Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/radeon/kms: Make GPU/CPU page size handling consistent in blit code (v2)
  drm/radeon/kms: fix typo in r100_blit_copy
  drm/radeon: Unreference GEM object outside of spinlock in page flip error path.
  drm/radeon: Don't read from CP ring write pointer registers.
  drm/ttm: request zeroed system memory pages for new TT buffer objects

drivers/gpu/drm/radeon/evergreen.c

@@ -1404,7 +1404,8 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 	/* Initialize the ring buffer's read and write pointers */
 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
 	WREG32(CP_RB_RPTR_WR, 0);
-	WREG32(CP_RB_WPTR, 0);
+	rdev->cp.wptr = 0;
+	WREG32(CP_RB_WPTR, rdev->cp.wptr);
 
 	/* set the wb address wether it's enabled or not */
 	WREG32(CP_RB_RPTR_ADDR,
@@ -1426,7 +1427,6 @@ int evergreen_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
 
 	rdev->cp.rptr = RREG32(CP_RB_RPTR);
-	rdev->cp.wptr = RREG32(CP_RB_WPTR);
 
 	evergreen_cp_start(rdev);
 	rdev->cp.ready = true;
@@ -3171,21 +3171,23 @@ int evergreen_suspend(struct radeon_device *rdev)
 }
 
 int evergreen_copy_blit(struct radeon_device *rdev,
-			uint64_t src_offset, uint64_t dst_offset,
-			unsigned num_pages, struct radeon_fence *fence)
+			uint64_t src_offset,
+			uint64_t dst_offset,
+			unsigned num_gpu_pages,
+			struct radeon_fence *fence)
 {
 	int r;
 
 	mutex_lock(&rdev->r600_blit.mutex);
 	rdev->r600_blit.vb_ib = NULL;
-	r = evergreen_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
+	r = evergreen_blit_prepare_copy(rdev, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
 	if (r) {
 		if (rdev->r600_blit.vb_ib)
 			radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
 		mutex_unlock(&rdev->r600_blit.mutex);
 		return r;
 	}
-	evergreen_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
+	evergreen_kms_blit_copy(rdev, src_offset, dst_offset, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
 	evergreen_blit_done_copy(rdev, fence);
 	mutex_unlock(&rdev->r600_blit.mutex);
 	return 0;

drivers/gpu/drm/radeon/ni.c

@@ -1187,7 +1187,8 @@ int cayman_cp_resume(struct radeon_device *rdev)
 
 	/* Initialize the ring buffer's read and write pointers */
 	WREG32(CP_RB0_CNTL, tmp | RB_RPTR_WR_ENA);
-	WREG32(CP_RB0_WPTR, 0);
+	rdev->cp.wptr = 0;
+	WREG32(CP_RB0_WPTR, rdev->cp.wptr);
 
 	/* set the wb address wether it's enabled or not */
 	WREG32(CP_RB0_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP_RPTR_OFFSET) & 0xFFFFFFFC);
@@ -1207,7 +1208,6 @@ int cayman_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB0_BASE, rdev->cp.gpu_addr >> 8);
 
 	rdev->cp.rptr = RREG32(CP_RB0_RPTR);
-	rdev->cp.wptr = RREG32(CP_RB0_WPTR);
 
 	/* ring1  - compute only */
 	/* Set ring buffer size */
@@ -1220,7 +1220,8 @@ int cayman_cp_resume(struct radeon_device *rdev)
 
 	/* Initialize the ring buffer's read and write pointers */
 	WREG32(CP_RB1_CNTL, tmp | RB_RPTR_WR_ENA);
-	WREG32(CP_RB1_WPTR, 0);
+	rdev->cp1.wptr = 0;
+	WREG32(CP_RB1_WPTR, rdev->cp1.wptr);
 
 	/* set the wb address wether it's enabled or not */
 	WREG32(CP_RB1_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP1_RPTR_OFFSET) & 0xFFFFFFFC);
@@ -1232,7 +1233,6 @@ int cayman_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB1_BASE, rdev->cp1.gpu_addr >> 8);
 
 	rdev->cp1.rptr = RREG32(CP_RB1_RPTR);
-	rdev->cp1.wptr = RREG32(CP_RB1_WPTR);
 
 	/* ring2 - compute only */
 	/* Set ring buffer size */
@@ -1245,7 +1245,8 @@ int cayman_cp_resume(struct radeon_device *rdev)
 
 	/* Initialize the ring buffer's read and write pointers */
 	WREG32(CP_RB2_CNTL, tmp | RB_RPTR_WR_ENA);
-	WREG32(CP_RB2_WPTR, 0);
+	rdev->cp2.wptr = 0;
+	WREG32(CP_RB2_WPTR, rdev->cp2.wptr);
 
