drm/radeon/kms/r6xx+: use new style fencing (v3)

On r6xx+ a newer fence mechanism was implemented to replace
the old wait_until plus scratch regs setup.  A single EOP event
will flush the destination caches, write a fence value, and generate
an interrupt.  This is the recommended fence mechanism on r6xx+ asics.

This requires my previous writeback patch.

v2: fix typo that enabled event fence checking on all asics
rather than just r6xx+.

v3: properly enable EOP interrupts
Should fix:
https://bugs.freedesktop.org/show_bug.cgi?id=29972

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

6 files changed, 64 insertions, 17 deletions

diff --git a/drivers/gpu/drm/radeon/evergreen.c b/drivers/gpu/drm/radeon/evergreen.c
index e47d221e24ac..4c82cc830271 100644
--- a/drivers/gpu/drm/radeon/evergreen.c
+++ b/drivers/gpu/drm/radeon/evergreen.c
@@ -1594,6 +1594,7 @@ int evergreen_irq_set(struct radeon_device *rdev)
 	if (rdev->irq.sw_int) {
 		DRM_DEBUG("evergreen_irq_set: sw int\n");
 		cp_int_cntl |= RB_INT_ENABLE;
+		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
 	}
 	if (rdev->irq.crtc_vblank_int[0]) {
 		DRM_DEBUG("evergreen_irq_set: vblank 0\n");
@@ -2012,6 +2013,7 @@ restart_ih:
 			break;
 		case 181: /* CP EOP event */
 			DRM_DEBUG("IH: CP EOP\n");
+			radeon_fence_process(rdev);
 			break;
 		case 233: /* GUI IDLE */
 			DRM_DEBUG("IH: CP EOP\n");
diff --git a/drivers/gpu/drm/radeon/r600.c b/drivers/gpu/drm/radeon/r600.c
index fbce58b2cd04..7c5f855a43e6 100644
--- a/drivers/gpu/drm/radeon/r600.c
+++ b/drivers/gpu/drm/radeon/r600.c
@@ -2279,21 +2279,31 @@ int r600_ring_test(struct radeon_device *rdev)
 void r600_fence_ring_emit(struct radeon_device *rdev,
 			  struct radeon_fence *fence)
 {
-	/* Also consider EVENT_WRITE_EOP.  it handles the interrupts + timestamps + events */
-
-	radeon_ring_write(rdev, PACKET3(PACKET3_EVENT_WRITE, 0));
-	radeon_ring_write(rdev, CACHE_FLUSH_AND_INV_EVENT);
-	/* wait for 3D idle clean */
-	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
-	radeon_ring_write(rdev, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
-	radeon_ring_write(rdev, WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
-	/* Emit fence sequence & fire IRQ */
-	radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
-	radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
-	radeon_ring_write(rdev, fence->seq);
-	/* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
-	radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
-	radeon_ring_write(rdev, RB_INT_STAT);
+	if (rdev->wb.use_event) {
+		u64 addr = rdev->wb.gpu_addr + R600_WB_EVENT_OFFSET +
+			(u64)(rdev->fence_drv.scratch_reg - rdev->scratch.reg_base);
+		/* EVENT_WRITE_EOP - flush caches, send int */
+		radeon_ring_write(rdev, PACKET3(PACKET3_EVENT_WRITE_EOP, 4));
+		radeon_ring_write(rdev, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT_TS) | EVENT_INDEX(5));
+		radeon_ring_write(rdev, addr & 0xffffffff);
+		radeon_ring_write(rdev, (upper_32_bits(addr) & 0xff) | DATA_SEL(1) | INT_SEL(2));
+		radeon_ring_write(rdev, fence->seq);
+		radeon_ring_write(rdev, 0);
+	} else {
+		radeon_ring_write(rdev, PACKET3(PACKET3_EVENT_WRITE, 0));
+		radeon_ring_write(rdev, EVENT_TYPE(CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0));
+		/* wait for 3D idle clean */
+		radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
+		radeon_ring_write(rdev, (WAIT_UNTIL - PACKET3_SET_CONFIG_REG_OFFSET) >> 2);
+		radeon_ring_write(rdev, WAIT_3D_IDLE_bit | WAIT_3D_IDLECLEAN_bit);
+		/* Emit fence sequence & fire IRQ */
+		radeon_ring_write(rdev, PACKET3(PACKET3_SET_CONFIG_REG, 1));
+		radeon_ring_write(rdev, ((rdev->fence_drv.scratch_reg - PACKET3_SET_CONFIG_REG_OFFSET) >> 2));
+		radeon_ring_write(rdev, fence->seq);
+		/* CP_INTERRUPT packet 3 no longer exists, use packet 0 */
+		radeon_ring_write(rdev, PACKET0(CP_INT_STATUS, 0));
+		radeon_ring_write(rdev, RB_INT_STAT);
+	}
 }
 
