static int radeon_benchmark_do_move(struct radeon_device *rdev, unsigned size,
uint64_t saddr, uint64_t daddr,
int flag, int n)
{
unsigned long start_jiffies;
unsigned long end_jiffies;
struct radeon_fence *fence = NULL;
int i, r;
start_jiffies = jiffies;
for (i = 0; i < n; i++) {
switch (flag) {
case RADEON_BENCHMARK_COPY_DMA:
fence = radeon_copy_dma(rdev, saddr, daddr,
size / RADEON_GPU_PAGE_SIZE,
NULL);
break;
case RADEON_BENCHMARK_COPY_BLIT:
fence = radeon_copy_blit(rdev, saddr, daddr,
size / RADEON_GPU_PAGE_SIZE,
NULL);
break;
default:
DRM_ERROR("Unknown copy method\n");
return -EINVAL;
}
if (IS_ERR(fence))
return PTR_ERR(fence);
r = radeon_fence_wait(fence, false);
radeon_fence_unref(&fence);
if (r)
return r;
}
end_jiffies = jiffies;
return jiffies_to_msecs(end_jiffies - start_jiffies);
}
static int radeon_sync_obj_wait(void *sync_obj, void *sync_arg,
bool lazy, bool interruptible)
{
return radeon_fence_wait((struct radeon_fence *)sync_obj, interruptible);
}
void radeon_test_ring_sync2(struct radeon_device *rdev,
struct radeon_ring *ringA,
struct radeon_ring *ringB,
struct radeon_ring *ringC)
{
struct radeon_fence *fenceA = NULL, *fenceB = NULL;
struct radeon_semaphore *semaphore = NULL;
int ridxA = radeon_ring_index(rdev, ringA);
int ridxB = radeon_ring_index(rdev, ringB);
int ridxC = radeon_ring_index(rdev, ringC);
bool sigA, sigB;
int i, r;
r = radeon_fence_create(rdev, &fenceA, ridxA);
if (r) {
DRM_ERROR("Failed to create sync fence 1\n");
goto out_cleanup;
}
r = radeon_fence_create(rdev, &fenceB, ridxB);
if (r) {
DRM_ERROR("Failed to create sync fence 2\n");
goto out_cleanup;
}
r = radeon_semaphore_create(rdev, &semaphore);
if (r) {
DRM_ERROR("Failed to create semaphore\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringA, 64);
if (r) {
DRM_ERROR("Failed to lock ring A %d\n", ridxA);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
radeon_fence_emit(rdev, fenceA);
radeon_ring_unlock_commit(rdev, ringA);
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %d\n", ridxB);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ridxB, semaphore);
radeon_fence_emit(rdev, fenceB);
radeon_ring_unlock_commit(rdev, ringB);
mdelay(1000);
if (radeon_fence_signaled(fenceA)) {
DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
goto out_cleanup;
}
if (radeon_fence_signaled(fenceB)) {
DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringC, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringC);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ridxC, semaphore);
radeon_ring_unlock_commit(rdev, ringC);
for (i = 0; i < 30; ++i) {
mdelay(100);
sigA = radeon_fence_signaled(fenceA);
sigB = radeon_fence_signaled(fenceB);
if (sigA || sigB)
break;
}
if (!sigA && !sigB) {
DRM_ERROR("Neither fence A nor B has been signaled\n");
goto out_cleanup;
} else if (sigA && sigB) {
DRM_ERROR("Both fence A and B has been signaled\n");
goto out_cleanup;
}
DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
r = radeon_ring_lock(rdev, ringC, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringC);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ridxC, semaphore);
radeon_ring_unlock_commit(rdev, ringC);
mdelay(1000);
r = radeon_fence_wait(fenceA, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence A\n");
goto out_cleanup;
}
r = radeon_fence_wait(fenceB, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence B\n");
goto out_cleanup;
}
out_cleanup:
if (semaphore)
radeon_semaphore_free(rdev, semaphore);
if (fenceA)
radeon_fence_unref(&fenceA);
if (fenceB)
radeon_fence_unref(&fenceB);
if (r)
printk(KERN_WARNING "Error while testing ring sync (%d).\n", 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_ring_sync(struct radeon_device *rdev,
struct radeon_ring *ringA,
struct radeon_ring *ringB)
{
struct radeon_fence *fence1 = NULL, *fence2 = NULL;
struct radeon_semaphore *semaphore = NULL;
int ridxA = radeon_ring_index(rdev, ringA);
int ridxB = radeon_ring_index(rdev, ringB);
int r;
r = radeon_fence_create(rdev, &fence1, ridxA);
if (r) {
DRM_ERROR("Failed to create sync fence 1\n");
goto out_cleanup;
}
r = radeon_fence_create(rdev, &fence2, ridxA);
if (r) {
DRM_ERROR("Failed to create sync fence 2\n");
goto out_cleanup;
}
r = radeon_semaphore_create(rdev, &semaphore);
if (r) {
DRM_ERROR("Failed to create semaphore\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringA, 64);
if (r) {
DRM_ERROR("Failed to lock ring A %d\n", ridxA);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
