbcon_copyarea(struct fb_info *info, const struct fb_copyarea *region)
 	OUT_RING(chan, 0);
 	OUT_RING(chan, region->sy);
 	FIRE_RING(chan);
+	return 0;
 }
 
-void
+int
 nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 {
 	struct nouveau_fbdev *nfbdev = info->par;
@@ -92,23 +80,14 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	uint32_t width, dwords, *data = (uint32_t *)image->data;
 	uint32_t mask = ~(~0 >> (32 - info->var.bits_per_pixel));
 	uint32_t *palette = info->pseudo_palette;
+	int ret;
 
-	if (info->state != FBINFO_STATE_RUNNING)
-		return;
-
-	if (image->depth != 1) {
-		cfb_imageblit(info, image);
-		return;
-	}
-
-	if (!(info->flags & FBINFO_HWACCEL_DISABLED) && RING_SPACE(chan, 11)) {
-		nouveau_fbcon_gpu_lockup(info);
-	}
+	if (image->depth != 1)
+		return -ENODEV;
 
-	if (info->flags & FBINFO_HWACCEL_DISABLED) {
-		cfb_imageblit(info, image);
-		return;
-	}
+	ret = RING_SPACE(chan, 11);
+	if (ret)
+		return ret;
 
 	width = ALIGN(image->width, 32);
 	dwords = (width * image->height) >> 5;
@@ -134,11 +113,9 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	while (dwords) {
 		int push = dwords > 2047 ? 2047 : dwords;
 
-		if (RING_SPACE(chan, push + 1)) {
-			nouveau_fbcon_gpu_lockup(info);
-			cfb_imageblit(info, image);
-			return;
-		}
+		ret = RING_SPACE(chan, push + 1);
+		if (ret)
+			return ret;
 
 		dwords -= push;
 
@@ -148,6 +125,7 @@ nv50_fbcon_imageblit(struct fb_info *info, const struct fb_image *image)
 	}
 
 	FIRE_RING(chan);
+	return 0;
 }
 
 int
@@ -157,12 +135,9 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	struct drm_device *dev = nfbdev->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nouveau_channel *chan = dev_priv->channel;
-	struct nouveau_gpuobj *eng2d = NULL;
-	uint64_t fb;
+	struct nouveau_bo *nvbo = nfbdev->nouveau_fb.nvbo;
 	int ret, format;
 
-	fb = info->fix.smem_start - dev_priv->fb_phys + dev_priv->vm_vram_base;
-
 	switch (info->var.bits_per_pixel) {
 	case 8:
 		format = 0xf3;
@@ -190,12 +165,7 @@ nv50_fbcon_accel_init(struct fb_info *info)
 		return -EINVAL;
 	}
 
-	ret = nouveau_gpuobj_gr_new(dev_priv->channel, 0x502d, &eng2d);
-	if (ret)
-		return ret;
-
-	ret = nouveau_ramht_insert(dev_priv->channel, Nv2D, eng2d);
-	nouveau_gpuobj_ref(NULL, &eng2d);
+	ret = nouveau_gpuobj_gr_new(dev_priv->channel, Nv2D, 0x502d);
 	if (ret)
 		return ret;
 
@@ -253,8 +223,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(fb));
-	OUT_RING(chan, lower_32_bits(fb));
+	OUT_RING(chan, upper_32_bits(nvbo->vma.offset));
+	OUT_RING(chan, lower_32_bits(nvbo->vma.offset));
 	BEGIN_RING(chan, NvSub2D, 0x0230, 2);
 	OUT_RING(chan, format);
 	OUT_RING(chan, 1);
@@ -262,8 +232,8 @@ nv50_fbcon_accel_init(struct fb_info *info)
 	OUT_RING(chan, info->fix.line_length);
 	OUT_RING(chan, info->var.xres_virtual);
 	OUT_RING(chan, info->var.yres_virtual);
-	OUT_RING(chan, upper_32_bits(fb));
-	OUT_RING(chan, lower_32_bits(fb));
+	OUT_RING(chan, upper_32_bits(nvbo->vma.offset));
+	OUT_RING(chan, lower_32_bits(nvbo->vma.offset));
 
 	return 0;
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_fifo.c b/drivers/gpu/drm/nouveau/nv50_fifo.c
index 1da65bd60c10..8dd04c5dac67 100644
--- a/drivers/gpu/drm/nouveau/nv50_fifo.c
+++ b/drivers/gpu/drm/nouveau/nv50_fifo.c
@@ -28,6 +28,7 @@
 #include "drm.h"
 #include "nouveau_drv.h"
 #include "nouveau_ramht.h"
+#include "nouveau_vm.h"
 
 static void
 nv50_fifo_playlist_update(struct drm_device *dev)
@@ -44,7 +45,8 @@ nv50_fifo_playlist_update(struct drm_device *dev)
 
 	/* We never schedule channel 0 or 127 */
 	for (i = 1, nr = 0; i < 127; i++) {
-		if (dev_priv->fifos[i] && dev_priv->fifos[i]->ramfc) {
+		if (dev_priv->channels.ptr[i] &&
+		    dev_priv->channels.ptr[i]->ramfc) {
 			nv_wo32(cur, (nr * 4), i);
 			nr++;
 		}
@@ -60,7 +62,7 @@ static void
 nv50_fifo_channel_enable(struct drm_device *dev, int channel)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_channel *chan = dev_priv->fifos[channel];
+	struct nouveau_channel *chan = dev_priv->channels.ptr[channel];
 	uint32_t inst;
 
 	NV_DEBUG(dev, "ch%d\n", channel);
@@ -105,6 +107,7 @@ nv50_fifo_init_intr(struct drm_device *dev)
 {
 	NV_DEBUG(dev, "\n");
 
+	nouveau_irq_register(dev, 8, nv04_fifo_isr);
 	nv_wr32(dev, NV03_PFIFO_INTR_0, 0xFFFFFFFF);
 	nv_wr32(dev, NV03_PFIFO_INTR_EN_0, 0xFFFFFFFF);
 }
@@ -118,7 +121,7 @@ nv50_fifo_init_context_table(struct drm_device *dev)
 	NV_DEBUG(dev, "\n");
 
 	for (i = 0; i < NV50_PFIFO_CTX_TABLE__SIZE; i++) {
-		if (dev_priv->fifos[i])
+		if (dev_priv->channels.ptr[i])
 			nv50_fifo_channel_enable(dev, i);
 		else
 			nv50_fifo_channel_disable(dev, i);
@@ -206,6 +209,9 @@ nv50_fifo_takedown(struct drm_device *dev)
 	if (!pfifo->playlist[0])
 		return;
 
+	nv_wr32(dev, 0x2140, 0x00000000);
+	nouveau_irq_unregister(dev, 8);
+
 	nouveau_gpuobj_ref(NULL, &pfifo->playlist[0]);
 	nouveau_gpuobj_ref(NULL, &pfifo->playlist[1]);
 }
@@ -256,6 +262,11 @@ nv50_fifo_create_context(struct nouveau_channel *chan)
 	}
 	ramfc = chan->ramfc;
 
+	chan->user = ioremap(pci_resource_start(dev->pdev, 0) +
+			     NV50_USER(chan->id), PAGE_SIZE);
+	if (!chan->user)
+		return -ENOMEM;
+
 	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
 
 	nv_wo32(ramfc, 0x48, chan->pushbuf->cinst >> 4);
@@ -291,10 +302,23 @@ void
 nv50_fifo_destroy_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
 	struct nouveau_gpuobj *ramfc = NULL;
+	unsigned long flags;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pfifo->reassign(dev, false);
+
+	/* Unload the context if it's the currently active one */
+	if (pfifo->channel_id(dev) == chan->id) {
+		pfifo->disable(dev);
+		pfifo->unload_context(dev);
+		pfifo->enable(dev);
+	}
+
 	/* This will ensure the channel is seen as disabled. */
 	nouveau_gpuobj_ref(chan->ramfc, &ramfc);
 	nouveau_gpuobj_ref(NULL, &chan->ramfc);
@@ -305,6 +329,14 @@ nv50_fifo_destroy_context(struct nouveau_channel *chan)
 		nv50_fifo_channel_disable(dev, 127);
 	nv50_fifo_playlist_update(dev);
 
+	pfifo->reassign(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
+	/* Free the channel resources */
+	if (chan->user) {
+		iounmap(chan->user);
+		chan->user = NULL;
+	}
 	nouveau_gpuobj_ref(NULL, &ramfc);
 	nouveau_gpuobj_ref(NULL, &chan->cache);
 }
@@ -392,7 +424,7 @@ nv50_fifo_unload_context(struct drm_device *dev)
 	if (chid < 1 || chid >= dev_priv->engine.fifo.channels - 1)
 		return 0;
 
-	chan = dev_priv->fifos[chid];
+	chan = dev_priv->channels.ptr[chid];
 	if (!chan) {
 		NV_ERROR(dev, "Inactive channel on PFIFO: %d\n", chid);
 		return -EINVAL;
@@ -467,5 +499,5 @@ nv50_fifo_unload_context(struct drm_device *dev)
 void
 nv50_fifo_tlb_flush(struct drm_device *dev)
 {
-	nv50_vm_flush(dev, 5);
+	nv50_vm_flush_engine(dev, 5);
 }
diff --git a/drivers/gpu/drm/nouveau/nv50_gpio.c b/drivers/gpu/drm/nouveau/nv50_gpio.c
index b2fab2bf3d61..6b149c0cc06d 100644
--- a/drivers/gpu/drm/nouveau/nv50_gpio.c
+++ b/drivers/gpu/drm/nouveau/nv50_gpio.c
@@ -26,6 +26,28 @@
 #include "nouveau_drv.h"
 #include "nouveau_hw.h"
 
+#include "nv50_display.h"
+
+static void nv50_gpio_isr(struct drm_device *dev);
+static void nv50_gpio_isr_bh(struct work_struct *work);
+
+struct nv50_gpio_priv {
+	struct list_head handlers;
+	spinlock_t lock;
+};
+
+struct nv50_gpio_handler {
+	struct drm_device *dev;
+	struct list_head head;
+	struct work_struct work;
+	bool inhibit;
+
+	struct dcb_gpio_entry *gpio;
+
+	void (*handler)(void *data, int state);
+	void *data;
+};
+
 static int
 nv50_gpio_location(struct dcb_gpio_entry *gpio, uint32_t *reg, uint32_t *shift)
 {
@@ -75,29 +97,123 @@ nv50_gpio_set(struct drm_device *dev, enum dcb_gpio_tag tag, int state)
 	return 0;
 }
 
+int
+nv50_gpio_irq_register(struct drm_device *dev, enum dcb_gpio_tag tag,
+		       void (*handler)(void *, int), void *data)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv = pgpio->priv;
+	struct nv50_gpio_handler *gpioh;
+	struct dcb_gpio_entry *gpio;
+	unsigned long flags;
+
+	gpio = nouveau_bios_gpio_entry(dev, tag);
+	if (!gpio)
+		return -ENOENT;
+
+	gpioh = kzalloc(sizeof(*gpioh), GFP_KERNEL);
+	if (!gpioh)
+		return -ENOMEM;
+
+	INIT_WORK(&gpioh->work, nv50_gpio_isr_bh);
+	gpioh->dev  = dev;
+	gpioh->gpio = gpio;
+	gpioh->handler = handler;
+	gpioh->data = data;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	list_add(&gpioh->head, &priv->handlers);
+	spin_unlock_irqrestore(&priv->lock, flags);
+	return 0;
+}
+
 void
-nv50_gpio_irq_enable(struct drm_device *dev, enum dcb_gpio_tag tag, bool on)
+nv50_gpio_irq_unregister(struct drm_device *dev, enum dcb_gpio_tag tag,
+			 void (*handler)(void *, int), void *data)
 {
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv = pgpio->priv;
+	struct nv50_gpio_handler *gpioh, *tmp;
 	struct dcb_gpio_entry *gpio;
-	u32 reg, mask;
+	unsigned long flags;
 
 	gpio = nouveau_bios_gpio_entry(dev, tag);
-	if (!gpio) {
-		NV_ERROR(dev, "gpio tag 0x%02x not found\n", tag);
+	if (!gpio)
 		return;
+
+	spin_lock_irqsave(&priv->lock, flags);
+	list_for_each_entry_safe(gpioh, tmp, &priv->handlers, head) {
+		if (gpioh->gpio != gpio ||
+		    gpioh->handler != handler ||
+		    gpioh->data != data)
+			continue;
+		list_del(&gpioh->head);
+		kfree(gpioh);
 	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+bool
+nv50_gpio_irq_enable(struct drm_device *dev, enum dcb_gpio_tag tag, bool on)
+{
+	struct dcb_gpio_entry *gpio;
+	u32 reg, mask;
+
+	gpio = nouveau_bios_gpio_entry(dev, tag);
+	if (!gpio)
+		return false;
 
 	reg  = gpio->line < 16 ? 0xe050 : 0xe070;
 	mask = 0x00010001 << (gpio->line & 0xf);
 
 	nv_wr32(dev, reg + 4, mask);
-	nv_mask(dev, reg + 0, mask, on ? mask : 0);
+	reg = nv_mask(dev, reg + 0, mask, on ? mask : 0);
+	return (reg & mask) == mask;
+}
+
+static int
+nv50_gpio_create(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv;
+
+	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
+	if (!priv)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&priv->handlers);
+	spin_lock_init(&priv->lock);
+	pgpio->priv = priv;
+	return 0;
+}
+
+static void
+nv50_gpio_destroy(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+
+	kfree(pgpio->priv);
+	pgpio->priv = NULL;
 }
 
 int
 nv50_gpio_init(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv;
+	int ret;
+
+	if (!pgpio->priv) {
+		ret = nv50_gpio_create(dev);
+		if (ret)
+			return ret;
+	}
+	priv = pgpio->priv;
 
 	/* disable, and ack any pending gpio interrupts */
 	nv_wr32(dev, 0xe050, 0x00000000);
@@ -107,5 +223,77 @@ nv50_gpio_init(struct drm_device *dev)
 		nv_wr32(dev, 0xe074, 0xffffffff);
 	}
 
+	nouveau_irq_register(dev, 21, nv50_gpio_isr);
 	return 0;
 }
+
+void
+nv50_gpio_fini(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	nv_wr32(dev, 0xe050, 0x00000000);
+	if (dev_priv->chipset >= 0x90)
+		nv_wr32(dev, 0xe070, 0x00000000);
+	nouveau_irq_unregister(dev, 21);
+
+	nv50_gpio_destroy(dev);
+}
+
+static void
+nv50_gpio_isr_bh(struct work_struct *work)
+{
+	struct nv50_gpio_handler *gpioh =
+		container_of(work, struct nv50_gpio_handler, work);
+	struct drm_nouveau_private *dev_priv = gpioh->dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv = pgpio->priv;
+	unsigned long flags;
+	int state;
+
+	state = pgpio->get(gpioh->dev, gpioh->gpio->tag);
+	if (state < 0)
+		return;
+
+	gpioh->handler(gpioh->data, state);
+
+	spin_lock_irqsave(&priv->lock, flags);
+	gpioh->inhibit = false;
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static void
+nv50_gpio_isr(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpio_engine *pgpio = &dev_priv->engine.gpio;
+	struct nv50_gpio_priv *priv = pgpio->priv;
+	struct nv50_gpio_handler *gpioh;
+	u32 intr0, intr1 = 0;
+	u32 hi, lo, ch;
+
+	intr0 = nv_rd32(dev, 0xe054) & nv_rd32(dev, 0xe050);
+	if (dev_priv->chipset >= 0x90)
+		intr1 = nv_rd32(dev, 0xe074) & nv_rd32(dev, 0xe070);
+
+	hi = (intr0 & 0x0000ffff) | (intr1 << 16);
+	lo = (intr0 >> 16) | (intr1 & 0xffff0000);
+	ch = hi | lo;
+
+	nv_wr32(dev, 0xe054, intr0);
+	if (dev_priv->chipset >= 0x90)
+		nv_wr32(dev, 0xe074, intr1);
+
+	spin_lock(&priv->lock);
+	list_for_each_entry(gpioh, &priv->handlers, head) {
+		if (!(ch & (1 << gpioh->gpio->line)))
+			continue;
+
+		if (gpioh->inhibit)
+			continue;
+		gpioh->inhibit = true;
+
+		queue_work(dev_priv->wq, &gpioh->work);
+	}
+	spin_unlock(&priv->lock);
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_graph.c b/drivers/gpu/drm/nouveau/nv50_graph.c
index 8b669d0af610..c510e74acf4d 100644
--- a/drivers/gpu/drm/nouveau/nv50_graph.c
+++ b/drivers/gpu/drm/nouveau/nv50_graph.c
@@ -29,6 +29,12 @@
 #include "nouveau_drv.h"
 #include "nouveau_ramht.h"
 #include "nouveau_grctx.h"
+#include "nouveau_dma.h"
+#include "nouveau_vm.h"
+#include "nv50_evo.h"
+
+static int  nv50_graph_register(struct drm_device *);
+static void nv50_graph_isr(struct drm_device *);
 
 static void
 nv50_graph_init_reset(struct drm_device *dev)
@@ -46,6 +52,7 @@ nv50_graph_init_intr(struct drm_device *dev)
 {
 	NV_DEBUG(dev, "\n");
 
+	nouveau_irq_register(dev, 12, nv50_graph_isr);
 	nv_wr32(dev, NV03_PGRAPH_INTR, 0xffffffff);
 	nv_wr32(dev, 0x400138, 0xffffffff);
 	nv_wr32(dev, NV40_PGRAPH_INTR_EN, 0xffffffff);
@@ -145,12 +152,15 @@ nv50_graph_init(struct drm_device *dev)
 	nv50_graph_init_reset(dev);
 	nv50_graph_init_regs__nv(dev);
 	nv50_graph_init_regs(dev);
-	nv50_graph_init_intr(dev);
 
 	ret = nv50_graph_init_ctxctl(dev);
 	if (ret)
 		return ret;
 
+	ret = nv50_graph_register(dev);
+	if (ret)
+		return ret;
+	nv50_graph_init_intr(dev);
 	return 0;
 }
 
@@ -158,6 +168,8 @@ void
 nv50_graph_takedown(struct drm_device *dev)
 {
 	NV_DEBUG(dev, "\n");
+	nv_wr32(dev, 0x40013c, 0x00000000);
+	nouveau_irq_unregister(dev, 12);
 }
 
 void
@@ -190,7 +202,7 @@ nv50_graph_channel(struct drm_device *dev)
 	inst = (inst & NV50_PGRAPH_CTXCTL_CUR_INSTANCE) << 12;
 
 	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
-		struct nouveau_channel *chan = dev_priv->fifos[i];
+		struct nouveau_channel *chan = dev_priv->channels.ptr[i];
 
 		if (chan && chan->ramin && chan->ramin->vinst == inst)
 			return chan;
@@ -211,7 +223,7 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
-	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0x1000,
+	ret = nouveau_gpuobj_new(dev, chan, pgraph->grctx_size, 0,
 				 NVOBJ_FLAG_ZERO_ALLOC |
 				 NVOBJ_FLAG_ZERO_FREE, &chan->ramin_grctx);
 	if (ret)
@@ -234,6 +246,7 @@ nv50_graph_create_context(struct nouveau_channel *chan)
 	nv_wo32(chan->ramin_grctx, 0x00000, chan->ramin->vinst >> 12);
 
 	dev_priv->engine.instmem.flush(dev);
+	atomic_inc(&chan->vm->pgraph_refs);
 	return 0;
 }
 
@@ -242,18 +255,31 @@ nv50_graph_destroy_context(struct nouveau_channel *chan)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
 	int i, hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20;
+	unsigned long flags;
 
 	NV_DEBUG(dev, "ch%d\n", chan->id);
 
 	if (!chan->ramin)
 		return;
 
+	spin_lock_irqsave(&dev_priv->context_switch_lock, flags);
+	pgraph->fifo_access(dev, false);
+
+	if (pgraph->channel(dev) == chan)
+		pgraph->unload_context(dev);
+
 	for (i = hdr; i < hdr + 24; i += 4)
 		nv_wo32(chan->ramin, i, 0);
 	dev_priv->engine.instmem.flush(dev);
 
+	pgraph->fifo_access(dev, true);
+	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
+
 	nouveau_gpuobj_ref(NULL, &chan->ramin_grctx);
+
+	atomic_dec(&chan->vm->pgraph_refs);
 }
 
 static int
@@ -306,7 +332,7 @@ nv50_graph_unload_context(struct drm_device *dev)
 	return 0;
 }
 
-void
+static void
 nv50_graph_context_switch(struct drm_device *dev)
 {
 	uint32_t inst;
@@ -322,8 +348,8 @@ nv50_graph_context_switch(struct drm_device *dev)
 }
 
 static int
-nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan, int grclass,
-			   int mthd, uint32_t data)
+nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan,
+			   u32 class, u32 mthd, u32 data)
 {
 	struct nouveau_gpuobj *gpuobj;
 
@@ -340,8 +366,8 @@ nv50_graph_nvsw_dma_vblsem(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan, int grclass,
-			      int mthd, uint32_t data)
+nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan,
+			      u32 class, u32 mthd, u32 data)
 {
 	if (nouveau_notifier_offset(chan->nvsw.vblsem, &data))
 		return -ERANGE;
@@ -351,16 +377,16 @@ nv50_graph_nvsw_vblsem_offset(struct nouveau_channel *chan, int grclass,
 }
 
 static int
-nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan, int grclass,
-				   int mthd, uint32_t data)
+nv50_graph_nvsw_vblsem_release_val(struct nouveau_channel *chan,
+				   u32 class, u32 mthd, u32 data)
 {
 	chan->nvsw.vblsem_rval = data;
 	return 0;
 }
 
 static int
-nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan, int grclass,
-			       int mthd, uint32_t data)
+nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan,
+			       u32 class, u32 mthd, u32 data)
 {
 	struct drm_device *dev = chan->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -368,45 +394,85 @@ nv50_graph_nvsw_vblsem_release(struct nouveau_channel *chan, int grclass,
 	if (!chan->nvsw.vblsem || chan->nvsw.vblsem_offset == ~0 || data > 1)
 		return -EINVAL;
 
-	if (!(nv_rd32(dev, NV50_PDISPLAY_INTR_EN) &
-		      NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_(data))) {
-		nv_wr32(dev, NV50_PDISPLAY_INTR_1,
-			NV50_PDISPLAY_INTR_1_VBLANK_CRTC_(data));
-		nv_wr32(dev, NV50_PDISPLAY_INTR_EN, nv_rd32(dev,
-			NV50_PDISPLAY_INTR_EN) |
-			NV50_PDISPLAY_INTR_EN_VBLANK_CRTC_(data));
-	}
+	drm_vblank_get(dev, data);
 
+	chan->nvsw.vblsem_head = data;
 	list_add(&chan->nvsw.vbl_wait, &dev_priv->vbl_waiting);
+
 	return 0;
 }
 
-static struct nouveau_pgraph_object_method nv50_graph_nvsw_methods[] = {
-	{ 0x018c, nv50_graph_nvsw_dma_vblsem },
-	{ 0x0400, nv50_graph_nvsw_vblsem_offset },
-	{ 0x0404, nv50_graph_nvsw_vblsem_release_val },
-	{ 0x0408, nv50_graph_nvsw_vblsem_release },
-	{}
-};
+static int
+nv50_graph_nvsw_mthd_page_flip(struct nouveau_channel *chan,
+			       u32 class, u32 mthd, u32 data)
+{
+	struct nouveau_page_flip_state s;
 
-struct nouveau_pgraph_object_class nv50_graph_grclass[] = {
-	{ 0x506e, true, nv50_graph_nvsw_methods }, /* nvsw */
-	{ 0x0030, false, NULL }, /* null */
-	{ 0x5039, false, NULL }, /* m2mf */
-	{ 0x502d, false, NULL }, /* 2d */
-	{ 0x50c0, false, NULL }, /* compute */
-	{ 0x85c0, false, NULL }, /* compute (nva3, nva5, nva8) */
-	{ 0x5097, false, NULL }, /* tesla (nv50) */
-	{ 0x8297, false, NULL }, /* tesla (nv8x/nv9x) */
-	{ 0x8397, false, NULL }, /* tesla (nva0, nvaa, nvac) */
-	{ 0x8597, false, NULL }, /* tesla (nva3, nva5, nva8) */
-	{}
-};
+	if (!nouveau_finish_page_flip(chan, &s)) {
+		/* XXX - Do something here */
+	}
+
+	return 0;
+}
+
+static int
+nv50_graph_register(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if (dev_priv->engine.graph.registered)
+		return 0;
+
+	NVOBJ_CLASS(dev, 0x506e, SW); /* nvsw */
+	NVOBJ_MTHD (dev, 0x506e, 0x018c, nv50_graph_nvsw_dma_vblsem);
+	NVOBJ_MTHD (dev, 0x506e, 0x0400, nv50_graph_nvsw_vblsem_offset);
+	NVOBJ_MTHD (dev, 0x506e, 0x0404, nv50_graph_nvsw_vblsem_release_val);
+	NVOBJ_MTHD (dev, 0x506e, 0x0408, nv50_graph_nvsw_vblsem_release);
+	NVOBJ_MTHD (dev, 0x506e, 0x0500, nv50_graph_nvsw_mthd_page_flip);
+
+	NVOBJ_CLASS(dev, 0x0030, GR); /* null */
+	NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */
+	NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */
+
+	/* tesla */
+	if (dev_priv->chipset == 0x50)
+		NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */
+	else
+	if (dev_priv->chipset < 0xa0)
+		NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */
+	else {
+		switch (dev_priv->chipset) {
+		case 0xa0:
+		case 0xaa:
+		case 0xac:
+			NVOBJ_CLASS(dev, 0x8397, GR);
+			break;
+		case 0xa3:
+		case 0xa5:
+		case 0xa8:
+			NVOBJ_CLASS(dev, 0x8597, GR);
+			break;
+		case 0xaf:
+			NVOBJ_CLASS(dev, 0x8697, GR);
+			break;
+		}
+	}
+
+	/* compute */
+	NVOBJ_CLASS(dev, 0x50c0, GR);
+	if (dev_priv->chipset  > 0xa0 &&
+	    dev_priv->chipset != 0xaa &&
+	    dev_priv->chipset != 0xac)
+		NVOBJ_CLASS(dev, 0x85c0, GR);
+
+	dev_priv->engine.graph.registered = true;
+	return 0;
+}
 
 void
 nv50_graph_tlb_flush(struct drm_device *dev)
 {
-	nv50_vm_flush(dev, 0);
+	nv50_vm_flush_engine(dev, 0);
 }
 
 void
@@ -449,8 +515,500 @@ nv86_graph_tlb_flush(struct drm_device *dev)
 			 nv_rd32(dev, 0x400384), nv_rd32(dev, 0x400388));
 	}
 
-	nv50_vm_flush(dev, 0);
+	nv50_vm_flush_engine(dev, 0);
 
 	nv_mask(dev, 0x400500, 0x00000001, 0x00000001);
 	spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags);
 }
+
+static struct nouveau_enum nv50_mp_exec_error_names[] =
+{
+	{ 3, "STACK_UNDERFLOW" },
+	{ 4, "QUADON_ACTIVE" },
+	{ 8, "TIMEOUT" },
+	{ 0x10, "INVALID_OPCODE" },
+	{ 0x40, "BREAKPOINT" },
+	{}
+};
+
+static struct nouveau_bitfield nv50_graph_trap_m2mf[] = {
+	{ 0x00000001, "NOTIFY" },
+	{ 0x00000002, "IN" },
+	{ 0x00000004, "OUT" },
+	{}
+};
+
+static struct nouveau_bitfield nv50_graph_trap_vfetch[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+static struct nouveau_bitfield nv50_graph_trap_strmout[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+static struct nouveau_bitfield nv50_graph_trap_ccache[] = {
+	{ 0x00000001, "FAULT" },
+	{}
+};
+
+/* There must be a *lot* of these. Will take some time to gather them up. */
+static struct nouveau_enum nv50_data_error_names[] = {
+	{ 4,	"INVALID_VALUE" },
+	{ 5,	"INVALID_ENUM" },
+	{ 8,	"INVALID_OBJECT" },
+	{ 0xc,	"INVALID_BITFIELD" },
+	{ 0x28,	"MP_NO_REG_SPACE" },
+	{ 0x2b,	"MP_BLOCK_SIZE_MISMATCH" },
+	{}
+};
+
+static struct nouveau_bitfield nv50_graph_intr[] = {
+	{ 0x00000001, "NOTIFY" },
+	{ 0x00000002, "COMPUTE_QUERY" },
+	{ 0x00000010, "ILLEGAL_MTHD" },
+	{ 0x00000020, "ILLEGAL_CLASS" },
+	{ 0x00000040, "DOUBLE_NOTIFY" },
+	{ 0x00001000, "CONTEXT_SWITCH" },
+	{ 0x00010000, "BUFFER_NOTIFY" },
+	{ 0x00100000, "DATA_ERROR" },
+	{ 0x00200000, "TRAP" },
+	{ 0x01000000, "SINGLE_STEP" },
+	{}
+};
+
+static void
+nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	uint32_t units = nv_rd32(dev, 0x1540);
+	uint32_t addr, mp10, status, pc, oplow, ophigh;
+	int i;
+	int mps = 0;
+	for (i = 0; i < 4; i++) {
+		if (!(units & 1 << (i+24)))
+			continue;
+		if (dev_priv->chipset < 0xa0)
+			addr = 0x408200 + (tpid << 12) + (i << 7);
+		else
+			addr = 0x408100 + (tpid << 11) + (i << 7);
+		mp10 = nv_rd32(dev, addr + 0x10);
+		status = nv_rd32(dev, addr + 0x14);
+		if (!status)
+			continue;
+		if (display) {
+			nv_rd32(dev, addr + 0x20);
+			pc = nv_rd32(dev, addr + 0x24);
+			oplow = nv_rd32(dev, addr + 0x70);
+			ophigh= nv_rd32(dev, addr + 0x74);
+			NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - "
+					"TP %d MP %d: ", tpid, i);
+			nouveau_enum_print(nv50_mp_exec_error_names, status);
+			printk(" at %06x warp %d, opcode %08x %08x\n",
+					pc&0xffffff, pc >> 24,
+					oplow, ophigh);
+		}
+		nv_wr32(dev, addr + 0x10, mp10);
+		nv_wr32(dev, addr + 0x14, 0);
+		mps++;
+	}
+	if (!mps && display)
+		NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - TP %d: "
+				"No MPs claiming errors?\n", tpid);
+}
+
+static void
+nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old,
+		uint32_t ustatus_new, int display, const char *name)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int tps = 0;
+	uint32_t units = nv_rd32(dev, 0x1540);
+	int i, r;
+	uint32_t ustatus_addr, ustatus;
+	for (i = 0; i < 16; i++) {
+		if (!(units & (1 << i)))
+			continue;
+		if (dev_priv->chipset < 0xa0)
+			ustatus_addr = ustatus_old + (i << 12);
+		else
+			ustatus_addr = ustatus_new + (i << 11);
+		ustatus = nv_rd32(dev, ustatus_addr) & 0x7fffffff;
+		if (!ustatus)
+			continue;
+		tps++;
+		switch (type) {
+		case 6: /* texture error... unknown for now */
+			nv50_fb_vm_trap(dev, display, name);
+			if (display) {
+				NV_ERROR(dev, "magic set %d:\n", i);
+				for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4)
+					NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r,
+						nv_rd32(dev, r));
+			}
+			break;
+		case 7: /* MP error */
+			if (ustatus & 0x00010000) {
+				nv50_pgraph_mp_trap(dev, i, display);
+				ustatus &= ~0x00010000;
+			}
+			break;
+		case 8: /* TPDMA error */
+			{
+			uint32_t e0c = nv_rd32(dev, ustatus_addr + 4);
+			uint32_t e10 = nv_rd32(dev, ustatus_addr + 8);
+			uint32_t e14 = nv_rd32(dev, ustatus_addr + 0xc);
+			uint32_t e18 = nv_rd32(dev, ustatus_addr + 0x10);
+			uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14);
+			uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18);
+			uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c);
+			nv50_fb_vm_trap(dev, display, name);
+			/* 2d engine destination */
+			if (ustatus & 0x00000010) {
+				if (display) {
+					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n",
+							i, e14, e10);
+					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000010;
+			}
+			/* Render target */
+			if (ustatus & 0x00000040) {
+				if (display) {
+					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n",
+							i, e14, e10);
+					NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000040;
+			}
+			/* CUDA memory: l[], g[] or stack. */
+			if (ustatus & 0x00000080) {
+				if (display) {
+					if (e18 & 0x80000000) {
+						/* g[] read fault? */
+						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n",
+								i, e14, e10 | ((e18 >> 24) & 0x1f));
+						e18 &= ~0x1f000000;
+					} else if (e18 & 0xc) {
+						/* g[] write fault? */
+						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n",
+								i, e14, e10 | ((e18 >> 7) & 0x1f));
+						e18 &= ~0x00000f80;
+					} else {
+						NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n",
+								i, e14, e10);
+					}
+					NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n",
+							i, e0c, e18, e1c, e20, e24);
+				}
+				ustatus &= ~0x00000080;
+			}
+			}
+			break;
+		}
+		if (ustatus) {
+			if (display)
+				NV_INFO(dev, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus);
+		}
+		nv_wr32(dev, ustatus_addr, 0xc0000000);
+	}
+
+	if (!tps && display)
+		NV_INFO(dev, "%s - No TPs claiming errors?\n", name);
+}
+
+static int
+nv50_pgraph_trap_handler(struct drm_device *dev, u32 display, u64 inst, u32 chid)
+{
+	u32 status = nv_rd32(dev, 0x400108);
+	u32 ustatus;
+
+	if (!status && display) {
+		NV_INFO(dev, "PGRAPH - TRAP: no units reporting traps?\n");
+		return 1;
+	}
+
+	/* DISPATCH: Relays commands to other units and handles NOTIFY,
+	 * COND, QUERY. If you get a trap from it, the command is still stuck
+	 * in DISPATCH and you need to do something about it. */
+	if (status & 0x001) {
+		ustatus = nv_rd32(dev, 0x400804) & 0x7fffffff;
+		if (!ustatus && display) {
+			NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - no ustatus?\n");
+		}
+
+		nv_wr32(dev, 0x400500, 0x00000000);
+
+		/* Known to be triggered by screwed up NOTIFY and COND... */
+		if (ustatus & 0x00000001) {
+			u32 addr = nv_rd32(dev, 0x400808);
+			u32 subc = (addr & 0x00070000) >> 16;
+			u32 mthd = (addr & 0x00001ffc);
+			u32 datal = nv_rd32(dev, 0x40080c);
+			u32 datah = nv_rd32(dev, 0x400810);
+			u32 class = nv_rd32(dev, 0x400814);
+			u32 r848 = nv_rd32(dev, 0x400848);
+
+			NV_INFO(dev, "PGRAPH - TRAP DISPATCH_FAULT\n");
+			if (display && (addr & 0x80000000)) {
+				NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) "
+					     "subc %d class 0x%04x mthd 0x%04x "
+					     "data 0x%08x%08x "
+					     "400808 0x%08x 400848 0x%08x\n",
+					chid, inst, subc, class, mthd, datah,
+					datal, addr, r848);
+			} else
+			if (display) {
+				NV_INFO(dev, "PGRAPH - no stuck command?\n");
+			}
+
+			nv_wr32(dev, 0x400808, 0);
+			nv_wr32(dev, 0x4008e8, nv_rd32(dev, 0x4008e8) & 3);
+			nv_wr32(dev, 0x400848, 0);
+			ustatus &= ~0x00000001;
+		}
+
+		if (ustatus & 0x00000002) {
+			u32 addr = nv_rd32(dev, 0x40084c);
+			u32 subc = (addr & 0x00070000) >> 16;
+			u32 mthd = (addr & 0x00001ffc);
+			u32 data = nv_rd32(dev, 0x40085c);
+			u32 class = nv_rd32(dev, 0x400814);
+
+			NV_INFO(dev, "PGRAPH - TRAP DISPATCH_QUERY\n");
+			if (display && (addr & 0x80000000)) {
+				NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) "
+					     "subc %d class 0x%04x mthd 0x%04x "
+					     "data 0x%08x 40084c 0x%08x\n",
+					chid, inst, subc, class, mthd,
+					data, addr);
+			} else
+			if (display) {
+				NV_INFO(dev, "PGRAPH - no stuck command?\n");
+			}
+
+			nv_wr32(dev, 0x40084c, 0);
+			ustatus &= ~0x00000002;
+		}
+
+		if (ustatus && display) {
+			NV_INFO(dev, "PGRAPH - TRAP_DISPATCH (unknown "
+				      "0x%08x)\n", ustatus);
+		}
+
+		nv_wr32(dev, 0x400804, 0xc0000000);
+		nv_wr32(dev, 0x400108, 0x001);
+		status &= ~0x001;
+		if (!status)
+			return 0;
+	}
+
+	/* M2MF: Memory to memory copy engine. */
+	if (status & 0x002) {
+		u32 ustatus = nv_rd32(dev, 0x406800) & 0x7fffffff;
+		if (display) {
+			NV_INFO(dev, "PGRAPH - TRAP_M2MF");
+			nouveau_bitfield_print(nv50_graph_trap_m2mf, ustatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - TRAP_M2MF %08x %08x %08x %08x\n",
+				nv_rd32(dev, 0x406804), nv_rd32(dev, 0x406808),
+				nv_rd32(dev, 0x40680c), nv_rd32(dev, 0x406810));
+
+		}
+
+		/* No sane way found yet -- just reset the bugger. */
+		nv_wr32(dev, 0x400040, 2);
+		nv_wr32(dev, 0x400040, 0);
+		nv_wr32(dev, 0x406800, 0xc0000000);
+		nv_wr32(dev, 0x400108, 0x002);
+		status &= ~0x002;
+	}
+
+	/* VFETCH: Fetches data from vertex buffers. */
+	if (status & 0x004) {
+		u32 ustatus = nv_rd32(dev, 0x400c04) & 0x7fffffff;
+		if (display) {
+			NV_INFO(dev, "PGRAPH - TRAP_VFETCH");
+			nouveau_bitfield_print(nv50_graph_trap_vfetch, ustatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - TRAP_VFETCH %08x %08x %08x %08x\n",
+				nv_rd32(dev, 0x400c00), nv_rd32(dev, 0x400c08),
+				nv_rd32(dev, 0x400c0c), nv_rd32(dev, 0x400c10));
+		}
+
+		nv_wr32(dev, 0x400c04, 0xc0000000);
+		nv_wr32(dev, 0x400108, 0x004);
+		status &= ~0x004;
+	}
+
+	/* STRMOUT: DirectX streamout / OpenGL transform feedback. */
+	if (status & 0x008) {
+		ustatus = nv_rd32(dev, 0x401800) & 0x7fffffff;
+		if (display) {
+			NV_INFO(dev, "PGRAPH - TRAP_STRMOUT");
+			nouveau_bitfield_print(nv50_graph_trap_strmout, ustatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - TRAP_STRMOUT %08x %08x %08x %08x\n",
+				nv_rd32(dev, 0x401804), nv_rd32(dev, 0x401808),
+				nv_rd32(dev, 0x40180c), nv_rd32(dev, 0x401810));
+
+		}
+
+		/* No sane way found yet -- just reset the bugger. */
+		nv_wr32(dev, 0x400040, 0x80);
+		nv_wr32(dev, 0x400040, 0);
+		nv_wr32(dev, 0x401800, 0xc0000000);
+		nv_wr32(dev, 0x400108, 0x008);
+		status &= ~0x008;
+	}
+
+	/* CCACHE: Handles code and c[] caches and fills them. */
+	if (status & 0x010) {
+		ustatus = nv_rd32(dev, 0x405018) & 0x7fffffff;
+		if (display) {
+			NV_INFO(dev, "PGRAPH - TRAP_CCACHE");
+			nouveau_bitfield_print(nv50_graph_trap_ccache, ustatus);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - TRAP_CCACHE %08x %08x %08x %08x"
+				     " %08x %08x %08x\n",
+				nv_rd32(dev, 0x405800), nv_rd32(dev, 0x405804),
+				nv_rd32(dev, 0x405808), nv_rd32(dev, 0x40580c),
+				nv_rd32(dev, 0x405810), nv_rd32(dev, 0x405814),
+				nv_rd32(dev, 0x40581c));
+
+		}
+
+		nv_wr32(dev, 0x405018, 0xc0000000);
+		nv_wr32(dev, 0x400108, 0x010);
+		status &= ~0x010;
+	}
+
+	/* Unknown, not seen yet... 0x402000 is the only trap status reg
+	 * remaining, so try to handle it anyway. Perhaps related to that
+	 * unknown DMA slot on tesla? */
+	if (status & 0x20) {
+		ustatus = nv_rd32(dev, 0x402000) & 0x7fffffff;
+		if (display)
+			NV_INFO(dev, "PGRAPH - TRAP_UNKC04 0x%08x\n", ustatus);
+		nv_wr32(dev, 0x402000, 0xc0000000);
+		/* no status modifiction on purpose */
+	}
+
+	/* TEXTURE: CUDA texturing units */
+	if (status & 0x040) {
+		nv50_pgraph_tp_trap(dev, 6, 0x408900, 0x408600, display,
+				    "PGRAPH - TRAP_TEXTURE");
+		nv_wr32(dev, 0x400108, 0x040);
+		status &= ~0x040;
+	}
+
+	/* MP: CUDA execution engines. */
+	if (status & 0x080) {
+		nv50_pgraph_tp_trap(dev, 7, 0x408314, 0x40831c, display,
+				    "PGRAPH - TRAP_MP");
+		nv_wr32(dev, 0x400108, 0x080);
+		status &= ~0x080;
+	}
+
+	/* TPDMA:  Handles TP-initiated uncached memory accesses:
+	 * l[], g[], stack, 2d surfaces, render targets. */
+	if (status & 0x100) {
+		nv50_pgraph_tp_trap(dev, 8, 0x408e08, 0x408708, display,
+				    "PGRAPH - TRAP_TPDMA");
+		nv_wr32(dev, 0x400108, 0x100);
+		status &= ~0x100;
+	}
+
+	if (status) {
+		if (display)
+			NV_INFO(dev, "PGRAPH - TRAP: unknown 0x%08x\n", status);
+		nv_wr32(dev, 0x400108, status);
+	}
+
+	return 1;
+}
+
+static int
+nv50_graph_isr_chid(struct drm_device *dev, u64 inst)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_channel *chan;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dev_priv->channels.lock, flags);
+	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
+		chan = dev_priv->channels.ptr[i];
+		if (!chan || !chan->ramin)
+			continue;
+
+		if (inst == chan->ramin->vinst)
+			break;
+	}
+	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
+	return i;
+}
+
+static void
+nv50_graph_isr(struct drm_device *dev)
+{
+	u32 stat;
+
+	while ((stat = nv_rd32(dev, 0x400100))) {
+		u64 inst = (u64)(nv_rd32(dev, 0x40032c) & 0x0fffffff) << 12;
+		u32 chid = nv50_graph_isr_chid(dev, inst);
+		u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR);
+		u32 subc = (addr & 0x00070000) >> 16;
+		u32 mthd = (addr & 0x00001ffc);
+		u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA);
+		u32 class = nv_rd32(dev, 0x400814);
+		u32 show = stat;
+
+		if (stat & 0x00000010) {
+			if (!nouveau_gpuobj_mthd_call2(dev, chid, class,
+						       mthd, data))
+				show &= ~0x00000010;
+		}
+
+		if (stat & 0x00001000) {
+			nv_wr32(dev, 0x400500, 0x00000000);
+			nv_wr32(dev, 0x400100, 0x00001000);
+			nv_mask(dev, 0x40013c, 0x00001000, 0x00000000);
+			nv50_graph_context_switch(dev);
+			stat &= ~0x00001000;
+			show &= ~0x00001000;
+		}
+
+		show = (show && nouveau_ratelimit()) ? show : 0;
+
+		if (show & 0x00100000) {
+			u32 ecode = nv_rd32(dev, 0x400110);
+			NV_INFO(dev, "PGRAPH - DATA_ERROR ");
+			nouveau_enum_print(nv50_data_error_names, ecode);
+			printk("\n");
+		}
+
+		if (stat & 0x00200000) {
+			if (!nv50_pgraph_trap_handler(dev, show, inst, chid))
+				show &= ~0x00200000;
+		}
+
+		nv_wr32(dev, 0x400100, stat);
+		nv_wr32(dev, 0x400500, 0x00010001);
+
+		if (show) {
+			NV_INFO(dev, "PGRAPH -");
+			nouveau_bitfield_print(nv50_graph_intr, show);
+			printk("\n");
+			NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) subc %d "
+				     "class 0x%04x mthd 0x%04x data 0x%08x\n",
+				chid, inst, subc, class, mthd, data);
+		}
+	}
+
+	if (nv_rd32(dev, 0x400824) & (1 << 31))
+		nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31));
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_instmem.c b/drivers/gpu/drm/nouveau/nv50_instmem.c
index b773229b7647..adac4da98f7e 100644
--- a/drivers/gpu/drm/nouveau/nv50_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv50_instmem.c
@@ -27,14 +27,20 @@
 
 #include "drmP.h"
 #include "drm.h"
+
 #include "nouveau_drv.h"
+#include "nouveau_vm.h"
+
+#define BAR1_VM_BASE 0x0020000000ULL
+#define BAR1_VM_SIZE pci_resource_len(dev->pdev, 1)
+#define BAR3_VM_BASE 0x0000000000ULL
+#define BAR3_VM_SIZE pci_resource_len(dev->pdev, 3)
 
 struct nv50_instmem_priv {
 	uint32_t save1700[5]; /* 0x1700->0x1710 */
 
-	struct nouveau_gpuobj *pramin_pt;
-	struct nouveau_gpuobj *pramin_bar;
-	struct nouveau_gpuobj *fb_bar;
+	struct nouveau_gpuobj *bar1_dmaobj;
+	struct nouveau_gpuobj *bar3_dmaobj;
 };
 
 static void
@@ -48,6 +54,7 @@ nv50_channel_del(struct nouveau_channel **pchan)
 		return;
 
 	nouveau_gpuobj_ref(NULL, &chan->ramfc);
+	nouveau_vm_ref(NULL, &chan->vm, chan->vm_pd);
 	nouveau_gpuobj_ref(NULL, &chan->vm_pd);
 	if (chan->ramin_heap.free_stack.next)
 		drm_mm_takedown(&chan->ramin_heap);
@@ -56,14 +63,14 @@ nv50_channel_del(struct nouveau_channel **pchan)
 }
 
 static int
-nv50_channel_new(struct drm_device *dev, u32 size,
+nv50_channel_new(struct drm_device *dev, u32 size, struct nouveau_vm *vm,
 		 struct nouveau_channel **pchan)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	u32 pgd = (dev_priv->chipset == 0x50) ? 0x1400 : 0x0200;
 	u32  fc = (dev_priv->chipset == 0x50) ? 0x0000 : 0x4200;
 	struct nouveau_channel *chan;
-	int ret;
+	int ret, i;
 
 	chan = kzalloc(sizeof(*chan), GFP_KERNEL);
 	if (!chan)
@@ -92,6 +99,17 @@ nv50_channel_new(struct drm_device *dev, u32 size,
 		return ret;
 	}
 
+	for (i = 0; i < 0x4000; i += 8) {
+		nv_wo32(chan->vm_pd, i + 0, 0x00000000);
+		nv_wo32(chan->vm_pd, i + 4, 0xdeadcafe);
+	}
+
+	ret = nouveau_vm_ref(vm, &chan->vm, chan->vm_pd);
+	if (ret) {
+		nv50_channel_del(&chan);
+		return ret;
+	}
+
 	ret = nouveau_gpuobj_new_fake(dev, chan->ramin->pinst == ~0 ? ~0 :
 				      chan->ramin->pinst + fc,
 				      chan->ramin->vinst + fc, 0x100,
@@ -111,6 +129,7 @@ nv50_instmem_init(struct drm_device *dev)
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv;
 	struct nouveau_channel *chan;
+	struct nouveau_vm *vm;
 	int ret, i;
 	u32 tmp;
 
@@ -127,112 +146,89 @@ nv50_instmem_init(struct drm_device *dev)
 	ret = drm_mm_init(&dev_priv->ramin_heap, 0, dev_priv->ramin_size);
 	if (ret) {
 		NV_ERROR(dev, "Failed to init RAMIN heap\n");
-		return -ENOMEM;
+		goto error;
 	}
 
-	/* we need a channel to plug into the hw to control the BARs */
-	ret = nv50_channel_new(dev, 128*1024, &dev_priv->fifos[0]);
+	/* BAR3 */
+	ret = nouveau_vm_new(dev, BAR3_VM_BASE, BAR3_VM_SIZE, BAR3_VM_BASE,
+			     29, 12, 16, &dev_priv->bar3_vm);
 	if (ret)
-		return ret;
-	chan = dev_priv->fifos[127] = dev_priv->fifos[0];
+		goto error;
 
-	/* allocate page table for PRAMIN BAR */
-	ret = nouveau_gpuobj_new(dev, chan, (dev_priv->ramin_size >> 12) * 8,
-				 0x1000, NVOBJ_FLAG_ZERO_ALLOC,
-				 &priv->pramin_pt);
+	ret = nouveau_gpuobj_new(dev, NULL, (BAR3_VM_SIZE >> 12) * 8,
+				 0x1000, NVOBJ_FLAG_DONT_MAP |
+				 NVOBJ_FLAG_ZERO_ALLOC,
+				 &dev_priv->bar3_vm->pgt[0].obj);
 	if (ret)
-		return ret;
+		goto error;
+	dev_priv->bar3_vm->pgt[0].page_shift = 12;
+	dev_priv->bar3_vm->pgt[0].refcount = 1;
 
-	nv_wo32(chan->vm_pd, 0x0000, priv->pramin_pt->vinst | 0x63);
-	nv_wo32(chan->vm_pd, 0x0004, 0);
+	nv50_instmem_map(dev_priv->bar3_vm->pgt[0].obj);
 
-	/* DMA object for PRAMIN BAR */
-	ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->pramin_bar);
+	ret = nv50_channel_new(dev, 128 * 1024, dev_priv->bar3_vm, &chan);
 	if (ret)
-		return ret;
-	nv_wo32(priv->pramin_bar, 0x00, 0x7fc00000);
-	nv_wo32(priv->pramin_bar, 0x04, dev_priv->ramin_size - 1);
-	nv_wo32(priv->pramin_bar, 0x08, 0x00000000);
-	nv_wo32(priv->pramin_bar, 0x0c, 0x00000000);
-	nv_wo32(priv->pramin_bar, 0x10, 0x00000000);
-	nv_wo32(priv->pramin_bar, 0x14, 0x00000000);
-
-	/* map channel into PRAMIN, gpuobj didn't do it for us */
-	ret = nv50_instmem_bind(dev, chan->ramin);
+		goto error;
+	dev_priv->channels.ptr[0] = dev_priv->channels.ptr[127] = chan;
+
+	ret = nv50_gpuobj_dma_new(chan, 0x0000, BAR3_VM_BASE, BAR3_VM_SIZE,
+				  NV_MEM_TARGET_VM, NV_MEM_ACCESS_VM,
+				  NV_MEM_TYPE_VM, NV_MEM_COMP_VM,
+				  &priv->bar3_dmaobj);
 	if (ret)
-		return ret;
+		goto error;
 
-	/* poke regs... */
 	nv_wr32(dev, 0x001704, 0x00000000 | (chan->ramin->vinst >> 12));
 	nv_wr32(dev, 0x001704, 0x40000000 | (chan->ramin->vinst >> 12));
-	nv_wr32(dev, 0x00170c, 0x80000000 | (priv->pramin_bar->cinst >> 4));
-
-	tmp = nv_ri32(dev, 0);
-	nv_wi32(dev, 0, ~tmp);
-	if (nv_ri32(dev, 0) != ~tmp) {
-		NV_ERROR(dev, "PRAMIN readback failed\n");
-		return -EIO;
-	}
-	nv_wi32(dev, 0, tmp);
+	nv_wr32(dev, 0x00170c, 0x80000000 | (priv->bar3_dmaobj->cinst >> 4));
 
+	dev_priv->engine.instmem.flush(dev);
 	dev_priv->ramin_available = true;
 
-	/* Determine VM layout */
-	dev_priv->vm_gart_base = roundup(NV50_VM_BLOCK, NV50_VM_BLOCK);
-	dev_priv->vm_gart_size = NV50_VM_BLOCK;
-
-	dev_priv->vm_vram_base = dev_priv->vm_gart_base + dev_priv->vm_gart_size;
-	dev_priv->vm_vram_size = dev_priv->vram_size;
-	if (dev_priv->vm_vram_size > NV50_VM_MAX_VRAM)
-		dev_priv->vm_vram_size = NV50_VM_MAX_VRAM;
-	dev_priv->vm_vram_size = roundup(dev_priv->vm_vram_size, NV50_VM_BLOCK);
-	dev_priv->vm_vram_pt_nr = dev_priv->vm_vram_size / NV50_VM_BLOCK;
-
-	dev_priv->vm_end = dev_priv->vm_vram_base + dev_priv->vm_vram_size;
-
-	NV_DEBUG(dev, "NV50VM: GART 0x%016llx-0x%016llx\n",
-		 dev_priv->vm_gart_base,
-		 dev_priv->vm_gart_base + dev_priv->vm_gart_size - 1);
-	NV_DEBUG(dev, "NV50VM: VRAM 0x%016llx-0x%016llx\n",
-		 dev_priv->vm_vram_base,
-		 dev_priv->vm_vram_base + dev_priv->vm_vram_size - 1);
-
-	/* VRAM page table(s), mapped into VM at +1GiB  */
-	for (i = 0; i < dev_priv->vm_vram_pt_nr; i++) {
-		ret = nouveau_gpuobj_new(dev, NULL, NV50_VM_BLOCK / 0x10000 * 8,
-					 0, NVOBJ_FLAG_ZERO_ALLOC,
-					 &chan->vm_vram_pt[i]);
-		if (ret) {
-			NV_ERROR(dev, "Error creating VRAM PGT: %d\n", ret);
-			dev_priv->vm_vram_pt_nr = i;
-			return ret;
-		}
-		dev_priv->vm_vram_pt[i] = chan->vm_vram_pt[i];
-
-		nv_wo32(chan->vm_pd, 0x10 + (i*8),
-			chan->vm_vram_pt[i]->vinst | 0x61);
-		nv_wo32(chan->vm_pd, 0x14 + (i*8), 0);
+	tmp = nv_ro32(chan->ramin, 0);
+	nv_wo32(chan->ramin, 0, ~tmp);
+	if (nv_ro32(chan->ramin, 0) != ~tmp) {
+		NV_ERROR(dev, "PRAMIN readback failed\n");
+		ret = -EIO;
+		goto error;
 	}
+	nv_wo32(chan->ramin, 0, tmp);
 
-	/* DMA object for FB BAR */
-	ret = nouveau_gpuobj_new(dev, chan, 6*4, 16, 0, &priv->fb_bar);
+	/* BAR1 */
+	ret = nouveau_vm_new(dev, BAR1_VM_BASE, BAR1_VM_SIZE, BAR1_VM_BASE,
+			     29, 12, 16, &vm);
 	if (ret)
-		return ret;
-	nv_wo32(priv->fb_bar, 0x00, 0x7fc00000);
-	nv_wo32(priv->fb_bar, 0x04, 0x40000000 +
-				    pci_resource_len(dev->pdev, 1) - 1);
-	nv_wo32(priv->fb_bar, 0x08, 0x40000000);
-	nv_wo32(priv->fb_bar, 0x0c, 0x00000000);
-	nv_wo32(priv->fb_bar, 0x10, 0x00000000);
-	nv_wo32(priv->fb_bar, 0x14, 0x00000000);
+		goto error;
 
