static int radeon_move_blit(struct ttm_buffer_object *bo,
bool evict, int no_wait_reserve, bool no_wait_gpu,
struct ttm_mem_reg *new_mem,
struct ttm_mem_reg *old_mem)
{
struct radeon_device *rdev;
uint64_t old_start, new_start;
struct radeon_fence *fence;
int r;
rdev = radeon_get_rdev(bo->bdev);
r = radeon_fence_create(rdev, &fence);
if (unlikely(r)) {
return r;
}
old_start = old_mem->start << PAGE_SHIFT;
new_start = new_mem->start << PAGE_SHIFT;
switch (old_mem->mem_type) {
case TTM_PL_VRAM:
old_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
old_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
switch (new_mem->mem_type) {
case TTM_PL_VRAM:
new_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
new_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
if (!rdev->cp.ready) {
DRM_ERROR("Trying to move memory with CP turned off.\n");
return -EINVAL;
}
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);
radeon_fence_unref(&fence);
return r;
}
static int radeon_move_blit(struct ttm_buffer_object *bo,
bool evict, bool no_wait_gpu,
struct ttm_mem_reg *new_mem,
struct ttm_mem_reg *old_mem)
{
struct radeon_device *rdev;
uint64_t old_start, new_start;
struct radeon_fence *fence;
int r, ridx;
rdev = radeon_get_rdev(bo->bdev);
ridx = radeon_copy_ring_index(rdev);
old_start = old_mem->start << PAGE_SHIFT;
new_start = new_mem->start << PAGE_SHIFT;
switch (old_mem->mem_type) {
case TTM_PL_VRAM:
old_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
old_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
switch (new_mem->mem_type) {
case TTM_PL_VRAM:
new_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
new_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
if (!rdev->ring[ridx].ready) {
DRM_ERROR("Trying to move memory with ring turned off.\n");
return -EINVAL;
}
CTASSERT((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) == 0);
/* sync other rings */
fence = bo->sync_obj;
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,
evict, no_wait_gpu, new_mem);
radeon_fence_unref(&fence);
return r;
}
static int radeon_move_blit(struct ttm_buffer_object *bo,
bool evict, int no_wait,
struct ttm_mem_reg *new_mem,
struct ttm_mem_reg *old_mem)
{
struct radeon_device *rdev;
uint64_t old_start, new_start;
struct radeon_fence *fence;
int r;
rdev = radeon_get_rdev(bo->bdev);
r = radeon_fence_create(rdev, &fence);
if (unlikely(r)) {
return r;
}
old_start = old_mem->mm_node->start << PAGE_SHIFT;
new_start = new_mem->mm_node->start << PAGE_SHIFT;
switch (old_mem->mem_type) {
case TTM_PL_VRAM:
old_start += rdev->mc.vram_location;
break;
case TTM_PL_TT:
old_start += rdev->mc.gtt_location;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
switch (new_mem->mem_type) {
case TTM_PL_VRAM:
new_start += rdev->mc.vram_location;
break;
case TTM_PL_TT:
new_start += rdev->mc.gtt_location;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
if (!rdev->cp.ready) {
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);
r = ttm_bo_move_accel_cleanup(bo, (void *)fence, NULL,
evict, no_wait, new_mem);
radeon_fence_unref(&fence);
return r;
}
static int radeon_move_blit(struct ttm_buffer_object *bo,
bool evict, int no_wait_reserve, bool no_wait_gpu,
struct ttm_mem_reg *new_mem,
struct ttm_mem_reg *old_mem)
{
struct radeon_device *rdev;
uint64_t old_start, new_start;
struct radeon_fence *fence;
int r, i;
rdev = radeon_get_rdev(bo->bdev);
r = radeon_fence_create(rdev, &fence, radeon_copy_ring_index(rdev));
if (unlikely(r)) {
return r;
}
old_start = old_mem->start << PAGE_SHIFT;
new_start = new_mem->start << PAGE_SHIFT;
switch (old_mem->mem_type) {
case TTM_PL_VRAM:
old_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
old_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
switch (new_mem->mem_type) {
case TTM_PL_VRAM:
new_start += rdev->mc.vram_start;
break;
case TTM_PL_TT:
new_start += rdev->mc.gtt_start;
break;
default:
DRM_ERROR("Unknown placement %d\n", old_mem->mem_type);
return -EINVAL;
}
if (!rdev->ring[radeon_copy_ring_index(rdev)].ready) {