 	/* set the wb address wether it's enabled or not */
 	WREG32(CP_RB2_RPTR_ADDR, (rdev->wb.gpu_addr + RADEON_WB_CP2_RPTR_OFFSET) & 0xFFFFFFFC);
@@ -1257,7 +1258,6 @@ int cayman_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_RB2_BASE, rdev->cp2.gpu_addr >> 8);
 
 	rdev->cp2.rptr = RREG32(CP_RB2_RPTR);
-	rdev->cp2.wptr = RREG32(CP_RB2_WPTR);
 
 	/* start the rings */
 	cayman_cp_start(rdev);

drivers/gpu/drm/radeon/r100.c

@@ -721,11 +721,11 @@ void r100_fence_ring_emit(struct radeon_device *rdev,
 int r100_copy_blit(struct radeon_device *rdev,
 		   uint64_t src_offset,
 		   uint64_t dst_offset,
-		   unsigned num_pages,
+		   unsigned num_gpu_pages,
 		   struct radeon_fence *fence)
 {
 	uint32_t cur_pages;
-	uint32_t stride_bytes = PAGE_SIZE;
+	uint32_t stride_bytes = RADEON_GPU_PAGE_SIZE;
 	uint32_t pitch;
 	uint32_t stride_pixels;
 	unsigned ndw;
@@ -737,7 +737,7 @@ int r100_copy_blit(struct radeon_device *rdev,
 	/* radeon pitch is /64 */
 	pitch = stride_bytes / 64;
 	stride_pixels = stride_bytes / 4;
-	num_loops = DIV_ROUND_UP(num_pages, 8191);
+	num_loops = DIV_ROUND_UP(num_gpu_pages, 8191);
 
 	/* Ask for enough room for blit + flush + fence */
 	ndw = 64 + (10 * num_loops);
@@ -746,12 +746,12 @@ int r100_copy_blit(struct radeon_device *rdev,
 		DRM_ERROR("radeon: moving bo (%d) asking for %u dw.\n", r, ndw);
 		return -EINVAL;
 	}
-	while (num_pages > 0) {
-		cur_pages = num_pages;
+	while (num_gpu_pages > 0) {
+		cur_pages = num_gpu_pages;
 		if (cur_pages > 8191) {
 			cur_pages = 8191;
 		}
-		num_pages -= cur_pages;
+		num_gpu_pages -= cur_pages;
 
 		/* pages are in Y direction - height
 		   page width in X direction - width */
@@ -773,8 +773,8 @@ int r100_copy_blit(struct radeon_device *rdev,
 		radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
 		radeon_ring_write(rdev, 0);
 		radeon_ring_write(rdev, (0x1fff) | (0x1fff << 16));
-		radeon_ring_write(rdev, num_pages);
-		radeon_ring_write(rdev, num_pages);
+		radeon_ring_write(rdev, cur_pages);
+		radeon_ring_write(rdev, cur_pages);
 		radeon_ring_write(rdev, cur_pages | (stride_pixels << 16));
 	}
 	radeon_ring_write(rdev, PACKET0(RADEON_DSTCACHE_CTLSTAT, 0));
@@ -990,7 +990,8 @@ int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 	/* Force read & write ptr to 0 */
 	WREG32(RADEON_CP_RB_CNTL, tmp | RADEON_RB_RPTR_WR_ENA | RADEON_RB_NO_UPDATE);
 	WREG32(RADEON_CP_RB_RPTR_WR, 0);
-	WREG32(RADEON_CP_RB_WPTR, 0);
+	rdev->cp.wptr = 0;
+	WREG32(RADEON_CP_RB_WPTR, rdev->cp.wptr);
 
 	/* set the wb address whether it's enabled or not */
 	WREG32(R_00070C_CP_RB_RPTR_ADDR,
@@ -1007,9 +1008,6 @@ int r100_cp_init(struct radeon_device *rdev, unsigned ring_size)
 	WREG32(RADEON_CP_RB_CNTL, tmp);
 	udelay(10);
 	rdev->cp.rptr = RREG32(RADEON_CP_RB_RPTR);
-	rdev->cp.wptr = RREG32(RADEON_CP_RB_WPTR);
-	/* protect against crazy HW on resume */
-	rdev->cp.wptr &= rdev->cp.ptr_mask;
 	/* Set cp mode to bus mastering & enable cp*/
 	WREG32(RADEON_CP_CSQ_MODE,
 	       REG_SET(RADEON_INDIRECT2_START, indirect2_start) |

drivers/gpu/drm/radeon/r200.c

@@ -84,7 +84,7 @@ static int r200_get_vtx_size_0(uint32_t vtx_fmt_0)
 int r200_copy_dma(struct radeon_device *rdev,
 		  uint64_t src_offset,
 		  uint64_t dst_offset,
-		  unsigned num_pages,
+		  unsigned num_gpu_pages,
 		  struct radeon_fence *fence)
 {
 	uint32_t size;
@@ -93,7 +93,7 @@ int r200_copy_dma(struct radeon_device *rdev,
 	int r = 0;
 