-	dev_priv->engine.instmem.flush(dev);
+	ret = nouveau_vm_ref(vm, &dev_priv->bar1_vm, chan->vm_pd);
+	if (ret)
+		goto error;
+	nouveau_vm_ref(NULL, &vm, NULL);
+
+	ret = nv50_gpuobj_dma_new(chan, 0x0000, BAR1_VM_BASE, BAR1_VM_SIZE,
+				  NV_MEM_TARGET_VM, NV_MEM_ACCESS_VM,
+				  NV_MEM_TYPE_VM, NV_MEM_COMP_VM,
+				  &priv->bar1_dmaobj);
+	if (ret)
+		goto error;
 
-	nv_wr32(dev, 0x001708, 0x80000000 | (priv->fb_bar->cinst >> 4));
+	nv_wr32(dev, 0x001708, 0x80000000 | (priv->bar1_dmaobj->cinst >> 4));
 	for (i = 0; i < 8; i++)
 		nv_wr32(dev, 0x1900 + (i*4), 0);
 
+	/* Create shared channel VM, space is reserved at the beginning
+	 * to catch "NULL pointer" references
+	 */
+	ret = nouveau_vm_new(dev, 0, (1ULL << 40), 0x0020000000ULL,
+			     29, 12, 16, &dev_priv->chan_vm);
+	if (ret)
+		return ret;
+
 	return 0;
+
+error:
+	nv50_instmem_takedown(dev);
+	return ret;
 }
 
 void
@@ -240,7 +236,7 @@ nv50_instmem_takedown(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	struct nouveau_channel *chan = dev_priv->fifos[0];
+	struct nouveau_channel *chan = dev_priv->channels.ptr[0];
 	int i;
 
 	NV_DEBUG(dev, "\n");
@@ -250,23 +246,23 @@ nv50_instmem_takedown(struct drm_device *dev)
 
 	dev_priv->ramin_available = false;
 
-	/* Restore state from before init */
+	nouveau_vm_ref(NULL, &dev_priv->chan_vm, NULL);
+
 	for (i = 0x1700; i <= 0x1710; i += 4)
 		nv_wr32(dev, i, priv->save1700[(i - 0x1700) / 4]);
 
-	nouveau_gpuobj_ref(NULL, &priv->fb_bar);
-	nouveau_gpuobj_ref(NULL, &priv->pramin_bar);
-	nouveau_gpuobj_ref(NULL, &priv->pramin_pt);
+	nouveau_gpuobj_ref(NULL, &priv->bar3_dmaobj);
+	nouveau_gpuobj_ref(NULL, &priv->bar1_dmaobj);
 
-	/* Destroy dummy channel */
-	if (chan) {
-		for (i = 0; i < dev_priv->vm_vram_pt_nr; i++)
-			nouveau_gpuobj_ref(NULL, &chan->vm_vram_pt[i]);
-		dev_priv->vm_vram_pt_nr = 0;
+	nouveau_vm_ref(NULL, &dev_priv->bar1_vm, chan->vm_pd);
+	dev_priv->channels.ptr[127] = 0;
+	nv50_channel_del(&dev_priv->channels.ptr[0]);
 
-		nv50_channel_del(&dev_priv->fifos[0]);
-		dev_priv->fifos[127] = NULL;
-	}
+	nouveau_gpuobj_ref(NULL, &dev_priv->bar3_vm->pgt[0].obj);
+	nouveau_vm_ref(NULL, &dev_priv->bar3_vm, NULL);
+
+	if (dev_priv->ramin_heap.free_stack.next)
+		drm_mm_takedown(&dev_priv->ramin_heap);
 
 	dev_priv->engine.instmem.priv = NULL;
 	kfree(priv);
@@ -276,16 +272,8 @@ int
 nv50_instmem_suspend(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nouveau_channel *chan = dev_priv->fifos[0];
-	struct nouveau_gpuobj *ramin = chan->ramin;
-	int i;
 
-	ramin->im_backing_suspend = vmalloc(ramin->size);
-	if (!ramin->im_backing_suspend)
-		return -ENOMEM;
-
-	for (i = 0; i < ramin->size; i += 4)
-		ramin->im_backing_suspend[i/4] = nv_ri32(dev, i);
+	dev_priv->ramin_available = false;
 	return 0;
 }
 
@@ -294,146 +282,121 @@ nv50_instmem_resume(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
 	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	struct nouveau_channel *chan = dev_priv->fifos[0];
-	struct nouveau_gpuobj *ramin = chan->ramin;
+	struct nouveau_channel *chan = dev_priv->channels.ptr[0];
 	int i;
 
-	dev_priv->ramin_available = false;
-	dev_priv->ramin_base = ~0;
-	for (i = 0; i < ramin->size; i += 4)
-		nv_wo32(ramin, i, ramin->im_backing_suspend[i/4]);
-	dev_priv->ramin_available = true;
-	vfree(ramin->im_backing_suspend);
-	ramin->im_backing_suspend = NULL;
-
 	/* Poke the relevant regs, and pray it works :) */
 	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12));
 	nv_wr32(dev, NV50_PUNK_UNK1710, 0);
 	nv_wr32(dev, NV50_PUNK_BAR_CFG_BASE, (chan->ramin->vinst >> 12) |
 					 NV50_PUNK_BAR_CFG_BASE_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->fb_bar->cinst >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR1_CTXDMA, (priv->bar1_dmaobj->cinst >> 4) |
 					NV50_PUNK_BAR1_CTXDMA_VALID);
-	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->pramin_bar->cinst >> 4) |
+	nv_wr32(dev, NV50_PUNK_BAR3_CTXDMA, (priv->bar3_dmaobj->cinst >> 4) |
 					NV50_PUNK_BAR3_CTXDMA_VALID);
 
 	for (i = 0; i < 8; i++)
 		nv_wr32(dev, 0x1900 + (i*4), 0);
+
+	dev_priv->ramin_available = true;
 }
 
+struct nv50_gpuobj_node {
+	struct nouveau_vram *vram;
+	struct nouveau_vma chan_vma;
+	u32 align;
+};
+
+
 int
-nv50_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj,
-		      uint32_t *sz)
+nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 {
+	struct drm_device *dev = gpuobj->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	struct nv50_gpuobj_node *node = NULL;
 	int ret;
 
-	if (gpuobj->im_backing)
-		return -EINVAL;
+	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+	node->align = align;
 
-	*sz = ALIGN(*sz, 4096);
-	if (*sz == 0)
-		return -EINVAL;
+	size  = (size + 4095) & ~4095;
+	align = max(align, (u32)4096);
 
-	ret = nouveau_bo_new(dev, NULL, *sz, 0, TTM_PL_FLAG_VRAM, 0, 0x0000,
-			     true, false, &gpuobj->im_backing);
+	ret = vram->get(dev, size, align, 0, 0, &node->vram);
 	if (ret) {
-		NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
+		kfree(node);
 		return ret;
 	}
 
-	ret = nouveau_bo_pin(gpuobj->im_backing, TTM_PL_FLAG_VRAM);
-	if (ret) {
-		NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret);
-		nouveau_bo_ref(NULL, &gpuobj->im_backing);
-		return ret;
+	gpuobj->vinst = node->vram->offset;
+
+	if (gpuobj->flags & NVOBJ_FLAG_VM) {
+		ret = nouveau_vm_get(dev_priv->chan_vm, size, 12,
+				     NV_MEM_ACCESS_RW | NV_MEM_ACCESS_SYS,
+				     &node->chan_vma);
+		if (ret) {
+			vram->put(dev, &node->vram);
+			kfree(node);
+			return ret;
+		}
+
+		nouveau_vm_map(&node->chan_vma, node->vram);
+		gpuobj->vinst = node->chan_vma.offset;
 	}
 
-	gpuobj->vinst = gpuobj->im_backing->bo.mem.start << PAGE_SHIFT;
+	gpuobj->size = size;
+	gpuobj->node = node;
 	return 0;
 }
 
 void
-nv50_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nv50_instmem_put(struct nouveau_gpuobj *gpuobj)
 {
+	struct drm_device *dev = gpuobj->dev;
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_vram_engine *vram = &dev_priv->engine.vram;
+	struct nv50_gpuobj_node *node;
+
+	node = gpuobj->node;
+	gpuobj->node = NULL;
 
-	if (gpuobj && gpuobj->im_backing) {
-		if (gpuobj->im_bound)
-			dev_priv->engine.instmem.unbind(dev, gpuobj);
-		nouveau_bo_unpin(gpuobj->im_backing);
-		nouveau_bo_ref(NULL, &gpuobj->im_backing);
-		gpuobj->im_backing = NULL;
+	if (node->chan_vma.node) {
+		nouveau_vm_unmap(&node->chan_vma);
+		nouveau_vm_put(&node->chan_vma);
 	}
+	vram->put(dev, &node->vram);
+	kfree(node);
 }
 
 int
-nv50_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nv50_instmem_map(struct nouveau_gpuobj *gpuobj)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	struct nouveau_gpuobj *pramin_pt = priv->pramin_pt;
-	uint32_t pte, pte_end;
-	uint64_t vram;
-
-	if (!gpuobj->im_backing || !gpuobj->im_pramin || gpuobj->im_bound)
-		return -EINVAL;
-
-	NV_DEBUG(dev, "st=0x%lx sz=0x%lx\n",
-		 gpuobj->im_pramin->start, gpuobj->im_pramin->size);
-
-	pte     = (gpuobj->im_pramin->start >> 12) << 1;
-	pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte;
-	vram    = gpuobj->vinst;
-
-	NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n",
-		 gpuobj->im_pramin->start, pte, pte_end);
-	NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst);
-
-	vram |= 1;
-	if (dev_priv->vram_sys_base) {
-		vram += dev_priv->vram_sys_base;
-		vram |= 0x30;
-	}
-
-	while (pte < pte_end) {
-		nv_wo32(pramin_pt, (pte * 4) + 0, lower_32_bits(vram));
-		nv_wo32(pramin_pt, (pte * 4) + 4, upper_32_bits(vram));
-		vram += 0x1000;
-		pte += 2;
-	}
-	dev_priv->engine.instmem.flush(dev);
+	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
+	struct nv50_gpuobj_node *node = gpuobj->node;
+	int ret;
 
-	nv50_vm_flush(dev, 6);
+	ret = nouveau_vm_get(dev_priv->bar3_vm, gpuobj->size, 12,
+			     NV_MEM_ACCESS_RW, &node->vram->bar_vma);
+	if (ret)
+		return ret;
 
-	gpuobj->im_bound = 1;
+	nouveau_vm_map(&node->vram->bar_vma, node->vram);
+	gpuobj->pinst = node->vram->bar_vma.offset;
 	return 0;
 }
 
-int
-nv50_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+void
+nv50_instmem_unmap(struct nouveau_gpuobj *gpuobj)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	struct nv50_instmem_priv *priv = dev_priv->engine.instmem.priv;
-	uint32_t pte, pte_end;
-
-	if (gpuobj->im_bound == 0)
-		return -EINVAL;
-
-	/* can happen during late takedown */
-	if (unlikely(!dev_priv->ramin_available))
-		return 0;
+	struct nv50_gpuobj_node *node = gpuobj->node;
 
-	pte     = (gpuobj->im_pramin->start >> 12) << 1;
-	pte_end = ((gpuobj->im_pramin->size >> 12) << 1) + pte;
-
-	while (pte < pte_end) {
-		nv_wo32(priv->pramin_pt, (pte * 4) + 0, 0x00000000);
-		nv_wo32(priv->pramin_pt, (pte * 4) + 4, 0x00000000);
-		pte += 2;
+	if (node->vram->bar_vma.node) {
+		nouveau_vm_unmap(&node->vram->bar_vma);
+		nouveau_vm_put(&node->vram->bar_vma);
 	}
-	dev_priv->engine.instmem.flush(dev);
-
-	gpuobj->im_bound = 0;
-	return 0;
 }
 
 void
@@ -452,11 +415,3 @@ nv84_instmem_flush(struct drm_device *dev)
 		NV_ERROR(dev, "PRAMIN flush timeout\n");
 }
 
-void
-nv50_vm_flush(struct drm_device *dev, int engine)
-{
-	nv_wr32(dev, 0x100c80, (engine << 16) | 1);
-	if (!nv_wait(dev, 0x100c80, 0x00000001, 0x00000000))
-		NV_ERROR(dev, "vm flush timeout: engine %d\n", engine);
-}
-
diff --git a/drivers/gpu/drm/nouveau/nv50_vm.c b/drivers/gpu/drm/nouveau/nv50_vm.c
new file mode 100644
index 000000000000..eebab95f59b2
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_vm.c
@@ -0,0 +1,178 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+
+#include "nouveau_drv.h"
+#include "nouveau_vm.h"
+
+void
+nv50_vm_map_pgt(struct nouveau_gpuobj *pgd, u32 type, u32 pde,
+		struct nouveau_gpuobj *pgt)
+{
+	struct drm_nouveau_private *dev_priv = pgd->dev->dev_private;
+	u32 coverage = (pgt->size >> 3) << type;
+	u64 phys;
+
+	phys  = pgt->vinst;
+	phys |= 0x01; /* present */
+	phys |= (type == 12) ? 0x02 : 0x00; /* 4KiB pages */
+	if (dev_priv->vram_sys_base) {
+		phys += dev_priv->vram_sys_base;
+		phys |= 0x30;
+	}
+
+	if (coverage <= 32 * 1024 * 1024)
+		phys |= 0x60;
+	else if (coverage <= 64 * 1024 * 1024)
+		phys |= 0x40;
+	else if (coverage < 128 * 1024 * 1024)
+		phys |= 0x20;
+
+	nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
+	nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
+}
+
+void
+nv50_vm_unmap_pgt(struct nouveau_gpuobj *pgd, u32 pde)
+{
+	nv_wo32(pgd, (pde * 8) + 0, 0x00000000);
+	nv_wo32(pgd, (pde * 8) + 4, 0xdeadcafe);
+}
+
+static inline u64
+nv50_vm_addr(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	     u64 phys, u32 memtype, u32 target)
+{
+	struct drm_nouveau_private *dev_priv = pgt->dev->dev_private;
+
+	phys |= 1; /* present */
+	phys |= (u64)memtype << 40;
+
+	/* IGPs don't have real VRAM, re-target to stolen system memory */
+	if (target == 0 && dev_priv->vram_sys_base) {
+		phys  += dev_priv->vram_sys_base;
+		target = 3;
+	}
+
+	phys |= target << 4;
+
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= (1 << 6);
+
+	if (!(vma->access & NV_MEM_ACCESS_WO))
+		phys |= (1 << 3);
+
+	return phys;
+}
+
+void
+nv50_vm_map(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	    struct nouveau_vram *mem, u32 pte, u32 cnt, u64 phys)
+{
+	u32 block, i;
+
+	phys  = nv50_vm_addr(vma, pgt, phys, mem->memtype, 0);
+	pte <<= 3;
+	cnt <<= 3;
+
+	while (cnt) {
+		u32 offset_h = upper_32_bits(phys);
+		u32 offset_l = lower_32_bits(phys);
+
+		for (i = 7; i >= 0; i--) {
+			block = 1 << (i + 3);
+			if (cnt >= block && !(pte & (block - 1)))
+				break;
+		}
+		offset_l |= (i << 7);
+
+		phys += block << (vma->node->type - 3);
+		cnt  -= block;
+
+		while (block) {
+			nv_wo32(pgt, pte + 0, offset_l);
+			nv_wo32(pgt, pte + 4, offset_h);
+			pte += 8;
+			block -= 8;
+		}
+	}
+}
+
+void
+nv50_vm_map_sg(struct nouveau_vma *vma, struct nouveau_gpuobj *pgt,
+	       u32 pte, dma_addr_t *list, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		u64 phys = nv50_vm_addr(vma, pgt, (u64)*list++, 0, 2);
+		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
+		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
+		pte += 8;
+	}
+}
+
+void
+nv50_vm_unmap(struct nouveau_gpuobj *pgt, u32 pte, u32 cnt)
+{
+	pte <<= 3;
+	while (cnt--) {
+		nv_wo32(pgt, pte + 0, 0x00000000);
+		nv_wo32(pgt, pte + 4, 0x00000000);
+		pte += 8;
+	}
+}
+
+void
+nv50_vm_flush(struct nouveau_vm *vm)
+{
+	struct drm_nouveau_private *dev_priv = vm->dev->dev_private;
+	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
+	struct nouveau_fifo_engine *pfifo = &dev_priv->engine.fifo;
+	struct nouveau_pgraph_engine *pgraph = &dev_priv->engine.graph;
+	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
+
+	pinstmem->flush(vm->dev);
+
+	/* BAR */
+	if (vm != dev_priv->chan_vm) {
+		nv50_vm_flush_engine(vm->dev, 6);
+		return;
+	}
+
+	pfifo->tlb_flush(vm->dev);
+
+	if (atomic_read(&vm->pgraph_refs))
+		pgraph->tlb_flush(vm->dev);
+	if (atomic_read(&vm->pcrypt_refs))
+		pcrypt->tlb_flush(vm->dev);
+}
+
+void
+nv50_vm_flush_engine(struct drm_device *dev, int engine)
+{
+	nv_wr32(dev, 0x100c80, (engine << 16) | 1);
+	if (!nv_wait(dev, 0x100c80, 0x00000001, 0x00000000))
+		NV_ERROR(dev, "vm flush timeout: engine %d\n", engine);
+}
diff --git a/drivers/gpu/drm/nouveau/nv50_vram.c b/drivers/gpu/drm/nouveau/nv50_vram.c
new file mode 100644
index 000000000000..47489ed0a5a8
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv50_vram.c
@@ -0,0 +1,190 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_mm.h"
+
+static int types[0x80] = {
+	1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 0, 0, 0, 0, 2, 2, 2, 2, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 0, 2, 2, 2, 2, 2, 2, 2, 0,
+	0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 0, 0,
+	0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+	1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 2, 2, 2, 2,
+	1, 0, 2, 0, 1, 0, 2, 0, 1, 1, 2, 2, 1, 1, 0, 0
+};
+
+bool
+nv50_vram_flags_valid(struct drm_device *dev, u32 tile_flags)
+{
+	int type = (tile_flags & NOUVEAU_GEM_TILE_LAYOUT_MASK) >> 8;
+
+	if (likely(type < sizeof(types) && types[type]))
+		return true;
+	return false;
+}
+
+void
+nv50_vram_del(struct drm_device *dev, struct nouveau_vram **pvram)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
+	struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
+	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm_node *this;
+	struct nouveau_vram *vram;
+
+	vram = *pvram;
+	*pvram = NULL;
+	if (unlikely(vram == NULL))
+		return;
+
+	mutex_lock(&mm->mutex);
+	while (!list_empty(&vram->regions)) {
+		this = list_first_entry(&vram->regions, struct nouveau_mm_node, rl_entry);
+
+		list_del(&this->rl_entry);
+		nouveau_mm_put(mm, this);
+	}
+	mutex_unlock(&mm->mutex);
+
+	kfree(vram);
+}
+
+int
+nv50_vram_new(struct drm_device *dev, u64 size, u32 align, u32 size_nc,
+	      u32 type, struct nouveau_vram **pvram)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct ttm_bo_device *bdev = &dev_priv->ttm.bdev;
+	struct ttm_mem_type_manager *man = &bdev->man[TTM_PL_VRAM];
+	struct nouveau_mm *mm = man->priv;
+	struct nouveau_mm_node *r;
+	struct nouveau_vram *vram;
+	int ret;
+
+	if (!types[type])
+		return -EINVAL;
+	size >>= 12;
+	align >>= 12;
+	size_nc >>= 12;
+
+	vram = kzalloc(sizeof(*vram), GFP_KERNEL);
+	if (!vram)
+		return -ENOMEM;
+
+	INIT_LIST_HEAD(&vram->regions);
+	vram->dev = dev_priv->dev;
+	vram->memtype = type;
+	vram->size = size;
+
+	mutex_lock(&mm->mutex);
+	do {
+		ret = nouveau_mm_get(mm, types[type], size, size_nc, align, &r);
+		if (ret) {
+			mutex_unlock(&mm->mutex);
+			nv50_vram_del(dev, &vram);
+			return ret;
+		}
+
+		list_add_tail(&r->rl_entry, &vram->regions);
+		size -= r->length;
+	} while (size);
+	mutex_unlock(&mm->mutex);
+
+	r = list_first_entry(&vram->regions, struct nouveau_mm_node, rl_entry);
+	vram->offset = (u64)r->offset << 12;
+	*pvram = vram;
+	return 0;
+}
+
+static u32
+nv50_vram_rblock(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	int i, parts, colbits, rowbitsa, rowbitsb, banks;
+	u64 rowsize, predicted;
+	u32 r0, r4, rt, ru, rblock_size;
+
+	r0 = nv_rd32(dev, 0x100200);
+	r4 = nv_rd32(dev, 0x100204);
+	rt = nv_rd32(dev, 0x100250);
+	ru = nv_rd32(dev, 0x001540);
+	NV_DEBUG(dev, "memcfg 0x%08x 0x%08x 0x%08x 0x%08x\n", r0, r4, rt, ru);
+
+	for (i = 0, parts = 0; i < 8; i++) {
+		if (ru & (0x00010000 << i))
+			parts++;
+	}
+
+	colbits  =  (r4 & 0x0000f000) >> 12;
+	rowbitsa = ((r4 & 0x000f0000) >> 16) + 8;
+	rowbitsb = ((r4 & 0x00f00000) >> 20) + 8;
+	banks    = ((r4 & 0x01000000) ? 8 : 4);
+
+	rowsize = parts * banks * (1 << colbits) * 8;
+	predicted = rowsize << rowbitsa;
+	if (r0 & 0x00000004)
+		predicted += rowsize << rowbitsb;
+
+	if (predicted != dev_priv->vram_size) {
+		NV_WARN(dev, "memory controller reports %dMiB VRAM\n",
+			(u32)(dev_priv->vram_size >> 20));
+		NV_WARN(dev, "we calculated %dMiB VRAM\n",
+			(u32)(predicted >> 20));
+	}
+
+	rblock_size = rowsize;
+	if (rt & 1)
+		rblock_size *= 3;
+
+	NV_DEBUG(dev, "rblock %d bytes\n", rblock_size);
+	return rblock_size;
+}
+
+int
+nv50_vram_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	dev_priv->vram_size  = nv_rd32(dev, 0x10020c);
+	dev_priv->vram_size |= (dev_priv->vram_size & 0xff) << 32;
+	dev_priv->vram_size &= 0xffffffff00ULL;
+
+	switch (dev_priv->chipset) {
+	case 0xaa:
+	case 0xac:
+	case 0xaf:
+		dev_priv->vram_sys_base = (u64)nv_rd32(dev, 0x100e10) << 12;
+		dev_priv->vram_rblock_size = 4096;
+		break;
+	default:
+		dev_priv->vram_rblock_size = nv50_vram_rblock(dev);
+		break;
+	}
+
+	return 0;
+}
diff --git a/drivers/gpu/drm/nouveau/nv84_crypt.c b/drivers/gpu/drm/nouveau/nv84_crypt.c
new file mode 100644
index 000000000000..ec18ae1c3886
--- /dev/null
+++ b/drivers/gpu/drm/nouveau/nv84_crypt.c
@@ -0,0 +1,140 @@
+/*
+ * Copyright 2010 Red Hat Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: Ben Skeggs
+ */
+
+#include "drmP.h"
+#include "nouveau_drv.h"
+#include "nouveau_util.h"
+#include "nouveau_vm.h"
+
+static void nv84_crypt_isr(struct drm_device *);
+
+int
+nv84_crypt_create_context(struct nouveau_channel *chan)
+{
+	struct drm_device *dev = chan->dev;
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_gpuobj *ramin = chan->ramin;
+	int ret;
+
+	NV_DEBUG(dev, "ch%d\n", chan->id);
+
+	ret = nouveau_gpuobj_new(dev, chan, 256, 0,
+				 NVOBJ_FLAG_ZERO_ALLOC | NVOBJ_FLAG_ZERO_FREE,
+				 &chan->crypt_ctx);
+	if (ret)
+		return ret;
+
+	nv_wo32(ramin, 0xa0, 0x00190000);
+	nv_wo32(ramin, 0xa4, chan->crypt_ctx->vinst + 0xff);
+	nv_wo32(ramin, 0xa8, chan->crypt_ctx->vinst);
+	nv_wo32(ramin, 0xac, 0);
+	nv_wo32(ramin, 0xb0, 0);
+	nv_wo32(ramin, 0xb4, 0);
+
+	dev_priv->engine.instmem.flush(dev);
+	atomic_inc(&chan->vm->pcrypt_refs);
+	return 0;
+}
+
+void
+nv84_crypt_destroy_context(struct nouveau_channel *chan)
+{
+	struct drm_device *dev = chan->dev;
+	u32 inst;
+
+	if (!chan->crypt_ctx)
+		return;
+
+	inst  = (chan->ramin->vinst >> 12);
+	inst |= 0x80000000;
+
+	/* mark context as invalid if still on the hardware, not
+	 * doing this causes issues the next time PCRYPT is used,
+	 * unsurprisingly :)
+	 */
+	nv_wr32(dev, 0x10200c, 0x00000000);
+	if (nv_rd32(dev, 0x102188) == inst)
+		nv_mask(dev, 0x102188, 0x80000000, 0x00000000);
+	if (nv_rd32(dev, 0x10218c) == inst)
+		nv_mask(dev, 0x10218c, 0x80000000, 0x00000000);
+	nv_wr32(dev, 0x10200c, 0x00000010);
+
+	nouveau_gpuobj_ref(NULL, &chan->crypt_ctx);
+	atomic_dec(&chan->vm->pcrypt_refs);
+}
+
+void
+nv84_crypt_tlb_flush(struct drm_device *dev)
+{
+	nv50_vm_flush_engine(dev, 0x0a);
+}
+
+int
+nv84_crypt_init(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nouveau_crypt_engine *pcrypt = &dev_priv->engine.crypt;
+
+	if (!pcrypt->registered) {
+		NVOBJ_CLASS(dev, 0x74c1, CRYPT);
+		pcrypt->registered = true;
+	}
+
+	nv_mask(dev, 0x000200, 0x00004000, 0x00000000);
+	nv_mask(dev, 0x000200, 0x00004000, 0x00004000);
+
+	nouveau_irq_register(dev, 14, nv84_crypt_isr);
+	nv_wr32(dev, 0x102130, 0xffffffff);
+	nv_wr32(dev, 0x102140, 0xffffffbf);
+
+	nv_wr32(dev, 0x10200c, 0x00000010);
+	return 0;
+}
+
+void
+nv84_crypt_fini(struct drm_device *dev)
+{
+	nv_wr32(dev, 0x102140, 0x00000000);
+	nouveau_irq_unregister(dev, 14);
+}
+
+static void
+nv84_crypt_isr(struct drm_device *dev)
+{
+	u32 stat = nv_rd32(dev, 0x102130);
+	u32 mthd = nv_rd32(dev, 0x102190);
+	u32 data = nv_rd32(dev, 0x102194);
+	u32 inst = nv_rd32(dev, 0x102188) & 0x7fffffff;
+	int show = nouveau_ratelimit();
+
+	if (show) {
+		NV_INFO(dev, "PCRYPT_INTR: 0x%08x 0x%08x 0x%08x 0x%08x\n",
+			     stat, mthd, data, inst);
+	}
+
+	nv_wr32(dev, 0x102130, stat);
+	nv_wr32(dev, 0x10200c, 0x10);
+
+	nv50_fb_vm_trap(dev, show, "PCRYPT");
+}
diff --git a/drivers/gpu/drm/nouveau/nvc0_instmem.c b/drivers/gpu/drm/nouveau/nvc0_instmem.c
index 13a0f78a9088..39232085193d 100644
--- a/drivers/gpu/drm/nouveau/nvc0_instmem.c
+++ b/drivers/gpu/drm/nouveau/nvc0_instmem.c
@@ -26,67 +26,89 @@
 
 #include "nouveau_drv.h"
 
+struct nvc0_gpuobj_node {
+	struct nouveau_bo *vram;
+	struct drm_mm_node *ramin;
+	u32 align;
+};
+
 int
-nvc0_instmem_populate(struct drm_device *dev, struct nouveau_gpuobj *gpuobj,
-		      uint32_t *size)
+nvc0_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 {
+	struct drm_device *dev = gpuobj->dev;
+	struct nvc0_gpuobj_node *node = NULL;
 	int ret;
 
-	*size = ALIGN(*size, 4096);
-	if (*size == 0)
-		return -EINVAL;
+	node = kzalloc(sizeof(*node), GFP_KERNEL);
+	if (!node)
+		return -ENOMEM;
+	node->align = align;
 
-	ret = nouveau_bo_new(dev, NULL, *size, 0, TTM_PL_FLAG_VRAM, 0, 0x0000,
-			     true, false, &gpuobj->im_backing);
+	ret = nouveau_bo_new(dev, NULL, size, align, TTM_PL_FLAG_VRAM,
+			     0, 0x0000, true, false, &node->vram);
 	if (ret) {
 		NV_ERROR(dev, "error getting PRAMIN backing pages: %d\n", ret);
 		return ret;
 	}
 
-	ret = nouveau_bo_pin(gpuobj->im_backing, TTM_PL_FLAG_VRAM);
+	ret = nouveau_bo_pin(node->vram, TTM_PL_FLAG_VRAM);
 	if (ret) {
 		NV_ERROR(dev, "error pinning PRAMIN backing VRAM: %d\n", ret);
-		nouveau_bo_ref(NULL, &gpuobj->im_backing);
+		nouveau_bo_ref(NULL, &node->vram);
 		return ret;
 	}
 
-	gpuobj->vinst = gpuobj->im_backing->bo.mem.start << PAGE_SHIFT;
+	gpuobj->vinst = node->vram->bo.mem.start << PAGE_SHIFT;
+	gpuobj->size  = node->vram->bo.mem.num_pages << PAGE_SHIFT;
+	gpuobj->node  = node;
 	return 0;
 }
 
 void
-nvc0_instmem_clear(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nvc0_instmem_put(struct nouveau_gpuobj *gpuobj)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
+	struct nvc0_gpuobj_node *node;
 
-	if (gpuobj && gpuobj->im_backing) {
-		if (gpuobj->im_bound)
-			dev_priv->engine.instmem.unbind(dev, gpuobj);
-		nouveau_bo_unpin(gpuobj->im_backing);
-		nouveau_bo_ref(NULL, &gpuobj->im_backing);
-		gpuobj->im_backing = NULL;
-	}
+	node = gpuobj->node;
+	gpuobj->node = NULL;
+
+	nouveau_bo_unpin(node->vram);
+	nouveau_bo_ref(NULL, &node->vram);
+	kfree(node);
 }
 
 int
-nvc0_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+nvc0_instmem_map(struct nouveau_gpuobj *gpuobj)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t pte, pte_end;
-	uint64_t vram;
-
-	if (!gpuobj->im_backing || !gpuobj->im_pramin || gpuobj->im_bound)
-		return -EINVAL;
-
-	NV_DEBUG(dev, "st=0x%lx sz=0x%lx\n",
-		 gpuobj->im_pramin->start, gpuobj->im_pramin->size);
-
-	pte     = gpuobj->im_pramin->start >> 12;
-	pte_end = (gpuobj->im_pramin->size >> 12) + pte;
+	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
+	struct nvc0_gpuobj_node *node = gpuobj->node;
+	struct drm_device *dev = gpuobj->dev;
+	struct drm_mm_node *ramin = NULL;
+	u32 pte, pte_end;
+	u64 vram;
+
+	do {
+		if (drm_mm_pre_get(&dev_priv->ramin_heap))
+			return -ENOMEM;
+
+		spin_lock(&dev_priv->ramin_lock);
+		ramin = drm_mm_search_free(&dev_priv->ramin_heap, gpuobj->size,
+					   node->align, 0);
+		if (ramin == NULL) {
+			spin_unlock(&dev_priv->ramin_lock);
+			return -ENOMEM;
+		}
+
+		ramin = drm_mm_get_block_atomic(ramin, gpuobj->size, node->align);
+		spin_unlock(&dev_priv->ramin_lock);
+	} while (ramin == NULL);
+
+	pte     = (ramin->start >> 12) << 1;
+	pte_end = ((ramin->size >> 12) << 1) + pte;
 	vram    = gpuobj->vinst;
 
 	NV_DEBUG(dev, "pramin=0x%lx, pte=%d, pte_end=%d\n",
-		 gpuobj->im_pramin->start, pte, pte_end);
+		 ramin->start, pte, pte_end);
 	NV_DEBUG(dev, "first vram page: 0x%010llx\n", gpuobj->vinst);
 
 	while (pte < pte_end) {
@@ -103,30 +125,35 @@ nvc0_instmem_bind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
 		nv_wr32(dev, 0x100cbc, 0x80000005);
 	}
 
-	gpuobj->im_bound = 1;
+	node->ramin   = ramin;
+	gpuobj->pinst = ramin->start;
 	return 0;
 }
 
-int
-nvc0_instmem_unbind(struct drm_device *dev, struct nouveau_gpuobj *gpuobj)
+void
+nvc0_instmem_unmap(struct nouveau_gpuobj *gpuobj)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t pte, pte_end;
+	struct drm_nouveau_private *dev_priv = gpuobj->dev->dev_private;
+	struct nvc0_gpuobj_node *node = gpuobj->node;
+	u32 pte, pte_end;
 
-	if (gpuobj->im_bound == 0)
-		return -EINVAL;
+	if (!node->ramin || !dev_priv->ramin_available)
+		return;
+
+	pte     = (node->ramin->start >> 12) << 1;
+	pte_end = ((node->ramin->size >> 12) << 1) + pte;
 
-	pte     = gpuobj->im_pramin->start >> 12;
-	pte_end = (gpuobj->im_pramin->size >> 12) + pte;
 	while (pte < pte_end) {
-		nv_wr32(dev, 0x702000 + (pte * 8), 0);
-		nv_wr32(dev, 0x702004 + (pte * 8), 0);
+		nv_wr32(gpuobj->dev, 0x702000 + (pte * 8), 0);
+		nv_wr32(gpuobj->dev, 0x702004 + (pte * 8), 0);
 		pte++;
 	}
-	dev_priv->engine.instmem.flush(dev);
+	dev_priv->engine.instmem.flush(gpuobj->dev);
 
-	gpuobj->im_bound = 0;
-	return 0;
+	spin_lock(&dev_priv->ramin_lock);
+	drm_mm_put_block(node->ramin);
+	node->ramin = NULL;
+	spin_unlock(&dev_priv->ramin_lock);
 }
 
 void
diff --git a/drivers/gpu/drm/nouveau/nvreg.h b/drivers/gpu/drm/nouveau/nvreg.h
index 881f8a585613..fe0f253089ac 100644
--- a/drivers/gpu/drm/nouveau/nvreg.h
+++ b/drivers/gpu/drm/nouveau/nvreg.h
@@ -153,7 +153,8 @@
 #define NV_PCRTC_START					0x00600800
 #define NV_PCRTC_CONFIG					0x00600804
 #	define NV_PCRTC_CONFIG_START_ADDRESS_NON_VGA		(1 << 0)
-#	define NV_PCRTC_CONFIG_START_ADDRESS_HSYNC		(2 << 0)
+#	define NV04_PCRTC_CONFIG_START_ADDRESS_HSYNC		(4 << 0)
+#	define NV10_PCRTC_CONFIG_START_ADDRESS_HSYNC		(2 << 0)
 #define NV_PCRTC_CURSOR_CONFIG				0x00600810
 #	define NV_PCRTC_CURSOR_CONFIG_ENABLE_ENABLE		(1 << 0)
 #	define NV_PCRTC_CURSOR_CONFIG_DOUBLE_SCAN_ENABLE	(1 << 4)
diff --git a/drivers/gpu/drm/radeon/Makefile b/drivers/gpu/drm/radeon/Makefile
index 6cae4f2028d2..e97e6f842699 100644
--- a/drivers/gpu/drm/radeon/Makefile
+++ b/drivers/gpu/drm/radeon/Makefile
@@ -65,10 +65,13 @@ radeon-y += radeon_device.o radeon_asic.o radeon_kms.o \
 	rs400.o rs600.o rs690.o rv515.o r520.o r600.o rv770.o radeon_test.o \
 	r200.o radeon_legacy_tv.o r600_cs.o r600_blit.o r600_blit_shaders.o \
 	r600_blit_kms.o radeon_pm.o atombios_dp.o r600_audio.o r600_hdmi.o \
-	evergreen.o evergreen_cs.o evergreen_blit_shaders.o evergreen_blit_kms.o
+	evergreen.o evergreen_cs.o evergreen_blit_shaders.o evergreen_blit_kms.o \
+	radeon_trace_points.o
 
 radeon-$(CONFIG_COMPAT) += radeon_ioc32.o
 radeon-$(CONFIG_VGA_SWITCHEROO) += radeon_atpx_handler.o
 radeon-$(CONFIG_ACPI) += radeon_acpi.o
 
 obj-$(CONFIG_DRM_RADEON)+= radeon.o
+
+CFLAGS_radeon_trace_points.o := -I$(src)
\ No newline at end of file
diff --git a/drivers/gpu/drm/radeon/ObjectID.h b/drivers/gpu/drm/radeon/ObjectID.h
index c714179d1bfa..c61c3fe9fb98 100644
--- a/drivers/gpu/drm/radeon/ObjectID.h
+++ b/drivers/gpu/drm/radeon/ObjectID.h
@@ -37,6 +37,8 @@
 #define GRAPH_OBJECT_TYPE_CONNECTOR               0x3
 #define GRAPH_OBJECT_TYPE_ROUTER                  0x4
 /* deleted */
+#define GRAPH_OBJECT_TYPE_DISPLAY_PATH            0x6  
+#define GRAPH_OBJECT_TYPE_GENERIC                 0x7
 
 /****************************************************/
 /* Encoder Object ID Definition                     */
@@ -64,6 +66,9 @@
 #define ENCODER_OBJECT_ID_VT1623                  0x10
 #define ENCODER_OBJECT_ID_HDMI_SI1930             0x11
 #define ENCODER_OBJECT_ID_HDMI_INTERNAL           0x12
+#define ENCODER_OBJECT_ID_ALMOND                  0x22
+#define ENCODER_OBJECT_ID_TRAVIS                  0x23
+#define ENCODER_OBJECT_ID_NUTMEG                  0x22
 /* Kaleidoscope (KLDSCP) Class Display Hardware (internal) */
 #define ENCODER_OBJECT_ID_INTERNAL_KLDSCP_TMDS1   0x13
 #define ENCODER_OBJECT_ID_INTERNAL_KLDSCP_DVO1    0x14
@@ -108,6 +113,7 @@
 #define CONNECTOR_OBJECT_ID_DISPLAYPORT           0x13
 #define CONNECTOR_OBJECT_ID_eDP                   0x14
 #define CONNECTOR_OBJECT_ID_MXM                   0x15
+#define CONNECTOR_OBJECT_ID_LVDS_eDP              0x16
 
 /* deleted */
 
@@ -124,6 +130,7 @@
 #define GENERIC_OBJECT_ID_GLSYNC                  0x01
 #define GENERIC_OBJECT_ID_PX2_NON_DRIVABLE        0x02
 #define GENERIC_OBJECT_ID_MXM_OPM                 0x03
+#define GENERIC_OBJECT_ID_STEREO_PIN              0x04        //This object could show up from Misc Object table, it follows ATOM_OBJECT format, and contains one ATOM_OBJECT_GPIO_CNTL_RECORD for the stereo pin
 
 /****************************************************/
 /* Graphics Object ENUM ID Definition               */
@@ -360,6 +367,26 @@
                                                   GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                   ENCODER_OBJECT_ID_GENERAL_EXTERNAL_DVO << OBJECT_ID_SHIFT)
 
+#define ENCODER_ALMOND_ENUM_ID1                  ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_ALMOND << OBJECT_ID_SHIFT)
+
+#define ENCODER_ALMOND_ENUM_ID2                  ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_ALMOND << OBJECT_ID_SHIFT)
+
+#define ENCODER_TRAVIS_ENUM_ID1                  ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_TRAVIS << OBJECT_ID_SHIFT)
+
+#define ENCODER_TRAVIS_ENUM_ID2                  ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_TRAVIS << OBJECT_ID_SHIFT)
+
+#define ENCODER_NUTMEG_ENUM_ID1                  ( GRAPH_OBJECT_TYPE_ENCODER << OBJECT_TYPE_SHIFT |\
+                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                  ENCODER_OBJECT_ID_NUTMEG << OBJECT_ID_SHIFT)
+
 /****************************************************/
 /* Connector Object ID definition - Shared with BIOS */
 /****************************************************/
@@ -421,6 +448,14 @@
                                                  GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_SINGLE_LINK_DVI_D << OBJECT_ID_SHIFT)
 
+#define CONNECTOR_SINGLE_LINK_DVI_D_ENUM_ID3   ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID3 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_SINGLE_LINK_DVI_D << OBJECT_ID_SHIFT)
+
+#define CONNECTOR_SINGLE_LINK_DVI_D_ENUM_ID4   ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID4 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_SINGLE_LINK_DVI_D << OBJECT_ID_SHIFT)
+
 #define CONNECTOR_DUAL_LINK_DVI_D_ENUM_ID1     ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_DUAL_LINK_DVI_D << OBJECT_ID_SHIFT)
@@ -512,6 +547,7 @@
 #define CONNECTOR_7PIN_DIN_ENUM_ID1            ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_7PIN_DIN << OBJECT_ID_SHIFT)
+
 #define CONNECTOR_7PIN_DIN_ENUM_ID2            ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
                                                  GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_7PIN_DIN << OBJECT_ID_SHIFT)
@@ -593,6 +629,14 @@
                                                  GRAPH_OBJECT_ENUM_ID7 << ENUM_ID_SHIFT |\
                                                  CONNECTOR_OBJECT_ID_MXM << OBJECT_ID_SHIFT)          //Mapping to MXM_DAC
 
+#define CONNECTOR_LVDS_eDP_ENUM_ID1            ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_LVDS_eDP << OBJECT_ID_SHIFT)
+
+#define CONNECTOR_LVDS_eDP_ENUM_ID2            ( GRAPH_OBJECT_TYPE_CONNECTOR << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID2 << ENUM_ID_SHIFT |\
+                                                 CONNECTOR_OBJECT_ID_LVDS_eDP << OBJECT_ID_SHIFT)
+
 /****************************************************/
 /* Router Object ID definition - Shared with BIOS   */
 /****************************************************/
@@ -621,6 +665,10 @@
                                                  GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
                                                  GENERIC_OBJECT_ID_MXM_OPM << OBJECT_ID_SHIFT)
 
+#define GENERICOBJECT_STEREO_PIN_ENUM_ID1        (GRAPH_OBJECT_TYPE_GENERIC << OBJECT_TYPE_SHIFT |\
+                                                 GRAPH_OBJECT_ENUM_ID1 << ENUM_ID_SHIFT |\
+                                                 GENERIC_OBJECT_ID_STEREO_PIN << OBJECT_ID_SHIFT)
+
 /****************************************************/
 /* Object Cap definition - Shared with BIOS         */
 /****************************************************/
diff --git a/drivers/gpu/drm/radeon/atombios.h b/drivers/gpu/drm/radeon/atombios.h
index fe359a239df3..58a0cd02c0a2 100644
--- a/drivers/gpu/drm/radeon/atombios.h
+++ b/drivers/gpu/drm/radeon/atombios.h
@@ -73,8 +73,18 @@
 #define ATOM_PPLL1            0
 #define ATOM_PPLL2            1
 #define ATOM_DCPLL            2
+#define ATOM_PPLL0            2
+#define ATOM_EXT_PLL1         8
+#define ATOM_EXT_PLL2         9
+#define ATOM_EXT_CLOCK        10
 #define ATOM_PPLL_INVALID     0xFF
 
+#define ENCODER_REFCLK_SRC_P1PLL       0       
+#define ENCODER_REFCLK_SRC_P2PLL       1
+#define ENCODER_REFCLK_SRC_DCPLL       2
+#define ENCODER_REFCLK_SRC_EXTCLK      3
+#define ENCODER_REFCLK_SRC_INVALID     0xFF
+
 #define ATOM_SCALER1          0
 #define ATOM_SCALER2          1
 
@@ -192,6 +202,9 @@ typedef struct _ATOM_COMMON_TABLE_HEADER
                                   /*Image can't be updated, while Driver needs to carry the new table! */
 }ATOM_COMMON_TABLE_HEADER;
 
+/****************************************************************************/	
+// Structure stores the ROM header.
+/****************************************************************************/	
 typedef struct _ATOM_ROM_HEADER
 {
   ATOM_COMMON_TABLE_HEADER		sHeader;
@@ -221,6 +234,9 @@ typedef struct _ATOM_ROM_HEADER
 	#define	USHORT	void*
 #endif
 
+/****************************************************************************/	
+// Structures used in Command.mtb 
+/****************************************************************************/	
 typedef struct _ATOM_MASTER_LIST_OF_COMMAND_TABLES{
   USHORT ASIC_Init;                              //Function Table, used by various SW components,latest version 1.1
   USHORT GetDisplaySurfaceSize;                  //Atomic Table,  Used by Bios when enabling HW ICON
@@ -312,6 +328,7 @@ typedef struct _ATOM_MASTER_LIST_OF_COMMAND_TABLES{
 #define SetUniphyInstance                        ASIC_StaticPwrMgtStatusChange
 #define HPDInterruptService                      ReadHWAssistedI2CStatus
 #define EnableVGA_Access                         GetSCLKOverMCLKRatio
+#define GetDispObjectInfo                        EnableYUV 
 
 typedef struct _ATOM_MASTER_COMMAND_TABLE
 {
@@ -357,6 +374,24 @@ typedef struct _ATOM_COMMON_ROM_COMMAND_TABLE_HEADER
 /****************************************************************************/	
 #define COMPUTE_MEMORY_PLL_PARAM        1
 #define COMPUTE_ENGINE_PLL_PARAM        2
+#define ADJUST_MC_SETTING_PARAM         3
+
+/****************************************************************************/	
+// Structures used by AdjustMemoryControllerTable
+/****************************************************************************/	
+typedef struct _ATOM_ADJUST_MEMORY_CLOCK_FREQ
+{
+#if ATOM_BIG_ENDIAN
+  ULONG ulPointerReturnFlag:1;      // BYTE_3[7]=1 - Return the pointer to the right Data Block; BYTE_3[7]=0 - Program the right Data Block 
+  ULONG ulMemoryModuleNumber:7;     // BYTE_3[6:0]
+  ULONG ulClockFreq:24;
+#else
+  ULONG ulClockFreq:24;
+  ULONG ulMemoryModuleNumber:7;     // BYTE_3[6:0]
+  ULONG ulPointerReturnFlag:1;      // BYTE_3[7]=1 - Return the pointer to the right Data Block; BYTE_3[7]=0 - Program the right Data Block 
+#endif
+}ATOM_ADJUST_MEMORY_CLOCK_FREQ;
+#define POINTER_RETURN_FLAG             0x80
 
 typedef struct _COMPUTE_MEMORY_ENGINE_PLL_PARAMETERS
 {
@@ -440,6 +475,26 @@ typedef struct _COMPUTE_MEMORY_ENGINE_PLL_PARAMETERS_V4
 #endif
 }COMPUTE_MEMORY_ENGINE_PLL_PARAMETERS_V4;
 
+typedef struct _COMPUTE_MEMORY_ENGINE_PLL_PARAMETERS_V5
+{
+  union
+  {
+    ATOM_COMPUTE_CLOCK_FREQ  ulClock;         //Input Parameter
+    ATOM_S_MPLL_FB_DIVIDER   ulFbDiv;         //Output Parameter
+  };
+  UCHAR   ucRefDiv;                           //Output Parameter      
+  UCHAR   ucPostDiv;                          //Output Parameter      
+  union
+  {
+    UCHAR   ucCntlFlag;                       //Output Flags
+    UCHAR   ucInputFlag;                      //Input Flags. ucInputFlag[0] - Strobe(1)/Performance(0) mode
+  };
+  UCHAR   ucReserved;                       
+}COMPUTE_MEMORY_ENGINE_PLL_PARAMETERS_V5;
+
+// ucInputFlag
+#define ATOM_PLL_INPUT_FLAG_PLL_STROBE_MODE_EN  1   // 1-StrobeMode, 0-PerformanceMode
+
 typedef struct _DYNAMICE_MEMORY_SETTINGS_PARAMETER
 {
   ATOM_COMPUTE_CLOCK_FREQ ulClock;
@@ -583,6 +638,7 @@ typedef struct _DIG_ENCODER_CONTROL_PARAMETERS
 #define ATOM_ENCODER_CONFIG_DPLINKRATE_MASK				0x01
 #define ATOM_ENCODER_CONFIG_DPLINKRATE_1_62GHZ		0x00
 #define ATOM_ENCODER_CONFIG_DPLINKRATE_2_70GHZ		0x01
+#define ATOM_ENCODER_CONFIG_DPLINKRATE_5_40GHZ		0x02
 #define ATOM_ENCODER_CONFIG_LINK_SEL_MASK				  0x04
 #define ATOM_ENCODER_CONFIG_LINKA								  0x00
 #define ATOM_ENCODER_CONFIG_LINKB								  0x04
@@ -608,6 +664,9 @@ typedef struct _DIG_ENCODER_CONTROL_PARAMETERS
 #define ATOM_ENCODER_MODE_TV											13
 #define ATOM_ENCODER_MODE_CV											14
 #define ATOM_ENCODER_MODE_CRT											15
+#define ATOM_ENCODER_MODE_DVO											16
+#define ATOM_ENCODER_MODE_DP_SST                  ATOM_ENCODER_MODE_DP    // For DP1.2
+#define ATOM_ENCODER_MODE_DP_MST                  5                       // For DP1.2
 
 typedef struct _ATOM_DIG_ENCODER_CONFIG_V2
 {
@@ -661,6 +720,7 @@ typedef struct _DIG_ENCODER_CONTROL_PARAMETERS_V2
 #define ATOM_ENCODER_CMD_DP_LINK_TRAINING_START       0x08
 #define ATOM_ENCODER_CMD_DP_LINK_TRAINING_PATTERN1    0x09
 #define ATOM_ENCODER_CMD_DP_LINK_TRAINING_PATTERN2    0x0a
+#define ATOM_ENCODER_CMD_DP_LINK_TRAINING_PATTERN3    0x13
 #define ATOM_ENCODER_CMD_DP_LINK_TRAINING_COMPLETE    0x0b
 #define ATOM_ENCODER_CMD_DP_VIDEO_OFF                 0x0c
 #define ATOM_ENCODER_CMD_DP_VIDEO_ON                  0x0d
@@ -671,24 +731,34 @@ typedef struct _DIG_ENCODER_CONTROL_PARAMETERS_V2
 #define ATOM_ENCODER_STATUS_LINK_TRAINING_COMPLETE    0x10
 #define ATOM_ENCODER_STATUS_LINK_TRAINING_INCOMPLETE  0x00
 
+//ucTableFormatRevision=1
+//ucTableContentRevision=3
 // Following function ENABLE sub-function will be used by driver when TMDS/HDMI/LVDS is used, disable function will be used by driver
 typedef struct _ATOM_DIG_ENCODER_CONFIG_V3
 {
 #if ATOM_BIG_ENDIAN
     UCHAR ucReserved1:1;
-    UCHAR ucDigSel:3;             // =0: DIGA/B/C/D/E/F
+    UCHAR ucDigSel:3;             // =0/1/2/3/4/5: DIG0/1/2/3/4/5 (In register spec also refered as DIGA/B/C/D/E/F)
     UCHAR ucReserved:3;
     UCHAR ucDPLinkRate:1;         // =0: 1.62Ghz, =1: 2.7Ghz
 #else
     UCHAR ucDPLinkRate:1;         // =0: 1.62Ghz, =1: 2.7Ghz
     UCHAR ucReserved:3;
-    UCHAR ucDigSel:3;             // =0: DIGA/B/C/D/E/F
+    UCHAR ucDigSel:3;             // =0/1/2/3/4/5: DIG0/1/2/3/4/5 (In register spec also refered as DIGA/B/C/D/E/F)
     UCHAR ucReserved1:1;
 #endif
 }ATOM_DIG_ENCODER_CONFIG_V3;
 
+#define ATOM_ENCODER_CONFIG_V3_DPLINKRATE_MASK				0x03
+#define ATOM_ENCODER_CONFIG_V3_DPLINKRATE_1_62GHZ		  0x00
+#define ATOM_ENCODER_CONFIG_V3_DPLINKRATE_2_70GHZ		  0x01
 #define ATOM_ENCODER_CONFIG_V3_ENCODER_SEL					  0x70
-
+#define ATOM_ENCODER_CONFIG_V3_DIG0_ENCODER					  0x00
+#define ATOM_ENCODER_CONFIG_V3_DIG1_ENCODER					  0x10
+#define ATOM_ENCODER_CONFIG_V3_DIG2_ENCODER					  0x20
+#define ATOM_ENCODER_CONFIG_V3_DIG3_ENCODER					  0x30
+#define ATOM_ENCODER_CONFIG_V3_DIG4_ENCODER					  0x40
+#define ATOM_ENCODER_CONFIG_V3_DIG5_ENCODER					  0x50
 
 typedef struct _DIG_ENCODER_CONTROL_PARAMETERS_V3
 {
@@ -707,6 +777,56 @@ typedef struct _DIG_ENCODER_CONTROL_PARAMETERS_V3
   UCHAR ucReserved;
 }DIG_ENCODER_CONTROL_PARAMETERS_V3;
 
+//ucTableFormatRevision=1
+//ucTableContentRevision=4
+// start from NI           
+// Following function ENABLE sub-function will be used by driver when TMDS/HDMI/LVDS is used, disable function will be used by driver
+typedef struct _ATOM_DIG_ENCODER_CONFIG_V4
+{
+#if ATOM_BIG_ENDIAN
+    UCHAR ucReserved1:1;
+    UCHAR ucDigSel:3;             // =0/1/2/3/4/5: DIG0/1/2/3/4/5 (In register spec also refered as DIGA/B/C/D/E/F)
+    UCHAR ucReserved:2;
+    UCHAR ucDPLinkRate:2;         // =0: 1.62Ghz, =1: 2.7Ghz, 2=5.4Ghz    <= Changed comparing to previous version
+#else
+    UCHAR ucDPLinkRate:2;         // =0: 1.62Ghz, =1: 2.7Ghz, 2=5.4Ghz    <= Changed comparing to previous version
+    UCHAR ucReserved:2;
+    UCHAR ucDigSel:3;             // =0/1/2/3/4/5: DIG0/1/2/3/4/5 (In register spec also refered as DIGA/B/C/D/E/F)
+    UCHAR ucReserved1:1;
+#endif
+}ATOM_DIG_ENCODER_CONFIG_V4;
+
+#define ATOM_ENCODER_CONFIG_V4_DPLINKRATE_MASK				0x03
+#define ATOM_ENCODER_CONFIG_V4_DPLINKRATE_1_62GHZ		  0x00
+#define ATOM_ENCODER_CONFIG_V4_DPLINKRATE_2_70GHZ		  0x01
+#define ATOM_ENCODER_CONFIG_V4_DPLINKRATE_5_40GHZ		  0x02
+#define ATOM_ENCODER_CONFIG_V4_ENCODER_SEL					  0x70
+#define ATOM_ENCODER_CONFIG_V4_DIG0_ENCODER					  0x00
+#define ATOM_ENCODER_CONFIG_V4_DIG1_ENCODER					  0x10
+#define ATOM_ENCODER_CONFIG_V4_DIG2_ENCODER					  0x20
+#define ATOM_ENCODER_CONFIG_V4_DIG3_ENCODER					  0x30
+#define ATOM_ENCODER_CONFIG_V4_DIG4_ENCODER					  0x40
+#define ATOM_ENCODER_CONFIG_V4_DIG5_ENCODER					  0x50
+
+typedef struct _DIG_ENCODER_CONTROL_PARAMETERS_V4
+{
+  USHORT usPixelClock;      // in 10KHz; for bios convenient
+  union{
+  ATOM_DIG_ENCODER_CONFIG_V4 acConfig;
+  UCHAR ucConfig;
+  };
+  UCHAR ucAction;                              
+  UCHAR ucEncoderMode;
+                            // =0: DP   encoder      
+                            // =1: LVDS encoder          
+                            // =2: DVI  encoder  
+                            // =3: HDMI encoder
+                            // =4: SDVO encoder
+                            // =5: DP audio
+  UCHAR ucLaneNum;          // how many lanes to enable
+  UCHAR ucBitPerColor;      // only valid for DP mode when ucAction = ATOM_ENCODER_CMD_SETUP
+  UCHAR ucHPD_ID;           // HPD ID (1-6). =0 means to skip HDP programming. New comparing to previous version
+}DIG_ENCODER_CONTROL_PARAMETERS_V4;
 