DRM_ERROR("Trying to move memory with ring turned off.\n");
return -EINVAL;
}
BUILD_BUG_ON((PAGE_SIZE % RADEON_GPU_PAGE_SIZE) != 0);
/* sync other rings */
if (rdev->family >= CHIP_R600) {
for (i = 0; i < RADEON_NUM_RINGS; ++i) {
/* no need to sync to our own or unused rings */
if (i == radeon_copy_ring_index(rdev) || !rdev->ring[i].ready)
continue;
if (!fence->semaphore) {
r = radeon_semaphore_create(rdev, &fence->semaphore);
/* FIXME: handle semaphore error */
if (r)
continue;
}
r = radeon_ring_lock(rdev, &rdev->ring[i], 3);
/* FIXME: handle ring lock error */
if (r)
continue;
radeon_semaphore_emit_signal(rdev, i, fence->semaphore);
radeon_ring_unlock_commit(rdev, &rdev->ring[i]);
r = radeon_ring_lock(rdev, &rdev->ring[radeon_copy_ring_index(rdev)], 3);
/* FIXME: handle ring lock error */
if (r)
continue;
radeon_semaphore_emit_wait(rdev, radeon_copy_ring_index(rdev), fence->semaphore);
radeon_ring_unlock_commit(rdev, &rdev->ring[radeon_copy_ring_index(rdev)]);
}
}
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);
radeon_fence_unref(&fence);
return r;
}
/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
void radeon_test_moves(struct radeon_device *rdev)
{
struct radeon_bo *vram_obj = NULL;
struct radeon_bo **gtt_obj = NULL;
struct radeon_fence *fence = NULL;
uint64_t gtt_addr, vram_addr;
unsigned i, n, size;
int r;
size = 1024 * 1024;
/* Number of tests =
* (Total GTT - IB pool - writeback page - ring buffers) / test size
*/
n = rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024;
for (i = 0; i < RADEON_NUM_RINGS; ++i)
n -= rdev->ring[i].ring_size;
if (rdev->wb.wb_obj)
n -= RADEON_GPU_PAGE_SIZE;
if (rdev->ih.ring_obj)
n -= rdev->ih.ring_size;
n /= size;
gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
if (!gtt_obj) {
DRM_ERROR("Failed to allocate %d pointers\n", n);
r = 1;
goto out_cleanup;
}
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, RADEON_GEM_DOMAIN_VRAM,
&vram_obj);
if (r) {
DRM_ERROR("Failed to create VRAM object\n");
goto out_cleanup;
}
r = radeon_bo_reserve(vram_obj, false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
if (r) {
DRM_ERROR("Failed to pin VRAM object\n");
goto out_cleanup;
}
for (i = 0; i < n; i++) {
void *gtt_map, *vram_map;
void **gtt_start, **gtt_end;
void **vram_start, **vram_end;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_GTT, gtt_obj + i);
if (r) {
DRM_ERROR("Failed to create GTT object %d\n", i);
goto out_cleanup;
}
r = radeon_bo_reserve(gtt_obj[i], false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
if (r) {
DRM_ERROR("Failed to pin GTT object %d\n", i);
goto out_cleanup;
}
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size;
gtt_start < gtt_end;
gtt_start++)
*gtt_start = gtt_start;
radeon_bo_kunmap(gtt_obj[i]);
r = radeon_fence_create(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
if (r) {
DRM_ERROR("Failed to create GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
r = radeon_bo_kmap(vram_obj, &vram_map);
if (r) {
DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
vram_start < vram_end;
gtt_start++, vram_start++) {
if (*vram_start != gtt_start) {
DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
"expected 0x%p (GTT/VRAM offset "
"0x%16llx/0x%16llx)\n",
i, *vram_start, gtt_start,
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)gtt_start - gtt_map),
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)gtt_start - gtt_map));
radeon_bo_kunmap(vram_obj);
goto out_cleanup;
}
*vram_start = vram_start;
}
radeon_bo_kunmap(vram_obj);
r = radeon_fence_create(rdev, &fence, RADEON_RING_TYPE_GFX_INDEX);
if (r) {
DRM_ERROR("Failed to create VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object after copy %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
gtt_start < gtt_end;
gtt_start++, vram_start++) {
if (*gtt_start != vram_start) {
DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
"expected 0x%p (VRAM/GTT offset "
"0x%16llx/0x%16llx)\n",
i, *gtt_start, vram_start,
(unsigned long long)