 	/* radeon pitch is /64 */
-	size = num_pages << PAGE_SHIFT;
+	size = num_gpu_pages << RADEON_GPU_PAGE_SHIFT;
 	num_loops = DIV_ROUND_UP(size, 0x1FFFFF);
 	r = radeon_ring_lock(rdev, num_loops * 4 + 64);
 	if (r) {

drivers/gpu/drm/radeon/r600.c

@@ -2209,7 +2209,8 @@ int r600_cp_resume(struct radeon_device *rdev)
 	/* Initialize the ring buffer's read and write pointers */
 	WREG32(CP_RB_CNTL, tmp | RB_RPTR_WR_ENA);
 	WREG32(CP_RB_RPTR_WR, 0);
-	WREG32(CP_RB_WPTR, 0);
+	rdev->cp.wptr = 0;
+	WREG32(CP_RB_WPTR, rdev->cp.wptr);
 
 	/* set the wb address whether it's enabled or not */
 	WREG32(CP_RB_RPTR_ADDR,
@@ -2231,7 +2232,6 @@ int r600_cp_resume(struct radeon_device *rdev)
 	WREG32(CP_DEBUG, (1 << 27) | (1 << 28));
 
 	rdev->cp.rptr = RREG32(CP_RB_RPTR);
-	rdev->cp.wptr = RREG32(CP_RB_WPTR);
 
 	r600_cp_start(rdev);
 	rdev->cp.ready = true;
@@ -2353,21 +2353,23 @@ void r600_fence_ring_emit(struct radeon_device *rdev,
 }
 
 int r600_copy_blit(struct radeon_device *rdev,
-		   uint64_t src_offset, uint64_t dst_offset,
-		   unsigned num_pages, struct radeon_fence *fence)
+		   uint64_t src_offset,
+		   uint64_t dst_offset,
+		   unsigned num_gpu_pages,
+		   struct radeon_fence *fence)
 {
 	int r;
 
 	mutex_lock(&rdev->r600_blit.mutex);
 	rdev->r600_blit.vb_ib = NULL;
-	r = r600_blit_prepare_copy(rdev, num_pages * RADEON_GPU_PAGE_SIZE);
+	r = r600_blit_prepare_copy(rdev, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
 	if (r) {
 		if (rdev->r600_blit.vb_ib)
 			radeon_ib_free(rdev, &rdev->r600_blit.vb_ib);
 		mutex_unlock(&rdev->r600_blit.mutex);
 		return r;
 	}
-	r600_kms_blit_copy(rdev, src_offset, dst_offset, num_pages * RADEON_GPU_PAGE_SIZE);
+	r600_kms_blit_copy(rdev, src_offset, dst_offset, num_gpu_pages * RADEON_GPU_PAGE_SIZE);
 	r600_blit_done_copy(rdev, fence);
 	mutex_unlock(&rdev->r600_blit.mutex);
 	return 0;

drivers/gpu/drm/radeon/radeon.h

@@ -322,6 +322,7 @@ union radeon_gart_table {
 
 #define RADEON_GPU_PAGE_SIZE 4096
 #define RADEON_GPU_PAGE_MASK (RADEON_GPU_PAGE_SIZE - 1)
+#define RADEON_GPU_PAGE_SHIFT 12
 
 struct radeon_gart {
 	dma_addr_t			table_addr;
@@ -914,17 +915,17 @@ struct radeon_asic {
 	int (*copy_blit)(struct radeon_device *rdev,
 			 uint64_t src_offset,
 			 uint64_t dst_offset,
-			 unsigned num_pages,
+			 unsigned num_gpu_pages,
 			 struct radeon_fence *fence);
 	int (*copy_dma)(struct radeon_device *rdev,
 			uint64_t src_offset,
 			uint64_t dst_offset,
-			unsigned num_pages,
+			unsigned num_gpu_pages,
 			struct radeon_fence *fence);
 	int (*copy)(struct radeon_device *rdev,
 		    uint64_t src_offset,
 		    uint64_t dst_offset,
-		    unsigned num_pages,
+		    unsigned num_gpu_pages,
 		    struct radeon_fence *fence);
 	uint32_t (*get_engine_clock)(struct radeon_device *rdev);
 	void (*set_engine_clock)(struct radeon_device *rdev, uint32_t eng_clock);

