11 files changed, 86 insertions, 197 deletions

diff --git a/drivers/gpu/drm/nouveau/nouveau_drv.h b/drivers/gpu/drm/nouveau/nouveau_drv.h
index 46e32573b3a3..01bffc4412d2 100644
--- a/drivers/gpu/drm/nouveau/nouveau_drv.h
+++ b/drivers/gpu/drm/nouveau/nouveau_drv.h
@@ -160,6 +160,7 @@ enum nouveau_flags {
 #define NVOBJ_FLAG_ZERO_ALLOC		(1 << 1)
 #define NVOBJ_FLAG_ZERO_FREE		(1 << 2)
 #define NVOBJ_FLAG_VM			(1 << 3)
+#define NVOBJ_FLAG_VM_USER		(1 << 4)
 
 #define NVOBJ_CINST_GLOBAL	0xdeadbeef
 
@@ -1576,6 +1577,20 @@ nv_match_device(struct drm_device *dev, unsigned device,
 		dev->pdev->subsystem_device == sub_device;
 }
 
+/* returns 1 if device is one of the nv4x using the 0x4497 object class,
+ * helpful to determine a number of other hardware features
+ */
+static inline int
+nv44_graph_class(struct drm_device *dev)
+{
+	struct drm_nouveau_private *dev_priv = dev->dev_private;
+
+	if ((dev_priv->chipset & 0xf0) == 0x60)
+		return 1;
+
+	return !(0x0baf & (1 << (dev_priv->chipset & 0x0f)));
+}
+
 /* memory type/access flags, do not match hardware values */
 #define NV_MEM_ACCESS_RO  1
 #define NV_MEM_ACCESS_WO  2
diff --git a/drivers/gpu/drm/nouveau/nouveau_fbcon.c b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
index 6d56a54b6e2e..60769d2f9a66 100644
--- a/drivers/gpu/drm/nouveau/nouveau_fbcon.c
+++ b/drivers/gpu/drm/nouveau/nouveau_fbcon.c
@@ -352,8 +352,8 @@ nouveau_fbcon_create(struct nouveau_fbdev *nfbdev,
 			      FBINFO_HWACCEL_IMAGEBLIT;
 	info->flags |= FBINFO_CAN_FORCE_OUTPUT;
 	info->fbops = &nouveau_fbcon_sw_ops;
-	info->fix.smem_start = dev->mode_config.fb_base +
-			       (nvbo->bo.mem.start << PAGE_SHIFT);
+	info->fix.smem_start = nvbo->bo.mem.bus.base +
+			       nvbo->bo.mem.bus.offset;
 	info->fix.smem_len = size;
 
 	info->screen_base = nvbo_kmap_obj_iovirtual(nouveau_fb->nvbo);
diff --git a/drivers/gpu/drm/nouveau/nouveau_mem.c b/drivers/gpu/drm/nouveau/nouveau_mem.c
index 69044eb104bb..26347b7cd872 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mem.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mem.c
@@ -742,30 +742,24 @@ 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;
+	u32 total = 0, free = 0;
 
 	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),
+		printk(KERN_DEBUG "%s %d: 0x%010llx 0x%010llx\n",
+		       prefix, 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;
+		if (!r->type)
+			free += 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);
-	}
+	printk(KERN_DEBUG "%s  total: 0x%010llx free: 0x%010llx\n",
+	       prefix, (u64)total << 12, (u64)free << 12);
+	printk(KERN_DEBUG "%s  block: 0x%08x\n",
+	       prefix, mm->block_size << 12);
 }
 
 const struct ttm_mem_type_manager_func nouveau_vram_manager = {
diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.c b/drivers/gpu/drm/nouveau/nouveau_mm.c
index cdbb11eb701b..8844b50c3e54 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mm.c
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.c
@@ -48,175 +48,76 @@ region_split(struct nouveau_mm *rmm, struct nouveau_mm_node *a, u32 size)
 
 	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)
+	if (b->type == 0)
 		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;
-}
+#define node(root, dir) ((root)->nl_entry.dir == &rmm->nodes) ? NULL : \
+	list_entry((root)->nl_entry.dir, struct nouveau_mm_node, nl_entry)
 
 void
 nouveau_mm_put(struct nouveau_mm *rmm, struct nouveau_mm_node *this)
 {
-	u32 block_s, block_l;
+	struct nouveau_mm_node *prev = node(this, prev);
+	struct nouveau_mm_node *next = node(this, next);
 
-	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;
+	this->type = 0;
 
-	/* split off any still-typed region at the start */
-	if (block_s != this->offset) {
-		if (!region_split(rmm, this, block_s - this->offset))
-			return;
+	if (prev && prev->type == 0) {
+		prev->length += this->length;
+		region_put(rmm, this);
+		this = prev;
 	}
 