(vram_addr - rdev->mc.vram_start +
(void*)vram_start - vram_map),
(unsigned long long)
(gtt_addr - rdev->mc.gtt_start +
(void*)vram_start - vram_map));
radeon_bo_kunmap(gtt_obj[i]);
goto out_cleanup;
}
}
radeon_bo_kunmap(gtt_obj[i]);
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
gtt_addr - rdev->mc.gtt_start);
}
out_cleanup:
if (vram_obj) {
if (radeon_bo_is_reserved(vram_obj)) {
radeon_bo_unpin(vram_obj);
radeon_bo_unreserve(vram_obj);
}
radeon_bo_unref(&vram_obj);
}
if (gtt_obj) {
for (i = 0; i < n; i++) {
if (gtt_obj[i]) {
if (radeon_bo_is_reserved(gtt_obj[i])) {
radeon_bo_unpin(gtt_obj[i]);
radeon_bo_unreserve(gtt_obj[i]);
}
radeon_bo_unref(>t_obj[i]);
}
}
kfree(gtt_obj);
}
if (fence) {
radeon_fence_unref(&fence);
}
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
void radeon_test_moves(struct radeon_device *rdev)
{
struct radeon_object *vram_obj = NULL;
struct radeon_object **gtt_obj = NULL;
struct radeon_fence *fence = NULL;
uint64_t gtt_addr, vram_addr;
unsigned i, n, size;
int r;
size = 1024 * 1024;
n = ((u32)(rdev->mc.gtt_size - RADEON_IB_POOL_SIZE*64*1024 - RADEON_GPU_PAGE_SIZE -
rdev->cp.ring_size)) / size;
gtt_obj = kzalloc(n * sizeof(*gtt_obj), GFP_KERNEL);
if (!gtt_obj) {
DRM_ERROR("Failed to allocate %d pointers\n", n);
r = 1;
goto out_cleanup;
}
r = radeon_object_create(rdev, NULL, size, true, RADEON_GEM_DOMAIN_VRAM,
false, &vram_obj);
if (r) {
DRM_ERROR("Failed to create VRAM object\n");
goto out_cleanup;
}
r = radeon_object_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
if (r) {
DRM_ERROR("Failed to pin VRAM object\n");
goto out_cleanup;
}
for (i = 0; i < n; i++) {
void *gtt_map, *vram_map;
void **gtt_start, **gtt_end;
void **vram_start, **vram_end;
r = radeon_object_create(rdev, NULL, size, true,
RADEON_GEM_DOMAIN_GTT, false, gtt_obj + i);
if (r) {
DRM_ERROR("Failed to create GTT object %d\n", i);
goto out_cleanup;
}
r = radeon_object_pin(gtt_obj[i], RADEON_GEM_DOMAIN_GTT, >t_addr);
if (r) {
DRM_ERROR("Failed to pin GTT object %d\n", i);
goto out_cleanup;
}
r = radeon_object_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size;
gtt_start < gtt_end;
gtt_start++)
*gtt_start = gtt_start;
radeon_object_kunmap(gtt_obj[i]);
r = radeon_fence_create(rdev, &fence);
if (r) {
DRM_ERROR("Failed to create GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
r = radeon_object_kmap(vram_obj, &vram_map);
if (r) {
DRM_ERROR("Failed to map VRAM object after copy %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
vram_start < vram_end;
gtt_start++, vram_start++) {
if (*vram_start != gtt_start) {
DRM_ERROR("Incorrect GTT->VRAM copy %d: Got 0x%p, "
"expected 0x%p (GTT map 0x%p-0x%p)\n",
i, *vram_start, gtt_start, gtt_map,
gtt_end);
radeon_object_kunmap(vram_obj);
goto out_cleanup;
}
*vram_start = vram_start;
}
radeon_object_kunmap(vram_obj);
r = radeon_fence_create(rdev, &fence);
if (r) {
DRM_ERROR("Failed to create VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
r = radeon_copy(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_cleanup;
}
radeon_fence_unref(&fence);
r = radeon_object_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object after copy %d\n", i);
goto out_cleanup;
}
for (gtt_start = gtt_map, gtt_end = gtt_map + size,
vram_start = vram_map, vram_end = vram_map + size;
gtt_start < gtt_end;
gtt_start++, vram_start++) {
if (*gtt_start != vram_start) {
DRM_ERROR("Incorrect VRAM->GTT copy %d: Got 0x%p, "
"expected 0x%p (VRAM map 0x%p-0x%p)\n",
i, *gtt_start, vram_start, vram_map,
vram_end);
radeon_object_kunmap(gtt_obj[i]);
goto out_cleanup;
}
}
radeon_object_kunmap(gtt_obj[i]);
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
gtt_addr - rdev->mc.gtt_location);
}
out_cleanup:
if (vram_obj) {
radeon_object_unpin(vram_obj);
radeon_object_unref(&vram_obj);
}
if (gtt_obj) {
for (i = 0; i < n; i++) {
if (gtt_obj[i]) {
radeon_object_unpin(gtt_obj[i]);
radeon_object_unref(>t_obj[i]);
}
}
kfree(gtt_obj);
}
if (fence) {
radeon_fence_unref(&fence);
}
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