 // define ucBitPerColor: 
 #define PANEL_BPC_UNDEFINE                               0x00
@@ -893,6 +1013,7 @@ typedef struct _ATOM_DIG_TRANSMITTER_CONFIG_V3
 #endif
 }ATOM_DIG_TRANSMITTER_CONFIG_V3;
 
+
 typedef struct _DIG_TRANSMITTER_CONTROL_PARAMETERS_V3
 {
 	union
@@ -936,6 +1057,149 @@ typedef struct _DIG_TRANSMITTER_CONTROL_PARAMETERS_V3
 #define ATOM_TRANSMITTER_CONFIG_V3_TRANSMITTER2           	0x40	//CD
 #define ATOM_TRANSMITTER_CONFIG_V3_TRANSMITTER3           	0x80	//EF
 
+
+/****************************************************************************/	
+// Structures used by UNIPHYTransmitterControlTable V1.4
+// ASIC Families: NI
+// ucTableFormatRevision=1
+// ucTableContentRevision=4
+/****************************************************************************/	
+typedef struct _ATOM_DP_VS_MODE_V4
+{
+  UCHAR ucLaneSel;
+ 	union
+ 	{  
+ 	  UCHAR ucLaneSet;
+ 	  struct {
+#if ATOM_BIG_ENDIAN
+ 		  UCHAR ucPOST_CURSOR2:2;         //Bit[7:6] Post Cursor2 Level      <= New in V4
+ 		  UCHAR ucPRE_EMPHASIS:3;         //Bit[5:3] Pre-emphasis Level
+ 		  UCHAR ucVOLTAGE_SWING:3;        //Bit[2:0] Voltage Swing Level
+#else
+ 		  UCHAR ucVOLTAGE_SWING:3;        //Bit[2:0] Voltage Swing Level
+ 		  UCHAR ucPRE_EMPHASIS:3;         //Bit[5:3] Pre-emphasis Level
+ 		  UCHAR ucPOST_CURSOR2:2;         //Bit[7:6] Post Cursor2 Level      <= New in V4
+#endif
+ 		};
+ 	}; 
+}ATOM_DP_VS_MODE_V4;
+ 
+typedef struct _ATOM_DIG_TRANSMITTER_CONFIG_V4
+{
+#if ATOM_BIG_ENDIAN
+  UCHAR ucTransmitterSel:2;         //bit7:6: =0 Dig Transmitter 1 ( Uniphy AB )
+                                    //        =1 Dig Transmitter 2 ( Uniphy CD )
+                                    //        =2 Dig Transmitter 3 ( Uniphy EF )
+  UCHAR ucRefClkSource:2;           //bit5:4: PPLL1 =0, PPLL2=1, DCPLL=2, EXT_CLK=3   <= New
+  UCHAR ucEncoderSel:1;             //bit3=0: Data/Clk path source from DIGA/C/E. =1: Data/clk path source from DIGB/D/F
+  UCHAR ucLinkSel:1;                //bit2=0: Uniphy LINKA or C or E when fDualLinkConnector=0. when fDualLinkConnector=1, it means master link of dual link is A or C or E
+                                    //    =1: Uniphy LINKB or D or F when fDualLinkConnector=0. when fDualLinkConnector=1, it means master link of dual link is B or D or F
+  UCHAR fCoherentMode:1;            //bit1=1: Coherent Mode ( for DVI/HDMI mode )
+  UCHAR fDualLinkConnector:1;       //bit0=1: Dual Link DVI connector
+#else
+  UCHAR fDualLinkConnector:1;       //bit0=1: Dual Link DVI connector
+  UCHAR fCoherentMode:1;            //bit1=1: Coherent Mode ( for DVI/HDMI mode )
+  UCHAR ucLinkSel:1;                //bit2=0: Uniphy LINKA or C or E when fDualLinkConnector=0. when fDualLinkConnector=1, it means master link of dual link is A or C or E
+                                    //    =1: Uniphy LINKB or D or F when fDualLinkConnector=0. when fDualLinkConnector=1, it means master link of dual link is B or D or F
+  UCHAR ucEncoderSel:1;             //bit3=0: Data/Clk path source from DIGA/C/E. =1: Data/clk path source from DIGB/D/F
+  UCHAR ucRefClkSource:2;           //bit5:4: PPLL1 =0, PPLL2=1, DCPLL=2, EXT_CLK=3   <= New
+  UCHAR ucTransmitterSel:2;         //bit7:6: =0 Dig Transmitter 1 ( Uniphy AB )
+                                    //        =1 Dig Transmitter 2 ( Uniphy CD )
+                                    //        =2 Dig Transmitter 3 ( Uniphy EF )
+#endif
+}ATOM_DIG_TRANSMITTER_CONFIG_V4;
+
+typedef struct _DIG_TRANSMITTER_CONTROL_PARAMETERS_V4
+{
+  union
+  {
+    USHORT usPixelClock;		// in 10KHz; for bios convenient
+    USHORT usInitInfo;			// when init uniphy,lower 8bit is used for connector type defined in objectid.h
+    ATOM_DP_VS_MODE_V4 asMode; // DP Voltage swing mode     Redefined comparing to previous version
+  };
+  union
+  {
+  ATOM_DIG_TRANSMITTER_CONFIG_V4 acConfig;
+  UCHAR ucConfig;
+  };
+  UCHAR ucAction;				    // define as ATOM_TRANSMITER_ACTION_XXX	                        
+  UCHAR ucLaneNum;
+  UCHAR ucReserved[3];
+}DIG_TRANSMITTER_CONTROL_PARAMETERS_V4;
+
+//ucConfig 
+//Bit0
+#define ATOM_TRANSMITTER_CONFIG_V4_DUAL_LINK_CONNECTOR			0x01
+//Bit1
+#define ATOM_TRANSMITTER_CONFIG_V4_COHERENT				          0x02
+//Bit2
+#define ATOM_TRANSMITTER_CONFIG_V4_LINK_SEL_MASK		        0x04
+#define ATOM_TRANSMITTER_CONFIG_V4_LINKA  			            0x00			
+#define ATOM_TRANSMITTER_CONFIG_V4_LINKB				            0x04
+// Bit3
+#define ATOM_TRANSMITTER_CONFIG_V4_ENCODER_SEL_MASK	        0x08
+#define ATOM_TRANSMITTER_CONFIG_V4_DIG1_ENCODER		          0x00				 
+#define ATOM_TRANSMITTER_CONFIG_V4_DIG2_ENCODER		          0x08				
+// Bit5:4
+#define ATOM_TRANSMITTER_CONFIG_V4_REFCLK_SEL_MASK 	        0x30
+#define ATOM_TRANSMITTER_CONFIG_V4_P1PLL         		        0x00
+#define ATOM_TRANSMITTER_CONFIG_V4_P2PLL		                0x10
+#define ATOM_TRANSMITTER_CONFIG_V4_DCPLL		                0x20   // New in _V4
+#define ATOM_TRANSMITTER_CONFIG_V4_REFCLK_SRC_EXT           0x30   // Changed comparing to V3
+// Bit7:6
+#define ATOM_TRANSMITTER_CONFIG_V4_TRANSMITTER_SEL_MASK     0xC0
+#define ATOM_TRANSMITTER_CONFIG_V4_TRANSMITTER1           	0x00	//AB
+#define ATOM_TRANSMITTER_CONFIG_V4_TRANSMITTER2           	0x40	//CD
+#define ATOM_TRANSMITTER_CONFIG_V4_TRANSMITTER3           	0x80	//EF
+
+
+/****************************************************************************/	
+// Structures used by ExternalEncoderControlTable V1.3
+// ASIC Families: Evergreen, Llano, NI
+// ucTableFormatRevision=1
+// ucTableContentRevision=3
+/****************************************************************************/	
+
+typedef struct _EXTERNAL_ENCODER_CONTROL_PARAMETERS_V3
+{
+  union{
+  USHORT usPixelClock;      // pixel clock in 10Khz, valid when ucAction=SETUP/ENABLE_OUTPUT 
+  USHORT usConnectorId;     // connector id, valid when ucAction = INIT
+  };
+  UCHAR  ucConfig;          // indicate which encoder, and DP link rate when ucAction = SETUP/ENABLE_OUTPUT  
+  UCHAR  ucAction;          // 
+  UCHAR  ucEncoderMode;     // encoder mode, only used when ucAction = SETUP/ENABLE_OUTPUT
+  UCHAR  ucLaneNum;         // lane number, only used when ucAction = SETUP/ENABLE_OUTPUT  
+  UCHAR  ucBitPerColor;     // output bit per color, only valid when ucAction = SETUP/ENABLE_OUTPUT and ucEncodeMode= DP
+  UCHAR  ucReserved;        
+}EXTERNAL_ENCODER_CONTROL_PARAMETERS_V3;
+
+// ucAction
+#define EXTERNAL_ENCODER_ACTION_V3_DISABLE_OUTPUT         0x00
+#define EXTERNAL_ENCODER_ACTION_V3_ENABLE_OUTPUT          0x01
+#define EXTERNAL_ENCODER_ACTION_V3_ENCODER_INIT           0x07
+#define EXTERNAL_ENCODER_ACTION_V3_ENCODER_SETUP          0x0f
+#define EXTERNAL_ENCODER_ACTION_V3_ENCODER_BLANKING_OFF   0x10
+#define EXTERNAL_ENCODER_ACTION_V3_ENCODER_BLANKING       0x11
+#define EXTERNAL_ENCODER_ACTION_V3_DACLOAD_DETECTION      0x12
+
+// ucConfig
+#define EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_MASK				0x03
+#define EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_1_62GHZ		  0x00
+#define EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_2_70GHZ		  0x01
+#define EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_5_40GHZ		  0x02
+#define EXTERNAL_ENCODER_CONFIG_V3_ENCODER_SEL_MASK		    0x70
+#define EXTERNAL_ENCODER_CONFIG_V3_ENCODER1		            0x00
+#define EXTERNAL_ENCODER_CONFIG_V3_ENCODER2		            0x10
+#define EXTERNAL_ENCODER_CONFIG_V3_ENCODER3		            0x20
+
+typedef struct _EXTERNAL_ENCODER_CONTROL_PS_ALLOCATION_V3
+{
+  EXTERNAL_ENCODER_CONTROL_PARAMETERS_V3 sExtEncoder;
+  ULONG ulReserved[2];
+}EXTERNAL_ENCODER_CONTROL_PS_ALLOCATION_V3;
+
+
 /****************************************************************************/	
 // Structures used by DAC1OuputControlTable
 //                    DAC2OuputControlTable
@@ -1142,6 +1406,7 @@ typedef struct _PIXEL_CLOCK_PARAMETERS_V2
 #define PIXEL_CLOCK_V4_MISC_SS_ENABLE               0x10
 #define PIXEL_CLOCK_V4_MISC_COHERENT_MODE           0x20
 
+
 typedef struct _PIXEL_CLOCK_PARAMETERS_V3
 {
   USHORT usPixelClock;                // in 10kHz unit; for bios convenient = (RefClk*FB_Div)/(Ref_Div*Post_Div)
@@ -1202,6 +1467,55 @@ typedef struct _PIXEL_CLOCK_PARAMETERS_V5
 #define PIXEL_CLOCK_V5_MISC_HDMI_32BPP              0x08
 #define PIXEL_CLOCK_V5_MISC_REF_DIV_SRC             0x10
 
+typedef struct _CRTC_PIXEL_CLOCK_FREQ
+{
+#if ATOM_BIG_ENDIAN
+  ULONG  ucCRTC:8;            // ATOM_CRTC1~6, indicate the CRTC controller to 
+                              // drive the pixel clock. not used for DCPLL case.
+  ULONG  ulPixelClock:24;     // target the pixel clock to drive the CRTC timing. 
+                              // 0 means disable PPLL/DCPLL. Expanded to 24 bits comparing to previous version.
+#else
+  ULONG  ulPixelClock:24;     // target the pixel clock to drive the CRTC timing. 
+                              // 0 means disable PPLL/DCPLL. Expanded to 24 bits comparing to previous version.
+  ULONG  ucCRTC:8;            // ATOM_CRTC1~6, indicate the CRTC controller to 
+                              // drive the pixel clock. not used for DCPLL case.
+#endif
+}CRTC_PIXEL_CLOCK_FREQ;
+
+typedef struct _PIXEL_CLOCK_PARAMETERS_V6
+{
+  union{
+    CRTC_PIXEL_CLOCK_FREQ ulCrtcPclkFreq;    // pixel clock and CRTC id frequency 
+    ULONG ulDispEngClkFreq;                  // dispclk frequency
+  };
+  USHORT usFbDiv;            // feedback divider integer part. 
+  UCHAR  ucPostDiv;          // post divider. 
+  UCHAR  ucRefDiv;           // Reference divider
+  UCHAR  ucPpll;             // ATOM_PPLL1/ATOM_PPLL2/ATOM_DCPLL
+  UCHAR  ucTransmitterID;    // ASIC encoder id defined in objectId.h, 
+                             // indicate which graphic encoder will be used. 
+  UCHAR  ucEncoderMode;      // Encoder mode: 
+  UCHAR  ucMiscInfo;         // bit[0]= Force program PPLL 
+                             // bit[1]= when VGA timing is used. 
+                             // bit[3:2]= HDMI panel bit depth: =0: 24bpp =1:30bpp, =2:32bpp
+                             // bit[4]= RefClock source for PPLL. 
+                             // =0: XTLAIN( default mode )
+	                           // =1: other external clock source, which is pre-defined                                            
+                             //     by VBIOS depend on the feature required.
+                             // bit[7:5]: reserved.
+  ULONG  ulFbDivDecFrac;     // 20 bit feedback divider decimal fraction part, range from 1~999999 ( 0.000001 to 0.999999 )
+
+}PIXEL_CLOCK_PARAMETERS_V6;
+
+#define PIXEL_CLOCK_V6_MISC_FORCE_PROG_PPLL					0x01
+#define PIXEL_CLOCK_V6_MISC_VGA_MODE								0x02
+#define PIXEL_CLOCK_V6_MISC_HDMI_BPP_MASK           0x0c
+#define PIXEL_CLOCK_V6_MISC_HDMI_24BPP              0x00
+#define PIXEL_CLOCK_V6_MISC_HDMI_36BPP              0x04
+#define PIXEL_CLOCK_V6_MISC_HDMI_30BPP              0x08
+#define PIXEL_CLOCK_V6_MISC_HDMI_48BPP              0x0c
+#define PIXEL_CLOCK_V6_MISC_REF_DIV_SRC             0x10
+
 typedef struct _GET_DISP_PLL_STATUS_INPUT_PARAMETERS_V2
 {
   PIXEL_CLOCK_PARAMETERS_V3 sDispClkInput;
@@ -1241,10 +1555,11 @@ typedef struct _ADJUST_DISPLAY_PLL_PARAMETERS
 typedef struct _ADJUST_DISPLAY_PLL_INPUT_PARAMETERS_V3
 {
 	USHORT usPixelClock;                    // target pixel clock
-	UCHAR ucTransmitterID;                  // transmitter id defined in objectid.h
+	UCHAR ucTransmitterID;                  // GPU transmitter id defined in objectid.h
 	UCHAR ucEncodeMode;                     // encoder mode: CRT, LVDS, DP, TMDS or HDMI
   UCHAR ucDispPllConfig;                 // display pll configure parameter defined as following DISPPLL_CONFIG_XXXX
-	UCHAR ucReserved[3];
+  UCHAR ucExtTransmitterID;               // external encoder id.
+	UCHAR ucReserved[2];
 }ADJUST_DISPLAY_PLL_INPUT_PARAMETERS_V3;
 
 // usDispPllConfig v1.2 for RoadRunner
@@ -1358,6 +1673,7 @@ typedef struct _SET_UP_HW_I2C_DATA_PARAMETERS
 /**************************************************************************/
 #define SPEED_FAN_CONTROL_PS_ALLOCATION   WRITE_ONE_BYTE_HW_I2C_DATA_PARAMETERS
 
+
 /****************************************************************************/	
 // Structures used by PowerConnectorDetectionTable
 /****************************************************************************/	
@@ -1438,6 +1754,31 @@ typedef struct _ENABLE_SPREAD_SPECTRUM_ON_PPLL_V2
 #define ATOM_PPLL_SS_AMOUNT_V2_NFRAC_MASK     0x0F00
 #define ATOM_PPLL_SS_AMOUNT_V2_NFRAC_SHIFT    8
 
+// Used by DCE5.0
+ typedef struct _ENABLE_SPREAD_SPECTRUM_ON_PPLL_V3
+{
+  USHORT  usSpreadSpectrumAmountFrac;   // SS_AMOUNT_DSFRAC New in DCE5.0
+  UCHAR   ucSpreadSpectrumType;	        // Bit[0]: 0-Down Spread,1-Center Spread. 
+                                        // Bit[1]: 1-Ext. 0-Int. 
+                                        // Bit[3:2]: =0 P1PLL =1 P2PLL =2 DCPLL
+                                        // Bits[7:4] reserved
+  UCHAR   ucEnable;	                    // ATOM_ENABLE or ATOM_DISABLE
+  USHORT  usSpreadSpectrumAmount;      	// Includes SS_AMOUNT_FBDIV[7:0] and SS_AMOUNT_NFRAC_SLIP[11:8]    
+  USHORT  usSpreadSpectrumStep;	        // SS_STEP_SIZE_DSFRAC
+}ENABLE_SPREAD_SPECTRUM_ON_PPLL_V3;
+    
+#define ATOM_PPLL_SS_TYPE_V3_DOWN_SPREAD      0x00
+#define ATOM_PPLL_SS_TYPE_V3_CENTRE_SPREAD    0x01
+#define ATOM_PPLL_SS_TYPE_V3_EXT_SPREAD       0x02
+#define ATOM_PPLL_SS_TYPE_V3_PPLL_SEL_MASK    0x0c
+#define ATOM_PPLL_SS_TYPE_V3_P1PLL            0x00
+#define ATOM_PPLL_SS_TYPE_V3_P2PLL            0x04
+#define ATOM_PPLL_SS_TYPE_V3_DCPLL            0x08
+#define ATOM_PPLL_SS_AMOUNT_V3_FBDIV_MASK     0x00FF
+#define ATOM_PPLL_SS_AMOUNT_V3_FBDIV_SHIFT    0
+#define ATOM_PPLL_SS_AMOUNT_V3_NFRAC_MASK     0x0F00
+#define ATOM_PPLL_SS_AMOUNT_V3_NFRAC_SHIFT    8
+
 #define ENABLE_SPREAD_SPECTRUM_ON_PPLL_PS_ALLOCATION  ENABLE_SPREAD_SPECTRUM_ON_PPLL
 
 /**************************************************************************/
@@ -1706,7 +2047,7 @@ typedef struct _ATOM_MASTER_LIST_OF_DATA_TABLES
   USHORT        StandardVESA_Timing;      // Only used by Bios
   USHORT        FirmwareInfo;             // Shared by various SW components,latest version 1.4
   USHORT        DAC_Info;                 // Will be obsolete from R600
-  USHORT        LVDS_Info;                // Shared by various SW components,latest version 1.1 
+  USHORT        LCD_Info;                 // Shared by various SW components,latest version 1.3, was called LVDS_Info 
   USHORT        TMDS_Info;                // Will be obsolete from R600
   USHORT        AnalogTV_Info;            // Shared by various SW components,latest version 1.1 
   USHORT        SupportedDevicesInfo;     // Will be obsolete from R600
@@ -1736,12 +2077,16 @@ typedef struct _ATOM_MASTER_LIST_OF_DATA_TABLES
 	USHORT				PowerSourceInfo;					// Shared by various SW components, latest versoin 1.1
 }ATOM_MASTER_LIST_OF_DATA_TABLES;
 
+// For backward compatible 
+#define LVDS_Info                LCD_Info
+
 typedef struct _ATOM_MASTER_DATA_TABLE
 { 
   ATOM_COMMON_TABLE_HEADER sHeader;  
   ATOM_MASTER_LIST_OF_DATA_TABLES   ListOfDataTables;
 }ATOM_MASTER_DATA_TABLE;
 
+
 /****************************************************************************/	
 // Structure used in MultimediaCapabilityInfoTable
 /****************************************************************************/	
@@ -1776,6 +2121,7 @@ typedef struct _ATOM_MULTIMEDIA_CONFIG_INFO
   UCHAR                    ucVideoInput4Info;// Video Input 4 Type (1:0) F/B setting (2) physical connector ID (5:3) reserved (7:6)
 }ATOM_MULTIMEDIA_CONFIG_INFO;
 
+
 /****************************************************************************/	
 // Structures used in FirmwareInfoTable
 /****************************************************************************/	
@@ -2031,8 +2377,47 @@ typedef struct _ATOM_FIRMWARE_INFO_V2_1
   UCHAR                           ucReserved4[3];
 }ATOM_FIRMWARE_INFO_V2_1;
 
+//the structure below to be used from NI
+//ucTableFormatRevision=2
+//ucTableContentRevision=2
+typedef struct _ATOM_FIRMWARE_INFO_V2_2
+{
+  ATOM_COMMON_TABLE_HEADER        sHeader; 
+  ULONG                           ulFirmwareRevision;
+  ULONG                           ulDefaultEngineClock;       //In 10Khz unit
+  ULONG                           ulDefaultMemoryClock;       //In 10Khz unit
+  ULONG                           ulReserved[2];
+  ULONG                           ulReserved1;                //Was ulMaxEngineClockPLL_Output; //In 10Khz unit*
+  ULONG                           ulReserved2;                //Was ulMaxMemoryClockPLL_Output; //In 10Khz unit*
+  ULONG                           ulMaxPixelClockPLL_Output;  //In 10Khz unit
+  ULONG                           ulBinaryAlteredInfo;        //Was ulASICMaxEngineClock  ?
+  ULONG                           ulDefaultDispEngineClkFreq; //In 10Khz unit. This is the frequency before DCDTO, corresponding to usBootUpVDDCVoltage.          
+  UCHAR                           ucReserved3;                //Was ucASICMaxTemperature;
+  UCHAR                           ucMinAllowedBL_Level;
+  USHORT                          usBootUpVDDCVoltage;        //In MV unit
+  USHORT                          usLcdMinPixelClockPLL_Output; // In MHz unit
+  USHORT                          usLcdMaxPixelClockPLL_Output; // In MHz unit
+  ULONG                           ulReserved4;                //Was ulAsicMaximumVoltage
+  ULONG                           ulMinPixelClockPLL_Output;  //In 10Khz unit
+  ULONG                           ulReserved5;                //Was usMinEngineClockPLL_Input and usMaxEngineClockPLL_Input
+  ULONG                           ulReserved6;                //Was usMinEngineClockPLL_Output and usMinMemoryClockPLL_Input
+  ULONG                           ulReserved7;                //Was usMaxMemoryClockPLL_Input and usMinMemoryClockPLL_Output
+  USHORT                          usReserved11;               //Was usMaxPixelClock;  //In 10Khz unit, Max.  Pclk used only for DAC
+  USHORT                          usMinPixelClockPLL_Input;   //In 10Khz unit
+  USHORT                          usMaxPixelClockPLL_Input;   //In 10Khz unit
+  USHORT                          usBootUpVDDCIVoltage;       //In unit of mv; Was usMinPixelClockPLL_Output;
+  ATOM_FIRMWARE_CAPABILITY_ACCESS usFirmwareCapability;
+  USHORT                          usCoreReferenceClock;       //In 10Khz unit	
+  USHORT                          usMemoryReferenceClock;     //In 10Khz unit	
+  USHORT                          usUniphyDPModeExtClkFreq;   //In 10Khz unit, if it is 0, In DP Mode Uniphy Input clock from internal PPLL, otherwise Input clock from external Spread clock
+  UCHAR                           ucMemoryModule_ID;          //Indicate what is the board design
+  UCHAR                           ucReserved9[3];
+  USHORT                          usBootUpMVDDCVoltage;       //In unit of mv; Was usMinPixelClockPLL_Output;
+  USHORT                          usReserved12;
+  ULONG                           ulReserved10[3];            // New added comparing to previous version
+}ATOM_FIRMWARE_INFO_V2_2;
 
-#define ATOM_FIRMWARE_INFO_LAST  ATOM_FIRMWARE_INFO_V2_1
+#define ATOM_FIRMWARE_INFO_LAST  ATOM_FIRMWARE_INFO_V2_2
 
 /****************************************************************************/	
 // Structures used in IntegratedSystemInfoTable
@@ -2212,7 +2597,7 @@ ulDockingPinCFGInfo: [15:0]-Bus/Device/Function # to CFG to read this Docking Pi
 ucDockingPinBit:     which bit in this register to read the pin status;
 ucDockingPinPolarity:Polarity of the pin when docked;
 
-ulCPUCapInfo:        [7:0]=1:Griffin;[7:0]=2:Greyhound;[7:0]=3:K8, other bits reserved for now and must be 0x0
+ulCPUCapInfo:        [7:0]=1:Griffin;[7:0]=2:Greyhound;[7:0]=3:K8, [7:0]=4:Pharaoh, other bits reserved for now and must be 0x0
 
 usNumberOfCyclesInPeriod:Indicate how many cycles when PWM duty is 100%.
 
@@ -2250,6 +2635,14 @@ usMinUpStreamHTLinkWidth:    Asymmetric link width support in the future, to rep
 usMinDownStreamHTLinkWidth:  same as above.
 */
 
+// ATOM_INTEGRATED_SYSTEM_INFO::ulCPUCapInfo  - CPU type definition 
+#define    INTEGRATED_SYSTEM_INFO__UNKNOWN_CPU             0
+#define    INTEGRATED_SYSTEM_INFO__AMD_CPU__GRIFFIN        1
+#define    INTEGRATED_SYSTEM_INFO__AMD_CPU__GREYHOUND      2
+#define    INTEGRATED_SYSTEM_INFO__AMD_CPU__K8             3
+#define    INTEGRATED_SYSTEM_INFO__AMD_CPU__PHARAOH        4
+
+#define    INTEGRATED_SYSTEM_INFO__AMD_CPU__MAX_CODE       INTEGRATED_SYSTEM_INFO__AMD_CPU__PHARAOH    // this deff reflects max defined CPU code
 
 #define SYSTEM_CONFIG_POWEREXPRESS_ENABLE                 0x00000001
 #define SYSTEM_CONFIG_RUN_AT_OVERDRIVE_ENGINE             0x00000002
@@ -2778,8 +3171,88 @@ typedef struct _ATOM_LVDS_INFO_V12
 #define PANEL_RANDOM_DITHER   0x80
 #define PANEL_RANDOM_DITHER_MASK   0x80
 
+#define ATOM_LVDS_INFO_LAST  ATOM_LVDS_INFO_V12   // no need to change this 
+
+/****************************************************************************/	
+// Structures used by LCD_InfoTable V1.3    Note: previous version was called ATOM_LVDS_INFO_V12
+// ASIC Families:  NI
+// ucTableFormatRevision=1
+// ucTableContentRevision=3
+/****************************************************************************/	
+typedef struct _ATOM_LCD_INFO_V13
+{
+  ATOM_COMMON_TABLE_HEADER sHeader;  
+  ATOM_DTD_FORMAT     sLCDTiming;
+  USHORT              usExtInfoTableOffset;
+  USHORT              usSupportedRefreshRate;     //Refer to panel info table in ATOMBIOS extension Spec.
+  ULONG               ulReserved0;
+  UCHAR               ucLCD_Misc;                // Reorganized in V13
+                                                 // Bit0: {=0:single, =1:dual},
+                                                 // Bit1: {=0:LDI format for RGB888, =1 FPDI format for RGB888}  // was {=0:666RGB, =1:888RGB},
+                                                 // Bit3:2: {Grey level}
+                                                 // Bit6:4 Color Bit Depth definition (see below definition in EDID V1.4 @BYTE 14h) 
+                                                 // Bit7   Reserved.  was for ATOM_PANEL_MISC_API_ENABLED, still need it?  
+  UCHAR               ucPanelDefaultRefreshRate;
+  UCHAR               ucPanelIdentification;
+  UCHAR               ucSS_Id;
+  USHORT              usLCDVenderID;
+  USHORT              usLCDProductID;
+  UCHAR               ucLCDPanel_SpecialHandlingCap;  // Reorganized in V13 
+                                                 // Bit0: Once DAL sees this CAP is set, it will read EDID from LCD on its own
+                                                 // Bit1: See LCDPANEL_CAP_DRR_SUPPORTED
+                                                 // Bit2: a quick reference whether an embadded panel (LCD1 ) is LVDS (0) or eDP (1)
+                                                 // Bit7-3: Reserved 
+  UCHAR               ucPanelInfoSize;					 //  start from ATOM_DTD_FORMAT to end of panel info, include ExtInfoTable
+  USHORT              usBacklightPWM;            //  Backlight PWM in Hz. New in _V13
+
+  UCHAR               ucPowerSequenceDIGONtoDE_in4Ms;
+  UCHAR               ucPowerSequenceDEtoVARY_BL_in4Ms;
+  UCHAR               ucPowerSequenceDEtoDIGON_in4Ms;
+  UCHAR               ucPowerSequenceVARY_BLtoDE_in4Ms;
+
+  UCHAR               ucOffDelay_in4Ms;
+  UCHAR               ucPowerSequenceVARY_BLtoBLON_in4Ms;
+  UCHAR               ucPowerSequenceBLONtoVARY_BL_in4Ms;
+  UCHAR               ucReserved1;
+
+  ULONG               ulReserved[4];
+}ATOM_LCD_INFO_V13;  
+
+#define ATOM_LCD_INFO_LAST  ATOM_LCD_INFO_V13    
+
+//Definitions for ucLCD_Misc
+#define ATOM_PANEL_MISC_V13_DUAL                   0x00000001
+#define ATOM_PANEL_MISC_V13_FPDI                   0x00000002
+#define ATOM_PANEL_MISC_V13_GREY_LEVEL             0x0000000C
+#define ATOM_PANEL_MISC_V13_GREY_LEVEL_SHIFT       2
+#define ATOM_PANEL_MISC_V13_COLOR_BIT_DEPTH_MASK   0x70
+#define ATOM_PANEL_MISC_V13_6BIT_PER_COLOR         0x10
+#define ATOM_PANEL_MISC_V13_8BIT_PER_COLOR         0x20
+
+//Color Bit Depth definition in EDID V1.4 @BYTE 14h
+//Bit 6  5  4
+                              //      0  0  0  -  Color bit depth is undefined
+                              //      0  0  1  -  6 Bits per Primary Color
+                              //      0  1  0  -  8 Bits per Primary Color
+                              //      0  1  1  - 10 Bits per Primary Color
+                              //      1  0  0  - 12 Bits per Primary Color
+                              //      1  0  1  - 14 Bits per Primary Color
+                              //      1  1  0  - 16 Bits per Primary Color
+                              //      1  1  1  - Reserved
+ 
+//Definitions for ucLCDPanel_SpecialHandlingCap:
+
+//Once DAL sees this CAP is set, it will read EDID from LCD on its own instead of using sLCDTiming in ATOM_LVDS_INFO_V12. 
+//Other entries in ATOM_LVDS_INFO_V12 are still valid/useful to DAL 
+#define	LCDPANEL_CAP_V13_READ_EDID              0x1        // = LCDPANEL_CAP_READ_EDID no change comparing to previous version
+
+//If a design supports DRR (dynamic refresh rate) on internal panels (LVDS or EDP), this cap is set in ucLCDPanel_SpecialHandlingCap together
+//with multiple supported refresh rates@usSupportedRefreshRate. This cap should not be set when only slow refresh rate is supported (static
+//refresh rate switch by SW. This is only valid from ATOM_LVDS_INFO_V12
+#define	LCDPANEL_CAP_V13_DRR_SUPPORTED          0x2        // = LCDPANEL_CAP_DRR_SUPPORTED no change comparing to previous version
 
-#define ATOM_LVDS_INFO_LAST  ATOM_LVDS_INFO_V12
+//Use this cap bit for a quick reference whether an embadded panel (LCD1 ) is LVDS or eDP.
+#define	LCDPANEL_CAP_V13_eDP                    0x4        // = LCDPANEL_CAP_eDP no change comparing to previous version
 
 typedef struct  _ATOM_PATCH_RECORD_MODE
 {
@@ -2944,9 +3417,9 @@ typedef struct _ATOM_DPCD_INFO
 #define MAX_DTD_MODE_IN_VRAM            6
 #define ATOM_DTD_MODE_SUPPORT_TBL_SIZE  (MAX_DTD_MODE_IN_VRAM*28)    //28= (SIZEOF ATOM_DTD_FORMAT) 
 #define ATOM_STD_MODE_SUPPORT_TBL_SIZE  32*8                         //32 is a predefined number,8= (SIZEOF ATOM_STD_FORMAT)
-#define DFP_ENCODER_TYPE_OFFSET					0x80
-#define DP_ENCODER_LANE_NUM_OFFSET			0x84
-#define DP_ENCODER_LINK_RATE_OFFSET			0x88
+//20 bytes for Encoder Type and DPCD in STD EDID area
+#define DFP_ENCODER_TYPE_OFFSET         (ATOM_EDID_RAW_DATASIZE + ATOM_DTD_MODE_SUPPORT_TBL_SIZE + ATOM_STD_MODE_SUPPORT_TBL_SIZE - 20)    
+#define ATOM_DP_DPCD_OFFSET             (DFP_ENCODER_TYPE_OFFSET + 4 )        
 
 #define ATOM_HWICON1_SURFACE_ADDR       0
 #define ATOM_HWICON2_SURFACE_ADDR       (ATOM_HWICON1_SURFACE_ADDR + ATOM_HWICON_SURFACE_SIZE)
@@ -2997,14 +3470,16 @@ typedef struct _ATOM_DPCD_INFO
 #define ATOM_DFP5_DTD_MODE_TBL_ADDR     (ATOM_DFP5_EDID_ADDR + ATOM_EDID_RAW_DATASIZE)
 #define ATOM_DFP5_STD_MODE_TBL_ADDR     (ATOM_DFP5_DTD_MODE_TBL_ADDR + ATOM_DTD_MODE_SUPPORT_TBL_SIZE)
 
-#define ATOM_DP_TRAINING_TBL_ADDR				(ATOM_DFP5_STD_MODE_TBL_ADDR+ATOM_STD_MODE_SUPPORT_TBL_SIZE)       
+#define ATOM_DP_TRAINING_TBL_ADDR       (ATOM_DFP5_STD_MODE_TBL_ADDR + ATOM_STD_MODE_SUPPORT_TBL_SIZE)
 
-#define ATOM_STACK_STORAGE_START        (ATOM_DP_TRAINING_TBL_ADDR+256)       
-#define ATOM_STACK_STORAGE_END          ATOM_STACK_STORAGE_START+512        
+#define ATOM_STACK_STORAGE_START        (ATOM_DP_TRAINING_TBL_ADDR + 1024)       
+#define ATOM_STACK_STORAGE_END          ATOM_STACK_STORAGE_START + 512        
 
 //The size below is in Kb!
 #define ATOM_VRAM_RESERVE_SIZE         ((((ATOM_STACK_STORAGE_END - ATOM_HWICON1_SURFACE_ADDR)>>10)+4)&0xFFFC)
    
+#define ATOM_VRAM_RESERVE_V2_SIZE      32
+
 #define	ATOM_VRAM_OPERATION_FLAGS_MASK         0xC0000000L
 #define ATOM_VRAM_OPERATION_FLAGS_SHIFT        30
 #define	ATOM_VRAM_BLOCK_NEEDS_NO_RESERVATION   0x1
@@ -3206,6 +3681,15 @@ typedef struct  _ATOM_DISPLAY_OBJECT_PATH
   USHORT    usGraphicObjIds[1];                             //1st Encoder Obj source from GPU to last Graphic Obj destinate to connector.
 }ATOM_DISPLAY_OBJECT_PATH;
 
+typedef struct  _ATOM_DISPLAY_EXTERNAL_OBJECT_PATH
+{
+  USHORT    usDeviceTag;                                   //supported device 
+  USHORT    usSize;                                        //the size of ATOM_DISPLAY_OBJECT_PATH
+  USHORT    usConnObjectId;                                //Connector Object ID 
+  USHORT    usGPUObjectId;                                 //GPU ID 
+  USHORT    usGraphicObjIds[2];                            //usGraphicObjIds[0]= GPU internal encoder, usGraphicObjIds[1]= external encoder 
+}ATOM_DISPLAY_EXTERNAL_OBJECT_PATH;
+
 typedef struct _ATOM_DISPLAY_OBJECT_PATH_TABLE
 {
   UCHAR                           ucNumOfDispPath;
@@ -3261,6 +3745,47 @@ typedef struct _ATOM_SRC_DST_TABLE_FOR_ONE_OBJECT         //usSrcDstTableOffset
 #define EXT_AUXDDC_LUTINDEX_7                   7
 #define MAX_NUMBER_OF_EXT_AUXDDC_LUT_ENTRIES   (EXT_AUXDDC_LUTINDEX_7+1)
 
+//ucChannelMapping are defined as following
+//for DP connector, eDP, DP to VGA/LVDS 
+//Bit[1:0]: Define which pin connect to DP connector DP_Lane0, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[3:2]: Define which pin connect to DP connector DP_Lane1, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[5:4]: Define which pin connect to DP connector DP_Lane2, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[7:6]: Define which pin connect to DP connector DP_Lane3, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+typedef struct _ATOM_DP_CONN_CHANNEL_MAPPING
+{
+#if ATOM_BIG_ENDIAN
+  UCHAR ucDP_Lane3_Source:2;
+  UCHAR ucDP_Lane2_Source:2;
+  UCHAR ucDP_Lane1_Source:2;
+  UCHAR ucDP_Lane0_Source:2;
+#else
+  UCHAR ucDP_Lane0_Source:2;
+  UCHAR ucDP_Lane1_Source:2;
+  UCHAR ucDP_Lane2_Source:2;
+  UCHAR ucDP_Lane3_Source:2;
+#endif
+}ATOM_DP_CONN_CHANNEL_MAPPING;
+
+//for DVI/HDMI, in dual link case, both links have to have same mapping. 
+//Bit[1:0]: Define which pin connect to DVI connector data Lane2, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[3:2]: Define which pin connect to DVI connector data Lane1, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[5:4]: Define which pin connect to DVI connector data Lane0, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+//Bit[7:6]: Define which pin connect to DVI connector clock lane, =0: source from GPU pin TX0, =1: from GPU pin TX1, =2: from GPU pin TX2, =3 from GPU pin TX3
+typedef struct _ATOM_DVI_CONN_CHANNEL_MAPPING
+{
+#if ATOM_BIG_ENDIAN
+  UCHAR ucDVI_CLK_Source:2;
+  UCHAR ucDVI_DATA0_Source:2;
+  UCHAR ucDVI_DATA1_Source:2;
+  UCHAR ucDVI_DATA2_Source:2;
+#else
+  UCHAR ucDVI_DATA2_Source:2;
+  UCHAR ucDVI_DATA1_Source:2;
+  UCHAR ucDVI_DATA0_Source:2;
+  UCHAR ucDVI_CLK_Source:2;
+#endif
+}ATOM_DVI_CONN_CHANNEL_MAPPING;
+
 typedef struct _EXT_DISPLAY_PATH
 {
   USHORT  usDeviceTag;                    //A bit vector to show what devices are supported 
@@ -3269,7 +3794,13 @@ typedef struct _EXT_DISPLAY_PATH
   UCHAR   ucExtAUXDDCLutIndex;            //An index into external AUX/DDC channel LUT
   UCHAR   ucExtHPDPINLutIndex;            //An index into external HPD pin LUT
   USHORT  usExtEncoderObjId;              //external encoder object id
-  USHORT  usReserved[3]; 
+  union{
+    UCHAR   ucChannelMapping;                  // if ucChannelMapping=0, using default one to one mapping
+    ATOM_DP_CONN_CHANNEL_MAPPING asDPMapping;
+    ATOM_DVI_CONN_CHANNEL_MAPPING asDVIMapping;
+  };
+  UCHAR   ucReserved;
+  USHORT  usReserved[2]; 
 }EXT_DISPLAY_PATH;
    
 #define NUMBER_OF_UCHAR_FOR_GUID          16
@@ -3281,7 +3812,8 @@ typedef  struct _ATOM_EXTERNAL_DISPLAY_CONNECTION_INFO
   UCHAR                    ucGuid [NUMBER_OF_UCHAR_FOR_GUID];     // a GUID is a 16 byte long string
   EXT_DISPLAY_PATH         sPath[MAX_NUMBER_OF_EXT_DISPLAY_PATH]; // total of fixed 7 entries.
   UCHAR                    ucChecksum;                            // a  simple Checksum of the sum of whole structure equal to 0x0. 
-  UCHAR                    Reserved [7];                          // for potential expansion
+  UCHAR                    uc3DStereoPinId;                       // use for eDP panel
+  UCHAR                    Reserved [6];                          // for potential expansion
 }ATOM_EXTERNAL_DISPLAY_CONNECTION_INFO;
 
 //Related definitions, all records are differnt but they have a commond header
@@ -3311,10 +3843,11 @@ typedef struct _ATOM_COMMON_RECORD_HEADER
 #define ATOM_CONNECTOR_AUXDDC_LUT_RECORD_TYPE          17 //This is for the case when connectors are not known to object table
 #define ATOM_OBJECT_LINK_RECORD_TYPE                   18 //Once this record is present under one object, it indicats the oobject is linked to another obj described by the record
 #define ATOM_CONNECTOR_REMOTE_CAP_RECORD_TYPE          19
+#define ATOM_ENCODER_CAP_RECORD_TYPE                   20
 
 
 //Must be updated when new record type is added,equal to that record definition!
-#define ATOM_MAX_OBJECT_RECORD_NUMBER             ATOM_CONNECTOR_REMOTE_CAP_RECORD_TYPE
+#define ATOM_MAX_OBJECT_RECORD_NUMBER             ATOM_ENCODER_CAP_RECORD_TYPE
 
 typedef struct  _ATOM_I2C_RECORD
 {
@@ -3441,6 +3974,26 @@ typedef struct  _ATOM_ENCODER_DVO_CF_RECORD
   UCHAR                       ucPadding[2];
 }ATOM_ENCODER_DVO_CF_RECORD;
 
+// Bit maps for ATOM_ENCODER_CAP_RECORD.ucEncoderCap
+#define ATOM_ENCODER_CAP_RECORD_HBR2     0x01         // DP1.2 HBR2 is supported by this path
+
+typedef struct  _ATOM_ENCODER_CAP_RECORD
+{
+  ATOM_COMMON_RECORD_HEADER   sheader;
+  union {
+    USHORT                    usEncoderCap;         
+    struct {
+#if ATOM_BIG_ENDIAN
+      USHORT                  usReserved:15;        // Bit1-15 may be defined for other capability in future
+      USHORT                  usHBR2Cap:1;          // Bit0 is for DP1.2 HBR2 capability. 
+#else
+      USHORT                  usHBR2Cap:1;          // Bit0 is for DP1.2 HBR2 capability. 
+      USHORT                  usReserved:15;        // Bit1-15 may be defined for other capability in future
+#endif
+    };
+  }; 
+}ATOM_ENCODER_CAP_RECORD;                             
+
 // value for ATOM_CONNECTOR_CF_RECORD.ucConnectedDvoBundle
 #define ATOM_CONNECTOR_CF_RECORD_CONNECTED_UPPER12BITBUNDLEA   1
 #define ATOM_CONNECTOR_CF_RECORD_CONNECTED_LOWER12BITBUNDLEB   2
@@ -3580,6 +4133,11 @@ typedef struct _ATOM_VOLTAGE_CONTROL
 #define	VOLTAGE_CONTROL_ID_DAC								0x02									//I2C control, used for R5xx/R6xx MVDDC,MVDDQ or VDDCI
 #define	VOLTAGE_CONTROL_ID_VT116xM						0x03									//I2C control, used for R6xx Core Voltage
 #define VOLTAGE_CONTROL_ID_DS4402							0x04									
+#define VOLTAGE_CONTROL_ID_UP6266 						0x05									
+#define VOLTAGE_CONTROL_ID_SCORPIO						0x06
+#define	VOLTAGE_CONTROL_ID_VT1556M						0x07									
+#define	VOLTAGE_CONTROL_ID_CHL822x						0x08									
+#define	VOLTAGE_CONTROL_ID_VT1586M						0x09
 
 typedef struct  _ATOM_VOLTAGE_OBJECT
 {
@@ -3670,66 +4228,157 @@ typedef struct _ATOM_POWER_SOURCE_INFO
 #define POWER_SENSOR_GPIO								0x01
 #define POWER_SENSOR_I2C								0x02
 
+typedef struct _ATOM_CLK_VOLT_CAPABILITY
+{
+  ULONG      ulVoltageIndex;                      // The Voltage Index indicated by FUSE, same voltage index shared with SCLK DPM fuse table        
+  ULONG      ulMaximumSupportedCLK;               // Maximum clock supported with specified voltage index, unit in 10kHz
+}ATOM_CLK_VOLT_CAPABILITY;
+
+typedef struct _ATOM_AVAILABLE_SCLK_LIST
+{
+  ULONG      ulSupportedSCLK;               // Maximum clock supported with specified voltage index,  unit in 10kHz
+  USHORT     usVoltageIndex;                // The Voltage Index indicated by FUSE for specified SCLK  
+  USHORT     usVoltageID;                   // The Voltage ID indicated by FUSE for specified SCLK 
+}ATOM_AVAILABLE_SCLK_LIST;
+
+// ATOM_INTEGRATED_SYSTEM_INFO_V6 ulSystemConfig cap definition
+#define ATOM_IGP_INFO_V6_SYSTEM_CONFIG__PCIE_POWER_GATING_ENABLE             1       // refer to ulSystemConfig bit[0]
+
+// this IntegrateSystemInfoTable is used for Liano/Ontario APU
 typedef struct _ATOM_INTEGRATED_SYSTEM_INFO_V6
 {
   ATOM_COMMON_TABLE_HEADER   sHeader;
   ULONG  ulBootUpEngineClock;
   ULONG  ulDentistVCOFreq;          
   ULONG  ulBootUpUMAClock;          
-  ULONG  ulReserved1[8];            
+  ATOM_CLK_VOLT_CAPABILITY   sDISPCLK_Voltage[4];            
   ULONG  ulBootUpReqDisplayVector;
   ULONG  ulOtherDisplayMisc;
   ULONG  ulGPUCapInfo;
-  ULONG  ulReserved2[3];            
+  ULONG  ulSB_MMIO_Base_Addr;
+  USHORT usRequestedPWMFreqInHz;
+  UCHAR  ucHtcTmpLmt;   
+  UCHAR  ucHtcHystLmt;
+  ULONG  ulMinEngineClock;           
   ULONG  ulSystemConfig;            
   ULONG  ulCPUCapInfo;              
-  USHORT usMaxNBVoltage;  
-  USHORT usMinNBVoltage;  
-  USHORT usBootUpNBVoltage;         
-  USHORT usExtDispConnInfoOffset;  
-  UCHAR  ucHtcTmpLmt;   
-  UCHAR  ucTjOffset;    
+  USHORT usNBP0Voltage;               
+  USHORT usNBP1Voltage;
+  USHORT usBootUpNBVoltage;                       
+  USHORT usExtDispConnInfoOffset;
+  USHORT usPanelRefreshRateRange;     
   UCHAR  ucMemoryType;  
   UCHAR  ucUMAChannelNumber;
   ULONG  ulCSR_M3_ARB_CNTL_DEFAULT[10];  
   ULONG  ulCSR_M3_ARB_CNTL_UVD[10]; 
   ULONG  ulCSR_M3_ARB_CNTL_FS3D[10];
-  ULONG  ulReserved3[42]; 
+  ATOM_AVAILABLE_SCLK_LIST   sAvail_SCLK[5];
+  ULONG  ulGMCRestoreResetTime;
+  ULONG  ulMinimumNClk;
+  ULONG  ulIdleNClk;
+  ULONG  ulDDR_DLL_PowerUpTime;
+  ULONG  ulDDR_PLL_PowerUpTime;
+  USHORT usPCIEClkSSPercentage;
+  USHORT usPCIEClkSSType;
+  USHORT usLvdsSSPercentage;
+  USHORT usLvdsSSpreadRateIn10Hz;
+  USHORT usHDMISSPercentage;
+  USHORT usHDMISSpreadRateIn10Hz;
+  USHORT usDVISSPercentage;
+  USHORT usDVISSpreadRateIn10Hz;
+  ULONG  ulReserved3[21]; 
   ATOM_EXTERNAL_DISPLAY_CONNECTION_INFO sExtDispConnInfo;   
 }ATOM_INTEGRATED_SYSTEM_INFO_V6;   
 
+// ulGPUCapInfo
+#define INTEGRATED_SYSTEM_INFO_V6_GPUCAPINFO__TMDSHDMI_COHERENT_SINGLEPLL_MODE       0x01
+#define INTEGRATED_SYSTEM_INFO_V6_GPUCAPINFO__DISABLE_AUX_HW_MODE_DETECTION          0x08
+
+// ulOtherDisplayMisc
+#define INTEGRATED_SYSTEM_INFO__GET_EDID_CALLBACK_FUNC_SUPPORT                       0x01
+
+
 /**********************************************************************************************************************
-// ATOM_INTEGRATED_SYSTEM_INFO_V6 Description
-//ulBootUpEngineClock:              VBIOS bootup Engine clock frequency, in 10kHz unit. 
-//ulDentistVCOFreq:                 Dentist VCO clock in 10kHz unit. 
-//ulBootUpUMAClock:                 System memory boot up clock frequency in 10Khz unit. 
-//ulReserved1[8]                    Reserved by now, must be 0x0. 
-//ulBootUpReqDisplayVector	        VBIOS boot up display IDs
-//                                  ATOM_DEVICE_CRT1_SUPPORT                  0x0001
-//                                  ATOM_DEVICE_CRT2_SUPPORT                  0x0010
-//                                  ATOM_DEVICE_DFP1_SUPPORT                  0x0008 
-//                                  ATOM_DEVICE_DFP6_SUPPORT                  0x0040 
-//                                  ATOM_DEVICE_DFP2_SUPPORT                  0x0080       
-//                                  ATOM_DEVICE_DFP3_SUPPORT                  0x0200       
-//                                  ATOM_DEVICE_DFP4_SUPPORT                  0x0400        
-//                                  ATOM_DEVICE_DFP5_SUPPORT                  0x0800
-//                                  ATOM_DEVICE_LCD1_SUPPORT                  0x0002
-//ulOtherDisplayMisc      	        Other display related flags, not defined yet. 
-//ulGPUCapInfo                      TBD
-//ulReserved2[3]                    must be 0x0 for the reserved.
-//ulSystemConfig                    TBD
-//ulCPUCapInfo                      TBD
-//usMaxNBVoltage                    High NB voltage in unit of mv, calculated using current VDDNB (D24F2xDC) and VDDNB offset fuse. 
-//usMinNBVoltage                    Low NB voltage in unit of mv, calculated using current VDDNB (D24F2xDC) and VDDNB offset fuse.
-//usBootUpNBVoltage                 Boot up NB voltage in unit of mv.
-//ucHtcTmpLmt                       Bit [22:16] of D24F3x64 Thermal Control (HTC) Register.
-//ucTjOffset                        Bit [28:22] of D24F3xE4 Thermtrip Status Register,may not be needed.
-//ucMemoryType                      [3:0]=1:DDR1;=2:DDR2;=3:DDR3.[7:4] is reserved.
-//ucUMAChannelNumber      	        System memory channel numbers. 
-//usExtDispConnectionInfoOffset     ATOM_EXTERNAL_DISPLAY_CONNECTION_INFO offset relative to beginning of this table. 
-//ulCSR_M3_ARB_CNTL_DEFAULT[10]     Arrays with values for CSR M3 arbiter for default
-//ulCSR_M3_ARB_CNTL_UVD[10]         Arrays with values for CSR M3 arbiter for UVD playback.
-//ulCSR_M3_ARB_CNTL_FS3D[10]        Arrays with values for CSR M3 arbiter for Full Screen 3D applications.
+  ATOM_INTEGRATED_SYSTEM_INFO_V6 Description
+ulBootUpEngineClock:              VBIOS bootup Engine clock frequency, in 10kHz unit. if it is equal 0, then VBIOS use pre-defined bootup engine clock
+ulDentistVCOFreq:                 Dentist VCO clock in 10kHz unit. 
+ulBootUpUMAClock:                 System memory boot up clock frequency in 10Khz unit. 
+sDISPCLK_Voltage:                 Report Display clock voltage requirement.
+ 
+ulBootUpReqDisplayVector:         VBIOS boot up display IDs, following are supported devices in Liano/Ontaio projects:
+                                  ATOM_DEVICE_CRT1_SUPPORT                  0x0001
+                                  ATOM_DEVICE_CRT2_SUPPORT                  0x0010
+                                  ATOM_DEVICE_DFP1_SUPPORT                  0x0008 
+                                  ATOM_DEVICE_DFP6_SUPPORT                  0x0040 
+                                  ATOM_DEVICE_DFP2_SUPPORT                  0x0080       
+                                  ATOM_DEVICE_DFP3_SUPPORT                  0x0200       
+                                  ATOM_DEVICE_DFP4_SUPPORT                  0x0400        
+                                  ATOM_DEVICE_DFP5_SUPPORT                  0x0800
+                                  ATOM_DEVICE_LCD1_SUPPORT                  0x0002
+ulOtherDisplayMisc:      	        Other display related flags, not defined yet. 
+ulGPUCapInfo:                     bit[0]=0: TMDS/HDMI Coherent Mode use cascade PLL mode.
+                                        =1: TMDS/HDMI Coherent Mode use signel PLL mode.
+                                  bit[3]=0: Enable HW AUX mode detection logic
+                                        =1: Disable HW AUX mode dettion logic
+ulSB_MMIO_Base_Addr:              Physical Base address to SB MMIO space. Driver needs to initialize it for SMU usage.
+
+usRequestedPWMFreqInHz:           When it's set to 0x0 by SBIOS: the LCD BackLight is not controlled by GPU(SW). 
+                                  Any attempt to change BL using VBIOS function or enable VariBri from PP table is not effective since ATOM_BIOS_INFO_BL_CONTROLLED_BY_GPU==0;
+                                  
+                                  When it's set to a non-zero frequency, the BackLight is controlled by GPU (SW) in one of two ways below:
+                                  1. SW uses the GPU BL PWM output to control the BL, in chis case, this non-zero frequency determines what freq GPU should use;
+                                  VBIOS will set up proper PWM frequency and ATOM_BIOS_INFO_BL_CONTROLLED_BY_GPU==1,as the result,
+                                  Changing BL using VBIOS function is functional in both driver and non-driver present environment; 
+                                  and enabling VariBri under the driver environment from PP table is optional.
+
+                                  2. SW uses other means to control BL (like DPCD),this non-zero frequency serves as a flag only indicating
+                                  that BL control from GPU is expected.
+                                  VBIOS will NOT set up PWM frequency but make ATOM_BIOS_INFO_BL_CONTROLLED_BY_GPU==1
+                                  Changing BL using VBIOS function could be functional in both driver and non-driver present environment,but
+                                  it's per platform 
+                                  and enabling VariBri under the driver environment from PP table is optional.
+
+ucHtcTmpLmt:                      Refer to D18F3x64 bit[22:16], HtcTmpLmt. 
+                                  Threshold on value to enter HTC_active state.
+ucHtcHystLmt:                     Refer to D18F3x64 bit[27:24], HtcHystLmt. 
+                                  To calculate threshold off value to exit HTC_active state, which is Threshold on vlaue minus ucHtcHystLmt.
+ulMinEngineClock:                 Minimum SCLK allowed in 10kHz unit. This is calculated based on WRCK Fuse settings.
+ulSystemConfig:                   Bit[0]=0: PCIE Power Gating Disabled 
+                                        =1: PCIE Power Gating Enabled
+                                  Bit[1]=0: DDR-DLL shut-down feature disabled.
+                                         1: DDR-DLL shut-down feature enabled.
+                                  Bit[2]=0: DDR-PLL Power down feature disabled.
+                                         1: DDR-PLL Power down feature enabled.                                 
+ulCPUCapInfo:                     TBD
+usNBP0Voltage:                    VID for voltage on NB P0 State
+usNBP1Voltage:                    VID for voltage on NB P1 State  
+usBootUpNBVoltage:                Voltage Index of GNB voltage configured by SBIOS, which is suffcient to support VBIOS DISPCLK requirement.
+usExtDispConnInfoOffset:          Offset to sExtDispConnInfo inside the structure
+usPanelRefreshRateRange:          Bit vector for LCD supported refresh rate range. If DRR is requestd by the platform, at least two bits need to be set
+                                  to indicate a range.
+                                  SUPPORTED_LCD_REFRESHRATE_30Hz          0x0004
+                                  SUPPORTED_LCD_REFRESHRATE_40Hz          0x0008
+                                  SUPPORTED_LCD_REFRESHRATE_50Hz          0x0010
+                                  SUPPORTED_LCD_REFRESHRATE_60Hz          0x0020
+ucMemoryType:                     [3:0]=1:DDR1;=2:DDR2;=3:DDR3.[7:4] is reserved.
+ucUMAChannelNumber:      	        System memory channel numbers. 
+ulCSR_M3_ARB_CNTL_DEFAULT[10]:    Arrays with values for CSR M3 arbiter for default
+ulCSR_M3_ARB_CNTL_UVD[10]:        Arrays with values for CSR M3 arbiter for UVD playback.
+ulCSR_M3_ARB_CNTL_FS3D[10]:       Arrays with values for CSR M3 arbiter for Full Screen 3D applications.
+sAvail_SCLK[5]:                   Arrays to provide availabe list of SLCK and corresponding voltage, order from low to high  
+ulGMCRestoreResetTime:            GMC power restore and GMC reset time to calculate data reconnection latency. Unit in ns. 
+ulMinimumNClk:                    Minimum NCLK speed among all NB-Pstates to calcualte data reconnection latency. Unit in 10kHz. 
+ulIdleNClk:                       NCLK speed while memory runs in self-refresh state. Unit in 10kHz.
+ulDDR_DLL_PowerUpTime:            DDR PHY DLL power up time. Unit in ns.
+ulDDR_PLL_PowerUpTime:            DDR PHY PLL power up time. Unit in ns.
+usPCIEClkSSPercentage:            PCIE Clock Spred Spectrum Percentage in unit 0.01%; 100 mean 1%.
+usPCIEClkSSType:                  PCIE Clock Spred Spectrum Type. 0 for Down spread(default); 1 for Center spread.
+usLvdsSSPercentage:               LVDS panel ( not include eDP ) Spread Spectrum Percentage in unit of 0.01%, =0, use VBIOS default setting. 
+usLvdsSSpreadRateIn10Hz:          LVDS panel ( not include eDP ) Spread Spectrum frequency in unit of 10Hz, =0, use VBIOS default setting. 
+usHDMISSPercentage:               HDMI Spread Spectrum Percentage in unit 0.01%; 100 mean 1%,  =0, use VBIOS default setting. 
+usHDMISSpreadRateIn10Hz:          HDMI Spread Spectrum frequency in unit of 10Hz,  =0, use VBIOS default setting. 
+usDVISSPercentage:                DVI Spread Spectrum Percentage in unit 0.01%; 100 mean 1%,  =0, use VBIOS default setting. 
+usDVISSpreadRateIn10Hz:           DVI Spread Spectrum frequency in unit of 10Hz,  =0, use VBIOS default setting. 
 **********************************************************************************************************************/
 
