drm/radeon: Try placing NO_CPU_ACCESS BOs outside of CPU accessible VRAM

This avoids them getting in the way of BOs which might be accessed by
the CPU. They can still go to the CPU accessible part of VRAM though if
there's no space outside of it.
Signed-off-by: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/radeon/radeon.h

@@ -474,7 +474,7 @@ struct radeon_bo {
 	struct list_head		list;
 	/* Protected by tbo.reserved */
 	u32				initial_domain;
-	struct ttm_place		placements[3];
+	struct ttm_place		placements[4];
 	struct ttm_placement		placement;
 	struct ttm_buffer_object	tbo;
 	struct ttm_bo_kmap_obj		kmap;

drivers/gpu/drm/radeon/radeon_object.c

@@ -99,22 +99,39 @@ void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
 
 	rbo->placement.placement = rbo->placements;
 	rbo->placement.busy_placement = rbo->placements;
-	if (domain & RADEON_GEM_DOMAIN_VRAM)
+	if (domain & RADEON_GEM_DOMAIN_VRAM) {
+		/* Try placing BOs which don't need CPU access outside of the
+		 * CPU accessible part of VRAM
+		 */
+		if ((rbo->flags & RADEON_GEM_NO_CPU_ACCESS) &&
+		    rbo->rdev->mc.visible_vram_size < rbo->rdev->mc.real_vram_size) {
+			rbo->placements[c].fpfn =
+				rbo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
+			rbo->placements[c++].flags = TTM_PL_FLAG_WC |
+						     TTM_PL_FLAG_UNCACHED |
+						     TTM_PL_FLAG_VRAM;
+		}
+
+		rbo->placements[c].fpfn = 0;
 		rbo->placements[c++].flags = TTM_PL_FLAG_WC |
 					     TTM_PL_FLAG_UNCACHED |
 					     TTM_PL_FLAG_VRAM;
+	}
 
 	if (domain & RADEON_GEM_DOMAIN_GTT) {
 		if (rbo->flags & RADEON_GEM_GTT_UC) {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_UNCACHED |
 				TTM_PL_FLAG_TT;
 
 		} else if ((rbo->flags & RADEON_GEM_GTT_WC) ||
 			   (rbo->rdev->flags & RADEON_IS_AGP)) {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_WC |
 				TTM_PL_FLAG_UNCACHED |
 				TTM_PL_FLAG_TT;
 		} else {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_CACHED |
 						     TTM_PL_FLAG_TT;
 		}
@@ -122,30 +139,35 @@ void radeon_ttm_placement_from_domain(struct radeon_bo *rbo, u32 domain)
 
 	if (domain & RADEON_GEM_DOMAIN_CPU) {
 		if (rbo->flags & RADEON_GEM_GTT_UC) {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_UNCACHED |
 				TTM_PL_FLAG_SYSTEM;
 
 		} else if ((rbo->flags & RADEON_GEM_GTT_WC) ||
 		    rbo->rdev->flags & RADEON_IS_AGP) {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_WC |
 				TTM_PL_FLAG_UNCACHED |
 				TTM_PL_FLAG_SYSTEM;
 		} else {
+			rbo->placements[c].fpfn = 0;
 			rbo->placements[c++].flags = TTM_PL_FLAG_CACHED |
 						     TTM_PL_FLAG_SYSTEM;
 		}
 	}
-	if (!c)
+	if (!c) {
+		rbo->placements[c].fpfn = 0;
 		rbo->placements[c++].flags = TTM_PL_MASK_CACHING |
 					     TTM_PL_FLAG_SYSTEM;
+	}
 
 	rbo->placement.num_placement = c;
 	rbo->placement.num_busy_placement = c;
 
 	for (i = 0; i < c; ++i) {
-		rbo->placements[i].fpfn = 0;
 		if ((rbo->flags & RADEON_GEM_CPU_ACCESS) &&
-		    (rbo->placements[i].flags & TTM_PL_FLAG_VRAM))
+		    (rbo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
+		    !rbo->placements[i].fpfn)
 			rbo->placements[i].lpfn =
 				rbo->rdev->mc.visible_vram_size >> PAGE_SHIFT;
 		else
@@ -743,8 +765,8 @@ int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 {
 	struct radeon_device *rdev;
 	struct radeon_bo *rbo;
-	unsigned long offset, size;
-	int r;
+	unsigned long offset, size, lpfn;
+	int i, r;
 
 	if (!radeon_ttm_bo_is_radeon_bo(bo))
 		return 0;
@@ -761,7 +783,13 @@ int radeon_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 
 	/* hurrah the memory is not visible ! */
 	radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_VRAM);
-	rbo->placements[0].lpfn = rdev->mc.visible_vram_size >> PAGE_SHIFT;
+	lpfn =	rdev->mc.visible_vram_size >> PAGE_SHIFT;
+	for (i = 0; i < rbo->placement.num_placement; i++) {
+		/* Force into visible VRAM */
+		if ((rbo->placements[i].flags & TTM_PL_FLAG_VRAM) &&
+		    (!rbo->placements[i].lpfn || rbo->placements[i].lpfn > lpfn))
+			rbo->placements[i].lpfn = lpfn;
+	}
 	r = ttm_bo_validate(bo, &rbo->placement, false, false);
 	if (unlikely(r == -ENOMEM)) {
 		radeon_ttm_placement_from_domain(rbo, RADEON_GEM_DOMAIN_GTT);