static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
uint64_t saddr, daddr;
int r, n;
int time;
n = RADEON_BENCHMARK_ITERATIONS;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, &sobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_reserve(sobj, false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(sobj, sdomain, &saddr);
radeon_bo_unreserve(sobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, &dobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_reserve(dobj, false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(dobj, ddomain, &daddr);
radeon_bo_unreserve(dobj);
if (r) {
goto out_cleanup;
}
/* r100 doesn't have dma engine so skip the test */
/* also, VRAM-to-VRAM test doesn't make much sense for DMA */
/* skip it as well if domains are the same */
if ((rdev->asic->copy_dma) && (sdomain != ddomain)) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_DMA, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "dma");
}
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_BLIT, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "blit");
out_cleanup:
if (sobj) {
r = radeon_bo_reserve(sobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(sobj);
radeon_bo_unreserve(sobj);
}
radeon_bo_unref(&sobj);
}
if (dobj) {
r = radeon_bo_reserve(dobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(dobj);
radeon_bo_unreserve(dobj);
}
radeon_bo_unref(&dobj);
}
if (r) {
DRM_ERROR("Error while benchmarking BO move.\n");
}
}
static void radeon_benchmark_move(struct radeon_device *rdev, unsigned size,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
uint64_t saddr, daddr;
int r, n;
int time;
n = RADEON_BENCHMARK_ITERATIONS;
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, sdomain, 0, NULL, &sobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_reserve(sobj, false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(sobj, sdomain, &saddr);
radeon_bo_unreserve(sobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_create(rdev, size, PAGE_SIZE, true, ddomain, 0, NULL, &dobj);
if (r) {
goto out_cleanup;
}
r = radeon_bo_reserve(dobj, false);
if (unlikely(r != 0))
goto out_cleanup;
r = radeon_bo_pin(dobj, ddomain, &daddr);
radeon_bo_unreserve(dobj);
if (r) {
goto out_cleanup;
}
if (rdev->asic->copy.dma) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_DMA, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "dma");
}
if (rdev->asic->copy.blit) {
time = radeon_benchmark_do_move(rdev, size, saddr, daddr,
RADEON_BENCHMARK_COPY_BLIT, n);
if (time < 0)
goto out_cleanup;
if (time > 0)
radeon_benchmark_log_results(n, size, time,
sdomain, ddomain, "blit");
}
out_cleanup:
if (sobj) {
r = radeon_bo_reserve(sobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(sobj);
radeon_bo_unreserve(sobj);
}
radeon_bo_unref(&sobj);
}
if (dobj) {
r = radeon_bo_reserve(dobj, false);
if (likely(r == 0)) {
radeon_bo_unpin(dobj);
radeon_bo_unreserve(dobj);
}
radeon_bo_unref(&dobj);
}
if (r) {
DRM_ERROR("Error while benchmarking BO move.\n");
}
}
/* 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,
NULL, &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, NULL, 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_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_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");
}
}
/* Test BO GTT->VRAM and VRAM->GTT GPU copies across the whole GTT aperture */
static void radeon_do_test_moves(struct radeon_device *rdev, int flag)
{
struct radeon_bo *vram_obj = NULL;
struct radeon_bo **gtt_obj = NULL;
struct radeon_fence *fence = NULL;
u64 gtt_addr, vram_addr;
unsigned n, size;
int i, r, ring;
switch (flag) {
case RADEON_TEST_COPY_DMA:
ring = radeon_copy_dma_ring_index(rdev);
break;
case RADEON_TEST_COPY_BLIT:
ring = radeon_copy_blit_ring_index(rdev);
break;
default:
DRM_ERROR("Unknown copy method\n");
return;
}
size = 1024 * 1024;
/* Number of tests =
* (Total GTT - IB pool - writeback page - ring buffers) / test size
*/
n = rdev->mc.gtt_size - rdev->gart_pin_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,
0, NULL, &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_unref;
r = radeon_bo_pin(vram_obj, RADEON_GEM_DOMAIN_VRAM, &vram_addr);
if (r) {
DRM_ERROR("Failed to pin VRAM object\n");
goto out_unres;
}
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, 0, NULL, gtt_obj + i);
if (r) {
DRM_ERROR("Failed to create GTT object %d\n", i);
goto out_lclean;
}
r = radeon_bo_reserve(gtt_obj[i], false);
if (unlikely(r != 0))
goto out_lclean_unref;
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_lclean_unres;
}
r = radeon_bo_kmap(gtt_obj[i], >t_map);
if (r) {
DRM_ERROR("Failed to map GTT object %d\n", i);