 /**************************************************************************/
@@ -3790,6 +4439,7 @@ typedef struct _ATOM_ASIC_SS_ASSIGNMENT
 #define ASIC_INTERNAL_SS_ON_LVDS    6
 #define ASIC_INTERNAL_SS_ON_DP      7
 #define ASIC_INTERNAL_SS_ON_DCPLL   8
+#define ASIC_EXTERNAL_SS_ON_DP_CLOCK 9
 
 typedef struct _ATOM_ASIC_SS_ASSIGNMENT_V2
 {
@@ -3903,6 +4553,7 @@ typedef struct _ATOM_ASIC_INTERNAL_SS_INFO_V3
 #define ATOM_S0_SYSTEM_POWER_STATE_VALUE_AC     1
 #define ATOM_S0_SYSTEM_POWER_STATE_VALUE_DC     2
 #define ATOM_S0_SYSTEM_POWER_STATE_VALUE_LITEAC 3
+#define ATOM_S0_SYSTEM_POWER_STATE_VALUE_LIT2AC 4
 
 //Byte aligned defintion for BIOS usage
 #define ATOM_S0_CRT1_MONOb0             0x01
@@ -4529,7 +5180,8 @@ typedef struct _ATOM_INIT_REG_BLOCK{
 #define INDEX_ACCESS_RANGE_BEGIN	    (VALUE_DWORD + 1)
 #define INDEX_ACCESS_RANGE_END		    (INDEX_ACCESS_RANGE_BEGIN + 1)
 #define VALUE_INDEX_ACCESS_SINGLE	    (INDEX_ACCESS_RANGE_END + 1)
-
+//#define ACCESS_MCIODEBUGIND            0x40       //defined in BIOS code
+#define ACCESS_PLACEHOLDER             0x80
 
 typedef struct _ATOM_MC_INIT_PARAM_TABLE
 { 
@@ -4554,6 +5206,10 @@ typedef struct _ATOM_MC_INIT_PARAM_TABLE
 #define _32Mx32             0x33
 #define _64Mx8              0x41
 #define _64Mx16             0x42
+#define _64Mx32             0x43
+#define _128Mx8             0x51
+#define _128Mx16            0x52
+#define _256Mx8             0x61
 
 #define SAMSUNG             0x1
 #define INFINEON            0x2
@@ -4569,10 +5225,11 @@ typedef struct _ATOM_MC_INIT_PARAM_TABLE
 #define QIMONDA             INFINEON
 #define PROMOS              MOSEL
 #define KRETON              INFINEON
+#define ELIXIR              NANYA
 
 /////////////Support for GDDR5 MC uCode to reside in upper 64K of ROM/////////////
 
-#define UCODE_ROM_START_ADDRESS		0x1c000
+#define UCODE_ROM_START_ADDRESS		0x1b800
 #define	UCODE_SIGNATURE			0x4375434d // 'MCuC' - MC uCode
 
 //uCode block header for reference
@@ -4903,7 +5560,34 @@ typedef struct _ATOM_VRAM_MODULE_V6
   ATOM_MEMORY_TIMING_FORMAT_V2  asMemTiming[5];//Memory Timing block sort from lower clock to higher clock
 }ATOM_VRAM_MODULE_V6;
 
-
+typedef struct _ATOM_VRAM_MODULE_V7
+{
+// Design Specific Values
+  ULONG	  ulChannelMapCfg;	                // mmMC_SHARED_CHREMAP
+  USHORT  usModuleSize;                     // Size of ATOM_VRAM_MODULE_V7
+  USHORT  usPrivateReserved;                // MC_ARB_RAMCFG (includes NOOFBANK,NOOFRANKS,NOOFROWS,NOOFCOLS)
+  USHORT  usReserved;
+  UCHAR   ucExtMemoryID;                    // Current memory module ID
+  UCHAR   ucMemoryType;                     // MEM_TYPE_DDR2/DDR3/GDDR3/GDDR5
+  UCHAR   ucChannelNum;                     // Number of mem. channels supported in this module
+  UCHAR   ucChannelWidth;                   // CHANNEL_16BIT/CHANNEL_32BIT/CHANNEL_64BIT
+  UCHAR   ucDensity;                        // _8Mx32, _16Mx32, _16Mx16, _32Mx16
+  UCHAR	  ucReserve;                        // Former container for Mx_FLAGS like DBI_AC_MODE_ENABLE_ASIC for GDDR4. Not used now.
+  UCHAR	  ucMisc;                           // RANK_OF_THISMEMORY etc.
+  UCHAR	  ucVREFI;                          // Not used.
+  UCHAR   ucNPL_RT;                         // Round trip delay (MC_SEQ_CAS_TIMING [28:24]:TCL=CL+NPL_RT-2). Always 2.
+  UCHAR	  ucPreamble;                       // [7:4] Write Preamble, [3:0] Read Preamble
+  UCHAR   ucMemorySize;                     // Total memory size in unit of 16MB for CONFIG_MEMSIZE - bit[23:0] zeros
+  UCHAR   ucReserved[3];
+// Memory Module specific values
+  USHORT  usEMRS2Value;                     // EMRS2/MR2 Value. 
+  USHORT  usEMRS3Value;                     // EMRS3/MR3 Value.
+  UCHAR   ucMemoryVenderID;                 // [7:4] Revision, [3:0] Vendor code
+  UCHAR	  ucRefreshRateFactor;              // [1:0]=RefreshFactor (00=8ms, 01=16ms, 10=32ms,11=64ms)
+  UCHAR	  ucFIFODepth;                      // FIFO depth can be detected during vendor detection, here is hardcoded per memory
+  UCHAR   ucCDR_Bandwidth;                  // [0:3]=Read CDR bandwidth, [4:7] - Write CDR Bandwidth
+  char    strMemPNString[20];               // part number end with '0'. 
+}ATOM_VRAM_MODULE_V7;
 
 typedef struct _ATOM_VRAM_INFO_V2
 {
@@ -4942,6 +5626,20 @@ typedef struct _ATOM_VRAM_INFO_V4
 																																						 //	ATOM_INIT_REG_BLOCK				 aMemAdjust;
 }ATOM_VRAM_INFO_V4;
 
+typedef struct _ATOM_VRAM_INFO_HEADER_V2_1
+{
+  ATOM_COMMON_TABLE_HEADER   sHeader;
+	USHORT										 usMemAdjustTblOffset;													 // offset of ATOM_INIT_REG_BLOCK structure for memory vendor specific MC adjust setting
+	USHORT										 usMemClkPatchTblOffset;												 //	offset of ATOM_INIT_REG_BLOCK structure for memory clock specific MC setting
+	USHORT										 usReserved[4];
+  UCHAR                      ucNumOfVRAMModule;                              // indicate number of VRAM module
+  UCHAR                      ucMemoryClkPatchTblVer;                         // version of memory AC timing register list
+  UCHAR                      ucVramModuleVer;                                // indicate ATOM_VRAM_MODUE version
+  UCHAR                      ucReserved; 
+  ATOM_VRAM_MODULE_V7		     aVramInfo[ATOM_MAX_NUMBER_OF_VRAM_MODULE];      // just for allocation, real number of blocks is in ucNumOfVRAMModule;
+}ATOM_VRAM_INFO_HEADER_V2_1;
+
+
 typedef struct _ATOM_VRAM_GPIO_DETECTION_INFO
 {
   ATOM_COMMON_TABLE_HEADER   sHeader;
@@ -5182,6 +5880,16 @@ typedef struct _ASIC_TRANSMITTER_INFO
 	UCHAR  ucReserved;
 }ASIC_TRANSMITTER_INFO;
 
+#define ASIC_TRANSMITTER_INFO_CONFIG__DVO_SDR_MODE          0x01
+#define ASIC_TRANSMITTER_INFO_CONFIG__COHERENT_MODE         0x02
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODEROBJ_ID_MASK    0xc4
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_A             0x00
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_B             0x04
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_C             0x40
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_D             0x44
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_E             0x80
+#define ASIC_TRANSMITTER_INFO_CONFIG__ENCODER_F             0x84
+
 typedef struct _ASIC_ENCODER_INFO
 {
 	UCHAR ucEncoderID;
@@ -5284,6 +5992,28 @@ typedef struct _DP_ENCODER_SERVICE_PARAMETERS
 /* /obselete */
 #define DP_ENCODER_SERVICE_PS_ALLOCATION				WRITE_ONE_BYTE_HW_I2C_DATA_PARAMETERS
 
+
+typedef struct _DP_ENCODER_SERVICE_PARAMETERS_V2
+{
+	USHORT usExtEncoderObjId;   // External Encoder Object Id, output parameter only, use when ucAction = DP_SERVICE_V2_ACTION_DET_EXT_CONNECTION
+  UCHAR  ucAuxId;
+  UCHAR  ucAction;
+  UCHAR  ucSinkType;          // Iput and Output parameters. 
+  UCHAR  ucHPDId;             // Input parameter, used when ucAction = DP_SERVICE_V2_ACTION_DET_EXT_CONNECTION
+	UCHAR  ucReserved[2];
+}DP_ENCODER_SERVICE_PARAMETERS_V2;
+
+typedef struct _DP_ENCODER_SERVICE_PS_ALLOCATION_V2
+{
+  DP_ENCODER_SERVICE_PARAMETERS_V2 asDPServiceParam;
+  PROCESS_AUX_CHANNEL_TRANSACTION_PARAMETERS_V2 asAuxParam;
+}DP_ENCODER_SERVICE_PS_ALLOCATION_V2;
+
+// ucAction
+#define DP_SERVICE_V2_ACTION_GET_SINK_TYPE							0x01
+#define DP_SERVICE_V2_ACTION_DET_LCD_CONNECTION			    0x02
+
+
 // DP_TRAINING_TABLE
 #define DPCD_SET_LINKRATE_LANENUM_PATTERN1_TBL_ADDR				ATOM_DP_TRAINING_TBL_ADDR		
 #define DPCD_SET_SS_CNTL_TBL_ADDR													(ATOM_DP_TRAINING_TBL_ADDR + 8 )
@@ -5339,6 +6069,7 @@ typedef struct _SET_HWBLOCK_INSTANCE_PARAMETER_V2
 #define SELECT_DCIO_IMPCAL            4
 #define SELECT_DCIO_DIG               6
 #define SELECT_CRTC_PIXEL_RATE        7
+#define SELECT_VGA_BLK                8
 
 /****************************************************************************/	
 //Portion VI: Definitinos for vbios MC scratch registers that driver used
@@ -5744,7 +6475,17 @@ typedef struct _ATOM_PPLIB_THERMALCONTROLLER
 #define ATOM_PP_THERMALCONTROLLER_ADT7473   9
 #define ATOM_PP_THERMALCONTROLLER_EXTERNAL_GPIO     11
 #define ATOM_PP_THERMALCONTROLLER_EVERGREEN 12
+#define ATOM_PP_THERMALCONTROLLER_EMC2103   13  /* 0x0D */ // Only fan control will be implemented, do NOT show this in PPGen.
+#define ATOM_PP_THERMALCONTROLLER_SUMO      14  /* 0x0E */ // Sumo type, used internally
+#define ATOM_PP_THERMALCONTROLLER_NISLANDS  15
+
+// Thermal controller 'combo type' to use an external controller for Fan control and an internal controller for thermal.
+// We probably should reserve the bit 0x80 for this use.
+// To keep the number of these types low we should also use the same code for all ASICs (i.e. do not distinguish RV6xx and RV7xx Internal here).
+// The driver can pick the correct internal controller based on the ASIC.
+
 #define ATOM_PP_THERMALCONTROLLER_ADT7473_WITH_INTERNAL   0x89    // ADT7473 Fan Control + Internal Thermal Controller
+#define ATOM_PP_THERMALCONTROLLER_EMC2103_WITH_INTERNAL   0x8D    // EMC2103 Fan Control + Internal Thermal Controller
 
 typedef struct _ATOM_PPLIB_STATE
 {
@@ -5841,6 +6582,29 @@ typedef struct _ATOM_PPLIB_POWERPLAYTABLE3
     USHORT                     usExtendendedHeaderOffset;
 } ATOM_PPLIB_POWERPLAYTABLE3, *LPATOM_PPLIB_POWERPLAYTABLE3;
 
+typedef struct _ATOM_PPLIB_POWERPLAYTABLE4
+{
+    ATOM_PPLIB_POWERPLAYTABLE3 basicTable3;
+    ULONG                      ulGoldenPPID;                    // PPGen use only     
+    ULONG                      ulGoldenRevision;                // PPGen use only
+    USHORT                     usVddcDependencyOnSCLKOffset;
+    USHORT                     usVddciDependencyOnMCLKOffset;
+    USHORT                     usVddcDependencyOnMCLKOffset;
+    USHORT                     usMaxClockVoltageOnDCOffset;
+    USHORT                     usReserved[2];  
+} ATOM_PPLIB_POWERPLAYTABLE4, *LPATOM_PPLIB_POWERPLAYTABLE4;
+
+typedef struct _ATOM_PPLIB_POWERPLAYTABLE5
+{
+    ATOM_PPLIB_POWERPLAYTABLE4 basicTable4;
+    ULONG                      ulTDPLimit;
+    ULONG                      ulNearTDPLimit;
+    ULONG                      ulSQRampingThreshold;
+    USHORT                     usCACLeakageTableOffset;         // Points to ATOM_PPLIB_CAC_Leakage_Table
+    ULONG                      ulCACLeakage;                    // TBD, this parameter is still under discussion.  Change to ulReserved if not needed.
+    ULONG                      ulReserved;
+} ATOM_PPLIB_POWERPLAYTABLE5, *LPATOM_PPLIB_POWERPLAYTABLE5;
+
 //// ATOM_PPLIB_NONCLOCK_INFO::usClassification
 #define ATOM_PPLIB_CLASSIFICATION_UI_MASK          0x0007
 #define ATOM_PPLIB_CLASSIFICATION_UI_SHIFT         0
@@ -5864,6 +6628,10 @@ typedef struct _ATOM_PPLIB_POWERPLAYTABLE3
 #define ATOM_PPLIB_CLASSIFICATION_HDSTATE                0x4000
 #define ATOM_PPLIB_CLASSIFICATION_SDSTATE                0x8000
 
+//// ATOM_PPLIB_NONCLOCK_INFO::usClassification2
+#define ATOM_PPLIB_CLASSIFICATION2_LIMITEDPOWERSOURCE_2     0x0001
+#define ATOM_PPLIB_CLASSIFICATION2_ULV                      0x0002
+
 //// ATOM_PPLIB_NONCLOCK_INFO::ulCapsAndSettings
 #define ATOM_PPLIB_SINGLE_DISPLAY_ONLY           0x00000001
 #define ATOM_PPLIB_SUPPORTS_VIDEO_PLAYBACK         0x00000002
@@ -5896,9 +6664,21 @@ typedef struct _ATOM_PPLIB_POWERPLAYTABLE3
 #define ATOM_PPLIB_M3ARB_MASK                       0x00060000
 #define ATOM_PPLIB_M3ARB_SHIFT                      17
 
+#define ATOM_PPLIB_ENABLE_DRR                       0x00080000
+
+// remaining 16 bits are reserved
+typedef struct _ATOM_PPLIB_THERMAL_STATE
+{
+    UCHAR   ucMinTemperature;
+    UCHAR   ucMaxTemperature;
+    UCHAR   ucThermalAction;
+}ATOM_PPLIB_THERMAL_STATE, *LPATOM_PPLIB_THERMAL_STATE;
+
 // Contained in an array starting at the offset
 // in ATOM_PPLIB_POWERPLAYTABLE::usNonClockInfoArrayOffset.
 // referenced from ATOM_PPLIB_STATE_INFO::ucNonClockStateIndex
+#define ATOM_PPLIB_NONCLOCKINFO_VER1      12
+#define ATOM_PPLIB_NONCLOCKINFO_VER2      24
 typedef struct _ATOM_PPLIB_NONCLOCK_INFO
 {
       USHORT usClassification;
@@ -5906,15 +6686,15 @@ typedef struct _ATOM_PPLIB_NONCLOCK_INFO
       UCHAR  ucMaxTemperature;
       ULONG  ulCapsAndSettings;
       UCHAR  ucRequiredPower;
-      UCHAR  ucUnused1[3];
+      USHORT usClassification2;
+      ULONG  ulVCLK;
+      ULONG  ulDCLK;
+      UCHAR  ucUnused[5];
 } ATOM_PPLIB_NONCLOCK_INFO;
 
 // Contained in an array starting at the offset
 // in ATOM_PPLIB_POWERPLAYTABLE::usClockInfoArrayOffset.
 // referenced from ATOM_PPLIB_STATE::ucClockStateIndices
-#define ATOM_PPLIB_NONCLOCKINFO_VER1      12
-#define ATOM_PPLIB_NONCLOCKINFO_VER2      24
-
 typedef struct _ATOM_PPLIB_R600_CLOCK_INFO
 {
       USHORT usEngineClockLow;
@@ -5985,6 +6765,93 @@ typedef struct _ATOM_PPLIB_RS780_CLOCK_INFO
 #define ATOM_PPLIB_RS780_HTLINKFREQ_LOW        1 
 #define ATOM_PPLIB_RS780_HTLINKFREQ_HIGH       2 
 
+typedef struct _ATOM_PPLIB_SUMO_CLOCK_INFO{
+      USHORT usEngineClockLow;  //clockfrequency & 0xFFFF. The unit is in 10khz
+      UCHAR  ucEngineClockHigh; //clockfrequency >> 16. 
+      UCHAR  vddcIndex;         //2-bit vddc index;
+      UCHAR  leakage;          //please use 8-bit absolute value, not the 6-bit % value 
+      //please initalize to 0
+      UCHAR  rsv;
+      //please initalize to 0
+      USHORT rsv1;
+      //please initialize to 0s
+      ULONG rsv2[2];
+}ATOM_PPLIB_SUMO_CLOCK_INFO;
+
+
+
+typedef struct _ATOM_PPLIB_STATE_V2
+{
+      //number of valid dpm levels in this state; Driver uses it to calculate the whole 
+      //size of the state: sizeof(ATOM_PPLIB_STATE_V2) + (ucNumDPMLevels - 1) * sizeof(UCHAR)
+      UCHAR ucNumDPMLevels;
+      
+      //a index to the array of nonClockInfos
+      UCHAR nonClockInfoIndex;
+      /**
+      * Driver will read the first ucNumDPMLevels in this array
+      */
+      UCHAR clockInfoIndex[1];
+} ATOM_PPLIB_STATE_V2;
+
+typedef struct StateArray{
+    //how many states we have 
+    UCHAR ucNumEntries;
+    
+    ATOM_PPLIB_STATE_V2 states[1];
+}StateArray;
+
+
+typedef struct ClockInfoArray{
+    //how many clock levels we have
+    UCHAR ucNumEntries;
+    
+    //sizeof(ATOM_PPLIB_SUMO_CLOCK_INFO)
+    UCHAR ucEntrySize;
+    
+    //this is for Sumo
+    ATOM_PPLIB_SUMO_CLOCK_INFO clockInfo[1];
+}ClockInfoArray;
+
+typedef struct NonClockInfoArray{
+
+    //how many non-clock levels we have. normally should be same as number of states
+    UCHAR ucNumEntries;
+    //sizeof(ATOM_PPLIB_NONCLOCK_INFO)
+    UCHAR ucEntrySize;
+    
+    ATOM_PPLIB_NONCLOCK_INFO nonClockInfo[1];
+}NonClockInfoArray;
+
+typedef struct _ATOM_PPLIB_Clock_Voltage_Dependency_Record
+{
+    USHORT usClockLow;
+    UCHAR  ucClockHigh;
+    USHORT usVoltage;
+}ATOM_PPLIB_Clock_Voltage_Dependency_Record;
+
+typedef struct _ATOM_PPLIB_Clock_Voltage_Dependency_Table
+{
+    UCHAR ucNumEntries;                                                // Number of entries.
+    ATOM_PPLIB_Clock_Voltage_Dependency_Record entries[1];             // Dynamically allocate entries.
+}ATOM_PPLIB_Clock_Voltage_Dependency_Table;
+
+typedef struct _ATOM_PPLIB_Clock_Voltage_Limit_Record
+{
+    USHORT usSclkLow;
+    UCHAR  ucSclkHigh;
+    USHORT usMclkLow;
+    UCHAR  ucMclkHigh;
+    USHORT usVddc;
+    USHORT usVddci;
+}ATOM_PPLIB_Clock_Voltage_Limit_Record;
+
+typedef struct _ATOM_PPLIB_Clock_Voltage_Limit_Table
+{
+    UCHAR ucNumEntries;                                                // Number of entries.
+    ATOM_PPLIB_Clock_Voltage_Limit_Record entries[1];                  // Dynamically allocate entries.
+}ATOM_PPLIB_Clock_Voltage_Limit_Table;
+
 /**************************************************************************/
 
 
diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index 4dc5b4714c5a..f7d7477daffb 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -40,6 +40,61 @@
 static void evergreen_gpu_init(struct radeon_device *rdev);
 void evergreen_fini(struct radeon_device *rdev);
 
+void evergreen_pre_page_flip(struct radeon_device *rdev, int crtc)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc];
+	u32 tmp;
+
+	/* make sure flip is at vb rather than hb */
+	tmp = RREG32(EVERGREEN_GRPH_FLIP_CONTROL + radeon_crtc->crtc_offset);
+	tmp &= ~EVERGREEN_GRPH_SURFACE_UPDATE_H_RETRACE_EN;
+	WREG32(EVERGREEN_GRPH_FLIP_CONTROL + radeon_crtc->crtc_offset, tmp);
+
+	/* set pageflip to happen anywhere in vblank interval */
+	WREG32(EVERGREEN_MASTER_UPDATE_MODE + radeon_crtc->crtc_offset, 0);
+
+	/* enable the pflip int */
+	radeon_irq_kms_pflip_irq_get(rdev, crtc);
+}
+
+void evergreen_post_page_flip(struct radeon_device *rdev, int crtc)
+{
+	/* disable the pflip int */
+	radeon_irq_kms_pflip_irq_put(rdev, crtc);
+}
+
+u32 evergreen_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
+	u32 tmp = RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset);
+
+	/* Lock the graphics update lock */
+	tmp |= EVERGREEN_GRPH_UPDATE_LOCK;
+	WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* update the scanout addresses */
+	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
+	       upper_32_bits(crtc_base));
+	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+
+	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + radeon_crtc->crtc_offset,
+	       upper_32_bits(crtc_base));
+	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+
+	/* Wait for update_pending to go high. */
+	while (!(RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING));
+	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
+
+	/* Unlock the lock, so double-buffering can take place inside vblank */
+	tmp &= ~EVERGREEN_GRPH_UPDATE_LOCK;
+	WREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* Return current update_pending status: */
+	return RREG32(EVERGREEN_GRPH_UPDATE + radeon_crtc->crtc_offset) & EVERGREEN_GRPH_SURFACE_UPDATE_PENDING;
+}
+
 /* get temperature in millidegrees */
 u32 evergreen_get_temp(struct radeon_device *rdev)
 {
@@ -57,6 +112,14 @@ u32 evergreen_get_temp(struct radeon_device *rdev)
 	return actual_temp * 1000;
 }
 
+u32 sumo_get_temp(struct radeon_device *rdev)
+{
+	u32 temp = RREG32(CG_THERMAL_STATUS) & 0xff;
+	u32 actual_temp = (temp >> 1) & 0xff;
+
+	return actual_temp * 1000;
+}
+
 void evergreen_pm_misc(struct radeon_device *rdev)
 {
 	int req_ps_idx = rdev->pm.requested_power_state_index;
@@ -888,31 +951,39 @@ static void evergreen_mc_stop(struct radeon_device *rdev, struct evergreen_mc_sa
 	save->vga_hdp_control = RREG32(VGA_HDP_CONTROL);
 	save->crtc_control[0] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET);
 	save->crtc_control[1] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET);
-	save->crtc_control[2] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET);
-	save->crtc_control[3] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET);
-	save->crtc_control[4] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET);
-	save->crtc_control[5] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		save->crtc_control[2] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET);
+		save->crtc_control[3] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET);
+		save->crtc_control[4] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET);
+		save->crtc_control[5] = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
+	}
 
 	/* Stop all video */
 	WREG32(VGA_RENDER_CONTROL, 0);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 1);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
+	}
 	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	}
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	}
 
 	WREG32(D1VGA_CONTROL, 0);
 	WREG32(D2VGA_CONTROL, 0);
@@ -942,41 +1013,43 @@ static void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_
 	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC1_REGISTER_OFFSET,
 	       (u32)rdev->mc.vram_start);
 
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
-	       upper_32_bits(rdev->mc.vram_start));
-	WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
-	WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
-	       (u32)rdev->mc.vram_start);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC2_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC2_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC3_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC3_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC4_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC4_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS_HIGH + EVERGREEN_CRTC5_REGISTER_OFFSET,
+		       upper_32_bits(rdev->mc.vram_start));
+		WREG32(EVERGREEN_GRPH_PRIMARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+		WREG32(EVERGREEN_GRPH_SECONDARY_SURFACE_ADDRESS + EVERGREEN_CRTC5_REGISTER_OFFSET,
+		       (u32)rdev->mc.vram_start);
+	}
 
 	WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS_HIGH, upper_32_bits(rdev->mc.vram_start));
 	WREG32(EVERGREEN_VGA_MEMORY_BASE_ADDRESS, (u32)rdev->mc.vram_start);
@@ -992,22 +1065,28 @@ static void evergreen_mc_resume(struct radeon_device *rdev, struct evergreen_mc_
 	WREG32(EVERGREEN_D6VGA_CONTROL, save->vga_control[5]);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 1);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 1);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 1);
+	}
 	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, save->crtc_control[0]);
 	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, save->crtc_control[1]);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, save->crtc_control[2]);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, save->crtc_control[3]);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, save->crtc_control[4]);
-	WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, save->crtc_control[5]);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, save->crtc_control[2]);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, save->crtc_control[3]);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, save->crtc_control[4]);
+		WREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, save->crtc_control[5]);
+	}
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
-	WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
+		WREG32(EVERGREEN_CRTC_UPDATE_LOCK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	}
 	WREG32(VGA_RENDER_CONTROL, save->vga_render_control);
 }
 
@@ -1055,6 +1134,12 @@ static void evergreen_mc_program(struct radeon_device *rdev)
 			rdev->mc.vram_end >> 12);
 	}
 	WREG32(MC_VM_SYSTEM_APERTURE_DEFAULT_ADDR, 0);
+	if (rdev->flags & RADEON_IS_IGP) {
+		tmp = RREG32(MC_FUS_VM_FB_OFFSET) & 0x000FFFFF;
+		tmp |= ((rdev->mc.vram_end >> 20) & 0xF) << 24;
+		tmp |= ((rdev->mc.vram_start >> 20) & 0xF) << 20;
+		WREG32(MC_FUS_VM_FB_OFFSET, tmp);
+	}
 	tmp = ((rdev->mc.vram_end >> 24) & 0xFFFF) << 16;
 	tmp |= ((rdev->mc.vram_start >> 24) & 0xFFFF);
 	WREG32(MC_VM_FB_LOCATION, tmp);
@@ -1283,6 +1368,7 @@ static u32 evergreen_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
 	switch (rdev->family) {
 	case CHIP_CEDAR:
 	case CHIP_REDWOOD:
+	case CHIP_PALM:
 		force_no_swizzle = false;
 		break;
 	case CHIP_CYPRESS:
@@ -1382,6 +1468,43 @@ static u32 evergreen_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
 	return backend_map;
 }
 
+static void evergreen_program_channel_remap(struct radeon_device *rdev)
+{
+	u32 tcp_chan_steer_lo, tcp_chan_steer_hi, mc_shared_chremap, tmp;
+
+	tmp = RREG32(MC_SHARED_CHMAP);
+	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
+	case 0:
+	case 1:
+	case 2:
+	case 3:
+	default:
+		/* default mapping */
+		mc_shared_chremap = 0x00fac688;
+		break;
+	}
+
+	switch (rdev->family) {
+	case CHIP_HEMLOCK:
+	case CHIP_CYPRESS:
+		tcp_chan_steer_lo = 0x54763210;
+		tcp_chan_steer_hi = 0x0000ba98;
+		break;
+	case CHIP_JUNIPER:
+	case CHIP_REDWOOD:
+	case CHIP_CEDAR:
+	case CHIP_PALM:
+	default:
+		tcp_chan_steer_lo = 0x76543210;
+		tcp_chan_steer_hi = 0x0000ba98;
+		break;
+	}
+
+	WREG32(TCP_CHAN_STEER_LO, tcp_chan_steer_lo);
+	WREG32(TCP_CHAN_STEER_HI, tcp_chan_steer_hi);
+	WREG32(MC_SHARED_CHREMAP, mc_shared_chremap);
+}
+
 static void evergreen_gpu_init(struct radeon_device *rdev)
 {
 	u32 cc_rb_backend_disable = 0;
@@ -1493,6 +1616,27 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
 		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
 		break;
+	case CHIP_PALM:
+		rdev->config.evergreen.num_ses = 1;
+		rdev->config.evergreen.max_pipes = 2;
+		rdev->config.evergreen.max_tile_pipes = 2;
+		rdev->config.evergreen.max_simds = 2;
+		rdev->config.evergreen.max_backends = 1 * rdev->config.evergreen.num_ses;
+		rdev->config.evergreen.max_gprs = 256;
+		rdev->config.evergreen.max_threads = 192;
+		rdev->config.evergreen.max_gs_threads = 16;
+		rdev->config.evergreen.max_stack_entries = 256;
+		rdev->config.evergreen.sx_num_of_sets = 4;
+		rdev->config.evergreen.sx_max_export_size = 128;
+		rdev->config.evergreen.sx_max_export_pos_size = 32;
+		rdev->config.evergreen.sx_max_export_smx_size = 96;
+		rdev->config.evergreen.max_hw_contexts = 4;
+		rdev->config.evergreen.sq_num_cf_insts = 1;
+
+		rdev->config.evergreen.sc_prim_fifo_size = 0x40;
+		rdev->config.evergreen.sc_hiz_tile_fifo_size = 0x30;
+		rdev->config.evergreen.sc_earlyz_tile_fifo_size = 0x130;
+		break;
 	}
 
 	/* Initialize HDP */
@@ -1685,6 +1829,8 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	WREG32(DMIF_ADDR_CONFIG, gb_addr_config);
 	WREG32(HDP_ADDR_CONFIG, gb_addr_config);
 
+	evergreen_program_channel_remap(rdev);
+
 	num_shader_engines = ((RREG32(GB_ADDR_CONFIG) & NUM_SHADER_ENGINES(3)) >> 12) + 1;
 	grbm_gfx_index = INSTANCE_BROADCAST_WRITES;
 
@@ -1767,9 +1913,15 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 		      GS_PRIO(2) |
 		      ES_PRIO(3));
 
-	if (rdev->family == CHIP_CEDAR)
+	switch (rdev->family) {
+	case CHIP_CEDAR:
+	case CHIP_PALM:
 		/* no vertex cache */
 		sq_config &= ~VC_ENABLE;
+		break;
+	default:
+		break;
+	}
 
 	sq_lds_resource_mgmt = RREG32(SQ_LDS_RESOURCE_MGMT);
 
@@ -1781,10 +1933,15 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	sq_gpr_resource_mgmt_3 = NUM_HS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
 	sq_gpr_resource_mgmt_3 |= NUM_LS_GPRS((rdev->config.evergreen.max_gprs - (4 * 2)) * 3 / 32);
 
-	if (rdev->family == CHIP_CEDAR)
+	switch (rdev->family) {
+	case CHIP_CEDAR:
+	case CHIP_PALM:
 		ps_thread_count = 96;
-	else
+		break;
+	default:
 		ps_thread_count = 128;
+		break;
+	}
 
 	sq_thread_resource_mgmt = NUM_PS_THREADS(ps_thread_count);
 	sq_thread_resource_mgmt |= NUM_VS_THREADS((((rdev->config.evergreen.max_threads - ps_thread_count) / 6) / 8) * 8);
@@ -1815,10 +1972,15 @@ static void evergreen_gpu_init(struct radeon_device *rdev)
 	WREG32(PA_SC_FORCE_EOV_MAX_CNTS, (FORCE_EOV_MAX_CLK_CNT(4095) |
 					  FORCE_EOV_MAX_REZ_CNT(255)));
 
-	if (rdev->family == CHIP_CEDAR)
+	switch (rdev->family) {
+	case CHIP_CEDAR:
+	case CHIP_PALM:
 		vgt_cache_invalidation = CACHE_INVALIDATION(TC_ONLY);
-	else
+		break;
+	default:
 		vgt_cache_invalidation = CACHE_INVALIDATION(VC_AND_TC);
+		break;
+	}
 	vgt_cache_invalidation |= AUTO_INVLD_EN(ES_AND_GS_AUTO);
 	WREG32(VGT_CACHE_INVALIDATION, vgt_cache_invalidation);
 
@@ -1902,12 +2064,18 @@ int evergreen_mc_init(struct radeon_device *rdev)
 	rdev->mc.aper_base = pci_resource_start(rdev->pdev, 0);
 	rdev->mc.aper_size = pci_resource_len(rdev->pdev, 0);
 	/* Setup GPU memory space */
-	/* size in MB on evergreen */
-	rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
-	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
+	if (rdev->flags & RADEON_IS_IGP) {
+		/* size in bytes on fusion */
+		rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE);
+		rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
+	} else {
+		/* size in MB on evergreen */
+		rdev->mc.mc_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
+		rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE) * 1024 * 1024;
+	}
 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
 	rdev->mc.active_vram_size = rdev->mc.visible_vram_size;
-	r600_vram_gtt_location(rdev, &rdev->mc);
+	r700_vram_gtt_location(rdev, &rdev->mc);
 	radeon_update_bandwidth_info(rdev);
 
 	return 0;
@@ -2024,17 +2192,21 @@ void evergreen_disable_interrupt_state(struct radeon_device *rdev)
 	WREG32(GRBM_INT_CNTL, 0);
 	WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
-	WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
-	WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
-	WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
-	WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
+		WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
+		WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
+		WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	}
 
 	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, 0);
 	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, 0);
-	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
-	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
-	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
-	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, 0);
+		WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, 0);
+		WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, 0);
+		WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, 0);
+	}
 
 	WREG32(DACA_AUTODETECT_INT_CONTROL, 0);
 	WREG32(DACB_AUTODETECT_INT_CONTROL, 0);
@@ -2060,6 +2232,7 @@ int evergreen_irq_set(struct radeon_device *rdev)
 	u32 crtc1 = 0, crtc2 = 0, crtc3 = 0, crtc4 = 0, crtc5 = 0, crtc6 = 0;
 	u32 hpd1, hpd2, hpd3, hpd4, hpd5, hpd6;
 	u32 grbm_int_cntl = 0;
+	u32 grph1 = 0, grph2 = 0, grph3 = 0, grph4 = 0, grph5 = 0, grph6 = 0;
 
 	if (!rdev->irq.installed) {
 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
@@ -2085,27 +2258,33 @@ int evergreen_irq_set(struct radeon_device *rdev)
 		cp_int_cntl |= RB_INT_ENABLE;
 		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
 	}
-	if (rdev->irq.crtc_vblank_int[0]) {
+	if (rdev->irq.crtc_vblank_int[0] ||
+	    rdev->irq.pflip[0]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 0\n");
 		crtc1 |= VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[1]) {
+	if (rdev->irq.crtc_vblank_int[1] ||
+	    rdev->irq.pflip[1]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 1\n");
 		crtc2 |= VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[2]) {
+	if (rdev->irq.crtc_vblank_int[2] ||
+	    rdev->irq.pflip[2]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 2\n");
 		crtc3 |= VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[3]) {
+	if (rdev->irq.crtc_vblank_int[3] ||
+	    rdev->irq.pflip[3]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 3\n");
 		crtc4 |= VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[4]) {
+	if (rdev->irq.crtc_vblank_int[4] ||
+	    rdev->irq.pflip[4]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 4\n");
 		crtc5 |= VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[5]) {
+	if (rdev->irq.crtc_vblank_int[5] ||
+	    rdev->irq.pflip[5]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 5\n");
 		crtc6 |= VBLANK_INT_MASK;
 	}
@@ -2143,10 +2322,19 @@ int evergreen_irq_set(struct radeon_device *rdev)
 
 	WREG32(INT_MASK + EVERGREEN_CRTC0_REGISTER_OFFSET, crtc1);
 	WREG32(INT_MASK + EVERGREEN_CRTC1_REGISTER_OFFSET, crtc2);
-	WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
-	WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
-	WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
-	WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
+	if (!(rdev->flags & RADEON_IS_IGP)) {
+		WREG32(INT_MASK + EVERGREEN_CRTC2_REGISTER_OFFSET, crtc3);
+		WREG32(INT_MASK + EVERGREEN_CRTC3_REGISTER_OFFSET, crtc4);
+		WREG32(INT_MASK + EVERGREEN_CRTC4_REGISTER_OFFSET, crtc5);
+		WREG32(INT_MASK + EVERGREEN_CRTC5_REGISTER_OFFSET, crtc6);
+	}
+
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET, grph1);
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET, grph2);
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET, grph3);
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET, grph4);
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET, grph5);
+	WREG32(GRPH_INT_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET, grph6);
 
 	WREG32(DC_HPD1_INT_CONTROL, hpd1);
 	WREG32(DC_HPD2_INT_CONTROL, hpd2);
@@ -2158,79 +2346,92 @@ int evergreen_irq_set(struct radeon_device *rdev)
 	return 0;
 }
 
-static inline void evergreen_irq_ack(struct radeon_device *rdev,
-				     u32 *disp_int,
-				     u32 *disp_int_cont,
-				     u32 *disp_int_cont2,
-				     u32 *disp_int_cont3,
-				     u32 *disp_int_cont4,
-				     u32 *disp_int_cont5)
+static inline void evergreen_irq_ack(struct radeon_device *rdev)
 {
 	u32 tmp;
 
-	*disp_int = RREG32(DISP_INTERRUPT_STATUS);
-	*disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
-	*disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
-	*disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
-	*disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
-	*disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
-
-	if (*disp_int & LB_D1_VBLANK_INTERRUPT)
+	rdev->irq.stat_regs.evergreen.disp_int = RREG32(DISP_INTERRUPT_STATUS);
+	rdev->irq.stat_regs.evergreen.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
+	rdev->irq.stat_regs.evergreen.disp_int_cont2 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE2);
+	rdev->irq.stat_regs.evergreen.disp_int_cont3 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE3);
+	rdev->irq.stat_regs.evergreen.disp_int_cont4 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE4);
+	rdev->irq.stat_regs.evergreen.disp_int_cont5 = RREG32(DISP_INTERRUPT_STATUS_CONTINUE5);
+	rdev->irq.stat_regs.evergreen.d1grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET);
+	rdev->irq.stat_regs.evergreen.d2grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET);
+	rdev->irq.stat_regs.evergreen.d3grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET);
+	rdev->irq.stat_regs.evergreen.d4grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET);
+	rdev->irq.stat_regs.evergreen.d5grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET);
+	rdev->irq.stat_regs.evergreen.d6grph_int = RREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET);
+
+	if (rdev->irq.stat_regs.evergreen.d1grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.evergreen.d2grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.evergreen.d3grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.evergreen.d4grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.evergreen.d5grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.evergreen.d6grph_int & GRPH_PFLIP_INT_OCCURRED)
+		WREG32(GRPH_INT_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, GRPH_PFLIP_INT_CLEAR);
+
+	if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int & LB_D1_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC0_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int_cont & LB_D2_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int_cont & LB_D2_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC1_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC2_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC3_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC4_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT)
 		WREG32(VBLANK_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VBLANK_ACK);
-	if (*disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT)
 		WREG32(VLINE_STATUS + EVERGREEN_CRTC5_REGISTER_OFFSET, VLINE_ACK);
 
-	if (*disp_int & DC_HPD1_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
 		tmp = RREG32(DC_HPD1_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD1_INT_CONTROL, tmp);
 	}
-	if (*disp_int_cont & DC_HPD2_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
 		tmp = RREG32(DC_HPD2_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD2_INT_CONTROL, tmp);
 	}
-	if (*disp_int_cont2 & DC_HPD3_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
 		tmp = RREG32(DC_HPD3_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD3_INT_CONTROL, tmp);
 	}
-	if (*disp_int_cont3 & DC_HPD4_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
 		tmp = RREG32(DC_HPD4_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD4_INT_CONTROL, tmp);
 	}
-	if (*disp_int_cont4 & DC_HPD5_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
 		tmp = RREG32(DC_HPD5_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD5_INT_CONTROL, tmp);
 	}
-	if (*disp_int_cont5 & DC_HPD6_INTERRUPT) {
+	if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
 		tmp = RREG32(DC_HPD5_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD6_INT_CONTROL, tmp);
@@ -2239,14 +2440,10 @@ static inline void evergreen_irq_ack(struct radeon_device *rdev,
 
 void evergreen_irq_disable(struct radeon_device *rdev)
 {
-	u32 disp_int, disp_int_cont, disp_int_cont2;
-	u32 disp_int_cont3, disp_int_cont4, disp_int_cont5;
-
 	r600_disable_interrupts(rdev);
 	/* Wait and acknowledge irq */
 	mdelay(1);
-	evergreen_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2,
-			  &disp_int_cont3, &disp_int_cont4, &disp_int_cont5);
+	evergreen_irq_ack(rdev);
 	evergreen_disable_interrupt_state(rdev);
 }
 
@@ -2286,8 +2483,6 @@ int evergreen_irq_process(struct radeon_device *rdev)
 	u32 rptr = rdev->ih.rptr;
 	u32 src_id, src_data;
 	u32 ring_index;
-	u32 disp_int, disp_int_cont, disp_int_cont2;
-	u32 disp_int_cont3, disp_int_cont4, disp_int_cont5;
 	unsigned long flags;
 	bool queue_hotplug = false;
 
@@ -2308,8 +2503,7 @@ int evergreen_irq_process(struct radeon_device *rdev)
 
 restart_ih:
 	/* display interrupts */
-	evergreen_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2,
-			  &disp_int_cont3, &disp_int_cont4, &disp_int_cont5);
+	evergreen_irq_ack(rdev);
 
 	rdev->ih.wptr = wptr;
 	while (rptr != wptr) {
@@ -2322,17 +2516,21 @@ restart_ih:
 		case 1: /* D1 vblank/vline */
 			switch (src_data) {
 			case 0: /* D1 vblank */
-				if (disp_int & LB_D1_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 0);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int &= ~LB_D1_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[0]) {
+						drm_handle_vblank(rdev->ddev, 0);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[0])
+						radeon_crtc_handle_flip(rdev, 0);
+					rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D1 vblank\n");
 				}
 				break;
 			case 1: /* D1 vline */
-				if (disp_int & LB_D1_VLINE_INTERRUPT) {
-					disp_int &= ~LB_D1_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int & LB_D1_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int &= ~LB_D1_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D1 vline\n");
 				}
 				break;
@@ -2344,17 +2542,21 @@ restart_ih:
 		case 2: /* D2 vblank/vline */
 			switch (src_data) {
 			case 0: /* D2 vblank */
-				if (disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 1);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[1]) {
+						drm_handle_vblank(rdev->ddev, 1);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[1])
+						radeon_crtc_handle_flip(rdev, 1);
+					rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D2 vblank\n");
 				}
 				break;
 			case 1: /* D2 vline */
-				if (disp_int_cont & LB_D2_VLINE_INTERRUPT) {
-					disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont & LB_D2_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont &= ~LB_D2_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D2 vline\n");
 				}
 				break;
@@ -2366,17 +2568,21 @@ restart_ih:
 		case 3: /* D3 vblank/vline */
 			switch (src_data) {
 			case 0: /* D3 vblank */
-				if (disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 2);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[2]) {
+						drm_handle_vblank(rdev->ddev, 2);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[2])
+						radeon_crtc_handle_flip(rdev, 2);
+					rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D3 vblank\n");
 				}
 				break;
 			case 1: /* D3 vline */
-				if (disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
-					disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & LB_D3_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~LB_D3_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D3 vline\n");
 				}
 				break;
@@ -2388,17 +2594,21 @@ restart_ih:
 		case 4: /* D4 vblank/vline */
 			switch (src_data) {
 			case 0: /* D4 vblank */
-				if (disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 3);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[3]) {
+						drm_handle_vblank(rdev->ddev, 3);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[3])
+						radeon_crtc_handle_flip(rdev, 3);
+					rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D4 vblank\n");
 				}
 				break;
 			case 1: /* D4 vline */
-				if (disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
-					disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & LB_D4_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~LB_D4_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D4 vline\n");
 				}
 				break;
@@ -2410,17 +2620,21 @@ restart_ih:
 		case 5: /* D5 vblank/vline */
 			switch (src_data) {
 			case 0: /* D5 vblank */
-				if (disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 4);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[4]) {
+						drm_handle_vblank(rdev->ddev, 4);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[4])
+						radeon_crtc_handle_flip(rdev, 4);
+					rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D5 vblank\n");
 				}
 				break;
 			case 1: /* D5 vline */
-				if (disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
-					disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & LB_D5_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~LB_D5_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D5 vline\n");
 				}
 				break;
@@ -2432,17 +2646,21 @@ restart_ih:
 		case 6: /* D6 vblank/vline */
 			switch (src_data) {
 			case 0: /* D6 vblank */
-				if (disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 5);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[5]) {
+						drm_handle_vblank(rdev->ddev, 5);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[5])
+						radeon_crtc_handle_flip(rdev, 5);
+					rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D6 vblank\n");
 				}
 				break;
 			case 1: /* D6 vline */
-				if (disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
-					disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & LB_D6_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~LB_D6_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D6 vline\n");
 				}
 				break;
@@ -2454,43 +2672,43 @@ restart_ih:
 		case 42: /* HPD hotplug */
 			switch (src_data) {
 			case 0:
-				if (disp_int & DC_HPD1_INTERRUPT) {
-					disp_int &= ~DC_HPD1_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int & DC_HPD1_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int &= ~DC_HPD1_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD1\n");
 				}
 				break;
 			case 1:
-				if (disp_int_cont & DC_HPD2_INTERRUPT) {
-					disp_int_cont &= ~DC_HPD2_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont & DC_HPD2_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont &= ~DC_HPD2_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD2\n");
 				}
 				break;
 			case 2:
-				if (disp_int_cont2 & DC_HPD3_INTERRUPT) {
-					disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont2 & DC_HPD3_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont2 &= ~DC_HPD3_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD3\n");
 				}
 				break;
 			case 3:
-				if (disp_int_cont3 & DC_HPD4_INTERRUPT) {
-					disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont3 & DC_HPD4_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont3 &= ~DC_HPD4_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD4\n");
 				}
 				break;
 			case 4:
-				if (disp_int_cont4 & DC_HPD5_INTERRUPT) {
-					disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont4 & DC_HPD5_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont4 &= ~DC_HPD5_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD5\n");
 				}
 				break;
 			case 5:
-				if (disp_int_cont5 & DC_HPD6_INTERRUPT) {
-					disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
+				if (rdev->irq.stat_regs.evergreen.disp_int_cont5 & DC_HPD6_INTERRUPT) {
+					rdev->irq.stat_regs.evergreen.disp_int_cont5 &= ~DC_HPD6_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD6\n");
 				}
@@ -2666,12 +2884,16 @@ static bool evergreen_card_posted(struct radeon_device *rdev)
 	u32 reg;
 
 	/* first check CRTCs */
-	reg = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) |
-		RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET) |
-		RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET) |
-		RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET) |
-		RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET) |
-		RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
+	if (rdev->flags & RADEON_IS_IGP)
+		reg = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET);
+	else
+		reg = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC3_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC4_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC5_REGISTER_OFFSET);
 	if (reg & EVERGREEN_CRTC_MASTER_EN)
 		return true;
 
diff --git a/drivers/gpu/drm/radeon/evergreen_blit_kms.c b/drivers/gpu/drm/radeon/evergreen_blit_kms.c
index e0e590110dd4..2ccd1f0545fe 100644
--- a/drivers/gpu/drm/radeon/evergreen_blit_kms.c
+++ b/drivers/gpu/drm/radeon/evergreen_blit_kms.c
@@ -147,7 +147,8 @@ set_vtx_resource(struct radeon_device *rdev, u64 gpu_addr)
 	radeon_ring_write(rdev, 0);
 	radeon_ring_write(rdev, SQ_TEX_VTX_VALID_BUFFER << 30);
 
-	if (rdev->family == CHIP_CEDAR)
+	if ((rdev->family == CHIP_CEDAR) ||
+	    (rdev->family == CHIP_PALM))
 		cp_set_surface_sync(rdev,
 				    PACKET3_TC_ACTION_ENA, 48, gpu_addr);
 	else
@@ -331,9 +332,31 @@ set_default_state(struct radeon_device *rdev)
 		num_hs_stack_entries = 85;
 		num_ls_stack_entries = 85;
 		break;
+	case CHIP_PALM:
+		num_ps_gprs = 93;
+		num_vs_gprs = 46;
+		num_temp_gprs = 4;
+		num_gs_gprs = 31;
+		num_es_gprs = 31;
+		num_hs_gprs = 23;
+		num_ls_gprs = 23;
+		num_ps_threads = 96;
+		num_vs_threads = 16;
+		num_gs_threads = 16;
+		num_es_threads = 16;
+		num_hs_threads = 16;
+		num_ls_threads = 16;
+		num_ps_stack_entries = 42;
+		num_vs_stack_entries = 42;
+		num_gs_stack_entries = 42;
+		num_es_stack_entries = 42;
+		num_hs_stack_entries = 42;
+		num_ls_stack_entries = 42;
+		break;
 	}
 
-	if (rdev->family == CHIP_CEDAR)
+	if ((rdev->family == CHIP_CEDAR) ||
+	    (rdev->family == CHIP_PALM))
 		sq_config = 0;
 	else
 		sq_config = VC_ENABLE;
diff --git a/drivers/gpu/drm/radeon/evergreen_reg.h b/drivers/gpu/drm/radeon/evergreen_reg.h
index 2330f3a36fd5..c781c92c3451 100644
--- a/drivers/gpu/drm/radeon/evergreen_reg.h
+++ b/drivers/gpu/drm/radeon/evergreen_reg.h
@@ -105,6 +105,11 @@
 #define EVERGREEN_GRPH_Y_START                          0x6830
 #define EVERGREEN_GRPH_X_END                            0x6834
 #define EVERGREEN_GRPH_Y_END                            0x6838
+#define EVERGREEN_GRPH_UPDATE                           0x6844
+#       define EVERGREEN_GRPH_SURFACE_UPDATE_PENDING    (1 << 2)
+#       define EVERGREEN_GRPH_UPDATE_LOCK               (1 << 16)
+#define EVERGREEN_GRPH_FLIP_CONTROL                     0x6848
+#       define EVERGREEN_GRPH_SURFACE_UPDATE_H_RETRACE_EN (1 << 0)
 
 /* CUR blocks at 0x6998, 0x7598, 0x10198, 0x10d98, 0x11998, 0x12598 */
 #define EVERGREEN_CUR_CONTROL                           0x6998
@@ -178,6 +183,7 @@
 #       define EVERGREEN_CRTC_DISP_READ_REQUEST_DISABLE (1 << 24)
 #define EVERGREEN_CRTC_STATUS                           0x6e8c
 #define EVERGREEN_CRTC_STATUS_POSITION                  0x6e90
+#define EVERGREEN_MASTER_UPDATE_MODE                    0x6ef8
 #define EVERGREEN_CRTC_UPDATE_LOCK                      0x6ed4
 