drivers/gpu/drm/radeon/radeon_asic.h

@@ -75,7 +75,7 @@ uint32_t r100_pll_rreg(struct radeon_device *rdev, uint32_t reg);
 int r100_copy_blit(struct radeon_device *rdev,
 		   uint64_t src_offset,
 		   uint64_t dst_offset,
-		   unsigned num_pages,
+		   unsigned num_gpu_pages,
 		   struct radeon_fence *fence);
 int r100_set_surface_reg(struct radeon_device *rdev, int reg,
 			 uint32_t tiling_flags, uint32_t pitch,
@@ -143,7 +143,7 @@ extern void r100_post_page_flip(struct radeon_device *rdev, int crtc);
 extern int r200_copy_dma(struct radeon_device *rdev,
 			 uint64_t src_offset,
 			 uint64_t dst_offset,
-			 unsigned num_pages,
+			 unsigned num_gpu_pages,
 			 struct radeon_fence *fence);
 void r200_set_safe_registers(struct radeon_device *rdev);
 
@@ -311,7 +311,7 @@ void r600_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib);
 int r600_ring_test(struct radeon_device *rdev);
 int r600_copy_blit(struct radeon_device *rdev,
 		   uint64_t src_offset, uint64_t dst_offset,
-		   unsigned num_pages, struct radeon_fence *fence);
+		   unsigned num_gpu_pages, struct radeon_fence *fence);
 void r600_hpd_init(struct radeon_device *rdev);
 void r600_hpd_fini(struct radeon_device *rdev);
 bool r600_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd);
@@ -403,7 +403,7 @@ void evergreen_bandwidth_update(struct radeon_device *rdev);
 void evergreen_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib);
 int evergreen_copy_blit(struct radeon_device *rdev,
 			uint64_t src_offset, uint64_t dst_offset,
-			unsigned num_pages, struct radeon_fence *fence);
+			unsigned num_gpu_pages, struct radeon_fence *fence);
 void evergreen_hpd_init(struct radeon_device *rdev);
 void evergreen_hpd_fini(struct radeon_device *rdev);
 bool evergreen_hpd_sense(struct radeon_device *rdev, enum radeon_hpd_id hpd);

drivers/gpu/drm/radeon/radeon_display.c

@@ -473,8 +473,8 @@ pflip_cleanup:
 	spin_lock_irqsave(&dev->event_lock, flags);
 	radeon_crtc->unpin_work = NULL;
 unlock_free:
-	drm_gem_object_unreference_unlocked(old_radeon_fb->obj);
 	spin_unlock_irqrestore(&dev->event_lock, flags);
+	drm_gem_object_unreference_unlocked(old_radeon_fb->obj);
 	radeon_fence_unref(&work->fence);
 	kfree(work);
 

drivers/gpu/drm/radeon/radeon_ttm.c

@@ -277,7 +277,12 @@ static int radeon_move_blit(struct ttm_buffer_object *bo,
 		DRM_ERROR("Trying to move memory with CP turned off.\n");
 		return -EINVAL;
 	}
-	r = radeon_copy(rdev, old_start, new_start, new_mem->num_pages, fence);
+
+	BUILD_BUG_ON((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) != 0);
+
+	r = radeon_copy(rdev, old_start, new_start,
+			new_mem->num_pages * (PAGE_SIZE / RADEON_GPU_PAGE_SIZE), /* GPU pages */
+			fence);
 	/* FIXME: handle copy error */
 	r = ttm_bo_move_accel_cleanup(bo, (void *)fence, NULL,
 				      evict, no_wait_reserve, no_wait_gpu, new_mem);

drivers/gpu/drm/ttm/ttm_bo.c

@@ -394,7 +394,8 @@ static int ttm_bo_handle_move_mem(struct ttm_buffer_object *bo,
 
 	if (!(new_man->flags & TTM_MEMTYPE_FLAG_FIXED)) {
 		if (bo->ttm == NULL) {
-			ret = ttm_bo_add_ttm(bo, false);
+			bool zero = !(old_man->flags & TTM_MEMTYPE_FLAG_FIXED);
+			ret = ttm_bo_add_ttm(bo, zero);
 			if (ret)
 				goto out_err;
 		}