goto out_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size);
gtt_start < gtt_end;
gtt_start++)
*gtt_start = gtt_start;
radeon_bo_kunmap(gtt_obj[i]);
if (ring == R600_RING_TYPE_DMA_INDEX)
r = radeon_copy_dma(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
else
r = radeon_copy_blit(rdev, gtt_addr, vram_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
if (r) {
DRM_ERROR("Failed GTT->VRAM copy %d\n", i);
goto out_lclean_unpin;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for GTT->VRAM fence %d\n", i);
goto out_lclean_unpin;
}
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_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size),
vram_start = vram_map, vram_end = (void *)((uintptr_t)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)
((uintptr_t)gtt_addr - (uintptr_t)rdev->mc.gtt_start +
(uintptr_t)gtt_start - (uintptr_t)gtt_map),
(unsigned long long)
((uintptr_t)vram_addr - (uintptr_t)rdev->mc.vram_start +
(uintptr_t)gtt_start - (uintptr_t)gtt_map));
radeon_bo_kunmap(vram_obj);
goto out_lclean_unpin;
}
*vram_start = vram_start;
}
radeon_bo_kunmap(vram_obj);
if (ring == R600_RING_TYPE_DMA_INDEX)
r = radeon_copy_dma(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
else
r = radeon_copy_blit(rdev, vram_addr, gtt_addr, size / RADEON_GPU_PAGE_SIZE, &fence);
if (r) {
DRM_ERROR("Failed VRAM->GTT copy %d\n", i);
goto out_lclean_unpin;
}
r = radeon_fence_wait(fence, false);
if (r) {
DRM_ERROR("Failed to wait for VRAM->GTT fence %d\n", i);
goto out_lclean_unpin;
}
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_lclean_unpin;
}
for (gtt_start = gtt_map, gtt_end = (void *)((uintptr_t)gtt_map + size),
vram_start = vram_map, vram_end = (void *)((uintptr_t)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)
((uintptr_t)vram_addr - (uintptr_t)rdev->mc.vram_start +
(uintptr_t)vram_start - (uintptr_t)vram_map),
(unsigned long long)
((uintptr_t)gtt_addr - (uintptr_t)rdev->mc.gtt_start +
(uintptr_t)vram_start - (uintptr_t)vram_map));
radeon_bo_kunmap(gtt_obj[i]);
goto out_lclean_unpin;
}
}
radeon_bo_kunmap(gtt_obj[i]);
DRM_INFO("Tested GTT->VRAM and VRAM->GTT copy for GTT offset 0x%llx\n",
(uintmax_t)gtt_addr - rdev->mc.gtt_start);
continue;
out_lclean_unpin:
radeon_bo_unpin(gtt_obj[i]);
out_lclean_unres:
radeon_bo_unreserve(gtt_obj[i]);
out_lclean_unref:
radeon_bo_unref(>t_obj[i]);
out_lclean:
for (--i; i >= 0; --i) {
radeon_bo_unpin(gtt_obj[i]);
radeon_bo_unreserve(gtt_obj[i]);
radeon_bo_unref(>t_obj[i]);
}
if (fence && !IS_ERR(fence))
radeon_fence_unref(&fence);
break;
}
radeon_bo_unpin(vram_obj);
out_unres:
radeon_bo_unreserve(vram_obj);
out_unref:
radeon_bo_unref(&vram_obj);
out_cleanup:
if (gtt_obj)
kfree(gtt_obj);
if (fence) {
radeon_fence_unref(&fence);
}
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
int radeon_ttm_init(struct radeon_device *rdev)
{
int r, r2;
r = radeon_ttm_global_init(rdev);
if (r) {
return r;
}
/* No others user of address space so set it to 0 */
r = ttm_bo_device_init(&rdev->mman.bdev,
rdev->mman.bo_global_ref.ref.object,
&radeon_bo_driver, DRM_FILE_PAGE_OFFSET,
rdev->need_dma32);
if (r) {
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
rdev->mman.initialized = true;
rdev->ddev->drm_ttm_bdev = &rdev->mman.bdev;
r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_VRAM,
rdev->mc.real_vram_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing VRAM heap.\n");
return r;
}
r = radeon_bo_create(rdev, 256 * 1024, PAGE_SIZE, true,
RADEON_GEM_DOMAIN_VRAM,
NULL, &rdev->stollen_vga_memory);
if (r) {
return r;
}
r = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (r) {
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
r = radeon_bo_pin(rdev->stollen_vga_memory, RADEON_GEM_DOMAIN_VRAM, NULL);
radeon_bo_unreserve(rdev->stollen_vga_memory);
if (r) {
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
DRM_INFO("radeon: %uM of VRAM memory ready\n",
(unsigned)rdev->mc.real_vram_size / (1024 * 1024));
r = ttm_bo_init_mm(&rdev->mman.bdev, TTM_PL_TT,
rdev->mc.gtt_size >> PAGE_SHIFT);
if (r) {
DRM_ERROR("Failed initializing GTT heap.\n");
r2 = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (likely(r2 == 0)) {
radeon_bo_unpin(rdev->stollen_vga_memory);
radeon_bo_unreserve(rdev->stollen_vga_memory);
}
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
DRM_INFO("radeon: %uM of GTT memory ready.\n",
(unsigned)(rdev->mc.gtt_size / (1024 * 1024)));
r = radeon_ttm_debugfs_init(rdev);
if (r) {
DRM_ERROR("Failed to init debugfs\n");
r2 = radeon_bo_reserve(rdev->stollen_vga_memory, false);
if (likely(r2 == 0)) {
radeon_bo_unpin(rdev->stollen_vga_memory);
radeon_bo_unreserve(rdev->stollen_vga_memory);
}
radeon_bo_unref(&rdev->stollen_vga_memory);
return r;
}
return 0;
}