 #define EVERGREEN_DC_GPIO_HPD_MASK                      0x64b0
diff --git a/drivers/gpu/drm/radeon/evergreend.h b/drivers/gpu/drm/radeon/evergreend.h
index 113c70cc8b39..5b869ce86917 100644
--- a/drivers/gpu/drm/radeon/evergreend.h
+++ b/drivers/gpu/drm/radeon/evergreend.h
@@ -164,11 +164,13 @@
 #define		SE_SC_BUSY					(1 << 29)
 #define		SE_DB_BUSY					(1 << 30)
 #define		SE_CB_BUSY					(1 << 31)
-
+/* evergreen */
 #define	CG_MULT_THERMAL_STATUS				0x740
 #define		ASIC_T(x)			        ((x) << 16)
 #define		ASIC_T_MASK			        0x7FF0000
 #define		ASIC_T_SHIFT			        16
+/* APU */
+#define	CG_THERMAL_STATUS			        0x678
 
 #define	HDP_HOST_PATH_CNTL				0x2C00
 #define	HDP_NONSURFACE_BASE				0x2C04
@@ -180,6 +182,7 @@
 #define MC_SHARED_CHMAP						0x2004
 #define		NOOFCHAN_SHIFT					12
 #define		NOOFCHAN_MASK					0x00003000
+#define MC_SHARED_CHREMAP					0x2008
 
 #define	MC_ARB_RAMCFG					0x2760
 #define		NOOFBANK_SHIFT					0
@@ -199,6 +202,7 @@
 #define	MC_VM_AGP_BOT					0x202C
 #define	MC_VM_AGP_BASE					0x2030
 #define	MC_VM_FB_LOCATION				0x2024
+#define	MC_FUS_VM_FB_OFFSET				0x2898
 #define	MC_VM_MB_L1_TLB0_CNTL				0x2234
 #define	MC_VM_MB_L1_TLB1_CNTL				0x2238
 #define	MC_VM_MB_L1_TLB2_CNTL				0x223C
@@ -348,6 +352,9 @@
 #define		SYNC_WALKER					(1 << 25)
 #define		SYNC_ALIGNER					(1 << 26)
 
+#define	TCP_CHAN_STEER_LO				0x960c
+#define	TCP_CHAN_STEER_HI				0x9610
+
 #define	VGT_CACHE_INVALIDATION				0x88C4
 #define		CACHE_INVALIDATION(x)				((x) << 0)
 #define			VC_ONLY						0
diff --git a/drivers/gpu/drm/radeon/r100.c b/drivers/gpu/drm/radeon/r100.c
index 8e10aa9f74b0..300b4a64d8fe 100644
--- a/drivers/gpu/drm/radeon/r100.c
+++ b/drivers/gpu/drm/radeon/r100.c
@@ -68,6 +68,56 @@ MODULE_FIRMWARE(FIRMWARE_R520);
  * r100,rv100,rs100,rv200,rs200,r200,rv250,rs300,rv280
  */
 
+void r100_pre_page_flip(struct radeon_device *rdev, int crtc)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc];
+	u32 tmp;
+
+	/* make sure flip is at vb rather than hb */
+	tmp = RREG32(RADEON_CRTC_OFFSET_CNTL + radeon_crtc->crtc_offset);
+	tmp &= ~RADEON_CRTC_OFFSET_FLIP_CNTL;
+	/* make sure pending bit is asserted */
+	tmp |= RADEON_CRTC_GUI_TRIG_OFFSET_LEFT_EN;
+	WREG32(RADEON_CRTC_OFFSET_CNTL + radeon_crtc->crtc_offset, tmp);
+
+	/* set pageflip to happen as late as possible in the vblank interval.
+	 * same field for crtc1/2
+	 */
+	tmp = RREG32(RADEON_CRTC_GEN_CNTL);
+	tmp &= ~RADEON_CRTC_VSTAT_MODE_MASK;
+	WREG32(RADEON_CRTC_GEN_CNTL, tmp);
+
+	/* enable the pflip int */
+	radeon_irq_kms_pflip_irq_get(rdev, crtc);
+}
+
+void r100_post_page_flip(struct radeon_device *rdev, int crtc)
+{
+	/* disable the pflip int */
+	radeon_irq_kms_pflip_irq_put(rdev, crtc);
+}
+
+u32 r100_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
+	u32 tmp = ((u32)crtc_base) | RADEON_CRTC_OFFSET__OFFSET_LOCK;
+
+	/* Lock the graphics update lock */
+	/* update the scanout addresses */
+	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
+
+	/* Wait for update_pending to go high. */
+	while (!(RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET));
+	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
+
+	/* Unlock the lock, so double-buffering can take place inside vblank */
+	tmp &= ~RADEON_CRTC_OFFSET__OFFSET_LOCK;
+	WREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset, tmp);
+
+	/* Return current update_pending status: */
+	return RREG32(RADEON_CRTC_OFFSET + radeon_crtc->crtc_offset) & RADEON_CRTC_OFFSET__GUI_TRIG_OFFSET;
+}
+
 void r100_pm_get_dynpm_state(struct radeon_device *rdev)
 {
 	int i;
@@ -526,10 +576,12 @@ int r100_irq_set(struct radeon_device *rdev)
 	if (rdev->irq.gui_idle) {
 		tmp |= RADEON_GUI_IDLE_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[0]) {
+	if (rdev->irq.crtc_vblank_int[0] ||
+	    rdev->irq.pflip[0]) {
 		tmp |= RADEON_CRTC_VBLANK_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[1]) {
+	if (rdev->irq.crtc_vblank_int[1] ||
+	    rdev->irq.pflip[1]) {
 		tmp |= RADEON_CRTC2_VBLANK_MASK;
 	}
 	if (rdev->irq.hpd[0]) {
@@ -600,14 +652,22 @@ int r100_irq_process(struct radeon_device *rdev)
 		}
 		/* Vertical blank interrupts */
 		if (status & RADEON_CRTC_VBLANK_STAT) {
-			drm_handle_vblank(rdev->ddev, 0);
-			rdev->pm.vblank_sync = true;
-			wake_up(&rdev->irq.vblank_queue);
+			if (rdev->irq.crtc_vblank_int[0]) {
+				drm_handle_vblank(rdev->ddev, 0);
+				rdev->pm.vblank_sync = true;
+				wake_up(&rdev->irq.vblank_queue);
+			}
+			if (rdev->irq.pflip[0])
+				radeon_crtc_handle_flip(rdev, 0);
 		}
 		if (status & RADEON_CRTC2_VBLANK_STAT) {
-			drm_handle_vblank(rdev->ddev, 1);
-			rdev->pm.vblank_sync = true;
-			wake_up(&rdev->irq.vblank_queue);
+			if (rdev->irq.crtc_vblank_int[1]) {
+				drm_handle_vblank(rdev->ddev, 1);
+				rdev->pm.vblank_sync = true;
+				wake_up(&rdev->irq.vblank_queue);
+			}
+			if (rdev->irq.pflip[1])
+				radeon_crtc_handle_flip(rdev, 1);
 		}
 		if (status & RADEON_FP_DETECT_STAT) {
 			queue_hotplug = true;
diff --git a/drivers/gpu/drm/radeon/r500_reg.h b/drivers/gpu/drm/radeon/r500_reg.h
index 6ac1f604e29b..fc437059918f 100644
--- a/drivers/gpu/drm/radeon/r500_reg.h
+++ b/drivers/gpu/drm/radeon/r500_reg.h
@@ -355,6 +355,8 @@
 #define AVIVO_D1CRTC_FRAME_COUNT                                0x60a4
 #define AVIVO_D1CRTC_STEREO_CONTROL                             0x60c4
 
+#define AVIVO_D1MODE_MASTER_UPDATE_MODE                         0x60e4
+
 /* master controls */
 #define AVIVO_DC_CRTC_MASTER_EN                                 0x60f8
 #define AVIVO_DC_CRTC_TV_CONTROL                                0x60fc
@@ -409,8 +411,10 @@
 #define AVIVO_D1GRPH_X_END                                      0x6134
 #define AVIVO_D1GRPH_Y_END                                      0x6138
 #define AVIVO_D1GRPH_UPDATE                                     0x6144
+#       define AVIVO_D1GRPH_SURFACE_UPDATE_PENDING              (1 << 2)
 #       define AVIVO_D1GRPH_UPDATE_LOCK                         (1 << 16)
 #define AVIVO_D1GRPH_FLIP_CONTROL                               0x6148
+#       define AVIVO_D1GRPH_SURFACE_UPDATE_H_RETRACE_EN         (1 << 0)
 
 #define AVIVO_D1CUR_CONTROL                     0x6400
 #       define AVIVO_D1CURSOR_EN                (1 << 0)
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index a322d4f647bd..c6a37e036f11 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -83,6 +83,9 @@ MODULE_FIRMWARE("radeon/JUNIPER_rlc.bin");
 MODULE_FIRMWARE("radeon/CYPRESS_pfp.bin");
 MODULE_FIRMWARE("radeon/CYPRESS_me.bin");
 MODULE_FIRMWARE("radeon/CYPRESS_rlc.bin");
+MODULE_FIRMWARE("radeon/PALM_pfp.bin");
+MODULE_FIRMWARE("radeon/PALM_me.bin");
+MODULE_FIRMWARE("radeon/SUMO_rlc.bin");
 
 int r600_debugfs_mc_info_init(struct radeon_device *rdev);
 
@@ -1161,7 +1164,7 @@ static void r600_mc_program(struct radeon_device *rdev)
  * Note: GTT start, end, size should be initialized before calling this
  * function on AGP platform.
  */
-void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
+static void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
 {
 	u64 size_bf, size_af;
 
@@ -2000,6 +2003,10 @@ int r600_init_microcode(struct radeon_device *rdev)
 		chip_name = "CYPRESS";
 		rlc_chip_name = "CYPRESS";
 		break;
+	case CHIP_PALM:
+		chip_name = "PALM";
+		rlc_chip_name = "SUMO";
+		break;
 	default: BUG();
 	}
 
@@ -2865,6 +2872,8 @@ static void r600_disable_interrupt_state(struct radeon_device *rdev)
 	WREG32(CP_INT_CNTL, CNTX_BUSY_INT_ENABLE | CNTX_EMPTY_INT_ENABLE);
 	WREG32(GRBM_INT_CNTL, 0);
 	WREG32(DxMODE_INT_MASK, 0);
+	WREG32(D1GRPH_INTERRUPT_CONTROL, 0);
+	WREG32(D2GRPH_INTERRUPT_CONTROL, 0);
 	if (ASIC_IS_DCE3(rdev)) {
 		WREG32(DCE3_DACA_AUTODETECT_INT_CONTROL, 0);
 		WREG32(DCE3_DACB_AUTODETECT_INT_CONTROL, 0);
@@ -2989,6 +2998,7 @@ int r600_irq_set(struct radeon_device *rdev)
 	u32 hpd1, hpd2, hpd3, hpd4 = 0, hpd5 = 0, hpd6 = 0;
 	u32 grbm_int_cntl = 0;
 	u32 hdmi1, hdmi2;
+	u32 d1grph = 0, d2grph = 0;
 
 	if (!rdev->irq.installed) {
 		WARN(1, "Can't enable IRQ/MSI because no handler is installed\n");
@@ -3025,11 +3035,13 @@ int r600_irq_set(struct radeon_device *rdev)
 		cp_int_cntl |= RB_INT_ENABLE;
 		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
 	}
-	if (rdev->irq.crtc_vblank_int[0]) {
+	if (rdev->irq.crtc_vblank_int[0] ||
+	    rdev->irq.pflip[0]) {
 		DRM_DEBUG("r600_irq_set: vblank 0\n");
 		mode_int |= D1MODE_VBLANK_INT_MASK;
 	}
-	if (rdev->irq.crtc_vblank_int[1]) {
+	if (rdev->irq.crtc_vblank_int[1] ||
+	    rdev->irq.pflip[1]) {
 		DRM_DEBUG("r600_irq_set: vblank 1\n");
 		mode_int |= D2MODE_VBLANK_INT_MASK;
 	}
@@ -3072,6 +3084,8 @@ int r600_irq_set(struct radeon_device *rdev)
 
 	WREG32(CP_INT_CNTL, cp_int_cntl);
 	WREG32(DxMODE_INT_MASK, mode_int);
+	WREG32(D1GRPH_INTERRUPT_CONTROL, d1grph);
+	WREG32(D2GRPH_INTERRUPT_CONTROL, d2grph);
 	WREG32(GRBM_INT_CNTL, grbm_int_cntl);
 	WREG32(R600_HDMI_BLOCK1 + R600_HDMI_CNTL, hdmi1);
 	if (ASIC_IS_DCE3(rdev)) {
@@ -3094,32 +3108,35 @@ int r600_irq_set(struct radeon_device *rdev)
 	return 0;
 }
 
-static inline void r600_irq_ack(struct radeon_device *rdev,
-				u32 *disp_int,
-				u32 *disp_int_cont,
-				u32 *disp_int_cont2)
+static inline void r600_irq_ack(struct radeon_device *rdev)
 {
 	u32 tmp;
 
 	if (ASIC_IS_DCE3(rdev)) {
-		*disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
-		*disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
-		*disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
+		rdev->irq.stat_regs.r600.disp_int = RREG32(DCE3_DISP_INTERRUPT_STATUS);
+		rdev->irq.stat_regs.r600.disp_int_cont = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE);
+		rdev->irq.stat_regs.r600.disp_int_cont2 = RREG32(DCE3_DISP_INTERRUPT_STATUS_CONTINUE2);
 	} else {
-		*disp_int = RREG32(DISP_INTERRUPT_STATUS);
-		*disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
-		*disp_int_cont2 = 0;
-	}
-
-	if (*disp_int & LB_D1_VBLANK_INTERRUPT)
+		rdev->irq.stat_regs.r600.disp_int = RREG32(DISP_INTERRUPT_STATUS);
+		rdev->irq.stat_regs.r600.disp_int_cont = RREG32(DISP_INTERRUPT_STATUS_CONTINUE);
+		rdev->irq.stat_regs.r600.disp_int_cont2 = 0;
+	}
+	rdev->irq.stat_regs.r600.d1grph_int = RREG32(D1GRPH_INTERRUPT_STATUS);
+	rdev->irq.stat_regs.r600.d2grph_int = RREG32(D2GRPH_INTERRUPT_STATUS);
+
+	if (rdev->irq.stat_regs.r600.d1grph_int & DxGRPH_PFLIP_INT_OCCURRED)
+		WREG32(D1GRPH_INTERRUPT_STATUS, DxGRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.r600.d2grph_int & DxGRPH_PFLIP_INT_OCCURRED)
+		WREG32(D2GRPH_INTERRUPT_STATUS, DxGRPH_PFLIP_INT_CLEAR);
+	if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VBLANK_INTERRUPT)
 		WREG32(D1MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
-	if (*disp_int & LB_D1_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VLINE_INTERRUPT)
 		WREG32(D1MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
-	if (*disp_int & LB_D2_VBLANK_INTERRUPT)
+	if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VBLANK_INTERRUPT)
 		WREG32(D2MODE_VBLANK_STATUS, DxMODE_VBLANK_ACK);
-	if (*disp_int & LB_D2_VLINE_INTERRUPT)
+	if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VLINE_INTERRUPT)
 		WREG32(D2MODE_VLINE_STATUS, DxMODE_VLINE_ACK);
-	if (*disp_int & DC_HPD1_INTERRUPT) {
+	if (rdev->irq.stat_regs.r600.disp_int & DC_HPD1_INTERRUPT) {
 		if (ASIC_IS_DCE3(rdev)) {
 			tmp = RREG32(DC_HPD1_INT_CONTROL);
 			tmp |= DC_HPDx_INT_ACK;
@@ -3130,7 +3147,7 @@ static inline void r600_irq_ack(struct radeon_device *rdev,
 			WREG32(DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
 		}
 	}
-	if (*disp_int & DC_HPD2_INTERRUPT) {
+	if (rdev->irq.stat_regs.r600.disp_int & DC_HPD2_INTERRUPT) {
 		if (ASIC_IS_DCE3(rdev)) {
 			tmp = RREG32(DC_HPD2_INT_CONTROL);
 			tmp |= DC_HPDx_INT_ACK;
@@ -3141,7 +3158,7 @@ static inline void r600_irq_ack(struct radeon_device *rdev,
 			WREG32(DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
 		}
 	}
-	if (*disp_int_cont & DC_HPD3_INTERRUPT) {
+	if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD3_INTERRUPT) {
 		if (ASIC_IS_DCE3(rdev)) {
 			tmp = RREG32(DC_HPD3_INT_CONTROL);
 			tmp |= DC_HPDx_INT_ACK;
@@ -3152,18 +3169,18 @@ static inline void r600_irq_ack(struct radeon_device *rdev,
 			WREG32(DC_HOT_PLUG_DETECT3_INT_CONTROL, tmp);
 		}
 	}
-	if (*disp_int_cont & DC_HPD4_INTERRUPT) {
+	if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD4_INTERRUPT) {
 		tmp = RREG32(DC_HPD4_INT_CONTROL);
 		tmp |= DC_HPDx_INT_ACK;
 		WREG32(DC_HPD4_INT_CONTROL, tmp);
 	}
 	if (ASIC_IS_DCE32(rdev)) {
-		if (*disp_int_cont2 & DC_HPD5_INTERRUPT) {
+		if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD5_INTERRUPT) {
 			tmp = RREG32(DC_HPD5_INT_CONTROL);
 			tmp |= DC_HPDx_INT_ACK;
 			WREG32(DC_HPD5_INT_CONTROL, tmp);
 		}
-		if (*disp_int_cont2 & DC_HPD6_INTERRUPT) {
+		if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD6_INTERRUPT) {
 			tmp = RREG32(DC_HPD5_INT_CONTROL);
 			tmp |= DC_HPDx_INT_ACK;
 			WREG32(DC_HPD6_INT_CONTROL, tmp);
@@ -3185,12 +3202,10 @@ static inline void r600_irq_ack(struct radeon_device *rdev,
 
 void r600_irq_disable(struct radeon_device *rdev)
 {
-	u32 disp_int, disp_int_cont, disp_int_cont2;
-
 	r600_disable_interrupts(rdev);
 	/* Wait and acknowledge irq */
 	mdelay(1);
-	r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
+	r600_irq_ack(rdev);
 	r600_disable_interrupt_state(rdev);
 }
 
@@ -3253,7 +3268,7 @@ int r600_irq_process(struct radeon_device *rdev)
 	u32 wptr = r600_get_ih_wptr(rdev);
 	u32 rptr = rdev->ih.rptr;
 	u32 src_id, src_data;
-	u32 ring_index, disp_int, disp_int_cont, disp_int_cont2;
+	u32 ring_index;
 	unsigned long flags;
 	bool queue_hotplug = false;
 
@@ -3274,7 +3289,7 @@ int r600_irq_process(struct radeon_device *rdev)
 
 restart_ih:
 	/* display interrupts */
-	r600_irq_ack(rdev, &disp_int, &disp_int_cont, &disp_int_cont2);
+	r600_irq_ack(rdev);
 
 	rdev->ih.wptr = wptr;
 	while (rptr != wptr) {
@@ -3287,17 +3302,21 @@ restart_ih:
 		case 1: /* D1 vblank/vline */
 			switch (src_data) {
 			case 0: /* D1 vblank */
-				if (disp_int & LB_D1_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 0);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int &= ~LB_D1_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[0]) {
+						drm_handle_vblank(rdev->ddev, 0);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[0])
+						radeon_crtc_handle_flip(rdev, 0);
+					rdev->irq.stat_regs.r600.disp_int &= ~LB_D1_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D1 vblank\n");
 				}
 				break;
 			case 1: /* D1 vline */
-				if (disp_int & LB_D1_VLINE_INTERRUPT) {
-					disp_int &= ~LB_D1_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & LB_D1_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int &= ~LB_D1_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D1 vline\n");
 				}
 				break;
@@ -3309,17 +3328,21 @@ restart_ih:
 		case 5: /* D2 vblank/vline */
 			switch (src_data) {
 			case 0: /* D2 vblank */
-				if (disp_int & LB_D2_VBLANK_INTERRUPT) {
-					drm_handle_vblank(rdev->ddev, 1);
-					rdev->pm.vblank_sync = true;
-					wake_up(&rdev->irq.vblank_queue);
-					disp_int &= ~LB_D2_VBLANK_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VBLANK_INTERRUPT) {
+					if (rdev->irq.crtc_vblank_int[1]) {
+						drm_handle_vblank(rdev->ddev, 1);
+						rdev->pm.vblank_sync = true;
+						wake_up(&rdev->irq.vblank_queue);
+					}
+					if (rdev->irq.pflip[1])
+						radeon_crtc_handle_flip(rdev, 1);
+					rdev->irq.stat_regs.r600.disp_int &= ~LB_D2_VBLANK_INTERRUPT;
 					DRM_DEBUG("IH: D2 vblank\n");
 				}
 				break;
 			case 1: /* D1 vline */
-				if (disp_int & LB_D2_VLINE_INTERRUPT) {
-					disp_int &= ~LB_D2_VLINE_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & LB_D2_VLINE_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int &= ~LB_D2_VLINE_INTERRUPT;
 					DRM_DEBUG("IH: D2 vline\n");
 				}
 				break;
@@ -3331,43 +3354,43 @@ restart_ih:
 		case 19: /* HPD/DAC hotplug */
 			switch (src_data) {
 			case 0:
-				if (disp_int & DC_HPD1_INTERRUPT) {
-					disp_int &= ~DC_HPD1_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & DC_HPD1_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int &= ~DC_HPD1_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD1\n");
 				}
 				break;
 			case 1:
-				if (disp_int & DC_HPD2_INTERRUPT) {
-					disp_int &= ~DC_HPD2_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int & DC_HPD2_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int &= ~DC_HPD2_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD2\n");
 				}
 				break;
 			case 4:
-				if (disp_int_cont & DC_HPD3_INTERRUPT) {
-					disp_int_cont &= ~DC_HPD3_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD3_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int_cont &= ~DC_HPD3_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD3\n");
 				}
 				break;
 			case 5:
-				if (disp_int_cont & DC_HPD4_INTERRUPT) {
-					disp_int_cont &= ~DC_HPD4_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int_cont & DC_HPD4_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int_cont &= ~DC_HPD4_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD4\n");
 				}
 				break;
 			case 10:
-				if (disp_int_cont2 & DC_HPD5_INTERRUPT) {
-					disp_int_cont2 &= ~DC_HPD5_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD5_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int_cont2 &= ~DC_HPD5_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD5\n");
 				}
 				break;
 			case 12:
-				if (disp_int_cont2 & DC_HPD6_INTERRUPT) {
-					disp_int_cont2 &= ~DC_HPD6_INTERRUPT;
+				if (rdev->irq.stat_regs.r600.disp_int_cont2 & DC_HPD6_INTERRUPT) {
+					rdev->irq.stat_regs.r600.disp_int_cont2 &= ~DC_HPD6_INTERRUPT;
 					queue_hotplug = true;
 					DRM_DEBUG("IH: HPD6\n");
 				}
diff --git a/drivers/gpu/drm/radeon/r600d.h b/drivers/gpu/drm/radeon/r600d.h
index bff4dc4f410f..c89cfa8e0c05 100644
--- a/drivers/gpu/drm/radeon/r600d.h
+++ b/drivers/gpu/drm/radeon/r600d.h
@@ -728,6 +728,15 @@
 /* DCE 3.2 */
 #       define DC_HPDx_EN                                 (1 << 28)
 
+#define D1GRPH_INTERRUPT_STATUS                           0x6158
+#define D2GRPH_INTERRUPT_STATUS                           0x6958
+#       define DxGRPH_PFLIP_INT_OCCURRED                  (1 << 0)
+#       define DxGRPH_PFLIP_INT_CLEAR                     (1 << 8)
+#define D1GRPH_INTERRUPT_CONTROL                          0x615c
+#define D2GRPH_INTERRUPT_CONTROL                          0x695c
+#       define DxGRPH_PFLIP_INT_MASK                      (1 << 0)
+#       define DxGRPH_PFLIP_INT_TYPE                      (1 << 8)
+
 /*
  * PM4
  */
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 3a7095743d44..431d4186ddf0 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -69,6 +69,7 @@
 #include <ttm/ttm_bo_driver.h>
 #include <ttm/ttm_placement.h>
 #include <ttm/ttm_module.h>
+#include <ttm/ttm_execbuf_util.h>
 
 #include "radeon_family.h"
 #include "radeon_mode.h"
@@ -180,6 +181,7 @@ void rs690_pm_info(struct radeon_device *rdev);
 extern u32 rv6xx_get_temp(struct radeon_device *rdev);
 extern u32 rv770_get_temp(struct radeon_device *rdev);
 extern u32 evergreen_get_temp(struct radeon_device *rdev);
+extern u32 sumo_get_temp(struct radeon_device *rdev);
 
 /*
  * Fences.
@@ -259,13 +261,12 @@ struct radeon_bo {
 };
 
 struct radeon_bo_list {
-	struct list_head	list;
+	struct ttm_validate_buffer tv;
 	struct radeon_bo	*bo;
 	uint64_t		gpu_offset;
 	unsigned		rdomain;
 	unsigned		wdomain;
 	u32			tiling_flags;
-	bool			reserved;
 };
 
 /*
@@ -377,11 +378,56 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg);
 /*
  * IRQS.
  */
+
+struct radeon_unpin_work {
+	struct work_struct work;
+	struct radeon_device *rdev;
+	int crtc_id;
+	struct radeon_fence *fence;
+	struct drm_pending_vblank_event *event;
+	struct radeon_bo *old_rbo;
+	u64 new_crtc_base;
+};
+
+struct r500_irq_stat_regs {
+	u32 disp_int;
+};
+
+struct r600_irq_stat_regs {
+	u32 disp_int;
+	u32 disp_int_cont;
+	u32 disp_int_cont2;
+	u32 d1grph_int;
+	u32 d2grph_int;
+};
+
+struct evergreen_irq_stat_regs {
+	u32 disp_int;
+	u32 disp_int_cont;
+	u32 disp_int_cont2;
+	u32 disp_int_cont3;
+	u32 disp_int_cont4;
+	u32 disp_int_cont5;
+	u32 d1grph_int;
+	u32 d2grph_int;
+	u32 d3grph_int;
+	u32 d4grph_int;
+	u32 d5grph_int;
+	u32 d6grph_int;
+};
+
+union radeon_irq_stat_regs {
+	struct r500_irq_stat_regs r500;
+	struct r600_irq_stat_regs r600;
+	struct evergreen_irq_stat_regs evergreen;
+};
+
 struct radeon_irq {
 	bool		installed;
 	bool		sw_int;
 	/* FIXME: use a define max crtc rather than hardcode it */
 	bool		crtc_vblank_int[6];
+	bool		pflip[6];
 	wait_queue_head_t	vblank_queue;
 	/* FIXME: use defines for max hpd/dacs */
 	bool            hpd[6];
@@ -392,12 +438,17 @@ struct radeon_irq {
 	bool		hdmi[2];
 	spinlock_t sw_lock;
 	int sw_refcount;
+	union radeon_irq_stat_regs stat_regs;
+	spinlock_t pflip_lock[6];
+	int pflip_refcount[6];
 };
 
 int radeon_irq_kms_init(struct radeon_device *rdev);
 void radeon_irq_kms_fini(struct radeon_device *rdev);
 void radeon_irq_kms_sw_irq_get(struct radeon_device *rdev);
 void radeon_irq_kms_sw_irq_put(struct radeon_device *rdev);
+void radeon_irq_kms_pflip_irq_get(struct radeon_device *rdev, int crtc);
+void radeon_irq_kms_pflip_irq_put(struct radeon_device *rdev, int crtc);
 
 /*
  * CP & ring.
@@ -687,6 +738,7 @@ enum radeon_int_thermal_type {
 	THERMAL_TYPE_RV6XX,
 	THERMAL_TYPE_RV770,
 	THERMAL_TYPE_EVERGREEN,
+	THERMAL_TYPE_SUMO,
 };
 
 struct radeon_voltage {
@@ -881,6 +933,10 @@ struct radeon_asic {
 	void (*pm_finish)(struct radeon_device *rdev);
 	void (*pm_init_profile)(struct radeon_device *rdev);
 	void (*pm_get_dynpm_state)(struct radeon_device *rdev);
+	/* pageflipping */
+	void (*pre_page_flip)(struct radeon_device *rdev, int crtc);
+	u32 (*page_flip)(struct radeon_device *rdev, int crtc, u64 crtc_base);
+	void (*post_page_flip)(struct radeon_device *rdev, int crtc);
 };
 
 /*
@@ -1269,6 +1325,7 @@ void r100_pll_errata_after_index(struct radeon_device *rdev);
 #define ASIC_IS_DCE3(rdev) ((rdev->family >= CHIP_RV620))
 #define ASIC_IS_DCE32(rdev) ((rdev->family >= CHIP_RV730))
 #define ASIC_IS_DCE4(rdev) ((rdev->family >= CHIP_CEDAR))
+#define ASIC_IS_DCE41(rdev) ((rdev->family >= CHIP_PALM))
 
 /*
  * BIOS helpers.
@@ -1344,6 +1401,9 @@ static inline void radeon_ring_write(struct radeon_device *rdev, uint32_t v)
 #define radeon_pm_finish(rdev) (rdev)->asic->pm_finish((rdev))
 #define radeon_pm_init_profile(rdev) (rdev)->asic->pm_init_profile((rdev))
 #define radeon_pm_get_dynpm_state(rdev) (rdev)->asic->pm_get_dynpm_state((rdev))
+#define radeon_pre_page_flip(rdev, crtc) rdev->asic->pre_page_flip((rdev), (crtc))
+#define radeon_page_flip(rdev, crtc, base) rdev->asic->page_flip((rdev), (crtc), (base))
+#define radeon_post_page_flip(rdev, crtc) rdev->asic->post_page_flip((rdev), (crtc))
 
 /* Common functions */
 /* AGP */
@@ -1432,7 +1492,6 @@ extern void rs690_line_buffer_adjust(struct radeon_device *rdev,
 					struct drm_display_mode *mode2);
 
 /* r600, rv610, rv630, rv620, rv635, rv670, rs780, rs880 */
-extern void r600_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
 extern bool r600_card_posted(struct radeon_device *rdev);
 extern void r600_cp_stop(struct radeon_device *rdev);
 extern int r600_cp_start(struct radeon_device *rdev);
@@ -1478,6 +1537,7 @@ extern void r600_hdmi_setmode(struct drm_encoder *encoder, struct drm_display_mo
 extern int r600_hdmi_buffer_status_changed(struct drm_encoder *encoder);
 extern void r600_hdmi_update_audio_settings(struct drm_encoder *encoder);
 
+extern void r700_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc);
 extern void r700_cp_stop(struct radeon_device *rdev);
 extern void r700_cp_fini(struct radeon_device *rdev);
 extern void evergreen_disable_interrupt_state(struct radeon_device *rdev);
diff --git a/drivers/gpu/drm/radeon/radeon_asic.c b/drivers/gpu/drm/radeon/radeon_asic.c
index 64fb89ecbf74..3d73fe484f42 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.c
+++ b/drivers/gpu/drm/radeon/radeon_asic.c
@@ -171,6 +171,9 @@ static struct radeon_asic r100_asic = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r100_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic r200_asic = {
@@ -215,6 +218,9 @@ static struct radeon_asic r200_asic = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r100_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic r300_asic = {
@@ -260,6 +266,9 @@ static struct radeon_asic r300_asic = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r100_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic r300_asic_pcie = {
@@ -304,6 +313,9 @@ static struct radeon_asic r300_asic_pcie = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r100_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic r420_asic = {
@@ -349,6 +361,9 @@ static struct radeon_asic r420_asic = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r420_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic rs400_asic = {
@@ -394,6 +409,9 @@ static struct radeon_asic rs400_asic = {
 	.pm_finish = &r100_pm_finish,
 	.pm_init_profile = &r100_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &r100_pre_page_flip,
+	.page_flip = &r100_page_flip,
+	.post_page_flip = &r100_post_page_flip,
 };
 
 static struct radeon_asic rs600_asic = {
@@ -439,6 +457,9 @@ static struct radeon_asic rs600_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r420_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic rs690_asic = {
@@ -484,6 +505,9 @@ static struct radeon_asic rs690_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r420_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic rv515_asic = {
@@ -529,6 +553,9 @@ static struct radeon_asic rv515_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r420_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic r520_asic = {
@@ -574,6 +601,9 @@ static struct radeon_asic r520_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r420_pm_init_profile,
 	.pm_get_dynpm_state = &r100_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic r600_asic = {
@@ -618,6 +648,9 @@ static struct radeon_asic r600_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r600_pm_init_profile,
 	.pm_get_dynpm_state = &r600_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic rs780_asic = {
@@ -662,6 +695,9 @@ static struct radeon_asic rs780_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &rs780_pm_init_profile,
 	.pm_get_dynpm_state = &r600_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rs600_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic rv770_asic = {
@@ -706,6 +742,9 @@ static struct radeon_asic rv770_asic = {
 	.pm_finish = &rs600_pm_finish,
 	.pm_init_profile = &r600_pm_init_profile,
 	.pm_get_dynpm_state = &r600_pm_get_dynpm_state,
+	.pre_page_flip = &rs600_pre_page_flip,
+	.page_flip = &rv770_page_flip,
+	.post_page_flip = &rs600_post_page_flip,
 };
 
 static struct radeon_asic evergreen_asic = {
@@ -749,6 +788,52 @@ static struct radeon_asic evergreen_asic = {
 	.pm_finish = &evergreen_pm_finish,
 	.pm_init_profile = &r600_pm_init_profile,
 	.pm_get_dynpm_state = &r600_pm_get_dynpm_state,
+	.pre_page_flip = &evergreen_pre_page_flip,
+	.page_flip = &evergreen_page_flip,
+	.post_page_flip = &evergreen_post_page_flip,
+};
+
+static struct radeon_asic sumo_asic = {
+	.init = &evergreen_init,
+	.fini = &evergreen_fini,
+	.suspend = &evergreen_suspend,
+	.resume = &evergreen_resume,
+	.cp_commit = &r600_cp_commit,
+	.gpu_is_lockup = &evergreen_gpu_is_lockup,
+	.asic_reset = &evergreen_asic_reset,
+	.vga_set_state = &r600_vga_set_state,
+	.gart_tlb_flush = &evergreen_pcie_gart_tlb_flush,
+	.gart_set_page = &rs600_gart_set_page,
+	.ring_test = &r600_ring_test,
+	.ring_ib_execute = &r600_ring_ib_execute,
+	.irq_set = &evergreen_irq_set,
+	.irq_process = &evergreen_irq_process,
+	.get_vblank_counter = &evergreen_get_vblank_counter,
+	.fence_ring_emit = &r600_fence_ring_emit,
+	.cs_parse = &evergreen_cs_parse,
+	.copy_blit = &evergreen_copy_blit,
+	.copy_dma = &evergreen_copy_blit,
+	.copy = &evergreen_copy_blit,
+	.get_engine_clock = &radeon_atom_get_engine_clock,
+	.set_engine_clock = &radeon_atom_set_engine_clock,
+	.get_memory_clock = NULL,
+	.set_memory_clock = NULL,
+	.get_pcie_lanes = NULL,
+	.set_pcie_lanes = NULL,
+	.set_clock_gating = NULL,
+	.set_surface_reg = r600_set_surface_reg,
+	.clear_surface_reg = r600_clear_surface_reg,
+	.bandwidth_update = &evergreen_bandwidth_update,
+	.hpd_init = &evergreen_hpd_init,
+	.hpd_fini = &evergreen_hpd_fini,
+	.hpd_sense = &evergreen_hpd_sense,
+	.hpd_set_polarity = &evergreen_hpd_set_polarity,
+	.gui_idle = &r600_gui_idle,
+	.pm_misc = &evergreen_pm_misc,
+	.pm_prepare = &evergreen_pm_prepare,
+	.pm_finish = &evergreen_pm_finish,
+	.pm_init_profile = &rs780_pm_init_profile,
+	.pm_get_dynpm_state = &r600_pm_get_dynpm_state,
 };
 
 int radeon_asic_init(struct radeon_device *rdev)
@@ -835,6 +920,9 @@ int radeon_asic_init(struct radeon_device *rdev)
 	case CHIP_HEMLOCK:
 		rdev->asic = &evergreen_asic;
 		break;
+	case CHIP_PALM:
+		rdev->asic = &sumo_asic;
+		break;
 	default:
 		/* FIXME: not supported yet */
 		return -EINVAL;
@@ -849,7 +937,9 @@ int radeon_asic_init(struct radeon_device *rdev)
 	if (rdev->flags & RADEON_SINGLE_CRTC)
 		rdev->num_crtc = 1;
 	else {
-		if (ASIC_IS_DCE4(rdev))
+		if (ASIC_IS_DCE41(rdev))
+			rdev->num_crtc = 2;
+		else if (ASIC_IS_DCE4(rdev))
 			rdev->num_crtc = 6;
 		else
 			rdev->num_crtc = 2;
diff --git a/drivers/gpu/drm/radeon/radeon_asic.h b/drivers/gpu/drm/radeon/radeon_asic.h
index 740988244143..4970eda1bd41 100644
--- a/drivers/gpu/drm/radeon/radeon_asic.h
+++ b/drivers/gpu/drm/radeon/radeon_asic.h
@@ -130,6 +130,9 @@ extern void r100_pm_prepare(struct radeon_device *rdev);
 extern void r100_pm_finish(struct radeon_device *rdev);
 extern void r100_pm_init_profile(struct radeon_device *rdev);
 extern void r100_pm_get_dynpm_state(struct radeon_device *rdev);
+extern void r100_pre_page_flip(struct radeon_device *rdev, int crtc);
+extern u32 r100_page_flip(struct radeon_device *rdev, int crtc, u64 crtc_base);
+extern void r100_post_page_flip(struct radeon_device *rdev, int crtc);
 
 /*
  * r200,rv250,rs300,rv280
@@ -205,6 +208,9 @@ void rs600_hpd_set_polarity(struct radeon_device *rdev,
 extern void rs600_pm_misc(struct radeon_device *rdev);
 extern void rs600_pm_prepare(struct radeon_device *rdev);
 extern void rs600_pm_finish(struct radeon_device *rdev);
+extern void rs600_pre_page_flip(struct radeon_device *rdev, int crtc);
+extern u32 rs600_page_flip(struct radeon_device *rdev, int crtc, u64 crtc_base);
+extern void rs600_post_page_flip(struct radeon_device *rdev, int crtc);
 
 /*
  * rs690,rs740
@@ -287,6 +293,7 @@ void rv770_fini(struct radeon_device *rdev);
 int rv770_suspend(struct radeon_device *rdev);
 int rv770_resume(struct radeon_device *rdev);
 extern void rv770_pm_misc(struct radeon_device *rdev);
+extern u32 rv770_page_flip(struct radeon_device *rdev, int crtc, u64 crtc_base);
 
 /*
  * evergreen
@@ -314,5 +321,8 @@ extern int evergreen_cs_parse(struct radeon_cs_parser *p);
 extern void evergreen_pm_misc(struct radeon_device *rdev);
 extern void evergreen_pm_prepare(struct radeon_device *rdev);
 extern void evergreen_pm_finish(struct radeon_device *rdev);
+extern void evergreen_pre_page_flip(struct radeon_device *rdev, int crtc);
+extern u32 evergreen_page_flip(struct radeon_device *rdev, int crtc, u64 crtc_base);
+extern void evergreen_post_page_flip(struct radeon_device *rdev, int crtc);
 
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon_atombios.c b/drivers/gpu/drm/radeon/radeon_atombios.c
index bc5a2c3382d9..8e82f672263f 100644
--- a/drivers/gpu/drm/radeon/radeon_atombios.c
+++ b/drivers/gpu/drm/radeon/radeon_atombios.c
@@ -1337,6 +1337,43 @@ bool radeon_atombios_get_ppll_ss_info(struct radeon_device *rdev,
 	return false;
 }
 
+static void radeon_atombios_get_igp_ss_overrides(struct radeon_device *rdev,
+						 struct radeon_atom_ss *ss,
+						 int id)
+{
+	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	int index = GetIndexIntoMasterTable(DATA, IntegratedSystemInfo);
+	u16 data_offset, size;
+	struct _ATOM_INTEGRATED_SYSTEM_INFO_V6 *igp_info;
+	u8 frev, crev;
+	u16 percentage = 0, rate = 0;
+
+	/* get any igp specific overrides */
+	if (atom_parse_data_header(mode_info->atom_context, index, &size,
+				   &frev, &crev, &data_offset)) {
+		igp_info = (struct _ATOM_INTEGRATED_SYSTEM_INFO_V6 *)
+			(mode_info->atom_context->bios + data_offset);
+		switch (id) {
+		case ASIC_INTERNAL_SS_ON_TMDS:
+			percentage = le16_to_cpu(igp_info->usDVISSPercentage);
+			rate = le16_to_cpu(igp_info->usDVISSpreadRateIn10Hz);
+			break;
+		case ASIC_INTERNAL_SS_ON_HDMI:
+			percentage = le16_to_cpu(igp_info->usHDMISSPercentage);
+			rate = le16_to_cpu(igp_info->usHDMISSpreadRateIn10Hz);
+			break;
+		case ASIC_INTERNAL_SS_ON_LVDS:
+			percentage = le16_to_cpu(igp_info->usLvdsSSPercentage);
+			rate = le16_to_cpu(igp_info->usLvdsSSpreadRateIn10Hz);
+			break;
+		}
+		if (percentage)
+			ss->percentage = percentage;
+		if (rate)
+			ss->rate = rate;
+	}
+}
+
 union asic_ss_info {
 	struct _ATOM_ASIC_INTERNAL_SS_INFO info;
 	struct _ATOM_ASIC_INTERNAL_SS_INFO_V2 info_2;
@@ -1401,6 +1438,8 @@ bool radeon_atombios_get_asic_ss_info(struct radeon_device *rdev,
 						le16_to_cpu(ss_info->info_3.asSpreadSpectrum[i].usSpreadSpectrumPercentage);
 					ss->type = ss_info->info_3.asSpreadSpectrum[i].ucSpreadSpectrumMode;
 					ss->rate = le16_to_cpu(ss_info->info_3.asSpreadSpectrum[i].usSpreadRateIn10Hz);
+					if (rdev->flags & RADEON_IS_IGP)
+						radeon_atombios_get_igp_ss_overrides(rdev, ss, id);
 					return true;
 				}
 			}
@@ -1740,495 +1779,600 @@ static const char *pp_lib_thermal_controller_names[] = {
 	"RV6xx",
 	"RV770",
 	"adt7473",
+	"NONE",
 	"External GPIO",
 	"Evergreen",
-	"adt7473 with internal",
-
+	"emc2103",
+	"Sumo",
 };
 
 union power_info {
 	struct _ATOM_POWERPLAY_INFO info;
 	struct _ATOM_POWERPLAY_INFO_V2 info_2;
 	struct _ATOM_POWERPLAY_INFO_V3 info_3;
-	struct _ATOM_PPLIB_POWERPLAYTABLE info_4;
+	struct _ATOM_PPLIB_POWERPLAYTABLE pplib;
+	struct _ATOM_PPLIB_POWERPLAYTABLE2 pplib2;
+	struct _ATOM_PPLIB_POWERPLAYTABLE3 pplib3;
 };
 
-void radeon_atombios_get_power_modes(struct radeon_device *rdev)
+union pplib_clock_info {
+	struct _ATOM_PPLIB_R600_CLOCK_INFO r600;
+	struct _ATOM_PPLIB_RS780_CLOCK_INFO rs780;
+	struct _ATOM_PPLIB_EVERGREEN_CLOCK_INFO evergreen;
+	struct _ATOM_PPLIB_SUMO_CLOCK_INFO sumo;
+};
+
+union pplib_power_state {
+	struct _ATOM_PPLIB_STATE v1;
+	struct _ATOM_PPLIB_STATE_V2 v2;
+};
+
+static void radeon_atombios_parse_misc_flags_1_3(struct radeon_device *rdev,
+						 int state_index,
+						 u32 misc, u32 misc2)
+{
+	rdev->pm.power_state[state_index].misc = misc;
+	rdev->pm.power_state[state_index].misc2 = misc2;
+	/* order matters! */
+	if (misc & ATOM_PM_MISCINFO_POWER_SAVING_MODE)
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_POWERSAVE;
+	if (misc & ATOM_PM_MISCINFO_DEFAULT_DC_STATE_ENTRY_TRUE)
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BATTERY;
+	if (misc & ATOM_PM_MISCINFO_DEFAULT_LOW_DC_STATE_ENTRY_TRUE)
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BATTERY;
+	if (misc & ATOM_PM_MISCINFO_LOAD_BALANCE_EN)
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BALANCED;
+	if (misc & ATOM_PM_MISCINFO_3D_ACCELERATION_EN) {
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_PERFORMANCE;
+		rdev->pm.power_state[state_index].flags &=
+			~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+	}
+	if (misc2 & ATOM_PM_MISCINFO2_SYSTEM_AC_LITE_MODE)
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BALANCED;
+	if (misc & ATOM_PM_MISCINFO_DRIVER_DEFAULT_MODE) {
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_DEFAULT;
+		rdev->pm.default_power_state_index = state_index;
+		rdev->pm.power_state[state_index].default_clock_mode =
+			&rdev->pm.power_state[state_index].clock_info[0];
+	} else if (state_index == 0) {
+		rdev->pm.power_state[state_index].clock_info[0].flags |=
+			RADEON_PM_MODE_NO_DISPLAY;
+	}
+}
+
+static int radeon_atombios_parse_power_table_1_3(struct radeon_device *rdev)
 {
 	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	u32 misc, misc2 = 0;
+	int num_modes = 0, i;
+	int state_index = 0;
+	struct radeon_i2c_bus_rec i2c_bus;
+	union power_info *power_info;
 	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
-	u16 data_offset;
+        u16 data_offset;
 	u8 frev, crev;
-	u32 misc, misc2 = 0, sclk, mclk;
-	union power_info *power_info;
-	struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info;
-	struct _ATOM_PPLIB_STATE *power_state;
-	int num_modes = 0, i, j;
-	int state_index = 0, mode_index = 0;
-	struct radeon_i2c_bus_rec i2c_bus;
-
-	rdev->pm.default_power_state_index = -1;
 