-	/* 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;
+	if (next && next->type == 0) {
+		next->offset  = this->offset;
+		next->length += this->length;
+		region_put(rmm, this);
 	}
-
-	/* 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)
+	struct nouveau_mm_node *prev, *this, *next;
+	u32 min = size_nc ? size_nc : size;
+	u32 align_mask = align - 1;
+	u32 splitoff;
+	u32 s, e;
+
+	list_for_each_entry(this, &rmm->free, fl_entry) {
+		e = this->offset + this->length;
+		s = this->offset;
+
+		prev = node(this, prev);
+		if (prev && prev->type != type)
+			s = roundup(s, rmm->block_size);
+
+		next = node(this, next);
+		if (next && next->type != type)
+			e = rounddown(e, rmm->block_size);
+
+		s  = (s + align_mask) & ~align_mask;
+		e &= ~align_mask;
+		if (s > e || e - s < min)
 			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;
-		}
+		splitoff = s - this->offset;
+		if (splitoff && !region_split(rmm, this, splitoff))
+			return -ENOMEM;
 
-		this = region_split(rmm, this, alloc);
-		if (this == NULL)
+		this = region_split(rmm, this, min(size, e - s));
+		if (!this)
 			return -ENOMEM;
 
-		this->free = false;
+		this->type = type;
 		list_del(&this->fl_entry);
 		*pnode = this;
 		return 0;
@@ -234,7 +135,6 @@ nouveau_mm_init(struct nouveau_mm **prmm, u32 offset, u32 length, u32 block)
 	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;
 
diff --git a/drivers/gpu/drm/nouveau/nouveau_mm.h b/drivers/gpu/drm/nouveau/nouveau_mm.h
index af3844933036..798eaf39691c 100644
--- a/drivers/gpu/drm/nouveau/nouveau_mm.h
+++ b/drivers/gpu/drm/nouveau/nouveau_mm.h
@@ -30,9 +30,7 @@ struct nouveau_mm_node {
 	struct list_head fl_entry;
 	struct list_head rl_entry;
 
-	bool free;
-	int  type;
-
+	u8  type;
 	u32 offset;
 	u32 length;
 };
diff --git a/drivers/gpu/drm/nouveau/nv40_graph.c b/drivers/gpu/drm/nouveau/nv40_graph.c
index 19ef92a0375a..8870d72388c8 100644
--- a/drivers/gpu/drm/nouveau/nv40_graph.c
+++ b/drivers/gpu/drm/nouveau/nv40_graph.c
@@ -451,8 +451,7 @@ nv40_graph_register(struct drm_device *dev)
 	NVOBJ_CLASS(dev, 0x309e, GR); /* swzsurf */
 
 	/* curie */
-	if (dev_priv->chipset >= 0x60 ||
-	    0x00005450 & (1 << (dev_priv->chipset & 0x0f)))
+	if (nv44_graph_class(dev))
 		NVOBJ_CLASS(dev, 0x4497, GR);
 	else
 		NVOBJ_CLASS(dev, 0x4097, GR);
diff --git a/drivers/gpu/drm/nouveau/nv40_grctx.c b/drivers/gpu/drm/nouveau/nv40_grctx.c
index ce585093264e..f70447d131d7 100644
--- a/drivers/gpu/drm/nouveau/nv40_grctx.c
+++ b/drivers/gpu/drm/nouveau/nv40_grctx.c
@@ -118,17 +118,6 @@
  */
 
 static int
-nv40_graph_4097(struct drm_device *dev)
-{
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-
-	if ((dev_priv->chipset & 0xf0) == 0x60)
-		return 0;
-
-	return !!(0x0baf & (1 << dev_priv->chipset));
-}
-
-static int
 nv40_graph_vs_count(struct drm_device *dev)
 {
 	struct drm_nouveau_private *dev_priv = dev->dev_private;
@@ -219,7 +208,7 @@ nv40_graph_construct_general(struct nouveau_grctx *ctx)
 		gr_def(ctx, 0x4009dc, 0x80000000);
 	} else {
 		cp_ctx(ctx, 0x400840, 20);
-		if (!nv40_graph_4097(ctx->dev)) {
+		if (nv44_graph_class(ctx->dev)) {
 			for (i = 0; i < 8; i++)
 				gr_def(ctx, 0x400860 + (i * 4), 0x00000001);
 		}
@@ -228,7 +217,7 @@ nv40_graph_construct_general(struct nouveau_grctx *ctx)
 		gr_def(ctx, 0x400888, 0x00000040);
 		cp_ctx(ctx, 0x400894, 11);
 		gr_def(ctx, 0x400894, 0x00000040);
-		if (nv40_graph_4097(ctx->dev)) {
+		if (!nv44_graph_class(ctx->dev)) {
 			for (i = 0; i < 8; i++)
 				gr_def(ctx, 0x4008a0 + (i * 4), 0x80000000);
 		}
@@ -546,7 +535,7 @@ nv40_graph_construct_state3d_2(struct nouveau_grctx *ctx)
 static void
 nv40_graph_construct_state3d_3(struct nouveau_grctx *ctx)
 {
-	int len = nv40_graph_4097(ctx->dev) ? 0x0684 : 0x0084;
+	int len = nv44_graph_class(ctx->dev) ? 0x0084 : 0x0684;
 