 int r600_copy_blit(struct radeon_device *rdev,
@@ -3012,6 +3022,7 @@ int r600_irq_set(struct radeon_device *rdev)
 	if (rdev->irq.sw_int) {
 		DRM_DEBUG("r600_irq_set: sw int\n");
 		cp_int_cntl |= RB_INT_ENABLE;
+		cp_int_cntl |= TIME_STAMP_INT_ENABLE;
 	}
 	if (rdev->irq.crtc_vblank_int[0]) {
 		DRM_DEBUG("r600_irq_set: vblank 0\n");
@@ -3377,6 +3388,7 @@ restart_ih:
 			break;
 		case 181: /* CP EOP event */
 			DRM_DEBUG("IH: CP EOP\n");
+			radeon_fence_process(rdev);
 			break;
 		case 233: /* GUI IDLE */
 			DRM_DEBUG("IH: CP EOP\n");
diff --git a/drivers/gpu/drm/radeon/r600d.h b/drivers/gpu/drm/radeon/r600d.h
index 858a1920c0d7..966a793e225b 100644
--- a/drivers/gpu/drm/radeon/r600d.h
+++ b/drivers/gpu/drm/radeon/r600d.h
@@ -474,6 +474,7 @@
 #define	VGT_VERTEX_REUSE_BLOCK_CNTL			0x28C58
 #define		VTX_REUSE_DEPTH_MASK				0x000000FF
 #define VGT_EVENT_INITIATOR                             0x28a90
+#       define CACHE_FLUSH_AND_INV_EVENT_TS                     (0x14 << 0)
 #       define CACHE_FLUSH_AND_INV_EVENT                        (0x16 << 0)
 
 #define VM_CONTEXT0_CNTL				0x1410
@@ -775,7 +776,27 @@
 #define		PACKET3_ME_INITIALIZE_DEVICE_ID(x) ((x) << 16)
 #define	PACKET3_COND_WRITE				0x45
 #define	PACKET3_EVENT_WRITE				0x46
+#define		EVENT_TYPE(x)                           ((x) << 0)
+#define		EVENT_INDEX(x)                          ((x) << 8)
+                /* 0 - any non-TS event
+		 * 1 - ZPASS_DONE
+		 * 2 - SAMPLE_PIPELINESTAT
+		 * 3 - SAMPLE_STREAMOUTSTAT*
+		 * 4 - *S_PARTIAL_FLUSH
+		 * 5 - TS events
+		 */
 #define	PACKET3_EVENT_WRITE_EOP				0x47
+#define		DATA_SEL(x)                             ((x) << 29)
+                /* 0 - discard
+		 * 1 - send low 32bit data
+		 * 2 - send 64bit data
+		 * 3 - send 64bit counter value
+		 */
+#define		INT_SEL(x)                              ((x) << 24)
+                /* 0 - none
+		 * 1 - interrupt only (DATA_SEL = 0)
+		 * 2 - interrupt when data write is confirmed
+		 */
 #define	PACKET3_ONE_REG_WRITE				0x57
 #define	PACKET3_SET_CONFIG_REG				0x68
 #define		PACKET3_SET_CONFIG_REG_OFFSET			0x00008000
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 4e10938d8dd1..2bfae5679135 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -595,11 +595,13 @@ struct radeon_wb {
 	volatile uint32_t	*wb;
 	uint64_t		gpu_addr;
 	bool                    enabled;
+	bool                    use_event;
 };
 
 #define RADEON_WB_SCRATCH_OFFSET 0
 #define RADEON_WB_CP_RPTR_OFFSET 1024
 #define R600_WB_IH_WPTR_OFFSET   2048
+#define R600_WB_EVENT_OFFSET     3072
 
 /**
  * struct radeon_pm - power management datas
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index cfc162d05010..8adfedfe547f 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -208,6 +208,8 @@ int radeon_wb_init(struct radeon_device *rdev)
 		return r;
 	}
 
+	/* disable event_write fences */
+	rdev->wb.use_event = false;
 	/* disabled via module param */
 	if (radeon_no_wb == 1)
 		rdev->wb.enabled = false;
@@ -215,8 +217,12 @@ int radeon_wb_init(struct radeon_device *rdev)
 		/* often unreliable on AGP */
 		if (rdev->flags & RADEON_IS_AGP) {
 			rdev->wb.enabled = false;
-		} else
+		} else {
 			rdev->wb.enabled = true;
+			/* event_write fences are only available on r600+ */
+			if (rdev->family >= CHIP_R600)
+				rdev->wb.use_event = true;
+		}
 	}
 
 	dev_info(rdev->dev, "WB %sabled\n", rdev->wb.enabled ? "en" : "dis");
diff --git a/drivers/gpu/drm/radeon/radeon_fence.c b/drivers/gpu/drm/radeon/radeon_fence.c
index 698a7ed3a6a1..216392d0353b 100644
--- a/drivers/gpu/drm/radeon/radeon_fence.c
+++ b/drivers/gpu/drm/radeon/radeon_fence.c
@@ -73,7 +73,11 @@ static bool radeon_fence_poll_locked(struct radeon_device *rdev)
 	unsigned long cjiffies;
 
 	if (rdev->wb.enabled) {
-		u32 scratch_index = RADEON_WB_SCRATCH_OFFSET + rdev->fence_drv.scratch_reg - rdev->scratch.reg_base;
+		u32 scratch_index;
+		if (rdev->wb.use_event)
+			scratch_index = R600_WB_EVENT_OFFSET + rdev->fence_drv.scratch_reg - rdev->scratch.reg_base;
+		else
+			scratch_index = RADEON_WB_SCRATCH_OFFSET + rdev->fence_drv.scratch_reg - rdev->scratch.reg_base;
 		seq = rdev->wb.wb[scratch_index/4];
 	} else
 		seq = RREG32(rdev->fence_drv.scratch_reg);