drm/amdgpu: add AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS flag v3

Add a flag noting that a BO must be created using linear VRAM
and set this flag on all in kernel users where appropriate.

Hopefully I haven't missed anything.

v2: add it in a few more places, fix CPU mapping.
v3: rename to VRAM_CONTIGUOUS, fix typo in CS code.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu_cgs.c

@@ -146,7 +146,8 @@ static int amdgpu_cgs_alloc_gpu_mem(struct cgs_device *cgs_device,
 	switch(type) {
 	case CGS_GPU_MEM_TYPE__VISIBLE_CONTIG_FB:
 	case CGS_GPU_MEM_TYPE__VISIBLE_FB:
-		flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED;
+		flags = AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
 		domain = AMDGPU_GEM_DOMAIN_VRAM;
 		if (max_offset > adev->mc.real_vram_size)
 			return -EINVAL;
@@ -157,7 +158,8 @@ static int amdgpu_cgs_alloc_gpu_mem(struct cgs_device *cgs_device,
 		break;
 	case CGS_GPU_MEM_TYPE__INVISIBLE_CONTIG_FB:
 	case CGS_GPU_MEM_TYPE__INVISIBLE_FB:
-		flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS;
+		flags = AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
+			AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
 		domain = AMDGPU_GEM_DOMAIN_VRAM;
 		if (adev->mc.visible_vram_size < adev->mc.real_vram_size) {
 			place.fpfn =

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -1195,6 +1195,15 @@ int amdgpu_cs_sysvm_access_required(struct amdgpu_cs_parser *parser)
 		r = amdgpu_ttm_bind(&bo->tbo, &bo->tbo.mem);
 		if (unlikely(r))
 			return r;
+
+		if (bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS)
+			continue;
+
+		bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
+		amdgpu_ttm_placement_from_domain(bo, bo->allowed_domains);
+		r = ttm_bo_validate(&bo->tbo, &bo->placement, false, false);
+		if (unlikely(r))
+			return r;
 	}
 
 	return 0;

drivers/gpu/drm/amd/amdgpu/amdgpu_device.c

@@ -264,7 +264,8 @@ static int amdgpu_vram_scratch_init(struct amdgpu_device *adev)
 	if (adev->vram_scratch.robj == NULL) {
 		r = amdgpu_bo_create(adev, AMDGPU_GPU_PAGE_SIZE,
 				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 				     NULL, NULL, &adev->vram_scratch.robj);
 		if (r) {
 			return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_fb.c

@@ -152,7 +152,8 @@ static int amdgpufb_create_pinned_object(struct amdgpu_fbdev *rfbdev,
 	aligned_size = ALIGN(size, PAGE_SIZE);
 	ret = amdgpu_gem_object_create(adev, aligned_size, 0,
 				       AMDGPU_GEM_DOMAIN_VRAM,
-				       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+				       AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+				       AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 				       true, &gobj);
 	if (ret) {
 		printk(KERN_ERR "failed to allocate framebuffer (%d)\n",

drivers/gpu/drm/amd/amdgpu/amdgpu_gart.c

@@ -126,7 +126,8 @@ int amdgpu_gart_table_vram_alloc(struct amdgpu_device *adev)
 	if (adev->gart.robj == NULL) {
 		r = amdgpu_bo_create(adev, adev->gart.table_size,
 				     PAGE_SIZE, true, AMDGPU_GEM_DOMAIN_VRAM,
-				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+				     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 				     NULL, NULL, &adev->gart.robj);
 		if (r) {
 			return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_object.c

@@ -245,7 +245,8 @@ int amdgpu_bo_create_kernel(struct amdgpu_device *adev,
 	int r;
 
 	r = amdgpu_bo_create(adev, size, align, true, domain,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, bo_ptr);
 	if (r) {
 		dev_err(adev->dev, "(%d) failed to allocate kernel bo\n", r);
@@ -643,6 +644,8 @@ int amdgpu_bo_pin_restricted(struct amdgpu_bo *bo, u32 domain,
 
 		return 0;
 	}
+
+	bo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
 	amdgpu_ttm_placement_from_domain(bo, domain);
 	for (i = 0; i < bo->placement.num_placement; i++) {
 		/* force to pin into visible video ram */
@@ -885,7 +888,9 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 
 	size = bo->mem.num_pages << PAGE_SHIFT;
 	offset = bo->mem.start << PAGE_SHIFT;
-	if ((offset + size) <= adev->mc.visible_vram_size)
+	/* TODO: figure out how to map scattered VRAM to the CPU */
+	if ((offset + size) <= adev->mc.visible_vram_size &&
+	    (abo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS))
 		return 0;
 
 	/* Can't move a pinned BO to visible VRAM */
@@ -893,6 +898,7 @@ int amdgpu_bo_fault_reserve_notify(struct ttm_buffer_object *bo)
 		return -EINVAL;
 