-	if (atom_parse_data_header(mode_info->atom_context, index, NULL,
-				   &frev, &crev, &data_offset)) {
-		power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
-		if (frev < 4) {
-			/* add the i2c bus for thermal/fan chip */
-			if (power_info->info.ucOverdriveThermalController > 0) {
-				DRM_INFO("Possible %s thermal controller at 0x%02x\n",
-					 thermal_controller_names[power_info->info.ucOverdriveThermalController],
-					 power_info->info.ucOverdriveControllerAddress >> 1);
-				i2c_bus = radeon_lookup_i2c_gpio(rdev, power_info->info.ucOverdriveI2cLine);
-				rdev->pm.i2c_bus = radeon_i2c_lookup(rdev, &i2c_bus);
-				if (rdev->pm.i2c_bus) {
-					struct i2c_board_info info = { };
-					const char *name = thermal_controller_names[power_info->info.
-										    ucOverdriveThermalController];
-					info.addr = power_info->info.ucOverdriveControllerAddress >> 1;
-					strlcpy(info.type, name, sizeof(info.type));
-					i2c_new_device(&rdev->pm.i2c_bus->adapter, &info);
-				}
+	if (!atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset))
+		return state_index;
+	power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
+
+	/* add the i2c bus for thermal/fan chip */
+	if (power_info->info.ucOverdriveThermalController > 0) {
+		DRM_INFO("Possible %s thermal controller at 0x%02x\n",
+			 thermal_controller_names[power_info->info.ucOverdriveThermalController],
+			 power_info->info.ucOverdriveControllerAddress >> 1);
+		i2c_bus = radeon_lookup_i2c_gpio(rdev, power_info->info.ucOverdriveI2cLine);
+		rdev->pm.i2c_bus = radeon_i2c_lookup(rdev, &i2c_bus);
+		if (rdev->pm.i2c_bus) {
+			struct i2c_board_info info = { };
+			const char *name = thermal_controller_names[power_info->info.
+								    ucOverdriveThermalController];
+			info.addr = power_info->info.ucOverdriveControllerAddress >> 1;
+			strlcpy(info.type, name, sizeof(info.type));
+			i2c_new_device(&rdev->pm.i2c_bus->adapter, &info);
+		}
+	}
+	num_modes = power_info->info.ucNumOfPowerModeEntries;
+	if (num_modes > ATOM_MAX_NUMBEROF_POWER_BLOCK)
+		num_modes = ATOM_MAX_NUMBEROF_POWER_BLOCK;
+	/* last mode is usually default, array is low to high */
+	for (i = 0; i < num_modes; i++) {
+		rdev->pm.power_state[state_index].clock_info[0].voltage.type = VOLTAGE_NONE;
+		switch (frev) {
+		case 1:
+			rdev->pm.power_state[state_index].num_clock_modes = 1;
+			rdev->pm.power_state[state_index].clock_info[0].mclk =
+				le16_to_cpu(power_info->info.asPowerPlayInfo[i].usMemoryClock);
+			rdev->pm.power_state[state_index].clock_info[0].sclk =
+				le16_to_cpu(power_info->info.asPowerPlayInfo[i].usEngineClock);
+			/* skip invalid modes */
+			if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
+			    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
+				continue;
+			rdev->pm.power_state[state_index].pcie_lanes =
+				power_info->info.asPowerPlayInfo[i].ucNumPciELanes;
+			misc = le32_to_cpu(power_info->info.asPowerPlayInfo[i].ulMiscInfo);
+			if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
+			    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_GPIO;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
+					radeon_lookup_gpio(rdev,
+							   power_info->info.asPowerPlayInfo[i].ucVoltageDropIndex);
+				if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						true;
+				else
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						false;
+			} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_VDDC;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
+					power_info->info.asPowerPlayInfo[i].ucVoltageDropIndex;
 			}
-			num_modes = power_info->info.ucNumOfPowerModeEntries;
-			if (num_modes > ATOM_MAX_NUMBEROF_POWER_BLOCK)
-				num_modes = ATOM_MAX_NUMBEROF_POWER_BLOCK;
-			/* last mode is usually default, array is low to high */
-			for (i = 0; i < num_modes; i++) {
-				rdev->pm.power_state[state_index].clock_info[0].voltage.type = VOLTAGE_NONE;
-				switch (frev) {
-				case 1:
-					rdev->pm.power_state[state_index].num_clock_modes = 1;
-					rdev->pm.power_state[state_index].clock_info[0].mclk =
-						le16_to_cpu(power_info->info.asPowerPlayInfo[i].usMemoryClock);
-					rdev->pm.power_state[state_index].clock_info[0].sclk =
-						le16_to_cpu(power_info->info.asPowerPlayInfo[i].usEngineClock);
-					/* skip invalid modes */
-					if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
-					    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
-						continue;
-					rdev->pm.power_state[state_index].pcie_lanes =
-						power_info->info.asPowerPlayInfo[i].ucNumPciELanes;
-					misc = le32_to_cpu(power_info->info.asPowerPlayInfo[i].ulMiscInfo);
-					if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
-					    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_GPIO;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
-							radeon_lookup_gpio(rdev,
-							power_info->info.asPowerPlayInfo[i].ucVoltageDropIndex);
-						if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								true;
-						else
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								false;
-					} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_VDDC;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
-							power_info->info.asPowerPlayInfo[i].ucVoltageDropIndex;
-					}
-					rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					rdev->pm.power_state[state_index].misc = misc;
-					/* order matters! */
-					if (misc & ATOM_PM_MISCINFO_POWER_SAVING_MODE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_POWERSAVE;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_LOW_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_LOAD_BALANCE_EN)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-					if (misc & ATOM_PM_MISCINFO_3D_ACCELERATION_EN) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_PERFORMANCE;
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					}
-					if (misc & ATOM_PM_MISCINFO_DRIVER_DEFAULT_MODE) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_DEFAULT;
-						rdev->pm.default_power_state_index = state_index;
-						rdev->pm.power_state[state_index].default_clock_mode =
-							&rdev->pm.power_state[state_index].clock_info[0];
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					} else if (state_index == 0) {
-						rdev->pm.power_state[state_index].clock_info[0].flags |=
-							RADEON_PM_MODE_NO_DISPLAY;
-					}
-					state_index++;
-					break;
-				case 2:
-					rdev->pm.power_state[state_index].num_clock_modes = 1;
-					rdev->pm.power_state[state_index].clock_info[0].mclk =
-						le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMemoryClock);
-					rdev->pm.power_state[state_index].clock_info[0].sclk =
-						le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulEngineClock);
-					/* skip invalid modes */
-					if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
-					    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
-						continue;
-					rdev->pm.power_state[state_index].pcie_lanes =
-						power_info->info_2.asPowerPlayInfo[i].ucNumPciELanes;
-					misc = le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMiscInfo);
-					misc2 = le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMiscInfo2);
-					if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
-					    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_GPIO;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
-							radeon_lookup_gpio(rdev,
-							power_info->info_2.asPowerPlayInfo[i].ucVoltageDropIndex);
-						if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								true;
-						else
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								false;
-					} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_VDDC;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
-							power_info->info_2.asPowerPlayInfo[i].ucVoltageDropIndex;
-					}
-					rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					rdev->pm.power_state[state_index].misc = misc;
-					rdev->pm.power_state[state_index].misc2 = misc2;
-					/* order matters! */
-					if (misc & ATOM_PM_MISCINFO_POWER_SAVING_MODE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_POWERSAVE;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_LOW_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_LOAD_BALANCE_EN)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-					if (misc & ATOM_PM_MISCINFO_3D_ACCELERATION_EN) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_PERFORMANCE;
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					}
-					if (misc2 & ATOM_PM_MISCINFO2_SYSTEM_AC_LITE_MODE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-					if (misc2 & ATOM_PM_MISCINFO2_MULTI_DISPLAY_SUPPORT)
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					if (misc & ATOM_PM_MISCINFO_DRIVER_DEFAULT_MODE) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_DEFAULT;
-						rdev->pm.default_power_state_index = state_index;
-						rdev->pm.power_state[state_index].default_clock_mode =
-							&rdev->pm.power_state[state_index].clock_info[0];
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					} else if (state_index == 0) {
-						rdev->pm.power_state[state_index].clock_info[0].flags |=
-							RADEON_PM_MODE_NO_DISPLAY;
-					}
-					state_index++;
-					break;
-				case 3:
-					rdev->pm.power_state[state_index].num_clock_modes = 1;
-					rdev->pm.power_state[state_index].clock_info[0].mclk =
-						le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMemoryClock);
-					rdev->pm.power_state[state_index].clock_info[0].sclk =
-						le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulEngineClock);
-					/* skip invalid modes */
-					if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
-					    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
-						continue;
-					rdev->pm.power_state[state_index].pcie_lanes =
-						power_info->info_3.asPowerPlayInfo[i].ucNumPciELanes;
-					misc = le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMiscInfo);
-					misc2 = le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMiscInfo2);
-					if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
-					    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_GPIO;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
-							radeon_lookup_gpio(rdev,
-							power_info->info_3.asPowerPlayInfo[i].ucVoltageDropIndex);
-						if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								true;
-						else
-							rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
-								false;
-					} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
-						rdev->pm.power_state[state_index].clock_info[0].voltage.type =
-							VOLTAGE_VDDC;
-						rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
-							power_info->info_3.asPowerPlayInfo[i].ucVoltageDropIndex;
-						if (misc2 & ATOM_PM_MISCINFO2_VDDCI_DYNAMIC_VOLTAGE_EN) {
-							rdev->pm.power_state[state_index].clock_info[0].voltage.vddci_enabled =
-								true;
-							rdev->pm.power_state[state_index].clock_info[0].voltage.vddci_id =
-							power_info->info_3.asPowerPlayInfo[i].ucVDDCI_VoltageDropIndex;
-						}
-					}
-					rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					rdev->pm.power_state[state_index].misc = misc;
-					rdev->pm.power_state[state_index].misc2 = misc2;
-					/* order matters! */
-					if (misc & ATOM_PM_MISCINFO_POWER_SAVING_MODE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_POWERSAVE;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_DEFAULT_LOW_DC_STATE_ENTRY_TRUE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-					if (misc & ATOM_PM_MISCINFO_LOAD_BALANCE_EN)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-					if (misc & ATOM_PM_MISCINFO_3D_ACCELERATION_EN) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_PERFORMANCE;
-						rdev->pm.power_state[state_index].flags &=
-							~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					}
-					if (misc2 & ATOM_PM_MISCINFO2_SYSTEM_AC_LITE_MODE)
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-					if (misc & ATOM_PM_MISCINFO_DRIVER_DEFAULT_MODE) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_DEFAULT;
-						rdev->pm.default_power_state_index = state_index;
-						rdev->pm.power_state[state_index].default_clock_mode =
-							&rdev->pm.power_state[state_index].clock_info[0];
-					} else if (state_index == 0) {
-						rdev->pm.power_state[state_index].clock_info[0].flags |=
-							RADEON_PM_MODE_NO_DISPLAY;
-					}
-					state_index++;
-					break;
-				}
+			rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+			radeon_atombios_parse_misc_flags_1_3(rdev, state_index, misc, 0);
+			state_index++;
+			break;
+		case 2:
+			rdev->pm.power_state[state_index].num_clock_modes = 1;
+			rdev->pm.power_state[state_index].clock_info[0].mclk =
+				le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMemoryClock);
+			rdev->pm.power_state[state_index].clock_info[0].sclk =
+				le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulEngineClock);
+			/* skip invalid modes */
+			if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
+			    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
+				continue;
+			rdev->pm.power_state[state_index].pcie_lanes =
+				power_info->info_2.asPowerPlayInfo[i].ucNumPciELanes;
+			misc = le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMiscInfo);
+			misc2 = le32_to_cpu(power_info->info_2.asPowerPlayInfo[i].ulMiscInfo2);
+			if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
+			    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_GPIO;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
+					radeon_lookup_gpio(rdev,
+							   power_info->info_2.asPowerPlayInfo[i].ucVoltageDropIndex);
+				if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						true;
+				else
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						false;
+			} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_VDDC;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
+					power_info->info_2.asPowerPlayInfo[i].ucVoltageDropIndex;
 			}
-			/* last mode is usually default */
-			if (rdev->pm.default_power_state_index == -1) {
-				rdev->pm.power_state[state_index - 1].type =
-					POWER_STATE_TYPE_DEFAULT;
-				rdev->pm.default_power_state_index = state_index - 1;
-				rdev->pm.power_state[state_index - 1].default_clock_mode =
-					&rdev->pm.power_state[state_index - 1].clock_info[0];
-				rdev->pm.power_state[state_index].flags &=
-					~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-				rdev->pm.power_state[state_index].misc = 0;
-				rdev->pm.power_state[state_index].misc2 = 0;
+			rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+			radeon_atombios_parse_misc_flags_1_3(rdev, state_index, misc, misc2);
+			state_index++;
+			break;
+		case 3:
+			rdev->pm.power_state[state_index].num_clock_modes = 1;
+			rdev->pm.power_state[state_index].clock_info[0].mclk =
+				le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMemoryClock);
+			rdev->pm.power_state[state_index].clock_info[0].sclk =
+				le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulEngineClock);
+			/* skip invalid modes */
+			if ((rdev->pm.power_state[state_index].clock_info[0].mclk == 0) ||
+			    (rdev->pm.power_state[state_index].clock_info[0].sclk == 0))
+				continue;
+			rdev->pm.power_state[state_index].pcie_lanes =
+				power_info->info_3.asPowerPlayInfo[i].ucNumPciELanes;
+			misc = le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMiscInfo);
+			misc2 = le32_to_cpu(power_info->info_3.asPowerPlayInfo[i].ulMiscInfo2);
+			if ((misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_SUPPORT) ||
+			    (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_GPIO;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.gpio =
+					radeon_lookup_gpio(rdev,
+							   power_info->info_3.asPowerPlayInfo[i].ucVoltageDropIndex);
+				if (misc & ATOM_PM_MISCINFO_VOLTAGE_DROP_ACTIVE_HIGH)
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						true;
+				else
+					rdev->pm.power_state[state_index].clock_info[0].voltage.active_high =
+						false;
+			} else if (misc & ATOM_PM_MISCINFO_PROGRAM_VOLTAGE) {
+				rdev->pm.power_state[state_index].clock_info[0].voltage.type =
+					VOLTAGE_VDDC;
+				rdev->pm.power_state[state_index].clock_info[0].voltage.vddc_id =
+					power_info->info_3.asPowerPlayInfo[i].ucVoltageDropIndex;
+				if (misc2 & ATOM_PM_MISCINFO2_VDDCI_DYNAMIC_VOLTAGE_EN) {
+					rdev->pm.power_state[state_index].clock_info[0].voltage.vddci_enabled =
+						true;
+					rdev->pm.power_state[state_index].clock_info[0].voltage.vddci_id =
+						power_info->info_3.asPowerPlayInfo[i].ucVDDCI_VoltageDropIndex;
+				}
 			}
+			rdev->pm.power_state[state_index].flags = RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+			radeon_atombios_parse_misc_flags_1_3(rdev, state_index, misc, misc2);
+			state_index++;
+			break;
+		}
+	}
+	/* last mode is usually default */
+	if (rdev->pm.default_power_state_index == -1) {
+		rdev->pm.power_state[state_index - 1].type =
+			POWER_STATE_TYPE_DEFAULT;
+		rdev->pm.default_power_state_index = state_index - 1;
+		rdev->pm.power_state[state_index - 1].default_clock_mode =
+			&rdev->pm.power_state[state_index - 1].clock_info[0];
+		rdev->pm.power_state[state_index].flags &=
+			~RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+		rdev->pm.power_state[state_index].misc = 0;
+		rdev->pm.power_state[state_index].misc2 = 0;
+	}
+	return state_index;
+}
+
+static void radeon_atombios_add_pplib_thermal_controller(struct radeon_device *rdev,
+							 ATOM_PPLIB_THERMALCONTROLLER *controller)
+{
+	struct radeon_i2c_bus_rec i2c_bus;
+
+	/* add the i2c bus for thermal/fan chip */
+	if (controller->ucType > 0) {
+		if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV6xx) {
+			DRM_INFO("Internal thermal controller %s fan control\n",
+				 (controller->ucFanParameters &
+				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
+			rdev->pm.int_thermal_type = THERMAL_TYPE_RV6XX;
+		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV770) {
+			DRM_INFO("Internal thermal controller %s fan control\n",
+				 (controller->ucFanParameters &
+				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
+			rdev->pm.int_thermal_type = THERMAL_TYPE_RV770;
+		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_EVERGREEN) {
+			DRM_INFO("Internal thermal controller %s fan control\n",
+				 (controller->ucFanParameters &
+				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
+			rdev->pm.int_thermal_type = THERMAL_TYPE_EVERGREEN;
+		} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_SUMO) {
+			DRM_INFO("Internal thermal controller %s fan control\n",
+				 (controller->ucFanParameters &
+				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
+			rdev->pm.int_thermal_type = THERMAL_TYPE_SUMO;
+		} else if ((controller->ucType ==
+			    ATOM_PP_THERMALCONTROLLER_EXTERNAL_GPIO) ||
+			   (controller->ucType ==
+			    ATOM_PP_THERMALCONTROLLER_ADT7473_WITH_INTERNAL) ||
+			   (controller->ucType ==
+			    ATOM_PP_THERMALCONTROLLER_EMC2103_WITH_INTERNAL)) {
+			DRM_INFO("Special thermal controller config\n");
 		} else {
-			int fw_index = GetIndexIntoMasterTable(DATA, FirmwareInfo);
-			uint8_t fw_frev, fw_crev;
-			uint16_t fw_data_offset, vddc = 0;
-			union firmware_info *firmware_info;
-			ATOM_PPLIB_THERMALCONTROLLER *controller = &power_info->info_4.sThermalController;
-
-			if (atom_parse_data_header(mode_info->atom_context, fw_index, NULL,
-						   &fw_frev, &fw_crev, &fw_data_offset)) {
-				firmware_info =
-					(union firmware_info *)(mode_info->atom_context->bios +
-								fw_data_offset);
-				vddc = firmware_info->info_14.usBootUpVDDCVoltage;
+			DRM_INFO("Possible %s thermal controller at 0x%02x %s fan control\n",
+				 pp_lib_thermal_controller_names[controller->ucType],
+				 controller->ucI2cAddress >> 1,
+				 (controller->ucFanParameters &
+				  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
+			i2c_bus = radeon_lookup_i2c_gpio(rdev, controller->ucI2cLine);
+			rdev->pm.i2c_bus = radeon_i2c_lookup(rdev, &i2c_bus);
+			if (rdev->pm.i2c_bus) {
+				struct i2c_board_info info = { };
+				const char *name = pp_lib_thermal_controller_names[controller->ucType];
+				info.addr = controller->ucI2cAddress >> 1;
+				strlcpy(info.type, name, sizeof(info.type));
+				i2c_new_device(&rdev->pm.i2c_bus->adapter, &info);
 			}
+		}
+	}
+}
 
-			/* add the i2c bus for thermal/fan chip */
-			if (controller->ucType > 0) {
-				if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV6xx) {
-					DRM_INFO("Internal thermal controller %s fan control\n",
-						 (controller->ucFanParameters &
-						  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-					rdev->pm.int_thermal_type = THERMAL_TYPE_RV6XX;
-				} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_RV770) {
-					DRM_INFO("Internal thermal controller %s fan control\n",
-						 (controller->ucFanParameters &
-						  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-					rdev->pm.int_thermal_type = THERMAL_TYPE_RV770;
-				} else if (controller->ucType == ATOM_PP_THERMALCONTROLLER_EVERGREEN) {
-					DRM_INFO("Internal thermal controller %s fan control\n",
-						 (controller->ucFanParameters &
-						  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-					rdev->pm.int_thermal_type = THERMAL_TYPE_EVERGREEN;
-				} else if ((controller->ucType ==
-					    ATOM_PP_THERMALCONTROLLER_EXTERNAL_GPIO) ||
-					   (controller->ucType ==
-					    ATOM_PP_THERMALCONTROLLER_ADT7473_WITH_INTERNAL)) {
-					DRM_INFO("Special thermal controller config\n");
-				} else {
-					DRM_INFO("Possible %s thermal controller at 0x%02x %s fan control\n",
-						 pp_lib_thermal_controller_names[controller->ucType],
-						 controller->ucI2cAddress >> 1,
-						 (controller->ucFanParameters &
-						  ATOM_PP_FANPARAMETERS_NOFAN) ? "without" : "with");
-					i2c_bus = radeon_lookup_i2c_gpio(rdev, controller->ucI2cLine);
-					rdev->pm.i2c_bus = radeon_i2c_lookup(rdev, &i2c_bus);
-					if (rdev->pm.i2c_bus) {
-						struct i2c_board_info info = { };
-						const char *name = pp_lib_thermal_controller_names[controller->ucType];
-						info.addr = controller->ucI2cAddress >> 1;
-						strlcpy(info.type, name, sizeof(info.type));
-						i2c_new_device(&rdev->pm.i2c_bus->adapter, &info);
-					}
+static u16 radeon_atombios_get_default_vddc(struct radeon_device *rdev)
+{
+	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	int index = GetIndexIntoMasterTable(DATA, FirmwareInfo);
+	u8 frev, crev;
+	u16 data_offset;
+	union firmware_info *firmware_info;
+	u16 vddc = 0;
 
-				}
-			}
-			/* first mode is usually default, followed by low to high */
-			for (i = 0; i < power_info->info_4.ucNumStates; i++) {
-				mode_index = 0;
-				power_state = (struct _ATOM_PPLIB_STATE *)
-					(mode_info->atom_context->bios +
-					 data_offset +
-					 le16_to_cpu(power_info->info_4.usStateArrayOffset) +
-					 i * power_info->info_4.ucStateEntrySize);
-				non_clock_info = (struct _ATOM_PPLIB_NONCLOCK_INFO *)
-					(mode_info->atom_context->bios +
-					 data_offset +
-					 le16_to_cpu(power_info->info_4.usNonClockInfoArrayOffset) +
-					 (power_state->ucNonClockStateIndex *
-					  power_info->info_4.ucNonClockSize));
-				for (j = 0; j < (power_info->info_4.ucStateEntrySize - 1); j++) {
-					if (rdev->flags & RADEON_IS_IGP) {
-						struct _ATOM_PPLIB_RS780_CLOCK_INFO *clock_info =
-							(struct _ATOM_PPLIB_RS780_CLOCK_INFO *)
-							(mode_info->atom_context->bios +
-							 data_offset +
-							 le16_to_cpu(power_info->info_4.usClockInfoArrayOffset) +
-							 (power_state->ucClockStateIndices[j] *
-							  power_info->info_4.ucClockInfoSize));
-						sclk = le16_to_cpu(clock_info->usLowEngineClockLow);
-						sclk |= clock_info->ucLowEngineClockHigh << 16;
-						rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
-						/* skip invalid modes */
-						if (rdev->pm.power_state[state_index].clock_info[mode_index].sclk == 0)
-							continue;
-						/* voltage works differently on IGPs */
-						mode_index++;
-					} else if (ASIC_IS_DCE4(rdev)) {
-						struct _ATOM_PPLIB_EVERGREEN_CLOCK_INFO *clock_info =
-							(struct _ATOM_PPLIB_EVERGREEN_CLOCK_INFO *)
-							(mode_info->atom_context->bios +
-							 data_offset +
-							 le16_to_cpu(power_info->info_4.usClockInfoArrayOffset) +
-							 (power_state->ucClockStateIndices[j] *
-							  power_info->info_4.ucClockInfoSize));
-						sclk = le16_to_cpu(clock_info->usEngineClockLow);
-						sclk |= clock_info->ucEngineClockHigh << 16;
-						mclk = le16_to_cpu(clock_info->usMemoryClockLow);
-						mclk |= clock_info->ucMemoryClockHigh << 16;
-						rdev->pm.power_state[state_index].clock_info[mode_index].mclk = mclk;
-						rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
-						/* skip invalid modes */
-						if ((rdev->pm.power_state[state_index].clock_info[mode_index].mclk == 0) ||
-						    (rdev->pm.power_state[state_index].clock_info[mode_index].sclk == 0))
-							continue;
-						rdev->pm.power_state[state_index].clock_info[mode_index].voltage.type =
-							VOLTAGE_SW;
-						rdev->pm.power_state[state_index].clock_info[mode_index].voltage.voltage =
-							clock_info->usVDDC;
-						/* XXX usVDDCI */
-						mode_index++;
-					} else {
-						struct _ATOM_PPLIB_R600_CLOCK_INFO *clock_info =
-							(struct _ATOM_PPLIB_R600_CLOCK_INFO *)
-							(mode_info->atom_context->bios +
-							 data_offset +
-							 le16_to_cpu(power_info->info_4.usClockInfoArrayOffset) +
-							 (power_state->ucClockStateIndices[j] *
-							  power_info->info_4.ucClockInfoSize));
-						sclk = le16_to_cpu(clock_info->usEngineClockLow);
-						sclk |= clock_info->ucEngineClockHigh << 16;
-						mclk = le16_to_cpu(clock_info->usMemoryClockLow);
-						mclk |= clock_info->ucMemoryClockHigh << 16;
-						rdev->pm.power_state[state_index].clock_info[mode_index].mclk = mclk;
-						rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
-						/* skip invalid modes */
-						if ((rdev->pm.power_state[state_index].clock_info[mode_index].mclk == 0) ||
-						    (rdev->pm.power_state[state_index].clock_info[mode_index].sclk == 0))
-							continue;
-						rdev->pm.power_state[state_index].clock_info[mode_index].voltage.type =
-							VOLTAGE_SW;
-						rdev->pm.power_state[state_index].clock_info[mode_index].voltage.voltage =
-							clock_info->usVDDC;
-						mode_index++;
-					}
-				}
-				rdev->pm.power_state[state_index].num_clock_modes = mode_index;
-				if (mode_index) {
-					misc = le32_to_cpu(non_clock_info->ulCapsAndSettings);
-					misc2 = le16_to_cpu(non_clock_info->usClassification);
-					rdev->pm.power_state[state_index].misc = misc;
-					rdev->pm.power_state[state_index].misc2 = misc2;
-					rdev->pm.power_state[state_index].pcie_lanes =
-						((misc & ATOM_PPLIB_PCIE_LINK_WIDTH_MASK) >>
-						ATOM_PPLIB_PCIE_LINK_WIDTH_SHIFT) + 1;
-					switch (misc2 & ATOM_PPLIB_CLASSIFICATION_UI_MASK) {
-					case ATOM_PPLIB_CLASSIFICATION_UI_BATTERY:
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BATTERY;
-						break;
-					case ATOM_PPLIB_CLASSIFICATION_UI_BALANCED:
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_BALANCED;
-						break;
-					case ATOM_PPLIB_CLASSIFICATION_UI_PERFORMANCE:
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_PERFORMANCE;
-						break;
-					case ATOM_PPLIB_CLASSIFICATION_UI_NONE:
-						if (misc2 & ATOM_PPLIB_CLASSIFICATION_3DPERFORMANCE)
-							rdev->pm.power_state[state_index].type =
-								POWER_STATE_TYPE_PERFORMANCE;
-						break;
-					}
-					rdev->pm.power_state[state_index].flags = 0;
-					if (misc & ATOM_PPLIB_SINGLE_DISPLAY_ONLY)
-						rdev->pm.power_state[state_index].flags |=
-							RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
-					if (misc2 & ATOM_PPLIB_CLASSIFICATION_BOOT) {
-						rdev->pm.power_state[state_index].type =
-							POWER_STATE_TYPE_DEFAULT;
-						rdev->pm.default_power_state_index = state_index;
-						rdev->pm.power_state[state_index].default_clock_mode =
-							&rdev->pm.power_state[state_index].clock_info[mode_index - 1];
-						/* patch the table values with the default slck/mclk from firmware info */
-						for (j = 0; j < mode_index; j++) {
-							rdev->pm.power_state[state_index].clock_info[j].mclk =
-								rdev->clock.default_mclk;
-							rdev->pm.power_state[state_index].clock_info[j].sclk =
-								rdev->clock.default_sclk;
-							if (vddc)
-								rdev->pm.power_state[state_index].clock_info[j].voltage.voltage =
-									vddc;
-						}
-					}
-					state_index++;
-				}
-			}
-			/* if multiple clock modes, mark the lowest as no display */
-			for (i = 0; i < state_index; i++) {
-				if (rdev->pm.power_state[i].num_clock_modes > 1)
-					rdev->pm.power_state[i].clock_info[0].flags |=
-						RADEON_PM_MODE_NO_DISPLAY;
-			}
-			/* first mode is usually default */
-			if (rdev->pm.default_power_state_index == -1) {
-				rdev->pm.power_state[0].type =
-					POWER_STATE_TYPE_DEFAULT;
-				rdev->pm.default_power_state_index = 0;
-				rdev->pm.power_state[0].default_clock_mode =
-					&rdev->pm.power_state[0].clock_info[0];
-			}
+	if (atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		firmware_info =
+			(union firmware_info *)(mode_info->atom_context->bios +
+						data_offset);
+		vddc = firmware_info->info_14.usBootUpVDDCVoltage;
+	}
+
+	return vddc;
+}
+
+static void radeon_atombios_parse_pplib_non_clock_info(struct radeon_device *rdev,
+						       int state_index, int mode_index,
+						       struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info)
+{
+	int j;
+	u32 misc = le32_to_cpu(non_clock_info->ulCapsAndSettings);
+	u32 misc2 = le16_to_cpu(non_clock_info->usClassification);
+	u16 vddc = radeon_atombios_get_default_vddc(rdev);
+
+	rdev->pm.power_state[state_index].misc = misc;
+	rdev->pm.power_state[state_index].misc2 = misc2;
+	rdev->pm.power_state[state_index].pcie_lanes =
+		((misc & ATOM_PPLIB_PCIE_LINK_WIDTH_MASK) >>
+		 ATOM_PPLIB_PCIE_LINK_WIDTH_SHIFT) + 1;
+	switch (misc2 & ATOM_PPLIB_CLASSIFICATION_UI_MASK) {
+	case ATOM_PPLIB_CLASSIFICATION_UI_BATTERY:
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BATTERY;
+		break;
+	case ATOM_PPLIB_CLASSIFICATION_UI_BALANCED:
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_BALANCED;
+		break;
+	case ATOM_PPLIB_CLASSIFICATION_UI_PERFORMANCE:
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_PERFORMANCE;
+		break;
+	case ATOM_PPLIB_CLASSIFICATION_UI_NONE:
+		if (misc2 & ATOM_PPLIB_CLASSIFICATION_3DPERFORMANCE)
+			rdev->pm.power_state[state_index].type =
+				POWER_STATE_TYPE_PERFORMANCE;
+		break;
+	}
+	rdev->pm.power_state[state_index].flags = 0;
+	if (misc & ATOM_PPLIB_SINGLE_DISPLAY_ONLY)
+		rdev->pm.power_state[state_index].flags |=
+			RADEON_PM_STATE_SINGLE_DISPLAY_ONLY;
+	if (misc2 & ATOM_PPLIB_CLASSIFICATION_BOOT) {
+		rdev->pm.power_state[state_index].type =
+			POWER_STATE_TYPE_DEFAULT;
+		rdev->pm.default_power_state_index = state_index;
+		rdev->pm.power_state[state_index].default_clock_mode =
+			&rdev->pm.power_state[state_index].clock_info[mode_index - 1];
+		/* patch the table values with the default slck/mclk from firmware info */
+		for (j = 0; j < mode_index; j++) {
+			rdev->pm.power_state[state_index].clock_info[j].mclk =
+				rdev->clock.default_mclk;
+			rdev->pm.power_state[state_index].clock_info[j].sclk =
+				rdev->clock.default_sclk;
+			if (vddc)
+				rdev->pm.power_state[state_index].clock_info[j].voltage.voltage =
+					vddc;
+		}
+	}
+}
+
+static bool radeon_atombios_parse_pplib_clock_info(struct radeon_device *rdev,
+						   int state_index, int mode_index,
+						   union pplib_clock_info *clock_info)
+{
+	u32 sclk, mclk;
+
+	if (rdev->flags & RADEON_IS_IGP) {
+		if (rdev->family >= CHIP_PALM) {
+			sclk = le16_to_cpu(clock_info->sumo.usEngineClockLow);
+			sclk |= clock_info->sumo.ucEngineClockHigh << 16;
+			rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
+		} else {
+			sclk = le16_to_cpu(clock_info->rs780.usLowEngineClockLow);
+			sclk |= clock_info->rs780.ucLowEngineClockHigh << 16;
+			rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
+		}
+	} else if (ASIC_IS_DCE4(rdev)) {
+		sclk = le16_to_cpu(clock_info->evergreen.usEngineClockLow);
+		sclk |= clock_info->evergreen.ucEngineClockHigh << 16;
+		mclk = le16_to_cpu(clock_info->evergreen.usMemoryClockLow);
+		mclk |= clock_info->evergreen.ucMemoryClockHigh << 16;
+		rdev->pm.power_state[state_index].clock_info[mode_index].mclk = mclk;
+		rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
+		rdev->pm.power_state[state_index].clock_info[mode_index].voltage.type =
+			VOLTAGE_SW;
+		rdev->pm.power_state[state_index].clock_info[mode_index].voltage.voltage =
+			clock_info->evergreen.usVDDC;
+	} else {
+		sclk = le16_to_cpu(clock_info->r600.usEngineClockLow);
+		sclk |= clock_info->r600.ucEngineClockHigh << 16;
+		mclk = le16_to_cpu(clock_info->r600.usMemoryClockLow);
+		mclk |= clock_info->r600.ucMemoryClockHigh << 16;
+		rdev->pm.power_state[state_index].clock_info[mode_index].mclk = mclk;
+		rdev->pm.power_state[state_index].clock_info[mode_index].sclk = sclk;
+		rdev->pm.power_state[state_index].clock_info[mode_index].voltage.type =
+			VOLTAGE_SW;
+		rdev->pm.power_state[state_index].clock_info[mode_index].voltage.voltage =
+			clock_info->r600.usVDDC;
+	}
+
+	if (rdev->flags & RADEON_IS_IGP) {
+		/* skip invalid modes */
+		if (rdev->pm.power_state[state_index].clock_info[mode_index].sclk == 0)
+			return false;
+	} else {
+		/* skip invalid modes */
+		if ((rdev->pm.power_state[state_index].clock_info[mode_index].mclk == 0) ||
+		    (rdev->pm.power_state[state_index].clock_info[mode_index].sclk == 0))
+			return false;
+	}
+	return true;
+}
+
+static int radeon_atombios_parse_power_table_4_5(struct radeon_device *rdev)
+{
+	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info;
+	union pplib_power_state *power_state;
+	int i, j;
+	int state_index = 0, mode_index = 0;
+	union pplib_clock_info *clock_info;
+	bool valid;
+	union power_info *power_info;
+	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
+        u16 data_offset;
+	u8 frev, crev;
+
+	if (!atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset))
+		return state_index;
+	power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
+
+	radeon_atombios_add_pplib_thermal_controller(rdev, &power_info->pplib.sThermalController);
+	/* first mode is usually default, followed by low to high */
+	for (i = 0; i < power_info->pplib.ucNumStates; i++) {
+		mode_index = 0;
+		power_state = (union pplib_power_state *)
+			(mode_info->atom_context->bios + data_offset +
+			 le16_to_cpu(power_info->pplib.usStateArrayOffset) +
+			 i * power_info->pplib.ucStateEntrySize);
+		non_clock_info = (struct _ATOM_PPLIB_NONCLOCK_INFO *)
+			(mode_info->atom_context->bios + data_offset +
+			 le16_to_cpu(power_info->pplib.usNonClockInfoArrayOffset) +
+			 (power_state->v1.ucNonClockStateIndex *
+			  power_info->pplib.ucNonClockSize));
+		for (j = 0; j < (power_info->pplib.ucStateEntrySize - 1); j++) {
+			clock_info = (union pplib_clock_info *)
+				(mode_info->atom_context->bios + data_offset +
+				 le16_to_cpu(power_info->pplib.usClockInfoArrayOffset) +
+				 (power_state->v1.ucClockStateIndices[j] *
+				  power_info->pplib.ucClockInfoSize));
+			valid = radeon_atombios_parse_pplib_clock_info(rdev,
+								       state_index, mode_index,
+								       clock_info);
+			if (valid)
+				mode_index++;
+		}
+		rdev->pm.power_state[state_index].num_clock_modes = mode_index;
+		if (mode_index) {
+			radeon_atombios_parse_pplib_non_clock_info(rdev, state_index, mode_index,
+								   non_clock_info);
+			state_index++;
+		}
+	}
+	/* if multiple clock modes, mark the lowest as no display */
+	for (i = 0; i < state_index; i++) {
+		if (rdev->pm.power_state[i].num_clock_modes > 1)
+			rdev->pm.power_state[i].clock_info[0].flags |=
+				RADEON_PM_MODE_NO_DISPLAY;
+	}
+	/* first mode is usually default */
+	if (rdev->pm.default_power_state_index == -1) {
+		rdev->pm.power_state[0].type =
+			POWER_STATE_TYPE_DEFAULT;
+		rdev->pm.default_power_state_index = 0;
+		rdev->pm.power_state[0].default_clock_mode =
+			&rdev->pm.power_state[0].clock_info[0];
+	}
+	return state_index;
+}
+
+static int radeon_atombios_parse_power_table_6(struct radeon_device *rdev)
+{
+	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	struct _ATOM_PPLIB_NONCLOCK_INFO *non_clock_info;
+	union pplib_power_state *power_state;
+	int i, j, non_clock_array_index, clock_array_index;
+	int state_index = 0, mode_index = 0;
+	union pplib_clock_info *clock_info;
+	struct StateArray *state_array;
+	struct ClockInfoArray *clock_info_array;
+	struct NonClockInfoArray *non_clock_info_array;
+	bool valid;
+	union power_info *power_info;
+	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
+        u16 data_offset;
+	u8 frev, crev;
+
+	if (!atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset))
+		return state_index;
+	power_info = (union power_info *)(mode_info->atom_context->bios + data_offset);
+
+	radeon_atombios_add_pplib_thermal_controller(rdev, &power_info->pplib.sThermalController);
+	state_array = (struct StateArray *)
+		(mode_info->atom_context->bios + data_offset +
+		 power_info->pplib.usStateArrayOffset);
+	clock_info_array = (struct ClockInfoArray *)
+		(mode_info->atom_context->bios + data_offset +
+		 power_info->pplib.usClockInfoArrayOffset);
+	non_clock_info_array = (struct NonClockInfoArray *)
+		(mode_info->atom_context->bios + data_offset +
+		 power_info->pplib.usNonClockInfoArrayOffset);
+	for (i = 0; i < state_array->ucNumEntries; i++) {
+		mode_index = 0;
+		power_state = (union pplib_power_state *)&state_array->states[i];
+		/* XXX this might be an inagua bug... */
+		non_clock_array_index = i; /* power_state->v2.nonClockInfoIndex */
+		non_clock_info = (struct _ATOM_PPLIB_NONCLOCK_INFO *)
+			&non_clock_info_array->nonClockInfo[non_clock_array_index];
+		for (j = 0; j < power_state->v2.ucNumDPMLevels; j++) {
+			clock_array_index = power_state->v2.clockInfoIndex[j];
+			/* XXX this might be an inagua bug... */
+			if (clock_array_index >= clock_info_array->ucNumEntries)
+				continue;
+			clock_info = (union pplib_clock_info *)
+				&clock_info_array->clockInfo[clock_array_index];
+			valid = radeon_atombios_parse_pplib_clock_info(rdev,
+								       state_index, mode_index,
+								       clock_info);
+			if (valid)
+				mode_index++;
+		}
+		rdev->pm.power_state[state_index].num_clock_modes = mode_index;
+		if (mode_index) {
+			radeon_atombios_parse_pplib_non_clock_info(rdev, state_index, mode_index,
+								   non_clock_info);
+			state_index++;
+		}
+	}
+	/* if multiple clock modes, mark the lowest as no display */
+	for (i = 0; i < state_index; i++) {
+		if (rdev->pm.power_state[i].num_clock_modes > 1)
+			rdev->pm.power_state[i].clock_info[0].flags |=
+				RADEON_PM_MODE_NO_DISPLAY;
+	}
+	/* first mode is usually default */
+	if (rdev->pm.default_power_state_index == -1) {
+		rdev->pm.power_state[0].type =
+			POWER_STATE_TYPE_DEFAULT;
+		rdev->pm.default_power_state_index = 0;
+		rdev->pm.power_state[0].default_clock_mode =
+			&rdev->pm.power_state[0].clock_info[0];
+	}
+	return state_index;
+}
+
+void radeon_atombios_get_power_modes(struct radeon_device *rdev)
+{
+	struct radeon_mode_info *mode_info = &rdev->mode_info;
+	int index = GetIndexIntoMasterTable(DATA, PowerPlayInfo);
+	u16 data_offset;
+	u8 frev, crev;
+	int state_index = 0;
+
+	rdev->pm.default_power_state_index = -1;
+
+	if (atom_parse_data_header(mode_info->atom_context, index, NULL,
+				   &frev, &crev, &data_offset)) {
+		switch (frev) {
+		case 1:
+		case 2:
+		case 3:
+			state_index = radeon_atombios_parse_power_table_1_3(rdev);
+			break;
+		case 4:
+		case 5:
+			state_index = radeon_atombios_parse_power_table_4_5(rdev);
+			break;
+		case 6:
+			state_index = radeon_atombios_parse_power_table_6(rdev);
+			break;
+		default:
+			break;
 		}
 	} else {
 		/* add the default mode */
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 6d64a2705f12..35b5eb8fbe2a 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -77,13 +77,13 @@ int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 			p->relocs_ptr[i] = &p->relocs[i];
 			p->relocs[i].robj = p->relocs[i].gobj->driver_private;
 			p->relocs[i].lobj.bo = p->relocs[i].robj;
-			p->relocs[i].lobj.rdomain = r->read_domains;
 			p->relocs[i].lobj.wdomain = r->write_domain;
+			p->relocs[i].lobj.rdomain = r->read_domains;
+			p->relocs[i].lobj.tv.bo = &p->relocs[i].robj->tbo;
 			p->relocs[i].handle = r->handle;
 			p->relocs[i].flags = r->flags;
-			INIT_LIST_HEAD(&p->relocs[i].lobj.list);
 			radeon_bo_list_add_object(&p->relocs[i].lobj,
-						&p->validated);
+						  &p->validated);
 		}
 	}
 	return radeon_bo_list_validate(&p->validated);
@@ -189,10 +189,13 @@ static void radeon_cs_parser_fini(struct radeon_cs_parser *parser, int error)
 {
 	unsigned i;
 
-	if (!error && parser->ib) {
-		radeon_bo_list_fence(&parser->validated, parser->ib->fence);
-	}
-	radeon_bo_list_unreserve(&parser->validated);
+
+	if (!error && parser->ib)
+		ttm_eu_fence_buffer_objects(&parser->validated,
+					    parser->ib->fence);
+	else
+		ttm_eu_backoff_reservation(&parser->validated);
+
 	if (parser->relocs != NULL) {
 		for (i = 0; i < parser->nrelocs; i++) {
 			if (parser->relocs[i].gobj)
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index e12e79326cb1..86660cb425ab 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -81,6 +81,7 @@ static const char radeon_family_name[][16] = {
 	"JUNIPER",
 	"CYPRESS",
 	"HEMLOCK",
+	"PALM",
 	"LAST",
 };
 
@@ -335,7 +336,12 @@ bool radeon_card_posted(struct radeon_device *rdev)
 	uint32_t reg;
 
 	/* first check CRTCs */
-	if (ASIC_IS_DCE4(rdev)) {
+	if (ASIC_IS_DCE41(rdev)) {
+		reg = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) |
+			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET);
+		if (reg & EVERGREEN_CRTC_MASTER_EN)
+			return true;
+	} else if (ASIC_IS_DCE4(rdev)) {
 		reg = RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC0_REGISTER_OFFSET) |
 			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC1_REGISTER_OFFSET) |
 			RREG32(EVERGREEN_CRTC_CONTROL + EVERGREEN_CRTC2_REGISTER_OFFSET) |
diff --git a/drivers/gpu/drm/radeon/radeon_display.c b/drivers/gpu/drm/radeon/radeon_display.c
index 1df4dc6c063c..7b17e639ab32 100644
--- a/drivers/gpu/drm/radeon/radeon_display.c
+++ b/drivers/gpu/drm/radeon/radeon_display.c
@@ -183,12 +183,272 @@ static void radeon_crtc_destroy(struct drm_crtc *crtc)
 	kfree(radeon_crtc);
 }
 
+/*
+ * Handle unpin events outside the interrupt handler proper.
+ */
+static void radeon_unpin_work_func(struct work_struct *__work)
+{
+	struct radeon_unpin_work *work =
+		container_of(__work, struct radeon_unpin_work, work);
+	int r;
+
+	/* unpin of the old buffer */
+	r = radeon_bo_reserve(work->old_rbo, false);
+	if (likely(r == 0)) {
+		r = radeon_bo_unpin(work->old_rbo);
+		if (unlikely(r != 0)) {
+			DRM_ERROR("failed to unpin buffer after flip\n");
+		}
+		radeon_bo_unreserve(work->old_rbo);
+	} else
+		DRM_ERROR("failed to reserve buffer after flip\n");
+	kfree(work);
+}
+
+void radeon_crtc_handle_flip(struct radeon_device *rdev, int crtc_id)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
+	struct radeon_unpin_work *work;
+	struct drm_pending_vblank_event *e;
+	struct timeval now;
+	unsigned long flags;
+	u32 update_pending;
+	int vpos, hpos;
+
+	spin_lock_irqsave(&rdev->ddev->event_lock, flags);
+	work = radeon_crtc->unpin_work;
+	if (work == NULL ||
+	    !radeon_fence_signaled(work->fence)) {
+		spin_unlock_irqrestore(&rdev->ddev->event_lock, flags);
+		return;
+	}
+	/* New pageflip, or just completion of a previous one? */
+	if (!radeon_crtc->deferred_flip_completion) {
+		/* do the flip (mmio) */
+		update_pending = radeon_page_flip(rdev, crtc_id, work->new_crtc_base);
+	} else {
+		/* This is just a completion of a flip queued in crtc
+		 * at last invocation. Make sure we go directly to
+		 * completion routine.
+		 */
+		update_pending = 0;
+		radeon_crtc->deferred_flip_completion = 0;
+	}
+
+	/* Has the pageflip already completed in crtc, or is it certain
+	 * to complete in this vblank?
+	 */
+	if (update_pending &&
+	    (DRM_SCANOUTPOS_VALID & radeon_get_crtc_scanoutpos(rdev->ddev, crtc_id,
+							       &vpos, &hpos)) &&
+	    (vpos >=0) &&
+	    (vpos < (99 * rdev->mode_info.crtcs[crtc_id]->base.hwmode.crtc_vdisplay)/100)) {
+		/* crtc didn't flip in this target vblank interval,
+		 * but flip is pending in crtc. It will complete it
+		 * in next vblank interval, so complete the flip at
+		 * next vblank irq.
+		 */
+		radeon_crtc->deferred_flip_completion = 1;
+		spin_unlock_irqrestore(&rdev->ddev->event_lock, flags);
+		return;
+	}
+
+	/* Pageflip (will be) certainly completed in this vblank. Clean up. */
+	radeon_crtc->unpin_work = NULL;
+
+	/* wakeup userspace */
+	if (work->event) {
+		e = work->event;
+		e->event.sequence = drm_vblank_count_and_time(rdev->ddev, crtc_id, &now);
+		e->event.tv_sec = now.tv_sec;
+		e->event.tv_usec = now.tv_usec;
+		list_add_tail(&e->base.link, &e->base.file_priv->event_list);
+		wake_up_interruptible(&e->base.file_priv->event_wait);
+	}
+	spin_unlock_irqrestore(&rdev->ddev->event_lock, flags);
+
+	drm_vblank_put(rdev->ddev, radeon_crtc->crtc_id);
+	radeon_fence_unref(&work->fence);
+	radeon_post_page_flip(work->rdev, work->crtc_id);
+	schedule_work(&work->work);
+}
+
+static int radeon_crtc_page_flip(struct drm_crtc *crtc,
+				 struct drm_framebuffer *fb,
+				 struct drm_pending_vblank_event *event)
+{
+	struct drm_device *dev = crtc->dev;
+	struct radeon_device *rdev = dev->dev_private;
+	struct radeon_crtc *radeon_crtc = to_radeon_crtc(crtc);
+	struct radeon_framebuffer *old_radeon_fb;
+	struct radeon_framebuffer *new_radeon_fb;
+	struct drm_gem_object *obj;
+	struct radeon_bo *rbo;
+	struct radeon_fence *fence;
+	struct radeon_unpin_work *work;
+	unsigned long flags;
+	u32 tiling_flags, pitch_pixels;
+	u64 base;
+	int r;
+
+	work = kzalloc(sizeof *work, GFP_KERNEL);
+	if (work == NULL)
+		return -ENOMEM;
+
+	r = radeon_fence_create(rdev, &fence);
+	if (unlikely(r != 0)) {
+		kfree(work);
+		DRM_ERROR("flip queue: failed to create fence.\n");
+		return -ENOMEM;
+	}
+	work->event = event;
+	work->rdev = rdev;
+	work->crtc_id = radeon_crtc->crtc_id;
+	work->fence = radeon_fence_ref(fence);
+	old_radeon_fb = to_radeon_framebuffer(crtc->fb);
+	new_radeon_fb = to_radeon_framebuffer(fb);
+	/* schedule unpin of the old buffer */
+	obj = old_radeon_fb->obj;
+	rbo = obj->driver_private;
+	work->old_rbo = rbo;
+	INIT_WORK(&work->work, radeon_unpin_work_func);
+
+	/* We borrow the event spin lock for protecting unpin_work */
+	spin_lock_irqsave(&dev->event_lock, flags);
+	if (radeon_crtc->unpin_work) {
+		spin_unlock_irqrestore(&dev->event_lock, flags);
+		kfree(work);
+		radeon_fence_unref(&fence);
+
+		DRM_DEBUG_DRIVER("flip queue: crtc already busy\n");
+		return -EBUSY;
+	}
+	radeon_crtc->unpin_work = work;
+	radeon_crtc->deferred_flip_completion = 0;
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+
+	/* pin the new buffer */
+	obj = new_radeon_fb->obj;
+	rbo = obj->driver_private;
+
+	DRM_DEBUG_DRIVER("flip-ioctl() cur_fbo = %p, cur_bbo = %p\n",
+			 work->old_rbo, rbo);
+
+	r = radeon_bo_reserve(rbo, false);
+	if (unlikely(r != 0)) {
+		DRM_ERROR("failed to reserve new rbo buffer before flip\n");
+		goto pflip_cleanup;
+	}
+	r = radeon_bo_pin(rbo, RADEON_GEM_DOMAIN_VRAM, &base);
+	if (unlikely(r != 0)) {
+		radeon_bo_unreserve(rbo);
+		r = -EINVAL;
+		DRM_ERROR("failed to pin new rbo buffer before flip\n");
+		goto pflip_cleanup;
+	}
+	radeon_bo_get_tiling_flags(rbo, &tiling_flags, NULL);
+	radeon_bo_unreserve(rbo);
+
+	if (!ASIC_IS_AVIVO(rdev)) {
+		/* crtc offset is from display base addr not FB location */
+		base -= radeon_crtc->legacy_display_base_addr;
+		pitch_pixels = fb->pitch / (fb->bits_per_pixel / 8);
+
+		if (tiling_flags & RADEON_TILING_MACRO) {
+			if (ASIC_IS_R300(rdev)) {
+				base &= ~0x7ff;
+			} else {
+				int byteshift = fb->bits_per_pixel >> 4;
+				int tile_addr = (((crtc->y >> 3) * pitch_pixels +  crtc->x) >> (8 - byteshift)) << 11;
+				base += tile_addr + ((crtc->x << byteshift) % 256) + ((crtc->y % 8) << 8);
+			}
+		} else {
+			int offset = crtc->y * pitch_pixels + crtc->x;
+			switch (fb->bits_per_pixel) {
+			case 8:
+			default:
+				offset *= 1;
+				break;
+			case 15:
+			case 16:
+				offset *= 2;
+				break;
+			case 24:
+				offset *= 3;
+				break;
+			case 32:
+				offset *= 4;
+				break;
+			}
+			base += offset;
+		}
+		base &= ~7;
+	}
+
+	spin_lock_irqsave(&dev->event_lock, flags);
+	work->new_crtc_base = base;
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+
+	/* update crtc fb */
+	crtc->fb = fb;
+
+	r = drm_vblank_get(dev, radeon_crtc->crtc_id);
+	if (r) {
+		DRM_ERROR("failed to get vblank before flip\n");
+		goto pflip_cleanup1;
+	}
+
+	/* 32 ought to cover us */
+	r = radeon_ring_lock(rdev, 32);
+	if (r) {
+		DRM_ERROR("failed to lock the ring before flip\n");
+		goto pflip_cleanup2;
+	}
+
+	/* emit the fence */
+	radeon_fence_emit(rdev, fence);
+	/* set the proper interrupt */
+	radeon_pre_page_flip(rdev, radeon_crtc->crtc_id);
+	/* fire the ring */
+	radeon_ring_unlock_commit(rdev);
+
+	return 0;
+
+pflip_cleanup2:
+	drm_vblank_put(dev, radeon_crtc->crtc_id);
+
+pflip_cleanup1:
+	r = radeon_bo_reserve(rbo, false);
+	if (unlikely(r != 0)) {
+		DRM_ERROR("failed to reserve new rbo in error path\n");
+		goto pflip_cleanup;
+	}
+	r = radeon_bo_unpin(rbo);
+	if (unlikely(r != 0)) {
+		radeon_bo_unreserve(rbo);
+		r = -EINVAL;
+		DRM_ERROR("failed to unpin new rbo in error path\n");
+		goto pflip_cleanup;
+	}
+	radeon_bo_unreserve(rbo);
+
+pflip_cleanup:
+	spin_lock_irqsave(&dev->event_lock, flags);
+	radeon_crtc->unpin_work = NULL;
+	spin_unlock_irqrestore(&dev->event_lock, flags);
+	radeon_fence_unref(&fence);
+	kfree(work);
+
+	return r;
+}
+
 static const struct drm_crtc_funcs radeon_crtc_funcs = {
 	.cursor_set = radeon_crtc_cursor_set,
 	.cursor_move = radeon_crtc_cursor_move,
 	.gamma_set = radeon_crtc_gamma_set,
 	.set_config = drm_crtc_helper_set_config,
 	.destroy = radeon_crtc_destroy,
+	.page_flip = radeon_crtc_page_flip,
 };
 
 static void radeon_crtc_init(struct drm_device *dev, int index)
@@ -225,7 +485,7 @@ static void radeon_crtc_init(struct drm_device *dev, int index)
 		radeon_legacy_init_crtc(dev, radeon_crtc);
 }
 
-static const char *encoder_names[34] = {
+static const char *encoder_names[36] = {
 	"NONE",
 	"INTERNAL_LVDS",
 	"INTERNAL_TMDS1",
@@ -260,6 +520,8 @@ static const char *encoder_names[34] = {
 	"INTERNAL_KLDSCP_LVTMA",
 	"INTERNAL_UNIPHY1",
 	"INTERNAL_UNIPHY2",
+	"NUTMEG",
+	"TRAVIS",
 };
 
 static const char *connector_names[15] = {
@@ -1019,7 +1281,7 @@ bool radeon_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
 /*
  * Retrieve current video scanout position of crtc on a given gpu.
  *
- * \param rdev Device to query.
+ * \param dev Device to query.
  * \param crtc Crtc to query.
  * \param *vpos Location where vertical scanout position should be stored.
  * \param *hpos Location where horizontal scanout position should go.
@@ -1031,72 +1293,74 @@ bool radeon_crtc_scaling_mode_fixup(struct drm_crtc *crtc,
  *
  * \return Flags, or'ed together as follows:
  *
- * RADEON_SCANOUTPOS_VALID = Query successfull.
- * RADEON_SCANOUTPOS_INVBL = Inside vblank.
- * RADEON_SCANOUTPOS_ACCURATE = Returned position is accurate. A lack of
+ * DRM_SCANOUTPOS_VALID = Query successfull.
+ * DRM_SCANOUTPOS_INVBL = Inside vblank.
+ * DRM_SCANOUTPOS_ACCURATE = Returned position is accurate. A lack of
  * this flag means that returned position may be offset by a constant but
  * unknown small number of scanlines wrt. real scanout position.
  *
  */
-int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos, int *hpos)
+int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc, int *vpos, int *hpos)
 {
 	u32 stat_crtc = 0, vbl = 0, position = 0;
 	int vbl_start, vbl_end, vtotal, ret = 0;
 	bool in_vbl = true;
 
+	struct radeon_device *rdev = dev->dev_private;
+
 	if (ASIC_IS_DCE4(rdev)) {
 		if (crtc == 0) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC0_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC0_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 1) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC1_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC1_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 2) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC2_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC2_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 3) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC3_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC3_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 4) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC4_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC4_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 5) {
 			vbl = RREG32(EVERGREEN_CRTC_V_BLANK_START_END +
 				     EVERGREEN_CRTC5_REGISTER_OFFSET);
 			position = RREG32(EVERGREEN_CRTC_STATUS_POSITION +
 					  EVERGREEN_CRTC5_REGISTER_OFFSET);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 	} else if (ASIC_IS_AVIVO(rdev)) {
 		if (crtc == 0) {
 			vbl = RREG32(AVIVO_D1CRTC_V_BLANK_START_END);
 			position = RREG32(AVIVO_D1CRTC_STATUS_POSITION);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 1) {
 			vbl = RREG32(AVIVO_D2CRTC_V_BLANK_START_END);
 			position = RREG32(AVIVO_D2CRTC_STATUS_POSITION);
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 	} else {
 		/* Pre-AVIVO: Different encoding of scanout pos and vblank interval. */
@@ -1112,7 +1376,7 @@ int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos,
 			if (!(stat_crtc & 1))
 				in_vbl = false;
 
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 		if (crtc == 1) {
 			vbl = (RREG32(RADEON_CRTC2_V_TOTAL_DISP) &
@@ -1122,7 +1386,7 @@ int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos,
 			if (!(stat_crtc & 1))
 				in_vbl = false;
 
-			ret |= RADEON_SCANOUTPOS_VALID;
+			ret |= DRM_SCANOUTPOS_VALID;
 		}
 	}
 
@@ -1133,13 +1397,13 @@ int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos,
 	/* Valid vblank area boundaries from gpu retrieved? */
 	if (vbl > 0) {
 		/* Yes: Decode. */
-		ret |= RADEON_SCANOUTPOS_ACCURATE;
+		ret |= DRM_SCANOUTPOS_ACCURATE;
 		vbl_start = vbl & 0x1fff;
 		vbl_end = (vbl >> 16) & 0x1fff;
 	}
 	else {
 		/* No: Fake something reasonable which gives at least ok results. */
-		vbl_start = rdev->mode_info.crtcs[crtc]->base.mode.crtc_vdisplay;
+		vbl_start = rdev->mode_info.crtcs[crtc]->base.hwmode.crtc_vdisplay;
 		vbl_end = 0;
 	}
 
@@ -1155,7 +1419,7 @@ int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos,
 
 	/* Inside "upper part" of vblank area? Apply corrective offset if so: */
 	if (in_vbl && (*vpos >= vbl_start)) {
-		vtotal = rdev->mode_info.crtcs[crtc]->base.mode.crtc_vtotal;
+		vtotal = rdev->mode_info.crtcs[crtc]->base.hwmode.crtc_vtotal;
 		*vpos = *vpos - vtotal;
 	}
 
@@ -1164,7 +1428,7 @@ int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos,
 
 	/* In vblank? */
 	if (in_vbl)
-		ret |= RADEON_SCANOUTPOS_INVBL;
+		ret |= DRM_SCANOUTPOS_INVBL;
 
 	return ret;
 }
diff --git a/drivers/gpu/drm/radeon/radeon_drv.c b/drivers/gpu/drm/radeon/radeon_drv.c
index 88e4ea925900..a92d2a5cea90 100644
--- a/drivers/gpu/drm/radeon/radeon_drv.c
+++ b/drivers/gpu/drm/radeon/radeon_drv.c
@@ -48,9 +48,10 @@
  * - 2.5.0 - add get accel 2 to work around ddx breakage for evergreen
  * - 2.6.0 - add tiling config query (r6xx+), add initial HiZ support (r300->r500)
  *   2.7.0 - fixups for r600 2D tiling support. (no external ABI change), add eg dyn gpr regs
+ *   2.8.0 - pageflip support
  */
 #define KMS_DRIVER_MAJOR	2
-#define KMS_DRIVER_MINOR	7
+#define KMS_DRIVER_MINOR	8
 #define KMS_DRIVER_PATCHLEVEL	0
 int radeon_driver_load_kms(struct drm_device *dev, unsigned long flags);
 int radeon_driver_unload_kms(struct drm_device *dev);
@@ -66,6 +67,10 @@ int radeon_resume_kms(struct drm_device *dev);
 u32 radeon_get_vblank_counter_kms(struct drm_device *dev, int crtc);
 int radeon_enable_vblank_kms(struct drm_device *dev, int crtc);
 void radeon_disable_vblank_kms(struct drm_device *dev, int crtc);
+int radeon_get_vblank_timestamp_kms(struct drm_device *dev, int crtc,
+				    int *max_error,
+				    struct timeval *vblank_time,
+				    unsigned flags);
 void radeon_driver_irq_preinstall_kms(struct drm_device *dev);
 int radeon_driver_irq_postinstall_kms(struct drm_device *dev);
 void radeon_driver_irq_uninstall_kms(struct drm_device *dev);
@@ -74,6 +79,8 @@ int radeon_dma_ioctl_kms(struct drm_device *dev, void *data,
 			 struct drm_file *file_priv);
 int radeon_gem_object_init(struct drm_gem_object *obj);
 void radeon_gem_object_free(struct drm_gem_object *obj);
+extern int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc,
+				      int *vpos, int *hpos);
 extern struct drm_ioctl_desc radeon_ioctls_kms[];
 extern int radeon_max_kms_ioctl;
 int radeon_mmap(struct file *filp, struct vm_area_struct *vma);
@@ -277,6 +284,8 @@ static struct drm_driver kms_driver = {
 	.get_vblank_counter = radeon_get_vblank_counter_kms,
 	.enable_vblank = radeon_enable_vblank_kms,
 	.disable_vblank = radeon_disable_vblank_kms,
+	.get_vblank_timestamp = radeon_get_vblank_timestamp_kms,
+	.get_scanout_position = radeon_get_crtc_scanoutpos,
 #if defined(CONFIG_DEBUG_FS)
 	.debugfs_init = radeon_debugfs_init,
 	.debugfs_cleanup = radeon_debugfs_cleanup,
diff --git a/drivers/gpu/drm/radeon/radeon_encoders.c b/drivers/gpu/drm/radeon/radeon_encoders.c
index 041943df966b..e4e64a80b58d 100644
--- a/drivers/gpu/drm/radeon/radeon_encoders.c
+++ b/drivers/gpu/drm/radeon/radeon_encoders.c
@@ -713,7 +713,7 @@ atombios_get_encoder_mode(struct drm_encoder *encoder)
  * DIG1/2 can drive UNIPHY0/1/2 link A or link B
  *
  * DCE 4.0
- * - 3 DIG transmitter blocks UNPHY0/1/2 (links A and B).
+ * - 3 DIG transmitter blocks UNIPHY0/1/2 (links A and B).
  * Supports up to 6 digital outputs
  * - 6 DIG encoder blocks.
  * - DIG to PHY mapping is hardcoded
@@ -724,6 +724,12 @@ atombios_get_encoder_mode(struct drm_encoder *encoder)
  * DIG5 drives UNIPHY2 link A, A+B
  * DIG6 drives UNIPHY2 link B
  *
+ * DCE 4.1
+ * - 3 DIG transmitter blocks UNIPHY0/1/2 (links A and B).
+ * Supports up to 6 digital outputs
+ * - 2 DIG encoder blocks.
+ * DIG1/2 can drive UNIPHY0/1/2 link A or link B
+ *
  * Routing
  * crtc -> dig encoder -> UNIPHY/LVTMA (1 or 2 links)
  * Examples:
@@ -904,9 +910,15 @@ atombios_dig_transmitter_setup(struct drm_encoder *encoder, int action, uint8_t
 		else
 			args.v3.ucLaneNum = 4;
 
-		if (dig->linkb) {
-			args.v3.acConfig.ucLinkSel = 1;
-			args.v3.acConfig.ucEncoderSel = 1;
+		if (ASIC_IS_DCE41(rdev)) {
+			args.v3.acConfig.ucEncoderSel = dig->dig_encoder;
+			if (dig->linkb)
+				args.v3.acConfig.ucLinkSel = 1;
+		} else {
+			if (dig->linkb) {
+				args.v3.acConfig.ucLinkSel = 1;
+				args.v3.acConfig.ucEncoderSel = 1;
+			}
 		}
 