 	cp_out (ctx, 0x300000);
 	cp_lsr (ctx, len - 4);
@@ -582,11 +571,11 @@ nv40_graph_construct_shader(struct nouveau_grctx *ctx)
 	} else {
 		b0_offset = 0x1d40/4; /* 2200 */
 		b1_offset = 0x3f40/4; /* 0b00 : 0a40 */
-		vs_len = nv40_graph_4097(dev) ? 0x4a40/4 : 0x4980/4;
+		vs_len = nv44_graph_class(dev) ? 0x4980/4 : 0x4a40/4;
 	}
 
 	cp_lsr(ctx, vs_len * vs_nr + 0x300/4);
-	cp_out(ctx, nv40_graph_4097(dev) ? 0x800041 : 0x800029);
+	cp_out(ctx, nv44_graph_class(dev) ? 0x800029 : 0x800041);
 
 	offset = ctx->ctxvals_pos;
 	ctx->ctxvals_pos += (0x0300/4 + (vs_nr * vs_len));
diff --git a/drivers/gpu/drm/nouveau/nv40_mc.c b/drivers/gpu/drm/nouveau/nv40_mc.c
index e4e72c12ab6a..03c0d4c3f355 100644
--- a/drivers/gpu/drm/nouveau/nv40_mc.c
+++ b/drivers/gpu/drm/nouveau/nv40_mc.c
@@ -6,27 +6,17 @@
 int
 nv40_mc_init(struct drm_device *dev)
 {
-	struct drm_nouveau_private *dev_priv = dev->dev_private;
-	uint32_t tmp;
-
 	/* Power up everything, resetting each individual unit will
 	 * be done later if needed.
 	 */
 	nv_wr32(dev, NV03_PMC_ENABLE, 0xFFFFFFFF);
 
-	switch (dev_priv->chipset) {
-	case 0x44:
-	case 0x46: /* G72 */
-	case 0x4e:
-	case 0x4c: /* C51_G7X */
-		tmp = nv_rd32(dev, NV04_PFB_FIFO_DATA);
+	if (nv44_graph_class(dev)) {
+		u32 tmp = nv_rd32(dev, NV04_PFB_FIFO_DATA);
 		nv_wr32(dev, NV40_PMC_1700, tmp);
 		nv_wr32(dev, NV40_PMC_1704, 0);
 		nv_wr32(dev, NV40_PMC_1708, 0);
 		nv_wr32(dev, NV40_PMC_170C, tmp);
-		break;
-	default:
-		break;
 	}
 
 	return 0;
diff --git a/drivers/gpu/drm/nouveau/nv50_instmem.c b/drivers/gpu/drm/nouveau/nv50_instmem.c
index 2e1b1cd19a4b..ea0041810ae3 100644
--- a/drivers/gpu/drm/nouveau/nv50_instmem.c
+++ b/drivers/gpu/drm/nouveau/nv50_instmem.c
@@ -332,8 +332,11 @@ nv50_instmem_get(struct nouveau_gpuobj *gpuobj, u32 size, u32 align)
 	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,
+		u32 flags = NV_MEM_ACCESS_RW;
+		if (!(gpuobj->flags & NVOBJ_FLAG_VM_USER))
+			flags |= NV_MEM_ACCESS_SYS;
+
+		ret = nouveau_vm_get(dev_priv->chan_vm, size, 12, flags,
 				     &node->chan_vma);
 		if (ret) {
 			vram->put(dev, &node->vram);
diff --git a/drivers/gpu/drm/nouveau/nvc0_graph.c b/drivers/gpu/drm/nouveau/nvc0_graph.c
index 5feacd5d5fa4..e6ea7d83187f 100644
--- a/drivers/gpu/drm/nouveau/nvc0_graph.c
+++ b/drivers/gpu/drm/nouveau/nvc0_graph.c
@@ -105,7 +105,8 @@ nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
 	if (ret)
 		return ret;
 
-	ret = nouveau_gpuobj_new(dev, NULL, 384 * 1024, 4096, NVOBJ_FLAG_VM,
+	ret = nouveau_gpuobj_new(dev, NULL, 384 * 1024, 4096,
+				 NVOBJ_FLAG_VM | NVOBJ_FLAG_VM_USER,
 				 &grch->unk418810);
 	if (ret)
 		return ret;
diff --git a/drivers/gpu/drm/nouveau/nvc0_vm.c b/drivers/gpu/drm/nouveau/nvc0_vm.c
index 4b9251bb0ff4..e4e83c2caf5b 100644
--- a/drivers/gpu/drm/nouveau/nvc0_vm.c
+++ b/drivers/gpu/drm/nouveau/nvc0_vm.c
@@ -48,8 +48,8 @@ nvc0_vm_addr(struct nouveau_vma *vma, u64 phys, u32 memtype, u32 target)
 	phys >>= 8;
 
 	phys |= 0x00000001; /* present */
-//	if (vma->access & NV_MEM_ACCESS_SYS)
-//		phys |= 0x00000002;
+	if (vma->access & NV_MEM_ACCESS_SYS)
+		phys |= 0x00000002;
 
 	phys |= ((u64)target  << 32);
 	phys |= ((u64)memtype << 36);