radeon_fence_emit(rdev, fence1);
radeon_semaphore_emit_wait(rdev, ridxA, semaphore);
radeon_fence_emit(rdev, fence2);
radeon_ring_unlock_commit(rdev, ringA);
mdelay(1000);
if (radeon_fence_signaled(fence1)) {
DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringB);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ridxB, semaphore);
radeon_ring_unlock_commit(rdev, ringB);
r = radeon_fence_wait(fence1, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
}
mdelay(1000);
if (radeon_fence_signaled(fence2)) {
DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringB);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ridxB, semaphore);
radeon_ring_unlock_commit(rdev, ringB);
r = radeon_fence_wait(fence2, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
}
out_cleanup:
if (semaphore)
radeon_semaphore_free(rdev, semaphore);
if (fence1)
radeon_fence_unref(&fence1);
if (fence2)
radeon_fence_unref(&fence2);
if (r)
printk(KERN_WARNING "Error while testing ring sync (%d).\n", r);
}
/* 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;
uint64_t 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 - 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_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;
struct radeon_fence *fence = NULL;
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_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 = 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 = 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_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 = 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_lclean_unpin;
}
}
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);
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)
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:
kfree(gtt_obj);
if (r) {
printk(KERN_WARNING "Error while testing BO move.\n");
}
}
/**
* radeon_semaphore_sync_rings - sync ring to all registered fences
*
* @rdev: radeon_device pointer
* @semaphore: semaphore object to use for sync
* @ring: ring that needs sync
*
* Ensure that all registered fences are signaled before letting
* the ring continue. The caller must hold the ring lock.
*/
int radeon_semaphore_sync_rings(struct radeon_device *rdev,
struct radeon_semaphore *semaphore,
int ring)
{
unsigned count = 0;
int i, r;
for (i = 0; i < RADEON_NUM_RINGS; ++i) {
struct radeon_fence *fence = semaphore->sync_to[i];
/* check if we really need to sync */
if (!radeon_fence_need_sync(fence, ring))
continue;
/* prevent GPU deadlocks */
if (!rdev->ring[i].ready) {
dev_err(rdev->dev, "Syncing to a disabled ring!");
return -EINVAL;
}
if (++count > RADEON_NUM_SYNCS) {
/* not enough room, wait manually */
r = radeon_fence_wait(fence, false);
if (r)
return r;
continue;
}
/* allocate enough space for sync command */
r = radeon_ring_alloc(rdev, &rdev->ring[i], 16);
if (r) {
return r;
}
/* emit the signal semaphore */
if (!radeon_semaphore_emit_signal(rdev, i, semaphore)) {
/* signaling wasn't successful wait manually */
radeon_ring_undo(&rdev->ring[i]);
r = radeon_fence_wait(fence, false);
if (r)
return r;
continue;
}
/* we assume caller has already allocated space on waiters ring */
if (!radeon_semaphore_emit_wait(rdev, ring, semaphore)) {
/* waiting wasn't successful wait manually */
radeon_ring_undo(&rdev->ring[i]);
r = radeon_fence_wait(fence, false);
if (r)
return r;
continue;
}
radeon_ring_commit(rdev, &rdev->ring[i], false);
radeon_fence_note_sync(fence, ring);
semaphore->gpu_addr += 8;
}
return 0;
}
void radeon_benchmark_move(struct radeon_device *rdev, unsigned bsize,
unsigned sdomain, unsigned ddomain)
{
struct radeon_bo *dobj = NULL;
struct radeon_bo *sobj = NULL;
struct radeon_fence *fence = NULL;
uint64_t saddr, daddr;
unsigned long start_jiffies;
unsigned long end_jiffies;
unsigned long time;
unsigned i, n, size;
int r;
size = bsize;
n = 1024;
r = radeon_bo_create(rdev, NULL, 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, NULL, 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 */
if (rdev->asic->copy_dma) {
start_jiffies = jiffies;
for (i = 0; i < n; i++) {
r = radeon_fence_create(rdev, &fence);
if (r) {
goto out_cleanup;
}
r = radeon_copy_dma(rdev, saddr, daddr,
size / RADEON_GPU_PAGE_SIZE, fence);
if (r) {
goto out_cleanup;
}
r = radeon_fence_wait(fence, false);
if (r) {
goto out_cleanup;
}
radeon_fence_unref(&fence);
}
end_jiffies = jiffies;
time = end_jiffies - start_jiffies;
time = jiffies_to_msecs(time);