 		/* Select the PLL for the PHY
@@ -1044,6 +1056,7 @@ atombios_set_edp_panel_power(struct drm_connector *connector, int action)
 
 union external_encoder_control {
 	EXTERNAL_ENCODER_CONTROL_PS_ALLOCATION v1;
+	EXTERNAL_ENCODER_CONTROL_PS_ALLOCATION_V3 v3;
 };
 
 static void
@@ -1054,6 +1067,7 @@ atombios_external_encoder_setup(struct drm_encoder *encoder,
 	struct drm_device *dev = encoder->dev;
 	struct radeon_device *rdev = dev->dev_private;
 	struct radeon_encoder *radeon_encoder = to_radeon_encoder(encoder);
+	struct radeon_encoder *ext_radeon_encoder = to_radeon_encoder(ext_encoder);
 	union external_encoder_control args;
 	struct drm_connector *connector = radeon_get_connector_for_encoder(encoder);
 	int index = GetIndexIntoMasterTable(COMMAND, ExternalEncoderControl);
@@ -1061,6 +1075,7 @@ atombios_external_encoder_setup(struct drm_encoder *encoder,
 	int dp_clock = 0;
 	int dp_lane_count = 0;
 	int connector_object_id = 0;
+	u32 ext_enum = (ext_radeon_encoder->encoder_enum & ENUM_ID_MASK) >> ENUM_ID_SHIFT;
 
 	if (connector) {
 		struct radeon_connector *radeon_connector = to_radeon_connector(connector);
@@ -1099,6 +1114,37 @@ atombios_external_encoder_setup(struct drm_encoder *encoder,
 			else
 				args.v1.sDigEncoder.ucLaneNum = 4;
 			break;
+		case 3:
+			args.v3.sExtEncoder.ucAction = action;
+			if (action == EXTERNAL_ENCODER_ACTION_V3_ENCODER_INIT)
+				args.v3.sExtEncoder.usConnectorId = connector_object_id;
+			else
+				args.v3.sExtEncoder.usPixelClock = cpu_to_le16(radeon_encoder->pixel_clock / 10);
+			args.v3.sExtEncoder.ucEncoderMode = atombios_get_encoder_mode(encoder);
+
+			if (args.v3.sExtEncoder.ucEncoderMode == ATOM_ENCODER_MODE_DP) {
+				if (dp_clock == 270000)
+					args.v3.sExtEncoder.ucConfig |= EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_2_70GHZ;
+				else if (dp_clock == 540000)
+					args.v3.sExtEncoder.ucConfig |= EXTERNAL_ENCODER_CONFIG_V3_DPLINKRATE_5_40GHZ;
+				args.v3.sExtEncoder.ucLaneNum = dp_lane_count;
+			} else if (radeon_encoder->pixel_clock > 165000)
+				args.v3.sExtEncoder.ucLaneNum = 8;
+			else
+				args.v3.sExtEncoder.ucLaneNum = 4;
+			switch (ext_enum) {
+			case GRAPH_OBJECT_ENUM_ID1:
+				args.v3.sExtEncoder.ucConfig |= EXTERNAL_ENCODER_CONFIG_V3_ENCODER1;
+				break;
+			case GRAPH_OBJECT_ENUM_ID2:
+				args.v3.sExtEncoder.ucConfig |= EXTERNAL_ENCODER_CONFIG_V3_ENCODER2;
+				break;
+			case GRAPH_OBJECT_ENUM_ID3:
+				args.v3.sExtEncoder.ucConfig |= EXTERNAL_ENCODER_CONFIG_V3_ENCODER3;
+				break;
+			}
+			args.v3.sExtEncoder.ucBitPerColor = PANEL_8BIT_PER_COLOR;
+			break;
 		default:
 			DRM_ERROR("Unknown table version: %d, %d\n", frev, crev);
 			return;
@@ -1289,12 +1335,18 @@ radeon_atom_encoder_dpms(struct drm_encoder *encoder, int mode)
 		switch (mode) {
 		case DRM_MODE_DPMS_ON:
 		default:
-			action = ATOM_ENABLE;
+			if (ASIC_IS_DCE41(rdev) && (rdev->flags & RADEON_IS_IGP))
+				action = EXTERNAL_ENCODER_ACTION_V3_ENABLE_OUTPUT;
+			else
+				action = ATOM_ENABLE;
 			break;
 		case DRM_MODE_DPMS_STANDBY:
 		case DRM_MODE_DPMS_SUSPEND:
 		case DRM_MODE_DPMS_OFF:
-			action = ATOM_DISABLE;
+			if (ASIC_IS_DCE41(rdev) && (rdev->flags & RADEON_IS_IGP))
+				action = EXTERNAL_ENCODER_ACTION_V3_DISABLE_OUTPUT;
+			else
+				action = ATOM_DISABLE;
 			break;
 		}
 		atombios_external_encoder_setup(encoder, ext_encoder, action);
@@ -1483,6 +1535,11 @@ static int radeon_atom_pick_dig_encoder(struct drm_encoder *encoder)
 	struct radeon_encoder_atom_dig *dig;
 	uint32_t dig_enc_in_use = 0;
 
+	/* on DCE41 and encoder can driver any phy so just crtc id */
+	if (ASIC_IS_DCE41(rdev)) {
+		return radeon_crtc->crtc_id;
+	}
+
 	if (ASIC_IS_DCE4(rdev)) {
 		dig = radeon_encoder->enc_priv;
 		switch (radeon_encoder->encoder_id) {
@@ -1610,7 +1667,13 @@ radeon_atom_encoder_mode_set(struct drm_encoder *encoder,
 	}
 
 	if (ext_encoder) {
-		atombios_external_encoder_setup(encoder, ext_encoder, ATOM_ENABLE);
+		if (ASIC_IS_DCE41(rdev) && (rdev->flags & RADEON_IS_IGP)) {
+			atombios_external_encoder_setup(encoder, ext_encoder,
+							EXTERNAL_ENCODER_ACTION_V3_ENCODER_INIT);
+			atombios_external_encoder_setup(encoder, ext_encoder,
+							EXTERNAL_ENCODER_ACTION_V3_ENCODER_SETUP);
+		} else
+			atombios_external_encoder_setup(encoder, ext_encoder, ATOM_ENABLE);
 	}
 
 	atombios_apply_encoder_quirks(encoder, adjusted_mode);
@@ -2029,6 +2092,8 @@ radeon_add_atom_encoder(struct drm_device *dev, uint32_t encoder_enum, uint32_t
 	case ENCODER_OBJECT_ID_TITFP513:
 	case ENCODER_OBJECT_ID_VT1623:
 	case ENCODER_OBJECT_ID_HDMI_SI1930:
+	case ENCODER_OBJECT_ID_TRAVIS:
+	case ENCODER_OBJECT_ID_NUTMEG:
 		/* these are handled by the primary encoders */
 		radeon_encoder->is_ext_encoder = true;
 		if (radeon_encoder->devices & (ATOM_DEVICE_LCD_SUPPORT))
diff --git a/drivers/gpu/drm/radeon/radeon_family.h b/drivers/gpu/drm/radeon/radeon_family.h
index e329066dcabd..4c222d5437d1 100644
--- a/drivers/gpu/drm/radeon/radeon_family.h
+++ b/drivers/gpu/drm/radeon/radeon_family.h
@@ -80,6 +80,7 @@ enum radeon_family {
 	CHIP_JUNIPER,
 	CHIP_CYPRESS,
 	CHIP_HEMLOCK,
+	CHIP_PALM,
 	CHIP_LAST,
 };
 
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index daacb281dfaf..171b0b2e3a64 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -38,6 +38,7 @@
 #include "drm.h"
 #include "radeon_reg.h"
 #include "radeon.h"
+#include "radeon_trace.h"
 
 int radeon_fence_emit(struct radeon_device *rdev, struct radeon_fence *fence)
 {
@@ -57,6 +58,7 @@ int radeon_fence_emit(struct radeon_device *rdev, struct radeon_fence *fence)
 	} else
 		radeon_fence_ring_emit(rdev, fence);
 
+	trace_radeon_fence_emit(rdev->ddev, fence->seq);
 	fence->emited = true;
 	list_del(&fence->list);
 	list_add_tail(&fence->list, &rdev->fence_drv.emited);
@@ -213,6 +215,7 @@ int radeon_fence_wait(struct radeon_fence *fence, bool intr)
 retry:
 	/* save current sequence used to check for GPU lockup */
 	seq = rdev->fence_drv.last_seq;
+	trace_radeon_fence_wait_begin(rdev->ddev, seq);
 	if (intr) {
 		radeon_irq_kms_sw_irq_get(rdev);
 		r = wait_event_interruptible_timeout(rdev->fence_drv.queue,
@@ -227,6 +230,7 @@ retry:
 			 radeon_fence_signaled(fence), timeout);
 		radeon_irq_kms_sw_irq_put(rdev);
 	}
+	trace_radeon_fence_wait_end(rdev->ddev, seq);
 	if (unlikely(!radeon_fence_signaled(fence))) {
 		/* we were interrupted for some reason and fence isn't
 		 * isn't signaled yet, resume wait
diff --git a/drivers/gpu/drm/radeon/radeon_irq_kms.c b/drivers/gpu/drm/radeon/radeon_irq_kms.c
index a108c7ed14f5..c6861bb751ad 100644
--- a/drivers/gpu/drm/radeon/radeon_irq_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_irq_kms.c
@@ -71,8 +71,10 @@ void radeon_driver_irq_preinstall_kms(struct drm_device *dev)
 	rdev->irq.gui_idle = false;
 	for (i = 0; i < rdev->num_crtc; i++)
 		rdev->irq.crtc_vblank_int[i] = false;
-	for (i = 0; i < 6; i++)
+	for (i = 0; i < 6; i++) {
 		rdev->irq.hpd[i] = false;
+		rdev->irq.pflip[i] = false;
+	}
 	radeon_irq_set(rdev);
 	/* Clear bits */
 	radeon_irq_process(rdev);
@@ -101,8 +103,10 @@ void radeon_driver_irq_uninstall_kms(struct drm_device *dev)
 	rdev->irq.gui_idle = false;
 	for (i = 0; i < rdev->num_crtc; i++)
 		rdev->irq.crtc_vblank_int[i] = false;
-	for (i = 0; i < 6; i++)
+	for (i = 0; i < 6; i++) {
 		rdev->irq.hpd[i] = false;
+		rdev->irq.pflip[i] = false;
+	}
 	radeon_irq_set(rdev);
 }
 
@@ -121,7 +125,7 @@ int radeon_irq_kms_init(struct radeon_device *rdev)
 	 * chips.  Disable MSI on them for now.
 	 */
 	if ((rdev->family >= CHIP_RV380) &&
-	    (!(rdev->flags & RADEON_IS_IGP)) &&
+	    ((!(rdev->flags & RADEON_IS_IGP)) || (rdev->family >= CHIP_PALM)) &&
 	    (!(rdev->flags & RADEON_IS_AGP))) {
 		int ret = pci_enable_msi(rdev->pdev);
 		if (!ret) {
@@ -175,3 +179,34 @@ void radeon_irq_kms_sw_irq_put(struct radeon_device *rdev)
 	spin_unlock_irqrestore(&rdev->irq.sw_lock, irqflags);
 }
 
+void radeon_irq_kms_pflip_irq_get(struct radeon_device *rdev, int crtc)
+{
+	unsigned long irqflags;
+
+	if (crtc < 0 || crtc >= rdev->num_crtc)
+		return;
+
+	spin_lock_irqsave(&rdev->irq.pflip_lock[crtc], irqflags);
+	if (rdev->ddev->irq_enabled && (++rdev->irq.pflip_refcount[crtc] == 1)) {
+		rdev->irq.pflip[crtc] = true;
+		radeon_irq_set(rdev);
+	}
+	spin_unlock_irqrestore(&rdev->irq.pflip_lock[crtc], irqflags);
+}
+
+void radeon_irq_kms_pflip_irq_put(struct radeon_device *rdev, int crtc)
+{
+	unsigned long irqflags;
+
+	if (crtc < 0 || crtc >= rdev->num_crtc)
+		return;
+
+	spin_lock_irqsave(&rdev->irq.pflip_lock[crtc], irqflags);
+	BUG_ON(rdev->ddev->irq_enabled && rdev->irq.pflip_refcount[crtc] <= 0);
+	if (rdev->ddev->irq_enabled && (--rdev->irq.pflip_refcount[crtc] == 0)) {
+		rdev->irq.pflip[crtc] = false;
+		radeon_irq_set(rdev);
+	}
+	spin_unlock_irqrestore(&rdev->irq.pflip_lock[crtc], irqflags);
+}
+
diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c
index 8fbbe1c6ebbd..4bf423ca4c12 100644
--- a/drivers/gpu/drm/radeon/radeon_kms.c
+++ b/drivers/gpu/drm/radeon/radeon_kms.c
@@ -277,6 +277,27 @@ void radeon_disable_vblank_kms(struct drm_device *dev, int crtc)
 	radeon_irq_set(rdev);
 }
 
+int radeon_get_vblank_timestamp_kms(struct drm_device *dev, int crtc,
+				    int *max_error,
+				    struct timeval *vblank_time,
+				    unsigned flags)
+{
+	struct drm_crtc *drmcrtc;
+	struct radeon_device *rdev = dev->dev_private;
+
+	if (crtc < 0 || crtc >= dev->num_crtcs) {
+		DRM_ERROR("Invalid crtc %d\n", crtc);
+		return -EINVAL;
+	}
+
+	/* Get associated drm_crtc: */
+	drmcrtc = &rdev->mode_info.crtcs[crtc]->base;
+
+	/* Helper routine in DRM core does all the work: */
+	return drm_calc_vbltimestamp_from_scanoutpos(dev, crtc, max_error,
+						     vblank_time, flags,
+						     drmcrtc);
+}
 
 /*
  * IOCTL.
diff --git a/drivers/gpu/drm/radeon/radeon_mode.h b/drivers/gpu/drm/radeon/radeon_mode.h
index e301c6f9e059..f406f02bf14e 100644
--- a/drivers/gpu/drm/radeon/radeon_mode.h
+++ b/drivers/gpu/drm/radeon/radeon_mode.h
@@ -277,6 +277,9 @@ struct radeon_crtc {
 	fixed20_12 hsc;
 	struct drm_display_mode native_mode;
 	int pll_id;
+	/* page flipping */
+	struct radeon_unpin_work *unpin_work;
+	int deferred_flip_completion;
 };
 
 struct radeon_encoder_primary_dac {
@@ -442,10 +445,6 @@ struct radeon_framebuffer {
 	struct drm_gem_object *obj;
 };
 
-/* radeon_get_crtc_scanoutpos() return flags */
-#define RADEON_SCANOUTPOS_VALID        (1 << 0)
-#define RADEON_SCANOUTPOS_INVBL        (1 << 1)
-#define RADEON_SCANOUTPOS_ACCURATE     (1 << 2)
 
 extern enum radeon_tv_std
 radeon_combios_get_tv_info(struct radeon_device *rdev);
@@ -562,7 +561,8 @@ extern int radeon_crtc_cursor_set(struct drm_crtc *crtc,
 extern int radeon_crtc_cursor_move(struct drm_crtc *crtc,
 				   int x, int y);
 
-extern int radeon_get_crtc_scanoutpos(struct radeon_device *rdev, int crtc, int *vpos, int *hpos);
+extern int radeon_get_crtc_scanoutpos(struct drm_device *dev, int crtc,
+				      int *vpos, int *hpos);
 
 extern bool radeon_combios_check_hardcoded_edid(struct radeon_device *rdev);
 extern struct edid *
@@ -662,4 +662,7 @@ int radeon_fbdev_total_size(struct radeon_device *rdev);
 bool radeon_fbdev_robj_is_fb(struct radeon_device *rdev, struct radeon_bo *robj);
 
 void radeon_fb_output_poll_changed(struct radeon_device *rdev);
+
+void radeon_crtc_handle_flip(struct radeon_device *rdev, int crtc_id);
+
 #endif
diff --git a/drivers/gpu/drm/radeon/radeon_object.c b/drivers/gpu/drm/radeon/radeon_object.c
index a598d0049aa5..7d6b8e88f746 100644
--- a/drivers/gpu/drm/radeon/radeon_object.c
+++ b/drivers/gpu/drm/radeon/radeon_object.c
@@ -34,6 +34,7 @@
 #include <drm/drmP.h>
 #include "radeon_drm.h"
 #include "radeon.h"
+#include "radeon_trace.h"
 
 
 int radeon_ttm_init(struct radeon_device *rdev);
@@ -146,6 +147,7 @@ retry:
 		list_add_tail(&bo->list, &rdev->gem.objects);
 		mutex_unlock(&bo->rdev->gem.mutex);
 	}
+	trace_radeon_bo_create(bo);
 	return 0;
 }
 
@@ -302,34 +304,9 @@ void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
 				struct list_head *head)
 {
 	if (lobj->wdomain) {
-		list_add(&lobj->list, head);
+		list_add(&lobj->tv.head, head);
 	} else {
-		list_add_tail(&lobj->list, head);
-	}
-}
-
-int radeon_bo_list_reserve(struct list_head *head)
-{
-	struct radeon_bo_list *lobj;
-	int r;
-
-	list_for_each_entry(lobj, head, list){
-		r = radeon_bo_reserve(lobj->bo, false);
-		if (unlikely(r != 0))
-			return r;
-		lobj->reserved = true;
-	}
-	return 0;
-}
-
-void radeon_bo_list_unreserve(struct list_head *head)
-{
-	struct radeon_bo_list *lobj;
-
-	list_for_each_entry(lobj, head, list) {
-		/* only unreserve object we successfully reserved */
-		if (lobj->reserved && radeon_bo_is_reserved(lobj->bo))
-			radeon_bo_unreserve(lobj->bo);
+		list_add_tail(&lobj->tv.head, head);
 	}
 }
 
@@ -340,14 +317,11 @@ int radeon_bo_list_validate(struct list_head *head)
 	u32 domain;
 	int r;
 
-	list_for_each_entry(lobj, head, list) {
-		lobj->reserved = false;
-	}
-	r = radeon_bo_list_reserve(head);
+	r = ttm_eu_reserve_buffers(head);
 	if (unlikely(r != 0)) {
 		return r;
 	}
-	list_for_each_entry(lobj, head, list) {
+	list_for_each_entry(lobj, head, tv.head) {
 		bo = lobj->bo;
 		if (!bo->pin_count) {
 			domain = lobj->wdomain ? lobj->wdomain : lobj->rdomain;
@@ -370,25 +344,6 @@ int radeon_bo_list_validate(struct list_head *head)
 	return 0;
 }
 
-void radeon_bo_list_fence(struct list_head *head, void *fence)
-{
-	struct radeon_bo_list *lobj;
-	struct radeon_bo *bo;
-	struct radeon_fence *old_fence = NULL;
-
-	list_for_each_entry(lobj, head, list) {
-		bo = lobj->bo;
-		spin_lock(&bo->tbo.lock);
-		old_fence = (struct radeon_fence *)bo->tbo.sync_obj;
-		bo->tbo.sync_obj = radeon_fence_ref(fence);
-		bo->tbo.sync_obj_arg = NULL;
-		spin_unlock(&bo->tbo.lock);
-		if (old_fence) {
-			radeon_fence_unref(&old_fence);
-		}
-	}
-}
-
 int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
 			     struct vm_area_struct *vma)
 {
diff --git a/drivers/gpu/drm/radeon/radeon_object.h b/drivers/gpu/drm/radeon/radeon_object.h
index d143702b244a..22d4c237dea5 100644
--- a/drivers/gpu/drm/radeon/radeon_object.h
+++ b/drivers/gpu/drm/radeon/radeon_object.h
@@ -126,12 +126,12 @@ static inline int radeon_bo_wait(struct radeon_bo *bo, u32 *mem_type,
 	r = ttm_bo_reserve(&bo->tbo, true, no_wait, false, 0);
 	if (unlikely(r != 0))
 		return r;
-	spin_lock(&bo->tbo.lock);
+	spin_lock(&bo->tbo.bdev->fence_lock);
 	if (mem_type)
 		*mem_type = bo->tbo.mem.mem_type;
 	if (bo->tbo.sync_obj)
 		r = ttm_bo_wait(&bo->tbo, true, true, no_wait);
-	spin_unlock(&bo->tbo.lock);
+	spin_unlock(&bo->tbo.bdev->fence_lock);
 	ttm_bo_unreserve(&bo->tbo);
 	return r;
 }
@@ -152,10 +152,7 @@ extern int radeon_bo_init(struct radeon_device *rdev);
 extern void radeon_bo_fini(struct radeon_device *rdev);
 extern void radeon_bo_list_add_object(struct radeon_bo_list *lobj,
 				struct list_head *head);
-extern int radeon_bo_list_reserve(struct list_head *head);
-extern void radeon_bo_list_unreserve(struct list_head *head);
 extern int radeon_bo_list_validate(struct list_head *head);
-extern void radeon_bo_list_fence(struct list_head *head, void *fence);
 extern int radeon_bo_fbdev_mmap(struct radeon_bo *bo,
 				struct vm_area_struct *vma);
 extern int radeon_bo_set_tiling_flags(struct radeon_bo *bo,
diff --git a/drivers/gpu/drm/radeon/radeon_pm.c b/drivers/gpu/drm/radeon/radeon_pm.c
index 8c9b2ef32c68..4de7776bd1c5 100644
--- a/drivers/gpu/drm/radeon/radeon_pm.c
+++ b/drivers/gpu/drm/radeon/radeon_pm.c
@@ -449,6 +449,9 @@ static ssize_t radeon_hwmon_show_temp(struct device *dev,
 	case THERMAL_TYPE_EVERGREEN:
 		temp = evergreen_get_temp(rdev);
 		break;
+	case THERMAL_TYPE_SUMO:
+		temp = sumo_get_temp(rdev);
+		break;
 	default:
 		temp = 0;
 		break;
@@ -487,6 +490,7 @@ static int radeon_hwmon_init(struct radeon_device *rdev)
 	case THERMAL_TYPE_RV6XX:
 	case THERMAL_TYPE_RV770:
 	case THERMAL_TYPE_EVERGREEN:
+	case THERMAL_TYPE_SUMO:
 		rdev->pm.int_hwmon_dev = hwmon_device_register(rdev->dev);
 		if (IS_ERR(rdev->pm.int_hwmon_dev)) {
 			err = PTR_ERR(rdev->pm.int_hwmon_dev);
@@ -720,9 +724,9 @@ static bool radeon_pm_in_vbl(struct radeon_device *rdev)
 	 */
 	for (crtc = 0; (crtc < rdev->num_crtc) && in_vbl; crtc++) {
 		if (rdev->pm.active_crtcs & (1 << crtc)) {
-			vbl_status = radeon_get_crtc_scanoutpos(rdev, crtc, &vpos, &hpos);
-			if ((vbl_status & RADEON_SCANOUTPOS_VALID) &&
-			    !(vbl_status & RADEON_SCANOUTPOS_INVBL))
+			vbl_status = radeon_get_crtc_scanoutpos(rdev->ddev, crtc, &vpos, &hpos);
+			if ((vbl_status & DRM_SCANOUTPOS_VALID) &&
+			    !(vbl_status & DRM_SCANOUTPOS_INVBL))
 				in_vbl = false;
 		}
 	}
diff --git a/drivers/gpu/drm/radeon/radeon_reg.h b/drivers/gpu/drm/radeon/radeon_reg.h
index 64928814de53..0a310b7f71c3 100644
--- a/drivers/gpu/drm/radeon/radeon_reg.h
+++ b/drivers/gpu/drm/radeon/radeon_reg.h
@@ -422,6 +422,7 @@
 #       define RADEON_CRTC_CSYNC_EN         (1 <<  4)
 #       define RADEON_CRTC_ICON_EN          (1 << 15)
 #       define RADEON_CRTC_CUR_EN           (1 << 16)
+#       define RADEON_CRTC_VSTAT_MODE_MASK  (3 << 17)
 #       define RADEON_CRTC_CUR_MODE_MASK    (7 << 20)
 #       define RADEON_CRTC_CUR_MODE_SHIFT   20
 #       define RADEON_CRTC_CUR_MODE_MONO    0
@@ -509,6 +510,8 @@
 #       define RADEON_CRTC_TILE_EN                      (1 << 15)
 #       define RADEON_CRTC_OFFSET_FLIP_CNTL             (1 << 16)
 #       define RADEON_CRTC_STEREO_OFFSET_EN             (1 << 17)
+#       define RADEON_CRTC_GUI_TRIG_OFFSET_LEFT_EN      (1 << 28)
+#       define RADEON_CRTC_GUI_TRIG_OFFSET_RIGHT_EN     (1 << 29)
 
 #define R300_CRTC_TILE_X0_Y0	            0x0350
 #define R300_CRTC2_TILE_X0_Y0	            0x0358
diff --git a/drivers/gpu/drm/radeon/radeon_trace.h b/drivers/gpu/drm/radeon/radeon_trace.h
new file mode 100644
index 000000000000..eafd8160a155
--- /dev/null
+++ b/drivers/gpu/drm/radeon/radeon_trace.h
@@ -0,0 +1,82 @@
+#if !defined(_RADEON_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _RADEON_TRACE_H_
+
+#include <linux/stringify.h>
+#include <linux/types.h>
+#include <linux/tracepoint.h>
+
+#include <drm/drmP.h>
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM radeon
+#define TRACE_SYSTEM_STRING __stringify(TRACE_SYSTEM)
+#define TRACE_INCLUDE_FILE radeon_trace
+
+TRACE_EVENT(radeon_bo_create,
+	    TP_PROTO(struct radeon_bo *bo),
+	    TP_ARGS(bo),
+	    TP_STRUCT__entry(
+			     __field(struct radeon_bo *, bo)
+			     __field(u32, pages)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->bo = bo;
+			   __entry->pages = bo->tbo.num_pages;
+			   ),
+	    TP_printk("bo=%p, pages=%u", __entry->bo, __entry->pages)
+);
+
+DECLARE_EVENT_CLASS(radeon_fence_request,
+
+	    TP_PROTO(struct drm_device *dev, u32 seqno),
+
+	    TP_ARGS(dev, seqno),
+
+	    TP_STRUCT__entry(
+			     __field(u32, dev)
+			     __field(u32, seqno)
+			     ),
+
+	    TP_fast_assign(
+			   __entry->dev = dev->primary->index;
+			   __entry->seqno = seqno;
+			   ),
+
+	    TP_printk("dev=%u, seqno=%u", __entry->dev, __entry->seqno)
+);
+
+DEFINE_EVENT(radeon_fence_request, radeon_fence_emit,
+
+	    TP_PROTO(struct drm_device *dev, u32 seqno),
+
+	    TP_ARGS(dev, seqno)
+);
+
+DEFINE_EVENT(radeon_fence_request, radeon_fence_retire,
+
+	    TP_PROTO(struct drm_device *dev, u32 seqno),
+
+	    TP_ARGS(dev, seqno)
+);
+
+DEFINE_EVENT(radeon_fence_request, radeon_fence_wait_begin,
+
+	    TP_PROTO(struct drm_device *dev, u32 seqno),
+
+	    TP_ARGS(dev, seqno)
+);
+
+DEFINE_EVENT(radeon_fence_request, radeon_fence_wait_end,
+
+	    TP_PROTO(struct drm_device *dev, u32 seqno),
+
+	    TP_ARGS(dev, seqno)
+);
+
+#endif
+
+/* This part must be outside protection */
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#include <trace/define_trace.h>
diff --git a/drivers/gpu/drm/radeon/radeon_trace_points.c b/drivers/gpu/drm/radeon/radeon_trace_points.c
new file mode 100644
index 000000000000..8175993df84d
--- /dev/null
+++ b/drivers/gpu/drm/radeon/radeon_trace_points.c
@@ -0,0 +1,9 @@
+/* Copyright Red Hat Inc 2010.
+ * Author : Dave Airlie <airlied@redhat.com>
+ */
+#include <drm/drmP.h>
+#include "radeon_drm.h"
+#include "radeon.h"
+
+#define CREATE_TRACE_POINTS
+#include "radeon_trace.h"
diff --git a/drivers/gpu/drm/radeon/rs600.c b/drivers/gpu/drm/radeon/rs600.c
index f1c6e02c2e6b..9a85b1614c86 100644
--- a/drivers/gpu/drm/radeon/rs600.c
+++ b/drivers/gpu/drm/radeon/rs600.c
@@ -46,6 +46,56 @@
 void rs600_gpu_init(struct radeon_device *rdev);
 int rs600_mc_wait_for_idle(struct radeon_device *rdev);
 
+void rs600_pre_page_flip(struct radeon_device *rdev, int crtc)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc];
+	u32 tmp;
+
+	/* make sure flip is at vb rather than hb */
+	tmp = RREG32(AVIVO_D1GRPH_FLIP_CONTROL + radeon_crtc->crtc_offset);
+	tmp &= ~AVIVO_D1GRPH_SURFACE_UPDATE_H_RETRACE_EN;
+	WREG32(AVIVO_D1GRPH_FLIP_CONTROL + radeon_crtc->crtc_offset, tmp);
+
+	/* set pageflip to happen anywhere in vblank interval */
+	WREG32(AVIVO_D1MODE_MASTER_UPDATE_MODE + radeon_crtc->crtc_offset, 0);
+
+	/* enable the pflip int */
+	radeon_irq_kms_pflip_irq_get(rdev, crtc);
+}
+
+void rs600_post_page_flip(struct radeon_device *rdev, int crtc)
+{
+	/* disable the pflip int */
+	radeon_irq_kms_pflip_irq_put(rdev, crtc);
+}
+
+u32 rs600_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
+	u32 tmp = RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset);
+
+	/* Lock the graphics update lock */
+	tmp |= AVIVO_D1GRPH_UPDATE_LOCK;
+	WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* update the scanout addresses */
+	WREG32(AVIVO_D1GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+	WREG32(AVIVO_D1GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+
+	/* Wait for update_pending to go high. */
+	while (!(RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING));
+	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
+
+	/* Unlock the lock, so double-buffering can take place inside vblank */
+	tmp &= ~AVIVO_D1GRPH_UPDATE_LOCK;
+	WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* Return current update_pending status: */
+	return RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING;
+}
+
 void rs600_pm_misc(struct radeon_device *rdev)
 {
 	int requested_index = rdev->pm.requested_power_state_index;
@@ -515,10 +565,12 @@ int rs600_irq_set(struct radeon_device *rdev)
 	if (rdev->irq.gui_idle) {
 		tmp |= S_000040_GUI_IDLE(1);
 	}
-	if (rdev->irq.crtc_vblank_int[0]) {
+	if (rdev->irq.crtc_vblank_int[0] ||
+	    rdev->irq.pflip[0]) {
 		mode_int |= S_006540_D1MODE_VBLANK_INT_MASK(1);
 	}
-	if (rdev->irq.crtc_vblank_int[1]) {
+	if (rdev->irq.crtc_vblank_int[1] ||
+	    rdev->irq.pflip[1]) {
 		mode_int |= S_006540_D2MODE_VBLANK_INT_MASK(1);
 	}
 	if (rdev->irq.hpd[0]) {
@@ -534,7 +586,7 @@ int rs600_irq_set(struct radeon_device *rdev)
 	return 0;
 }
 
-static inline uint32_t rs600_irq_ack(struct radeon_device *rdev, u32 *r500_disp_int)
+static inline u32 rs600_irq_ack(struct radeon_device *rdev)
 {
 	uint32_t irqs = RREG32(R_000044_GEN_INT_STATUS);
 	uint32_t irq_mask = S_000044_SW_INT(1);
@@ -547,27 +599,27 @@ static inline uint32_t rs600_irq_ack(struct radeon_device *rdev, u32 *r500_disp_
 	}
 
 	if (G_000044_DISPLAY_INT_STAT(irqs)) {
-		*r500_disp_int = RREG32(R_007EDC_DISP_INTERRUPT_STATUS);
-		if (G_007EDC_LB_D1_VBLANK_INTERRUPT(*r500_disp_int)) {
+		rdev->irq.stat_regs.r500.disp_int = RREG32(R_007EDC_DISP_INTERRUPT_STATUS);
+		if (G_007EDC_LB_D1_VBLANK_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			WREG32(R_006534_D1MODE_VBLANK_STATUS,
 				S_006534_D1MODE_VBLANK_ACK(1));
 		}
-		if (G_007EDC_LB_D2_VBLANK_INTERRUPT(*r500_disp_int)) {
+		if (G_007EDC_LB_D2_VBLANK_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			WREG32(R_006D34_D2MODE_VBLANK_STATUS,
 				S_006D34_D2MODE_VBLANK_ACK(1));
 		}
-		if (G_007EDC_DC_HOT_PLUG_DETECT1_INTERRUPT(*r500_disp_int)) {
+		if (G_007EDC_DC_HOT_PLUG_DETECT1_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			tmp = RREG32(R_007D08_DC_HOT_PLUG_DETECT1_INT_CONTROL);
 			tmp |= S_007D08_DC_HOT_PLUG_DETECT1_INT_ACK(1);
 			WREG32(R_007D08_DC_HOT_PLUG_DETECT1_INT_CONTROL, tmp);
 		}
-		if (G_007EDC_DC_HOT_PLUG_DETECT2_INTERRUPT(*r500_disp_int)) {
+		if (G_007EDC_DC_HOT_PLUG_DETECT2_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			tmp = RREG32(R_007D18_DC_HOT_PLUG_DETECT2_INT_CONTROL);
 			tmp |= S_007D18_DC_HOT_PLUG_DETECT2_INT_ACK(1);
 			WREG32(R_007D18_DC_HOT_PLUG_DETECT2_INT_CONTROL, tmp);
 		}
 	} else {
-		*r500_disp_int = 0;
+		rdev->irq.stat_regs.r500.disp_int = 0;
 	}
 
 	if (irqs) {
@@ -578,32 +630,30 @@ static inline uint32_t rs600_irq_ack(struct radeon_device *rdev, u32 *r500_disp_
 
 void rs600_irq_disable(struct radeon_device *rdev)
 {
-	u32 tmp;
-
 	WREG32(R_000040_GEN_INT_CNTL, 0);
 	WREG32(R_006540_DxMODE_INT_MASK, 0);
 	/* Wait and acknowledge irq */
 	mdelay(1);
-	rs600_irq_ack(rdev, &tmp);
+	rs600_irq_ack(rdev);
 }
 
 int rs600_irq_process(struct radeon_device *rdev)
 {
-	uint32_t status, msi_rearm;
-	uint32_t r500_disp_int;
+	u32 status, msi_rearm;
 	bool queue_hotplug = false;
 
 	/* reset gui idle ack.  the status bit is broken */
 	rdev->irq.gui_idle_acked = false;
 
-	status = rs600_irq_ack(rdev, &r500_disp_int);
-	if (!status && !r500_disp_int) {
+	status = rs600_irq_ack(rdev);
+	if (!status && !rdev->irq.stat_regs.r500.disp_int) {
 		return IRQ_NONE;
 	}
-	while (status || r500_disp_int) {
+	while (status || rdev->irq.stat_regs.r500.disp_int) {
 		/* SW interrupt */
-		if (G_000044_SW_INT(status))
+		if (G_000044_SW_INT(status)) {
 			radeon_fence_process(rdev);
+		}
 		/* GUI idle */
 		if (G_000040_GUI_IDLE(status)) {
 			rdev->irq.gui_idle_acked = true;
@@ -611,25 +661,33 @@ int rs600_irq_process(struct radeon_device *rdev)
 			wake_up(&rdev->irq.idle_queue);
 		}
 		/* Vertical blank interrupts */
-		if (G_007EDC_LB_D1_VBLANK_INTERRUPT(r500_disp_int)) {
-			drm_handle_vblank(rdev->ddev, 0);
-			rdev->pm.vblank_sync = true;
-			wake_up(&rdev->irq.vblank_queue);
+		if (G_007EDC_LB_D1_VBLANK_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
+			if (rdev->irq.crtc_vblank_int[0]) {
+				drm_handle_vblank(rdev->ddev, 0);
+				rdev->pm.vblank_sync = true;
+				wake_up(&rdev->irq.vblank_queue);
+			}
+			if (rdev->irq.pflip[0])
+				radeon_crtc_handle_flip(rdev, 0);
 		}
-		if (G_007EDC_LB_D2_VBLANK_INTERRUPT(r500_disp_int)) {
-			drm_handle_vblank(rdev->ddev, 1);
-			rdev->pm.vblank_sync = true;
-			wake_up(&rdev->irq.vblank_queue);
+		if (G_007EDC_LB_D2_VBLANK_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
+			if (rdev->irq.crtc_vblank_int[1]) {
+				drm_handle_vblank(rdev->ddev, 1);
+				rdev->pm.vblank_sync = true;
+				wake_up(&rdev->irq.vblank_queue);
+			}
+			if (rdev->irq.pflip[1])
+				radeon_crtc_handle_flip(rdev, 1);
 		}
-		if (G_007EDC_DC_HOT_PLUG_DETECT1_INTERRUPT(r500_disp_int)) {
+		if (G_007EDC_DC_HOT_PLUG_DETECT1_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			queue_hotplug = true;
 			DRM_DEBUG("HPD1\n");
 		}
-		if (G_007EDC_DC_HOT_PLUG_DETECT2_INTERRUPT(r500_disp_int)) {
+		if (G_007EDC_DC_HOT_PLUG_DETECT2_INTERRUPT(rdev->irq.stat_regs.r500.disp_int)) {
 			queue_hotplug = true;
 			DRM_DEBUG("HPD2\n");
 		}
-		status = rs600_irq_ack(rdev, &r500_disp_int);
+		status = rs600_irq_ack(rdev);
 	}
 	/* reset gui idle ack.  the status bit is broken */
 	rdev->irq.gui_idle_acked = false;
diff --git a/drivers/gpu/drm/radeon/rv770.c b/drivers/gpu/drm/radeon/rv770.c
index 4dfead8cee33..645aa1fd7611 100644
--- a/drivers/gpu/drm/radeon/rv770.c
+++ b/drivers/gpu/drm/radeon/rv770.c
@@ -42,6 +42,40 @@
 static void rv770_gpu_init(struct radeon_device *rdev);
 void rv770_fini(struct radeon_device *rdev);
 
+u32 rv770_page_flip(struct radeon_device *rdev, int crtc_id, u64 crtc_base)
+{
+	struct radeon_crtc *radeon_crtc = rdev->mode_info.crtcs[crtc_id];
+	u32 tmp = RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset);
+
+	/* Lock the graphics update lock */
+	tmp |= AVIVO_D1GRPH_UPDATE_LOCK;
+	WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* update the scanout addresses */
+	if (radeon_crtc->crtc_id) {
+		WREG32(D2GRPH_SECONDARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base));
+		WREG32(D2GRPH_PRIMARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base));
+	} else {
+		WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base));
+		WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS_HIGH, upper_32_bits(crtc_base));
+	}
+	WREG32(D1GRPH_SECONDARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+	WREG32(D1GRPH_PRIMARY_SURFACE_ADDRESS + radeon_crtc->crtc_offset,
+	       (u32)crtc_base);
+
+	/* Wait for update_pending to go high. */
+	while (!(RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING));
+	DRM_DEBUG("Update pending now high. Unlocking vupdate_lock.\n");
+
+	/* Unlock the lock, so double-buffering can take place inside vblank */
+	tmp &= ~AVIVO_D1GRPH_UPDATE_LOCK;
+	WREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset, tmp);
+
+	/* Return current update_pending status: */
+	return RREG32(AVIVO_D1GRPH_UPDATE + radeon_crtc->crtc_offset) & AVIVO_D1GRPH_SURFACE_UPDATE_PENDING;
+}
+
 /* get temperature in millidegrees */
 u32 rv770_get_temp(struct radeon_device *rdev)
 {
@@ -489,6 +523,49 @@ static u32 r700_get_tile_pipe_to_backend_map(struct radeon_device *rdev,
 	return backend_map;
 }
 
+static void rv770_program_channel_remap(struct radeon_device *rdev)
+{
+	u32 tcp_chan_steer, mc_shared_chremap, tmp;
+	bool force_no_swizzle;
+
+	switch (rdev->family) {
+	case CHIP_RV770:
+	case CHIP_RV730:
+		force_no_swizzle = false;
+		break;
+	case CHIP_RV710:
+	case CHIP_RV740:
+	default:
+		force_no_swizzle = true;
+		break;
+	}
+
+	tmp = RREG32(MC_SHARED_CHMAP);
+	switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
+	case 0:
+	case 1:
+	default:
+		/* default mapping */
+		mc_shared_chremap = 0x00fac688;
+		break;
+	case 2:
+	case 3:
+		if (force_no_swizzle)
+			mc_shared_chremap = 0x00fac688;
+		else
+			mc_shared_chremap = 0x00bbc298;
+		break;
+	}
+
+	if (rdev->family == CHIP_RV740)
+		tcp_chan_steer = 0x00ef2a60;
+	else
+		tcp_chan_steer = 0x00fac688;
+
+	WREG32(TCP_CHAN_STEER, tcp_chan_steer);
+	WREG32(MC_SHARED_CHREMAP, mc_shared_chremap);
+}
+
 static void rv770_gpu_init(struct radeon_device *rdev)
 {
 	int i, j, num_qd_pipes;
@@ -688,6 +765,8 @@ static void rv770_gpu_init(struct radeon_device *rdev)
 	WREG32(DCP_TILING_CONFIG, (gb_tiling_config & 0xffff));
 	WREG32(HDP_TILING_CONFIG, (gb_tiling_config & 0xffff));
 
+	rv770_program_channel_remap(rdev);
+
 	WREG32(CC_RB_BACKEND_DISABLE,      cc_rb_backend_disable);
 	WREG32(CC_GC_SHADER_PIPE_CONFIG,   cc_gc_shader_pipe_config);
 	WREG32(GC_USER_SHADER_PIPE_CONFIG, cc_gc_shader_pipe_config);
@@ -956,6 +1035,45 @@ static void rv770_vram_scratch_fini(struct radeon_device *rdev)
 	radeon_bo_unref(&rdev->vram_scratch.robj);
 }
 
+void r700_vram_gtt_location(struct radeon_device *rdev, struct radeon_mc *mc)
+{
+	u64 size_bf, size_af;
+
+	if (mc->mc_vram_size > 0xE0000000) {
+		/* leave room for at least 512M GTT */
+		dev_warn(rdev->dev, "limiting VRAM\n");
+		mc->real_vram_size = 0xE0000000;
+		mc->mc_vram_size = 0xE0000000;
+	}
+	if (rdev->flags & RADEON_IS_AGP) {
+		size_bf = mc->gtt_start;
+		size_af = 0xFFFFFFFF - mc->gtt_end + 1;
+		if (size_bf > size_af) {
+			if (mc->mc_vram_size > size_bf) {
+				dev_warn(rdev->dev, "limiting VRAM\n");
+				mc->real_vram_size = size_bf;
+				mc->mc_vram_size = size_bf;
+			}
+			mc->vram_start = mc->gtt_start - mc->mc_vram_size;
+		} else {
+			if (mc->mc_vram_size > size_af) {
+				dev_warn(rdev->dev, "limiting VRAM\n");
+				mc->real_vram_size = size_af;
+				mc->mc_vram_size = size_af;
+			}
+			mc->vram_start = mc->gtt_end;
+		}
+		mc->vram_end = mc->vram_start + mc->mc_vram_size - 1;
+		dev_info(rdev->dev, "VRAM: %lluM 0x%08llX - 0x%08llX (%lluM used)\n",
+				mc->mc_vram_size >> 20, mc->vram_start,
+				mc->vram_end, mc->real_vram_size >> 20);
+	} else {
+		radeon_vram_location(rdev, &rdev->mc, 0);
+		rdev->mc.gtt_base_align = 0;
+		radeon_gtt_location(rdev, mc);
+	}
+}
+
 int rv770_mc_init(struct radeon_device *rdev)
 {
 	u32 tmp;
@@ -996,7 +1114,7 @@ int rv770_mc_init(struct radeon_device *rdev)
 	rdev->mc.real_vram_size = RREG32(CONFIG_MEMSIZE);
 	rdev->mc.visible_vram_size = rdev->mc.aper_size;
 	rdev->mc.active_vram_size = rdev->mc.visible_vram_size;
-	r600_vram_gtt_location(rdev, &rdev->mc);
+	r700_vram_gtt_location(rdev, &rdev->mc);
 	radeon_update_bandwidth_info(rdev);
 
 	return 0;
diff --git a/drivers/gpu/drm/radeon/rv770d.h b/drivers/gpu/drm/radeon/rv770d.h
index b7a5a20e81dc..fc77e1e1a179 100644
--- a/drivers/gpu/drm/radeon/rv770d.h
+++ b/drivers/gpu/drm/radeon/rv770d.h
@@ -138,6 +138,7 @@
 #define MC_SHARED_CHMAP						0x2004
 #define		NOOFCHAN_SHIFT					12
 #define		NOOFCHAN_MASK					0x00003000
+#define MC_SHARED_CHREMAP					0x2008
 
 #define	MC_ARB_RAMCFG					0x2760
 #define		NOOFBANK_SHIFT					0
@@ -303,6 +304,7 @@
 #define		BILINEAR_PRECISION_8_BIT			(1 << 31)
 
 #define	TCP_CNTL					0x9610
+#define	TCP_CHAN_STEER					0x9614
 
 #define	VGT_CACHE_INVALIDATION				0x88C4
 #define		CACHE_INVALIDATION(x)				((x)<<0)
@@ -351,4 +353,11 @@
 
 #define	SRBM_STATUS				        0x0E50
 
+#define D1GRPH_PRIMARY_SURFACE_ADDRESS                    0x6110
+#define D1GRPH_PRIMARY_SURFACE_ADDRESS_HIGH               0x6914
+#define D2GRPH_PRIMARY_SURFACE_ADDRESS_HIGH               0x6114
+#define D1GRPH_SECONDARY_SURFACE_ADDRESS                  0x6118
+#define D1GRPH_SECONDARY_SURFACE_ADDRESS_HIGH             0x691c
+#define D2GRPH_SECONDARY_SURFACE_ADDRESS_HIGH             0x611c
+
 #endif
diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 148a322d8f5d..cf2ec562550e 100644
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -169,7 +169,7 @@ int ttm_bo_wait_unreserved(struct ttm_buffer_object *bo, bool interruptible)
 }
 EXPORT_SYMBOL(ttm_bo_wait_unreserved);
 
-static void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
+void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_mem_type_manager *man;
@@ -191,11 +191,7 @@ static void ttm_bo_add_to_lru(struct ttm_buffer_object *bo)
 	}
 }
 
-/**
- * Call with the lru_lock held.
- */
-
-static int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
+int ttm_bo_del_from_lru(struct ttm_buffer_object *bo)
 {
 	int put_count = 0;
 
@@ -227,9 +223,18 @@ int ttm_bo_reserve_locked(struct ttm_buffer_object *bo,
 		/**
 		 * Deadlock avoidance for multi-bo reserving.
 		 */
-		if (use_sequence && bo->seq_valid &&
-			(sequence - bo->val_seq < (1 << 31))) {
-			return -EAGAIN;
+		if (use_sequence && bo->seq_valid) {
+			/**
+			 * We've already reserved this one.
+			 */
+			if (unlikely(sequence == bo->val_seq))
+				return -EDEADLK;
+			/**
+			 * Already reserved by a thread that will not back
+			 * off for us. We need to back off.
+			 */
+			if (unlikely(sequence - bo->val_seq < (1 << 31)))
+				return -EAGAIN;
 		}
 
 		if (no_wait)
@@ -267,6 +272,13 @@ static void ttm_bo_ref_bug(struct kref *list_kref)
 	BUG();
 }
 
+void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count,
+			 bool never_free)
+{
+	kref_sub(&bo->list_kref, count,
+		 (never_free) ? ttm_bo_ref_bug : ttm_bo_release_list);
+}
+
 int ttm_bo_reserve(struct ttm_buffer_object *bo,
 		   bool interruptible,
 		   bool no_wait, bool use_sequence, uint32_t sequence)
@@ -282,20 +294,24 @@ int ttm_bo_reserve(struct ttm_buffer_object *bo,
 		put_count = ttm_bo_del_from_lru(bo);
 	spin_unlock(&glob->lru_lock);
 
-	while (put_count--)
-		kref_put(&bo->list_kref, ttm_bo_ref_bug);
+	ttm_bo_list_ref_sub(bo, put_count, true);
 
 	return ret;
 }
 
+void ttm_bo_unreserve_locked(struct ttm_buffer_object *bo)
+{
+	ttm_bo_add_to_lru(bo);
+	atomic_set(&bo->reserved, 0);
+	wake_up_all(&bo->event_queue);
+}
+
 void ttm_bo_unreserve(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_global *glob = bo->glob;
 
 	spin_lock(&glob->lru_lock);
-	ttm_bo_add_to_lru(bo);
-	atomic_set(&bo->reserved, 0);
-	wake_up_all(&bo->event_queue);
+	ttm_bo_unreserve_locked(bo);
 	spin_unlock(&glob->lru_lock);
 }
 EXPORT_SYMBOL(ttm_bo_unreserve);
@@ -362,8 +378,13 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 	int ret = 0;
 
 	if (old_is_pci || new_is_pci ||
-	    ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0))
-		ttm_bo_unmap_virtual(bo);
+	    ((mem->placement & bo->mem.placement & TTM_PL_MASK_CACHING) == 0)) {
+		ret = ttm_mem_io_lock(old_man, true);
+		if (unlikely(ret != 0))
+			goto out_err;
+		ttm_bo_unmap_virtual_locked(bo);
+		ttm_mem_io_unlock(old_man);
+	}
 
 	/*
 	 * Create and bind a ttm if required.
@@ -416,11 +437,9 @@ moved:
 	}
 
 	if (bo->mem.mm_node) {
-		spin_lock(&bo->lock);
 		bo->offset = (bo->mem.start << PAGE_SHIFT) +
 		    bdev->man[bo->mem.mem_type].gpu_offset;
 		bo->cur_placement = bo->mem.placement;
-		spin_unlock(&bo->lock);
 	} else
 		bo->offset = 0;
 
@@ -452,7 +471,6 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
 		ttm_tt_destroy(bo->ttm);
 		bo->ttm = NULL;
 	}
-
 	ttm_bo_mem_put(bo, &bo->mem);
 
 	atomic_set(&bo->reserved, 0);
@@ -474,14 +492,14 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 	int put_count;
 	int ret;
 
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	(void) ttm_bo_wait(bo, false, false, true);
 	if (!bo->sync_obj) {
 
 		spin_lock(&glob->lru_lock);
 
 		/**
-		 * Lock inversion between bo::reserve and bo::lock here,
+		 * Lock inversion between bo:reserve and bdev::fence_lock here,
 		 * but that's OK, since we're only trylocking.
 		 */
 
@@ -490,14 +508,13 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
 		if (unlikely(ret == -EBUSY))
 			goto queue;
 
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 		put_count = ttm_bo_del_from_lru(bo);
 
 		spin_unlock(&glob->lru_lock);
 		ttm_bo_cleanup_memtype_use(bo);
 
-		while (put_count--)
-			kref_put(&bo->list_kref, ttm_bo_ref_bug);
+		ttm_bo_list_ref_sub(bo, put_count, true);
 
 		return;
 	} else {
@@ -512,7 +529,7 @@ queue:
 	kref_get(&bo->list_kref);
 	list_add_tail(&bo->ddestroy, &bdev->ddestroy);
 	spin_unlock(&glob->lru_lock);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bdev->fence_lock);
 
 	if (sync_obj) {
 		driver->sync_obj_flush(sync_obj, sync_obj_arg);
@@ -537,14 +554,15 @@ static int ttm_bo_cleanup_refs(struct ttm_buffer_object *bo,
 			       bool no_wait_reserve,
 			       bool no_wait_gpu)
 {
+	struct ttm_bo_device *bdev = bo->bdev;
 	struct ttm_bo_global *glob = bo->glob;
 	int put_count;
 	int ret = 0;
 
 retry:
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bdev->fence_lock);
 
 	if (unlikely(ret != 0))
 		return ret;
@@ -580,8 +598,7 @@ retry:
 	spin_unlock(&glob->lru_lock);
 	ttm_bo_cleanup_memtype_use(bo);
 
-	while (put_count--)
-		kref_put(&bo->list_kref, ttm_bo_ref_bug);
+	ttm_bo_list_ref_sub(bo, put_count, true);
 
 	return 0;
 }
@@ -652,6 +669,7 @@ static void ttm_bo_release(struct kref *kref)
 	struct ttm_buffer_object *bo =
 	    container_of(kref, struct ttm_buffer_object, kref);
 	struct ttm_bo_device *bdev = bo->bdev;
+	struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
 
 	if (likely(bo->vm_node != NULL)) {
 		rb_erase(&bo->vm_rb, &bdev->addr_space_rb);
@@ -659,6 +677,9 @@ static void ttm_bo_release(struct kref *kref)
 		bo->vm_node = NULL;
 	}
 	write_unlock(&bdev->vm_lock);
+	ttm_mem_io_lock(man, false);
+	ttm_mem_io_free_vm(bo);
+	ttm_mem_io_unlock(man);
 	ttm_bo_cleanup_refs_or_queue(bo);
 	kref_put(&bo->list_kref, ttm_bo_release_list);
 	write_lock(&bdev->vm_lock);
@@ -698,9 +719,9 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 	struct ttm_placement placement;
 	int ret = 0;
 
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bdev->fence_lock);
 
 	if (unlikely(ret != 0)) {
 		if (ret != -ERESTARTSYS) {
@@ -715,7 +736,8 @@ static int ttm_bo_evict(struct ttm_buffer_object *bo, bool interruptible,
 
 	evict_mem = bo->mem;
 	evict_mem.mm_node = NULL;
-	evict_mem.bus.io_reserved = false;
+	evict_mem.bus.io_reserved_vm = false;
+	evict_mem.bus.io_reserved_count = 0;
 
 	placement.fpfn = 0;
 	placement.lpfn = 0;
@@ -802,8 +824,7 @@ retry:
 
 	BUG_ON(ret != 0);
 
-	while (put_count--)
-		kref_put(&bo->list_kref, ttm_bo_ref_bug);
+	ttm_bo_list_ref_sub(bo, put_count, true);
 
 	ret = ttm_bo_evict(bo, interruptible, no_wait_reserve, no_wait_gpu);
 	ttm_bo_unreserve(bo);
@@ -1036,6 +1057,7 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 {
 	int ret = 0;
 	struct ttm_mem_reg mem;
+	struct ttm_bo_device *bdev = bo->bdev;
 
 	BUG_ON(!atomic_read(&bo->reserved));
 
@@ -1044,15 +1066,16 @@ int ttm_bo_move_buffer(struct ttm_buffer_object *bo,
 	 * Have the driver move function wait for idle when necessary,
 	 * instead of doing it here.
 	 */
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, interruptible, no_wait_gpu);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bdev->fence_lock);
 	if (ret)
 		return ret;
 	mem.num_pages = bo->num_pages;
 	mem.size = mem.num_pages << PAGE_SHIFT;
 	mem.page_alignment = bo->mem.page_alignment;
-	mem.bus.io_reserved = false;
+	mem.bus.io_reserved_vm = false;
+	mem.bus.io_reserved_count = 0;
 	/*
 	 * Determine where to move the buffer.
 	 */
@@ -1163,7 +1186,6 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	}
 	bo->destroy = destroy;
 