 	/* hurrah the memory is not visible ! */
+	abo->flags |= AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS;
 	amdgpu_ttm_placement_from_domain(abo, AMDGPU_GEM_DOMAIN_VRAM);
 	lpfn =	adev->mc.visible_vram_size >> PAGE_SHIFT;
 	for (i = 0; i < abo->placement.num_placement; i++) {
@@ -954,6 +960,8 @@ u64 amdgpu_bo_gpu_offset(struct amdgpu_bo *bo)
 	WARN_ON_ONCE(!ww_mutex_is_locked(&bo->tbo.resv->lock) &&
 		     !bo->pin_count);
 	WARN_ON_ONCE(bo->tbo.mem.start == AMDGPU_BO_INVALID_OFFSET);
+	WARN_ON_ONCE(bo->tbo.mem.mem_type == TTM_PL_VRAM &&
+		     !(bo->flags & AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS));
 
 	return bo->tbo.offset;
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c

@@ -1119,7 +1119,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
 
 	r = amdgpu_bo_create(adev, 256 * 1024, PAGE_SIZE, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, &adev->stollen_vga_memory);
 	if (r) {
 		return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c

@@ -1002,7 +1002,8 @@ int amdgpu_uvd_get_create_msg(struct amdgpu_ring *ring, uint32_t handle,
 
 	r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, &bo);
 	if (r)
 		return r;
@@ -1051,7 +1052,8 @@ int amdgpu_uvd_get_destroy_msg(struct amdgpu_ring *ring, uint32_t handle,
 
 	r = amdgpu_bo_create(adev, 1024, PAGE_SIZE, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, &bo);
 	if (r)
 		return r;

drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c

@@ -157,7 +157,8 @@ int amdgpu_vce_sw_init(struct amdgpu_device *adev, unsigned long size)
 
 	r = amdgpu_bo_create(adev, size, PAGE_SIZE, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
-			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+			     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, &adev->vce.vcpu_bo);
 	if (r) {
 		dev_err(adev->dev, "(%d) failed to allocate VCE bo\n", r);

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -1416,7 +1416,8 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
 				     AMDGPU_GPU_PAGE_SIZE, true,
 				     AMDGPU_GEM_DOMAIN_VRAM,
 				     AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
-				     AMDGPU_GEM_CREATE_SHADOW,
+				     AMDGPU_GEM_CREATE_SHADOW |
+				     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 				     NULL, resv, &pt);
 		if (r)
 			goto error_free;
@@ -1626,7 +1627,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm)
 	r = amdgpu_bo_create(adev, pd_size, align, true,
 			     AMDGPU_GEM_DOMAIN_VRAM,
 			     AMDGPU_GEM_CREATE_NO_CPU_ACCESS |
-			     AMDGPU_GEM_CREATE_SHADOW,
+			     AMDGPU_GEM_CREATE_SHADOW |
+			     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 			     NULL, NULL, &vm->page_directory);
 	if (r)
 		goto error_free_sched_entity;

drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c

@@ -3391,7 +3391,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 		if (adev->gfx.rlc.save_restore_obj == NULL) {
 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
 					     AMDGPU_GEM_DOMAIN_VRAM,
-					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 					     NULL, NULL,
 					     &adev->gfx.rlc.save_restore_obj);
 			if (r) {
@@ -3435,7 +3436,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 		if (adev->gfx.rlc.clear_state_obj == NULL) {
 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
 					     AMDGPU_GEM_DOMAIN_VRAM,
-					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 					     NULL, NULL,
 					     &adev->gfx.rlc.clear_state_obj);
 			if (r) {
@@ -3475,7 +3477,8 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev)
 		if (adev->gfx.rlc.cp_table_obj == NULL) {
 			r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
 					     AMDGPU_GEM_DOMAIN_VRAM,
-					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 					     NULL, NULL,
 					     &adev->gfx.rlc.cp_table_obj);
 			if (r) {

drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c

@@ -1273,7 +1273,8 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
 		if (adev->gfx.rlc.clear_state_obj == NULL) {
 			r = amdgpu_bo_create(adev, dws * 4, PAGE_SIZE, true,
 					     AMDGPU_GEM_DOMAIN_VRAM,
-					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 					     NULL, NULL,
 					     &adev->gfx.rlc.clear_state_obj);
 			if (r) {
@@ -1315,7 +1316,8 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev)
 		if (adev->gfx.rlc.cp_table_obj == NULL) {
 			r = amdgpu_bo_create(adev, adev->gfx.rlc.cp_table_size, PAGE_SIZE, true,
 					     AMDGPU_GEM_DOMAIN_VRAM,
-					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED,
+					     AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED |
+					     AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS,
 					     NULL, NULL,
 					     &adev->gfx.rlc.cp_table_obj);
 			if (r) {

include/uapi/drm/amdgpu_drm.h

@@ -81,6 +81,8 @@ extern "C" {
 #define AMDGPU_GEM_CREATE_VRAM_CLEARED		(1 << 3)
 /* Flag that create shadow bo(GTT) while allocating vram bo */
 #define AMDGPU_GEM_CREATE_SHADOW		(1 << 4)
+/* Flag that allocating the BO should use linear VRAM */
+#define AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS	(1 << 5)
 
 struct drm_amdgpu_gem_create_in  {
 	/** the requested memory size */