if (time > 0) {
i = ((n * size) >> 10) / time;
printk(KERN_INFO "radeon: dma %u bo moves of %ukb from"
" %d to %d in %lums (%ukb/ms %ukb/s %uM/s)\n",
n, size >> 10,
sdomain, ddomain, time,
i, i * 1000, (i * 1000) / 1024);
}
void radeon_test_ring_sync(struct radeon_device *rdev,
struct radeon_ring *ringA,
struct radeon_ring *ringB)
{
struct radeon_fence *fence1 = NULL, *fence2 = NULL;
struct radeon_semaphore *semaphore = NULL;
int r;
r = radeon_semaphore_create(rdev, &semaphore);
if (r) {
DRM_ERROR("Failed to create semaphore\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringA, 64);
if (r) {
DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
r = radeon_fence_emit(rdev, &fence1, ringA->idx);
if (r) {
DRM_ERROR("Failed to emit fence 1\n");
radeon_ring_unlock_undo(rdev, ringA);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
r = radeon_fence_emit(rdev, &fence2, ringA->idx);
if (r) {
DRM_ERROR("Failed to emit fence 2\n");
radeon_ring_unlock_undo(rdev, ringA);
goto out_cleanup;
}
radeon_ring_unlock_commit(rdev, ringA);
DRM_MDELAY(1000);
if (radeon_fence_signaled(fence1)) {
DRM_ERROR("Fence 1 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringB);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringB);
r = radeon_fence_wait(fence1, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
}
DRM_MDELAY(1000);
if (radeon_fence_signaled(fence2)) {
DRM_ERROR("Fence 2 signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringB);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ringB->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringB);
r = radeon_fence_wait(fence2, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence 1\n");
goto out_cleanup;
}
out_cleanup:
radeon_semaphore_free(rdev, &semaphore, NULL);
if (fence1)
radeon_fence_unref(&fence1);
if (fence2)
radeon_fence_unref(&fence2);
if (r)
DRM_ERROR("Error while testing ring sync (%d).\n", r);
}
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");
}
}
static void radeon_test_ring_sync2(struct radeon_device *rdev,
struct radeon_ring *ringA,
struct radeon_ring *ringB,
struct radeon_ring *ringC)
{
struct radeon_fence *fenceA = NULL, *fenceB = NULL;
struct radeon_semaphore *semaphore = NULL;
bool sigA, sigB;
int i, r;
r = radeon_semaphore_create(rdev, &semaphore);
if (r) {
DRM_ERROR("Failed to create semaphore\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringA, 64);
if (r) {
DRM_ERROR("Failed to lock ring A %d\n", ringA->idx);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ringA->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringA, false);
r = radeon_test_create_and_emit_fence(rdev, ringA, &fenceA);
if (r)
goto out_cleanup;
r = radeon_ring_lock(rdev, ringB, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %d\n", ringB->idx);
goto out_cleanup;
}
radeon_semaphore_emit_wait(rdev, ringB->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringB, false);
r = radeon_test_create_and_emit_fence(rdev, ringB, &fenceB);
if (r)
goto out_cleanup;
msleep(1000);
if (radeon_fence_signaled(fenceA)) {
DRM_ERROR("Fence A signaled without waiting for semaphore.\n");
goto out_cleanup;
}
if (radeon_fence_signaled(fenceB)) {
DRM_ERROR("Fence B signaled without waiting for semaphore.\n");
goto out_cleanup;
}
r = radeon_ring_lock(rdev, ringC, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringC);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringC, false);
for (i = 0; i < 30; ++i) {
msleep(100);
sigA = radeon_fence_signaled(fenceA);
sigB = radeon_fence_signaled(fenceB);
if (sigA || sigB)
break;
}
if (!sigA && !sigB) {
DRM_ERROR("Neither fence A nor B has been signaled\n");
goto out_cleanup;
} else if (sigA && sigB) {
DRM_ERROR("Both fence A and B has been signaled\n");
goto out_cleanup;
}
DRM_INFO("Fence %c was first signaled\n", sigA ? 'A' : 'B');
r = radeon_ring_lock(rdev, ringC, 64);
if (r) {
DRM_ERROR("Failed to lock ring B %p\n", ringC);
goto out_cleanup;
}
radeon_semaphore_emit_signal(rdev, ringC->idx, semaphore);
radeon_ring_unlock_commit(rdev, ringC, false);
msleep(1000);
r = radeon_fence_wait(fenceA, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence A\n");
goto out_cleanup;
}
r = radeon_fence_wait(fenceB, false);
if (r) {
DRM_ERROR("Failed to wait for sync fence B\n");
goto out_cleanup;
}
out_cleanup:
radeon_semaphore_free(rdev, &semaphore, NULL);
if (fenceA)
radeon_fence_unref(&fenceA);
if (fenceB)
radeon_fence_unref(&fenceB);
if (r)
pr_warn("Error while testing ring sync (%d)\n", r);
}