-	spin_lock_init(&bo->lock);
 	kref_init(&bo->kref);
 	kref_init(&bo->list_kref);
 	atomic_set(&bo->cpu_writers, 0);
@@ -1172,6 +1194,7 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	INIT_LIST_HEAD(&bo->lru);
 	INIT_LIST_HEAD(&bo->ddestroy);
 	INIT_LIST_HEAD(&bo->swap);
+	INIT_LIST_HEAD(&bo->io_reserve_lru);
 	bo->bdev = bdev;
 	bo->glob = bdev->glob;
 	bo->type = type;
@@ -1181,7 +1204,8 @@ int ttm_bo_init(struct ttm_bo_device *bdev,
 	bo->mem.num_pages = bo->num_pages;
 	bo->mem.mm_node = NULL;
 	bo->mem.page_alignment = page_alignment;
-	bo->mem.bus.io_reserved = false;
+	bo->mem.bus.io_reserved_vm = false;
+	bo->mem.bus.io_reserved_count = 0;
 	bo->buffer_start = buffer_start & PAGE_MASK;
 	bo->priv_flags = 0;
 	bo->mem.placement = (TTM_PL_FLAG_SYSTEM | TTM_PL_FLAG_CACHED);
@@ -1355,6 +1379,10 @@ int ttm_bo_init_mm(struct ttm_bo_device *bdev, unsigned type,
 	BUG_ON(type >= TTM_NUM_MEM_TYPES);
 	man = &bdev->man[type];
 	BUG_ON(man->has_type);
+	man->io_reserve_fastpath = true;
+	man->use_io_reserve_lru = false;
+	mutex_init(&man->io_reserve_mutex);
+	INIT_LIST_HEAD(&man->io_reserve_lru);
 
 	ret = bdev->driver->init_mem_type(bdev, type, man);
 	if (ret)
@@ -1527,7 +1555,8 @@ int ttm_bo_device_init(struct ttm_bo_device *bdev,
 	bdev->dev_mapping = NULL;
 	bdev->glob = glob;
 	bdev->need_dma32 = need_dma32;
-
+	bdev->val_seq = 0;
+	spin_lock_init(&bdev->fence_lock);
 	mutex_lock(&glob->device_list_mutex);
 	list_add_tail(&bdev->device_list, &glob->device_list);
 	mutex_unlock(&glob->device_list_mutex);
@@ -1561,7 +1590,7 @@ bool ttm_mem_reg_is_pci(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
 	return true;
 }
 
-void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
+void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo)
 {
 	struct ttm_bo_device *bdev = bo->bdev;
 	loff_t offset = (loff_t) bo->addr_space_offset;
@@ -1570,8 +1599,20 @@ void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
 	if (!bdev->dev_mapping)
 		return;
 	unmap_mapping_range(bdev->dev_mapping, offset, holelen, 1);
-	ttm_mem_io_free(bdev, &bo->mem);
+	ttm_mem_io_free_vm(bo);
+}
+
+void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo)
+{
+	struct ttm_bo_device *bdev = bo->bdev;
+	struct ttm_mem_type_manager *man = &bdev->man[bo->mem.mem_type];
+
+	ttm_mem_io_lock(man, false);
+	ttm_bo_unmap_virtual_locked(bo);
+	ttm_mem_io_unlock(man);
 }
+
+
 EXPORT_SYMBOL(ttm_bo_unmap_virtual);
 
 static void ttm_bo_vm_insert_rb(struct ttm_buffer_object *bo)
@@ -1651,6 +1692,7 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
 		bool lazy, bool interruptible, bool no_wait)
 {
 	struct ttm_bo_driver *driver = bo->bdev->driver;
+	struct ttm_bo_device *bdev = bo->bdev;
 	void *sync_obj;
 	void *sync_obj_arg;
 	int ret = 0;
@@ -1664,9 +1706,9 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
 			void *tmp_obj = bo->sync_obj;
 			bo->sync_obj = NULL;
 			clear_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
-			spin_unlock(&bo->lock);
+			spin_unlock(&bdev->fence_lock);
 			driver->sync_obj_unref(&tmp_obj);
-			spin_lock(&bo->lock);
+			spin_lock(&bdev->fence_lock);
 			continue;
 		}
 
@@ -1675,29 +1717,29 @@ int ttm_bo_wait(struct ttm_buffer_object *bo,
 
 		sync_obj = driver->sync_obj_ref(bo->sync_obj);
 		sync_obj_arg = bo->sync_obj_arg;
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 		ret = driver->sync_obj_wait(sync_obj, sync_obj_arg,
 					    lazy, interruptible);
 		if (unlikely(ret != 0)) {
 			driver->sync_obj_unref(&sync_obj);
-			spin_lock(&bo->lock);
+			spin_lock(&bdev->fence_lock);
 			return ret;
 		}
-		spin_lock(&bo->lock);
+		spin_lock(&bdev->fence_lock);
 		if (likely(bo->sync_obj == sync_obj &&
 			   bo->sync_obj_arg == sync_obj_arg)) {
 			void *tmp_obj = bo->sync_obj;
 			bo->sync_obj = NULL;
 			clear_bit(TTM_BO_PRIV_FLAG_MOVING,
 				  &bo->priv_flags);
-			spin_unlock(&bo->lock);
+			spin_unlock(&bdev->fence_lock);
 			driver->sync_obj_unref(&sync_obj);
 			driver->sync_obj_unref(&tmp_obj);
-			spin_lock(&bo->lock);
+			spin_lock(&bdev->fence_lock);
 		} else {
-			spin_unlock(&bo->lock);
+			spin_unlock(&bdev->fence_lock);
 			driver->sync_obj_unref(&sync_obj);
-			spin_lock(&bo->lock);
+			spin_lock(&bdev->fence_lock);
 		}
 	}
 	return 0;
@@ -1706,6 +1748,7 @@ EXPORT_SYMBOL(ttm_bo_wait);
 
 int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
 {
+	struct ttm_bo_device *bdev = bo->bdev;
 	int ret = 0;
 
 	/*
@@ -1715,9 +1758,9 @@ int ttm_bo_synccpu_write_grab(struct ttm_buffer_object *bo, bool no_wait)
 	ret = ttm_bo_reserve(bo, true, no_wait, false, 0);
 	if (unlikely(ret != 0))
 		return ret;
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, true, no_wait);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bdev->fence_lock);
 	if (likely(ret == 0))
 		atomic_inc(&bo->cpu_writers);
 	ttm_bo_unreserve(bo);
@@ -1783,16 +1826,15 @@ static int ttm_bo_swapout(struct ttm_mem_shrink *shrink)
 	put_count = ttm_bo_del_from_lru(bo);
 	spin_unlock(&glob->lru_lock);
 
-	while (put_count--)
-		kref_put(&bo->list_kref, ttm_bo_ref_bug);
+	ttm_bo_list_ref_sub(bo, put_count, true);
 
 	/**
 	 * Wait for GPU, then move to system cached.
 	 */
 
-	spin_lock(&bo->lock);
+	spin_lock(&bo->bdev->fence_lock);
 	ret = ttm_bo_wait(bo, false, false, false);
-	spin_unlock(&bo->lock);
+	spin_unlock(&bo->bdev->fence_lock);
 
 	if (unlikely(ret != 0))
 		goto out;
diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index 3106d5bcce32..77dbf408c0d0 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -75,37 +75,123 @@ int ttm_bo_move_ttm(struct ttm_buffer_object *bo,
 }
 EXPORT_SYMBOL(ttm_bo_move_ttm);
 
-int ttm_mem_io_reserve(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+int ttm_mem_io_lock(struct ttm_mem_type_manager *man, bool interruptible)
 {
-	int ret;
+	if (likely(man->io_reserve_fastpath))
+		return 0;
+
+	if (interruptible)
+		return mutex_lock_interruptible(&man->io_reserve_mutex);
+
+	mutex_lock(&man->io_reserve_mutex);
+	return 0;
+}
 
-	if (!mem->bus.io_reserved) {
-		mem->bus.io_reserved = true;
+void ttm_mem_io_unlock(struct ttm_mem_type_manager *man)
+{
+	if (likely(man->io_reserve_fastpath))
+		return;
+
+	mutex_unlock(&man->io_reserve_mutex);
+}
+
+static int ttm_mem_io_evict(struct ttm_mem_type_manager *man)
+{
+	struct ttm_buffer_object *bo;
+
+	if (!man->use_io_reserve_lru || list_empty(&man->io_reserve_lru))
+		return -EAGAIN;
+
+	bo = list_first_entry(&man->io_reserve_lru,
+			      struct ttm_buffer_object,
+			      io_reserve_lru);
+	list_del_init(&bo->io_reserve_lru);
+	ttm_bo_unmap_virtual_locked(bo);
+
+	return 0;
+}
+
+static int ttm_mem_io_reserve(struct ttm_bo_device *bdev,
+			      struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+	int ret = 0;
+
+	if (!bdev->driver->io_mem_reserve)
+		return 0;
+	if (likely(man->io_reserve_fastpath))
+		return bdev->driver->io_mem_reserve(bdev, mem);
+
+	if (bdev->driver->io_mem_reserve &&
+	    mem->bus.io_reserved_count++ == 0) {
+retry:
 		ret = bdev->driver->io_mem_reserve(bdev, mem);
+		if (ret == -EAGAIN) {
+			ret = ttm_mem_io_evict(man);
+			if (ret == 0)
+				goto retry;
+		}
+	}
+	return ret;
+}
+
+static void ttm_mem_io_free(struct ttm_bo_device *bdev,
+			    struct ttm_mem_reg *mem)
+{
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
+
+	if (likely(man->io_reserve_fastpath))
+		return;
+
+	if (bdev->driver->io_mem_reserve &&
+	    --mem->bus.io_reserved_count == 0 &&
+	    bdev->driver->io_mem_free)
+		bdev->driver->io_mem_free(bdev, mem);
+
+}
+
+int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo)
+{
+	struct ttm_mem_reg *mem = &bo->mem;
+	int ret;
+
+	if (!mem->bus.io_reserved_vm) {
+		struct ttm_mem_type_manager *man =
+			&bo->bdev->man[mem->mem_type];
+
+		ret = ttm_mem_io_reserve(bo->bdev, mem);
 		if (unlikely(ret != 0))
 			return ret;
+		mem->bus.io_reserved_vm = true;
+		if (man->use_io_reserve_lru)
+			list_add_tail(&bo->io_reserve_lru,
+				      &man->io_reserve_lru);
 	}
 	return 0;
 }
 
-void ttm_mem_io_free(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem)
+void ttm_mem_io_free_vm(struct ttm_buffer_object *bo)
 {
-	if (bdev->driver->io_mem_reserve) {
-		if (mem->bus.io_reserved) {
-			mem->bus.io_reserved = false;
-			bdev->driver->io_mem_free(bdev, mem);
-		}
+	struct ttm_mem_reg *mem = &bo->mem;
+
+	if (mem->bus.io_reserved_vm) {
+		mem->bus.io_reserved_vm = false;
+		list_del_init(&bo->io_reserve_lru);
+		ttm_mem_io_free(bo->bdev, mem);
 	}
 }
 
 int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 			void **virtual)
 {
+	struct ttm_mem_type_manager *man = &bdev->man[mem->mem_type];
 	int ret;
 	void *addr;
 
 	*virtual = NULL;
+	(void) ttm_mem_io_lock(man, false);
 	ret = ttm_mem_io_reserve(bdev, mem);
+	ttm_mem_io_unlock(man);
 	if (ret || !mem->bus.is_iomem)
 		return ret;
 
@@ -117,7 +203,9 @@ int ttm_mem_reg_ioremap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 		else
 			addr = ioremap_nocache(mem->bus.base + mem->bus.offset, mem->bus.size);
 		if (!addr) {
+			(void) ttm_mem_io_lock(man, false);
 			ttm_mem_io_free(bdev, mem);
+			ttm_mem_io_unlock(man);
 			return -ENOMEM;
 		}
 	}
@@ -134,7 +222,9 @@ void ttm_mem_reg_iounmap(struct ttm_bo_device *bdev, struct ttm_mem_reg *mem,
 
 	if (virtual && mem->bus.addr == NULL)
 		iounmap(virtual);
+	(void) ttm_mem_io_lock(man, false);
 	ttm_mem_io_free(bdev, mem);
+	ttm_mem_io_unlock(man);
 }
 
 static int ttm_copy_io_page(void *dst, void *src, unsigned long page)
@@ -231,7 +321,7 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 	struct ttm_mem_type_manager *man = &bdev->man[new_mem->mem_type];
 	struct ttm_tt *ttm = bo->ttm;
 	struct ttm_mem_reg *old_mem = &bo->mem;
-	struct ttm_mem_reg old_copy = *old_mem;
+	struct ttm_mem_reg old_copy;
 	void *old_iomap;
 	void *new_iomap;
 	int ret;
@@ -280,8 +370,7 @@ int ttm_bo_move_memcpy(struct ttm_buffer_object *bo,
 	}
 	mb();
 out2:
-	ttm_bo_free_old_node(bo);
-
+	old_copy = *old_mem;
 	*old_mem = *new_mem;
 	new_mem->mm_node = NULL;
 
@@ -292,9 +381,10 @@ out2:
 	}
 
 out1:
-	ttm_mem_reg_iounmap(bdev, new_mem, new_iomap);
+	ttm_mem_reg_iounmap(bdev, old_mem, new_iomap);
 out:
 	ttm_mem_reg_iounmap(bdev, &old_copy, old_iomap);
+	ttm_bo_mem_put(bo, &old_copy);
 	return ret;
 }
 EXPORT_SYMBOL(ttm_bo_move_memcpy);
@@ -337,11 +427,11 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 	 * TODO: Explicit member copy would probably be better here.
 	 */
 
-	spin_lock_init(&fbo->lock);
 	init_waitqueue_head(&fbo->event_queue);
 	INIT_LIST_HEAD(&fbo->ddestroy);
 	INIT_LIST_HEAD(&fbo->lru);
 	INIT_LIST_HEAD(&fbo->swap);
+	INIT_LIST_HEAD(&fbo->io_reserve_lru);
 	fbo->vm_node = NULL;
 	atomic_set(&fbo->cpu_writers, 0);
 
@@ -453,6 +543,8 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
 		unsigned long start_page, unsigned long num_pages,
 		struct ttm_bo_kmap_obj *map)
 {
+	struct ttm_mem_type_manager *man =
+		&bo->bdev->man[bo->mem.mem_type];
 	unsigned long offset, size;
 	int ret;
 
@@ -467,7 +559,9 @@ int ttm_bo_kmap(struct ttm_buffer_object *bo,
 	if (num_pages > 1 && !DRM_SUSER(DRM_CURPROC))
 		return -EPERM;
 #endif
+	(void) ttm_mem_io_lock(man, false);
 	ret = ttm_mem_io_reserve(bo->bdev, &bo->mem);
+	ttm_mem_io_unlock(man);
 	if (ret)
 		return ret;
 	if (!bo->mem.bus.is_iomem) {
@@ -482,12 +576,15 @@ EXPORT_SYMBOL(ttm_bo_kmap);
 
 void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 {
+	struct ttm_buffer_object *bo = map->bo;
+	struct ttm_mem_type_manager *man =
+		&bo->bdev->man[bo->mem.mem_type];
+
 	if (!map->virtual)
 		return;
 	switch (map->bo_kmap_type) {
 	case ttm_bo_map_iomap:
 		iounmap(map->virtual);
-		ttm_mem_io_free(map->bo->bdev, &map->bo->mem);
 		break;
 	case ttm_bo_map_vmap:
 		vunmap(map->virtual);
@@ -500,6 +597,9 @@ void ttm_bo_kunmap(struct ttm_bo_kmap_obj *map)
 	default:
 		BUG();
 	}
+	(void) ttm_mem_io_lock(man, false);
+	ttm_mem_io_free(map->bo->bdev, &map->bo->mem);
+	ttm_mem_io_unlock(man);
 	map->virtual = NULL;
 	map->page = NULL;
 }
@@ -520,7 +620,7 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 	struct ttm_buffer_object *ghost_obj;
 	void *tmp_obj = NULL;
 
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	if (bo->sync_obj) {
 		tmp_obj = bo->sync_obj;
 		bo->sync_obj = NULL;
@@ -529,7 +629,7 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 	bo->sync_obj_arg = sync_obj_arg;
 	if (evict) {
 		ret = ttm_bo_wait(bo, false, false, false);
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
 		if (ret)
@@ -552,7 +652,7 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 		 */
 
 		set_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags);
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
 
diff --git a/drivers/gpu/drm/ttm/ttm_bo_vm.c b/drivers/gpu/drm/ttm/ttm_bo_vm.c
index fe6cb77899f4..221b924acebe 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_vm.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_vm.c
@@ -83,6 +83,8 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	int i;
 	unsigned long address = (unsigned long)vmf->virtual_address;
 	int retval = VM_FAULT_NOPAGE;
+	struct ttm_mem_type_manager *man =
+		&bdev->man[bo->mem.mem_type];
 
 	/*
 	 * Work around locking order reversal in fault / nopfn
@@ -118,24 +120,28 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 	 * move.
 	 */
 
-	spin_lock(&bo->lock);
+	spin_lock(&bdev->fence_lock);
 	if (test_bit(TTM_BO_PRIV_FLAG_MOVING, &bo->priv_flags)) {
 		ret = ttm_bo_wait(bo, false, true, false);
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 		if (unlikely(ret != 0)) {
 			retval = (ret != -ERESTARTSYS) ?
 			    VM_FAULT_SIGBUS : VM_FAULT_NOPAGE;
 			goto out_unlock;
 		}
 	} else
-		spin_unlock(&bo->lock);
+		spin_unlock(&bdev->fence_lock);
 
-
-	ret = ttm_mem_io_reserve(bdev, &bo->mem);
-	if (ret) {
-		retval = VM_FAULT_SIGBUS;
+	ret = ttm_mem_io_lock(man, true);
+	if (unlikely(ret != 0)) {
+		retval = VM_FAULT_NOPAGE;
 		goto out_unlock;
 	}
+	ret = ttm_mem_io_reserve_vm(bo);
+	if (unlikely(ret != 0)) {
+		retval = VM_FAULT_SIGBUS;
+		goto out_io_unlock;
+	}
 
 	page_offset = ((address - vma->vm_start) >> PAGE_SHIFT) +
 	    bo->vm_node->start - vma->vm_pgoff;
@@ -144,7 +150,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 	if (unlikely(page_offset >= bo->num_pages)) {
 		retval = VM_FAULT_SIGBUS;
-		goto out_unlock;
+		goto out_io_unlock;
 	}
 
 	/*
@@ -182,7 +188,7 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 			page = ttm_tt_get_page(ttm, page_offset);
 			if (unlikely(!page && i == 0)) {
 				retval = VM_FAULT_OOM;
-				goto out_unlock;
+				goto out_io_unlock;
 			} else if (unlikely(!page)) {
 				break;
 			}
@@ -200,14 +206,15 @@ static int ttm_bo_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 		else if (unlikely(ret != 0)) {
 			retval =
 			    (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
-			goto out_unlock;
+			goto out_io_unlock;
 		}
 
 		address += PAGE_SIZE;
 		if (unlikely(++page_offset >= page_last))
 			break;
 	}
-
+out_io_unlock:
+	ttm_mem_io_unlock(man);
 out_unlock:
 	ttm_bo_unreserve(bo);
 	return retval;
diff --git a/drivers/gpu/drm/ttm/ttm_execbuf_util.c b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
index c285c2902d15..3832fe10b4df 100644
--- a/drivers/gpu/drm/ttm/ttm_execbuf_util.c
+++ b/drivers/gpu/drm/ttm/ttm_execbuf_util.c
@@ -32,7 +32,7 @@
 #include <linux/sched.h>
 #include <linux/module.h>
 
-void ttm_eu_backoff_reservation(struct list_head *list)
+static void ttm_eu_backoff_reservation_locked(struct list_head *list)
 {
 	struct ttm_validate_buffer *entry;
 
@@ -41,10 +41,77 @@ void ttm_eu_backoff_reservation(struct list_head *list)
 		if (!entry->reserved)
 			continue;
 
+		if (entry->removed) {
+			ttm_bo_add_to_lru(bo);
+			entry->removed = false;
+
+		}
 		entry->reserved = false;
-		ttm_bo_unreserve(bo);
+		atomic_set(&bo->reserved, 0);
+		wake_up_all(&bo->event_queue);
+	}
+}
+
+static void ttm_eu_del_from_lru_locked(struct list_head *list)
+{
+	struct ttm_validate_buffer *entry;
+
+	list_for_each_entry(entry, list, head) {
+		struct ttm_buffer_object *bo = entry->bo;
+		if (!entry->reserved)
+			continue;
+
+		if (!entry->removed) {
+			entry->put_count = ttm_bo_del_from_lru(bo);
+			entry->removed = true;
+		}
 	}
 }
+
+static void ttm_eu_list_ref_sub(struct list_head *list)
+{
+	struct ttm_validate_buffer *entry;
+
+	list_for_each_entry(entry, list, head) {
+		struct ttm_buffer_object *bo = entry->bo;
+
+		if (entry->put_count) {
+			ttm_bo_list_ref_sub(bo, entry->put_count, true);
+			entry->put_count = 0;
+		}
+	}
+}
+
+static int ttm_eu_wait_unreserved_locked(struct list_head *list,
+					 struct ttm_buffer_object *bo)
+{
+	struct ttm_bo_global *glob = bo->glob;
+	int ret;
+
+	ttm_eu_del_from_lru_locked(list);
+	spin_unlock(&glob->lru_lock);
+	ret = ttm_bo_wait_unreserved(bo, true);
+	spin_lock(&glob->lru_lock);
+	if (unlikely(ret != 0))
+		ttm_eu_backoff_reservation_locked(list);
+	return ret;
+}
+
+
+void ttm_eu_backoff_reservation(struct list_head *list)
+{
+	struct ttm_validate_buffer *entry;
+	struct ttm_bo_global *glob;
+
+	if (list_empty(list))
+		return;
+
+	entry = list_first_entry(list, struct ttm_validate_buffer, head);
+	glob = entry->bo->glob;
+	spin_lock(&glob->lru_lock);
+	ttm_eu_backoff_reservation_locked(list);
+	spin_unlock(&glob->lru_lock);
+}
 EXPORT_SYMBOL(ttm_eu_backoff_reservation);
 
 /*
@@ -59,37 +126,76 @@ EXPORT_SYMBOL(ttm_eu_backoff_reservation);
  * buffers in different orders.
  */
 
-int ttm_eu_reserve_buffers(struct list_head *list, uint32_t val_seq)
+int ttm_eu_reserve_buffers(struct list_head *list)
 {
+	struct ttm_bo_global *glob;
 	struct ttm_validate_buffer *entry;
 	int ret;
+	uint32_t val_seq;
+
+	if (list_empty(list))
+		return 0;
+
+	list_for_each_entry(entry, list, head) {
+		entry->reserved = false;
+		entry->put_count = 0;
+		entry->removed = false;
+	}
+
+	entry = list_first_entry(list, struct ttm_validate_buffer, head);
+	glob = entry->bo->glob;
 
 retry:
+	spin_lock(&glob->lru_lock);
+	val_seq = entry->bo->bdev->val_seq++;
+
 	list_for_each_entry(entry, list, head) {
 		struct ttm_buffer_object *bo = entry->bo;
 
-		entry->reserved = false;
-		ret = ttm_bo_reserve(bo, true, false, true, val_seq);
-		if (ret != 0) {
-			ttm_eu_backoff_reservation(list);
-			if (ret == -EAGAIN) {
-				ret = ttm_bo_wait_unreserved(bo, true);
-				if (unlikely(ret != 0))
-					return ret;
-				goto retry;
-			} else
+retry_this_bo:
+		ret = ttm_bo_reserve_locked(bo, true, true, true, val_seq);
+		switch (ret) {
+		case 0:
+			break;
+		case -EBUSY:
+			ret = ttm_eu_wait_unreserved_locked(list, bo);
+			if (unlikely(ret != 0)) {
+				spin_unlock(&glob->lru_lock);
+				ttm_eu_list_ref_sub(list);
 				return ret;
+			}
+			goto retry_this_bo;
+		case -EAGAIN:
+			ttm_eu_backoff_reservation_locked(list);
+			spin_unlock(&glob->lru_lock);
+			ttm_eu_list_ref_sub(list);
+			ret = ttm_bo_wait_unreserved(bo, true);
+			if (unlikely(ret != 0))
+				return ret;
+			goto retry;
+		default:
+			ttm_eu_backoff_reservation_locked(list);
+			spin_unlock(&glob->lru_lock);
+			ttm_eu_list_ref_sub(list);
+			return ret;
 		}
 
 		entry->reserved = true;
 		if (unlikely(atomic_read(&bo->cpu_writers) > 0)) {
-			ttm_eu_backoff_reservation(list);
+			ttm_eu_backoff_reservation_locked(list);
+			spin_unlock(&glob->lru_lock);
+			ttm_eu_list_ref_sub(list);
 			ret = ttm_bo_wait_cpu(bo, false);
 			if (ret)
 				return ret;
 			goto retry;
 		}
 	}
+
+	ttm_eu_del_from_lru_locked(list);
+	spin_unlock(&glob->lru_lock);
+	ttm_eu_list_ref_sub(list);
+
 	return 0;
 }
 EXPORT_SYMBOL(ttm_eu_reserve_buffers);
@@ -97,21 +203,36 @@ EXPORT_SYMBOL(ttm_eu_reserve_buffers);
 void ttm_eu_fence_buffer_objects(struct list_head *list, void *sync_obj)
 {
 	struct ttm_validate_buffer *entry;
+	struct ttm_buffer_object *bo;
+	struct ttm_bo_global *glob;
+	struct ttm_bo_device *bdev;
+	struct ttm_bo_driver *driver;
 
-	list_for_each_entry(entry, list, head) {
-		struct ttm_buffer_object *bo = entry->bo;
-		struct ttm_bo_driver *driver = bo->bdev->driver;
-		void *old_sync_obj;
+	if (list_empty(list))
+		return;
+
+	bo = list_first_entry(list, struct ttm_validate_buffer, head)->bo;
+	bdev = bo->bdev;
+	driver = bdev->driver;
+	glob = bo->glob;
 
-		spin_lock(&bo->lock);
-		old_sync_obj = bo->sync_obj;
+	spin_lock(&bdev->fence_lock);
+	spin_lock(&glob->lru_lock);
+
+	list_for_each_entry(entry, list, head) {
+		bo = entry->bo;
+		entry->old_sync_obj = bo->sync_obj;
 		bo->sync_obj = driver->sync_obj_ref(sync_obj);
 		bo->sync_obj_arg = entry->new_sync_obj_arg;
-		spin_unlock(&bo->lock);
-		ttm_bo_unreserve(bo);
+		ttm_bo_unreserve_locked(bo);
 		entry->reserved = false;
-		if (old_sync_obj)
-			driver->sync_obj_unref(&old_sync_obj);
+	}
+	spin_unlock(&glob->lru_lock);
+	spin_unlock(&bdev->fence_lock);
+
+	list_for_each_entry(entry, list, head) {
+		if (entry->old_sync_obj)
+			driver->sync_obj_unref(&entry->old_sync_obj);
 	}
 }
 EXPORT_SYMBOL(ttm_eu_fence_buffer_objects);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
index e7a58d055041..10fc01f69c40 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_drv.h
@@ -264,7 +264,6 @@ struct vmw_private {
 	 */
 
 	struct vmw_sw_context ctx;
-	uint32_t val_seq;
 	struct mutex cmdbuf_mutex;
 
 	/**
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 76954e3528c1..41b95ed6dbcd 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -653,8 +653,7 @@ int vmw_execbuf_ioctl(struct drm_device *dev, void *data,
 	ret = vmw_cmd_check_all(dev_priv, sw_context, cmd, arg->command_size);
 	if (unlikely(ret != 0))
 		goto out_err;
-	ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes,
-				     dev_priv->val_seq++);
+	ret = ttm_eu_reserve_buffers(&sw_context->validate_nodes);
 	if (unlikely(ret != 0))
 		goto out_err;
 
diff --git a/include/drm/drmP.h b/include/drm/drmP.h
index 628f76772d22..0f14f94ed8f4 100644
--- a/include/drm/drmP.h
+++ b/include/drm/drmP.h
@@ -683,6 +683,21 @@ struct drm_master {
 	void *driver_priv; /**< Private structure for driver to use */
 };
 
+/* Size of ringbuffer for vblank timestamps. Just double-buffer
+ * in initial implementation.
+ */
+#define DRM_VBLANKTIME_RBSIZE 2
+
+/* Flags and return codes for get_vblank_timestamp() driver function. */
+#define DRM_CALLED_FROM_VBLIRQ 1
+#define DRM_VBLANKTIME_SCANOUTPOS_METHOD (1 << 0)
+#define DRM_VBLANKTIME_INVBL             (1 << 1)
+
+/* get_scanout_position() return flags */
+#define DRM_SCANOUTPOS_VALID        (1 << 0)
+#define DRM_SCANOUTPOS_INVBL        (1 << 1)
+#define DRM_SCANOUTPOS_ACCURATE     (1 << 2)
+
 /**
  * DRM driver structure. This structure represent the common code for
  * a family of cards. There will one drm_device for each card present
@@ -760,6 +775,68 @@ struct drm_driver {
 	 */
 	int (*device_is_agp) (struct drm_device *dev);
 
+	/**
+	 * Called by vblank timestamping code.
+	 *
+	 * Return the current display scanout position from a crtc.
+	 *
+	 * \param dev  DRM device.
+	 * \param crtc Id of the crtc to query.
+	 * \param *vpos Target location for current vertical scanout position.
+	 * \param *hpos Target location for current horizontal scanout position.
+	 *
+	 * Returns vpos as a positive number while in active scanout area.
+	 * Returns vpos as a negative number inside vblank, counting the number
+	 * of scanlines to go until end of vblank, e.g., -1 means "one scanline
+	 * until start of active scanout / end of vblank."
+	 *
+	 * \return Flags, or'ed together as follows:
+	 *
+	 * DRM_SCANOUTPOS_VALID = Query successfull.
+	 * DRM_SCANOUTPOS_INVBL = Inside vblank.
+	 * DRM_SCANOUTPOS_ACCURATE = Returned position is accurate. A lack of
+	 * this flag means that returned position may be offset by a constant
+	 * but unknown small number of scanlines wrt. real scanout position.
+	 *
+	 */
+	int (*get_scanout_position) (struct drm_device *dev, int crtc,
+				     int *vpos, int *hpos);
+
+	/**
+	 * Called by \c drm_get_last_vbltimestamp. Should return a precise
+	 * timestamp when the most recent VBLANK interval ended or will end.
+	 *
+	 * Specifically, the timestamp in @vblank_time should correspond as
+	 * closely as possible to the time when the first video scanline of
+	 * the video frame after the end of VBLANK will start scanning out,
+	 * the time immmediately after end of the VBLANK interval. If the
+	 * @crtc is currently inside VBLANK, this will be a time in the future.
+	 * If the @crtc is currently scanning out a frame, this will be the
+	 * past start time of the current scanout. This is meant to adhere
+	 * to the OpenML OML_sync_control extension specification.
+	 *
+	 * \param dev dev DRM device handle.
+	 * \param crtc crtc for which timestamp should be returned.
+	 * \param *max_error Maximum allowable timestamp error in nanoseconds.
+	 *                   Implementation should strive to provide timestamp
+	 *                   with an error of at most *max_error nanoseconds.
+	 *                   Returns true upper bound on error for timestamp.
+	 * \param *vblank_time Target location for returned vblank timestamp.
+	 * \param flags 0 = Defaults, no special treatment needed.
+	 * \param       DRM_CALLED_FROM_VBLIRQ = Function is called from vblank
+	 *	        irq handler. Some drivers need to apply some workarounds
+	 *              for gpu-specific vblank irq quirks if flag is set.
+	 *
+	 * \returns
+	 * Zero if timestamping isn't supported in current display mode or a
+	 * negative number on failure. A positive status code on success,
+	 * which describes how the vblank_time timestamp was computed.
+	 */
+	int (*get_vblank_timestamp) (struct drm_device *dev, int crtc,
+				     int *max_error,
+				     struct timeval *vblank_time,
+				     unsigned flags);
+
 	/* these have to be filled in */
 
 	irqreturn_t(*irq_handler) (DRM_IRQ_ARGS);
@@ -983,6 +1060,8 @@ struct drm_device {
 
 	wait_queue_head_t *vbl_queue;   /**< VBLANK wait queue */
 	atomic_t *_vblank_count;        /**< number of VBLANK interrupts (driver must alloc the right number of counters) */
+	struct timeval *_vblank_time;   /**< timestamp of current vblank_count (drivers must alloc right number of fields) */
+	spinlock_t vblank_time_lock;    /**< Protects vblank count and time updates during vblank enable/disable */
 	spinlock_t vbl_lock;
 	atomic_t *vblank_refcount;      /* number of users of vblank interruptsper crtc */
 	u32 *last_vblank;               /* protected by dev->vbl_lock, used */
@@ -1282,11 +1361,22 @@ extern int drm_wait_vblank(struct drm_device *dev, void *data,
 			   struct drm_file *filp);
 extern int drm_vblank_wait(struct drm_device *dev, unsigned int *vbl_seq);
 extern u32 drm_vblank_count(struct drm_device *dev, int crtc);
+extern u32 drm_vblank_count_and_time(struct drm_device *dev, int crtc,
+				     struct timeval *vblanktime);
 extern void drm_handle_vblank(struct drm_device *dev, int crtc);
 extern int drm_vblank_get(struct drm_device *dev, int crtc);
 extern void drm_vblank_put(struct drm_device *dev, int crtc);
 extern void drm_vblank_off(struct drm_device *dev, int crtc);
 extern void drm_vblank_cleanup(struct drm_device *dev);
+extern u32 drm_get_last_vbltimestamp(struct drm_device *dev, int crtc,
+				     struct timeval *tvblank, unsigned flags);
+extern int drm_calc_vbltimestamp_from_scanoutpos(struct drm_device *dev,
+						 int crtc, int *max_error,
+						 struct timeval *vblank_time,
+						 unsigned flags,
+						 struct drm_crtc *refcrtc);
+extern void drm_calc_timestamping_constants(struct drm_crtc *crtc);
+
 /* Modesetting support */
 extern void drm_vblank_pre_modeset(struct drm_device *dev, int crtc);
 extern void drm_vblank_post_modeset(struct drm_device *dev, int crtc);
@@ -1337,6 +1427,9 @@ extern void drm_put_dev(struct drm_device *dev);
 extern int drm_put_minor(struct drm_minor **minor);
 extern unsigned int drm_debug;
 
+extern unsigned int drm_vblank_offdelay;
+extern unsigned int drm_timestamp_precision;
+
 extern struct class *drm_class;
 extern struct proc_dir_entry *drm_proc_root;
 extern struct dentry *drm_debugfs_root;
diff --git a/include/drm/drm_crtc.h b/include/drm/drm_crtc.h
index 029aa688e787..acd7fade160d 100644
--- a/include/drm/drm_crtc.h
+++ b/include/drm/drm_crtc.h
@@ -351,8 +351,14 @@ struct drm_crtc {
 
 	bool enabled;
 
+	/* Requested mode from modesetting. */
 	struct drm_display_mode mode;
 
+	/* Programmed mode in hw, after adjustments for encoders,
+	 * crtc, panel scaling etc. Needed for timestamping etc.
+	 */
+	struct drm_display_mode hwmode;
+
 	int x, y;
 	const struct drm_crtc_funcs *funcs;
 
@@ -360,6 +366,9 @@ struct drm_crtc {
 	uint32_t gamma_size;
 	uint16_t *gamma_store;
 
+	/* Constants needed for precise vblank and swap timestamping. */
+	s64 framedur_ns, linedur_ns, pixeldur_ns;
+
 	/* if you are using the helper */
 	void *helper_private;
 };
diff --git a/include/drm/drm_pciids.h b/include/drm/drm_pciids.h
index 883c1d439899..e6b28a39942f 100644
--- a/include/drm/drm_pciids.h
+++ b/include/drm/drm_pciids.h
@@ -419,6 +419,10 @@
 	{0x1002, 0x9713, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS880|RADEON_IS_MOBILITY|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
 	{0x1002, 0x9714, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS880|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
 	{0x1002, 0x9715, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_RS880|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
+	{0x1002, 0x9802, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_PALM|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
+	{0x1002, 0x9803, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_PALM|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
+	{0x1002, 0x9804, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_PALM|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
+	{0x1002, 0x9805, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_PALM|RADEON_NEW_MEMMAP|RADEON_IS_IGP}, \
 	{0, 0, 0}
 
 #define r128_PCI_IDS \
diff --git a/include/drm/nouveau_drm.h b/include/drm/nouveau_drm.h
index bc5590b1a1ac..e2cfe80f6fca 100644
--- a/include/drm/nouveau_drm.h
+++ b/include/drm/nouveau_drm.h
@@ -71,16 +71,14 @@ struct drm_nouveau_gpuobj_free {
 #define NOUVEAU_GETPARAM_PCI_VENDOR      3
 #define NOUVEAU_GETPARAM_PCI_DEVICE      4
 #define NOUVEAU_GETPARAM_BUS_TYPE        5
-#define NOUVEAU_GETPARAM_FB_PHYSICAL     6
-#define NOUVEAU_GETPARAM_AGP_PHYSICAL    7
 #define NOUVEAU_GETPARAM_FB_SIZE         8
 #define NOUVEAU_GETPARAM_AGP_SIZE        9
-#define NOUVEAU_GETPARAM_PCI_PHYSICAL    10
 #define NOUVEAU_GETPARAM_CHIPSET_ID      11
 #define NOUVEAU_GETPARAM_VM_VRAM_BASE    12
 #define NOUVEAU_GETPARAM_GRAPH_UNITS     13
 #define NOUVEAU_GETPARAM_PTIMER_TIME     14
 #define NOUVEAU_GETPARAM_HAS_BO_USAGE    15
+#define NOUVEAU_GETPARAM_HAS_PAGEFLIP    16
 struct drm_nouveau_getparam {
 	uint64_t param;
 	uint64_t value;
@@ -171,7 +169,6 @@ struct drm_nouveau_gem_pushbuf {
 };
 
 #define NOUVEAU_GEM_CPU_PREP_NOWAIT                                  0x00000001
-#define NOUVEAU_GEM_CPU_PREP_NOBLOCK                                 0x00000002
 #define NOUVEAU_GEM_CPU_PREP_WRITE                                   0x00000004
 struct drm_nouveau_gem_cpu_prep {
 	uint32_t handle;
diff --git a/include/drm/ttm/ttm_bo_api.h b/include/drm/ttm/ttm_bo_api.h
index beafc156a535..50852aad260a 100644
--- a/include/drm/ttm/ttm_bo_api.h
+++ b/include/drm/ttm/ttm_bo_api.h
@@ -74,6 +74,8 @@ struct ttm_placement {
  * @is_iomem:		is this io memory ?
  * @size:		size in byte
  * @offset:		offset from the base address
+ * @io_reserved_vm:     The VM system has a refcount in @io_reserved_count
+ * @io_reserved_count:  Refcounting the numbers of callers to ttm_mem_io_reserve
  *
  * Structure indicating the bus placement of an object.
  */
@@ -83,7 +85,8 @@ struct ttm_bus_placement {
 	unsigned long	size;
 	unsigned long	offset;
 	bool		is_iomem;
-	bool		io_reserved;
+	bool		io_reserved_vm;
+	uint64_t        io_reserved_count;
 };
 
 
@@ -154,7 +157,6 @@ struct ttm_tt;
  * keeps one refcount. When this refcount reaches zero,
  * the object is destroyed.
  * @event_queue: Queue for processes waiting on buffer object status change.
- * @lock: spinlock protecting mostly synchronization members.
  * @mem: structure describing current placement.
  * @persistant_swap_storage: Usually the swap storage is deleted for buffers
  * pinned in physical memory. If this behaviour is not desired, this member
@@ -213,7 +215,6 @@ struct ttm_buffer_object {
 	struct kref kref;
 	struct kref list_kref;
 	wait_queue_head_t event_queue;
-	spinlock_t lock;
 
 	/**
 	 * Members protected by the bo::reserved lock.
@@ -237,6 +238,7 @@ struct ttm_buffer_object {
 	struct list_head lru;
 	struct list_head ddestroy;
 	struct list_head swap;
+	struct list_head io_reserve_lru;
 	uint32_t val_seq;
 	bool seq_valid;
 
@@ -248,10 +250,10 @@ struct ttm_buffer_object {
 	atomic_t reserved;
 
 	/**
-	 * Members protected by the bo::lock
+	 * Members protected by struct buffer_object_device::fence_lock
 	 * In addition, setting sync_obj to anything else
 	 * than NULL requires bo::reserved to be held. This allows for
-	 * checking NULL while reserved but not holding bo::lock.
+	 * checking NULL while reserved but not holding the mentioned lock.
 	 */
 
 	void *sync_obj_arg;
@@ -364,6 +366,44 @@ extern int ttm_bo_validate(struct ttm_buffer_object *bo,
  */
 extern void ttm_bo_unref(struct ttm_buffer_object **bo);
 
+
+/**
+ * ttm_bo_list_ref_sub
+ *
+ * @bo: The buffer object.
+ * @count: The number of references with which to decrease @bo::list_kref;
+ * @never_free: The refcount should not reach zero with this operation.
+ *
+ * Release @count lru list references to this buffer object.
+ */
+extern void ttm_bo_list_ref_sub(struct ttm_buffer_object *bo, int count,
+				bool never_free);
+
+/**
+ * ttm_bo_add_to_lru
+ *
+ * @bo: The buffer object.
+ *
+ * Add this bo to the relevant mem type lru and, if it's backed by
+ * system pages (ttms) to the swap list.
+ * This function must be called with struct ttm_bo_global::lru_lock held, and
+ * is typically called immediately prior to unreserving a bo.
+ */
+extern void ttm_bo_add_to_lru(struct ttm_buffer_object *bo);
+
+/**
+ * ttm_bo_del_from_lru
+ *
+ * @bo: The buffer object.
+ *
+ * Remove this bo from all lru lists used to lookup and reserve an object.
+ * This function must be called with struct ttm_bo_global::lru_lock held,
+ * and is usually called just immediately after the bo has been reserved to
+ * avoid recursive reservation from lru lists.
+ */
+extern int ttm_bo_del_from_lru(struct ttm_buffer_object *bo);
+
+
 /**
  * ttm_bo_lock_delayed_workqueue
  *
diff --git a/include/drm/ttm/ttm_bo_driver.h b/include/drm/ttm/ttm_bo_driver.h
index 8e0c848326b6..1da8af6ac884 100644
--- a/include/drm/ttm/ttm_bo_driver.h
+++ b/include/drm/ttm/ttm_bo_driver.h
@@ -179,30 +179,6 @@ struct ttm_tt {
 #define TTM_MEMTYPE_FLAG_MAPPABLE      (1 << 1)	/* Memory mappable */
 #define TTM_MEMTYPE_FLAG_CMA           (1 << 3)	/* Can't map aperture */
 
-/**
- * struct ttm_mem_type_manager
- *
- * @has_type: The memory type has been initialized.
- * @use_type: The memory type is enabled.
- * @flags: TTM_MEMTYPE_XX flags identifying the traits of the memory
- * managed by this memory type.
- * @gpu_offset: If used, the GPU offset of the first managed page of
- * fixed memory or the first managed location in an aperture.
- * @size: Size of the managed region.
- * @available_caching: A mask of available caching types, TTM_PL_FLAG_XX,
- * as defined in ttm_placement_common.h
- * @default_caching: The default caching policy used for a buffer object
- * placed in this memory type if the user doesn't provide one.
- * @manager: The range manager used for this memory type. FIXME: If the aperture
- * has a page size different from the underlying system, the granularity
- * of this manager should take care of this. But the range allocating code
- * in ttm_bo.c needs to be modified for this.
- * @lru: The lru list for this memory type.
- *
- * This structure is used to identify and manage memory types for a device.
- * It's set up by the ttm_bo_driver::init_mem_type method.
- */
-
 struct ttm_mem_type_manager;
 
 struct ttm_mem_type_manager_func {
@@ -287,6 +263,36 @@ struct ttm_mem_type_manager_func {
 	void (*debug)(struct ttm_mem_type_manager *man, const char *prefix);
 };
 
+/**
+ * struct ttm_mem_type_manager
+ *
+ * @has_type: The memory type has been initialized.
+ * @use_type: The memory type is enabled.
+ * @flags: TTM_MEMTYPE_XX flags identifying the traits of the memory
+ * managed by this memory type.
+ * @gpu_offset: If used, the GPU offset of the first managed page of
+ * fixed memory or the first managed location in an aperture.
+ * @size: Size of the managed region.
+ * @available_caching: A mask of available caching types, TTM_PL_FLAG_XX,
+ * as defined in ttm_placement_common.h
+ * @default_caching: The default caching policy used for a buffer object
+ * placed in this memory type if the user doesn't provide one.
+ * @func: structure pointer implementing the range manager. See above
+ * @priv: Driver private closure for @func.
+ * @io_reserve_mutex: Mutex optionally protecting shared io_reserve structures
+ * @use_io_reserve_lru: Use an lru list to try to unreserve io_mem_regions
+ * reserved by the TTM vm system.
+ * @io_reserve_lru: Optional lru list for unreserving io mem regions.
+ * @io_reserve_fastpath: Only use bdev::driver::io_mem_reserve to obtain
+ * static information. bdev::driver::io_mem_free is never used.
+ * @lru: The lru list for this memory type.
+ *
+ * This structure is used to identify and manage memory types for a device.
+ * It's set up by the ttm_bo_driver::init_mem_type method.
+ */
+
+
+
 struct ttm_mem_type_manager {
 	struct ttm_bo_device *bdev;
 
@@ -303,6 +309,15 @@ struct ttm_mem_type_manager {
 	uint32_t default_caching;
 	const struct ttm_mem_type_manager_func *func;
 	void *priv;
+	struct mutex io_reserve_mutex;
+	bool use_io_reserve_lru;
+	bool io_reserve_fastpath;
+
+	/*
+	 * Protected by @io_reserve_mutex:
+	 */
+
+	struct list_head io_reserve_lru;
 
 	/*
 	 * Protected by the global->lru_lock.
@@ -510,9 +525,12 @@ struct ttm_bo_global {
  *
  * @driver: Pointer to a struct ttm_bo_driver struct setup by the driver.
  * @man: An array of mem_type_managers.
+ * @fence_lock: Protects the synchronizing members on *all* bos belonging
+ * to this device.
  * @addr_space_mm: Range manager for the device address space.
  * lru_lock: Spinlock that protects the buffer+device lru lists and
  * ddestroy lists.
+ * @val_seq: Current validation sequence.
  * @nice_mode: Try nicely to wait for buffer idle when cleaning a manager.
  * If a GPU lockup has been detected, this is forced to 0.
  * @dev_mapping: A pointer to the struct address_space representing the
@@ -531,6 +549,7 @@ struct ttm_bo_device {
 	struct ttm_bo_driver *driver;
 	rwlock_t vm_lock;
 	struct ttm_mem_type_manager man[TTM_NUM_MEM_TYPES];
+	spinlock_t fence_lock;
 	/*
 	 * Protected by the vm lock.
 	 */
@@ -541,6 +560,7 @@ struct ttm_bo_device {
 	 * Protected by the global:lru lock.
 	 */
 	struct list_head ddestroy;
+	uint32_t val_seq;
 
 	/*
 	 * Protected by load / firstopen / lastclose /unload sync.
@@ -753,31 +773,6 @@ extern void ttm_bo_mem_put_locked(struct ttm_buffer_object *bo,
 
 extern int ttm_bo_wait_cpu(struct ttm_buffer_object *bo, bool no_wait);
 
-/**
- * ttm_bo_pci_offset - Get the PCI offset for the buffer object memory.
- *
- * @bo Pointer to a struct ttm_buffer_object.
- * @bus_base On return the base of the PCI region
- * @bus_offset On return the byte offset into the PCI region
- * @bus_size On return the byte size of the buffer object or zero if
- * the buffer object memory is not accessible through a PCI region.
- *
- * Returns:
- * -EINVAL if the buffer object is currently not mappable.
- * 0 otherwise.
- */
-
-extern int ttm_bo_pci_offset(struct ttm_bo_device *bdev,
-			     struct ttm_mem_reg *mem,
-			     unsigned long *bus_base,
-			     unsigned long *bus_offset,
-			     unsigned long *bus_size);
-
-extern int ttm_mem_io_reserve(struct ttm_bo_device *bdev,
-				struct ttm_mem_reg *mem);
-extern void ttm_mem_io_free(struct ttm_bo_device *bdev,
-				struct ttm_mem_reg *mem);
-
 extern void ttm_bo_global_release(struct drm_global_reference *ref);
 extern int ttm_bo_global_init(struct drm_global_reference *ref);
 
@@ -810,6 +805,22 @@ extern int ttm_bo_device_init(struct ttm_bo_device *bdev,
 extern void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo);
 
 /**
+ * ttm_bo_unmap_virtual
+ *
+ * @bo: tear down the virtual mappings for this BO
+ *
+ * The caller must take ttm_mem_io_lock before calling this function.
+ */
+extern void ttm_bo_unmap_virtual_locked(struct ttm_buffer_object *bo);
+
+extern int ttm_mem_io_reserve_vm(struct ttm_buffer_object *bo);
+extern void ttm_mem_io_free_vm(struct ttm_buffer_object *bo);
+extern int ttm_mem_io_lock(struct ttm_mem_type_manager *man,
+			   bool interruptible);
+extern void ttm_mem_io_unlock(struct ttm_mem_type_manager *man);
+
+
+/**
  * ttm_bo_reserve:
  *
  * @bo: A pointer to a struct ttm_buffer_object.
@@ -859,11 +870,44 @@ extern void ttm_bo_unmap_virtual(struct ttm_buffer_object *bo);
  * try again. (only if use_sequence == 1).
  * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
  * a signal. Release all buffer reservations and return to user-space.
+ * -EBUSY: The function needed to sleep, but @no_wait was true
+ * -EDEADLK: Bo already reserved using @sequence. This error code will only
+ * be returned if @use_sequence is set to true.
  */
 extern int ttm_bo_reserve(struct ttm_buffer_object *bo,
 			  bool interruptible,
 			  bool no_wait, bool use_sequence, uint32_t sequence);
 
+
+/**
+ * ttm_bo_reserve_locked:
+ *
+ * @bo: A pointer to a struct ttm_buffer_object.
+ * @interruptible: Sleep interruptible if waiting.
+ * @no_wait: Don't sleep while trying to reserve, rather return -EBUSY.
+ * @use_sequence: If @bo is already reserved, Only sleep waiting for
+ * it to become unreserved if @sequence < (@bo)->sequence.
+ *
+ * Must be called with struct ttm_bo_global::lru_lock held,
+ * and will not remove reserved buffers from the lru lists.
+ * The function may release the LRU spinlock if it needs to sleep.
+ * Otherwise identical to ttm_bo_reserve.
+ *
+ * Returns:
+ * -EAGAIN: The reservation may cause a deadlock.
+ * Release all buffer reservations, wait for @bo to become unreserved and
+ * try again. (only if use_sequence == 1).
+ * -ERESTARTSYS: A wait for the buffer to become unreserved was interrupted by
+ * a signal. Release all buffer reservations and return to user-space.
+ * -EBUSY: The function needed to sleep, but @no_wait was true
+ * -EDEADLK: Bo already reserved using @sequence. This error code will only
+ * be returned if @use_sequence is set to true.
+ */
+extern int ttm_bo_reserve_locked(struct ttm_buffer_object *bo,
+				 bool interruptible,
+				 bool no_wait, bool use_sequence,
+				 uint32_t sequence);
+
 /**
  * ttm_bo_unreserve
  *
@@ -874,6 +918,16 @@ extern int ttm_bo_reserve(struct ttm_buffer_object *bo,
 extern void ttm_bo_unreserve(struct ttm_buffer_object *bo);
 
 /**
+ * ttm_bo_unreserve_locked
+ *
+ * @bo: A pointer to a struct ttm_buffer_object.
+ *
+ * Unreserve a previous reservation of @bo.
+ * Needs to be called with struct ttm_bo_global::lru_lock held.
+ */
+extern void ttm_bo_unreserve_locked(struct ttm_buffer_object *bo);
+
+/**
  * ttm_bo_wait_unreserved
  *
  * @bo: A pointer to a struct ttm_buffer_object.
diff --git a/include/drm/ttm/ttm_execbuf_util.h b/include/drm/ttm/ttm_execbuf_util.h
index cd2c475da9ea..26cc7f9ffa41 100644
--- a/include/drm/ttm/ttm_execbuf_util.h
+++ b/include/drm/ttm/ttm_execbuf_util.h
@@ -41,7 +41,10 @@
  * @bo:             refcounted buffer object pointer.
  * @new_sync_obj_arg: New sync_obj_arg for @bo, to be used once
  * adding a new sync object.
- * @reservied:      Indicates whether @bo has been reserved for validation.
+ * @reserved:       Indicates whether @bo has been reserved for validation.
+ * @removed:        Indicates whether @bo has been removed from lru lists.
+ * @put_count:      Number of outstanding references on bo::list_kref.
+ * @old_sync_obj:   Pointer to a sync object about to be unreferenced
  */
 
 struct ttm_validate_buffer {
@@ -49,6 +52,9 @@ struct ttm_validate_buffer {
 	struct ttm_buffer_object *bo;
 	void *new_sync_obj_arg;
 	bool reserved;
+	bool removed;
+	int put_count;
+	void *old_sync_obj;
 };
 
 /**
@@ -66,7 +72,6 @@ extern void ttm_eu_backoff_reservation(struct list_head *list);
  * function ttm_eu_reserve_buffers
  *
  * @list:    thread private list of ttm_validate_buffer structs.
- * @val_seq: A unique sequence number.
  *
  * Tries to reserve bos pointed to by the list entries for validation.
  * If the function returns 0, all buffers are marked as "unfenced",
@@ -88,7 +93,7 @@ extern void ttm_eu_backoff_reservation(struct list_head *list);
  * has failed.
  */
 
-extern int ttm_eu_reserve_buffers(struct list_head *list, uint32_t val_seq);
+extern int ttm_eu_reserve_buffers(struct list_head *list);
 
 /**
  * function ttm_eu_fence_buffer_objects.
diff --git a/include/linux/kref.h b/include/linux/kref.h
index 6cc38fc07ab7..d4a62ab2ee5e 100644
--- a/include/linux/kref.h
+++ b/include/linux/kref.h
@@ -24,5 +24,7 @@ struct kref {
 void kref_init(struct kref *kref);
 void kref_get(struct kref *kref);
 int kref_put(struct kref *kref, void (*release) (struct kref *kref));
+int kref_sub(struct kref *kref, unsigned int count,
+	     void (*release) (struct kref *kref));
 
 #endif /* _KREF_H_ */
diff --git a/lib/kref.c b/lib/kref.c
index d3d227a08a4b..3efb882b11db 100644
--- a/lib/kref.c
+++ b/lib/kref.c
@@ -62,6 +62,36 @@ int kref_put(struct kref *kref, void (*release)(struct kref *kref))
 	return 0;
 }
 
+
+/**
+ * kref_sub - subtract a number of refcounts for object.
+ * @kref: object.
+ * @count: Number of recounts to subtract.
+ * @release: pointer to the function that will clean up the object when the
+ *	     last reference to the object is released.
+ *	     This pointer is required, and it is not acceptable to pass kfree
+ *	     in as this function.
+ *
+ * Subtract @count from the refcount, and if 0, call release().
+ * Return 1 if the object was removed, otherwise return 0.  Beware, if this
+ * function returns 0, you still can not count on the kref from remaining in
+ * memory.  Only use the return value if you want to see if the kref is now
+ * gone, not present.
+ */
+int kref_sub(struct kref *kref, unsigned int count,
+	     void (*release)(struct kref *kref))
+{
+	WARN_ON(release == NULL);
+	WARN_ON(release == (void (*)(struct kref *))kfree);
+
+	if (atomic_sub_and_test((int) count, &kref->refcount)) {
+		release(kref);
+		return 1;
+	}
+	return 0;
+}
+
 EXPORT_SYMBOL(kref_init);
 EXPORT_SYMBOL(kref_get);
 EXPORT_SYMBOL(kref_put);
+EXPORT_SYMBOL(kref_sub);