On 11/13/25 17:05, Pierre-Eric Pelloux-Prayer wrote:

Until now ttm stored a single pipelined eviction fence which means drivers had to use a single entity for these evictions.

To lift this requirement, this commit allows up to 8 entities to be used.

Ideally a dma_resv object would have been used as a container of the eviction fences, but the locking rules makes it complex. dma_resv all have the same ww_class, which means "Attempting to lock more mutexes after ww_acquire_done." is an error.

One alternative considered was to introduced a 2nd ww_class for specific resv to hold a single "transient" lock (= the resv lock would only be held for a short period, without taking any other locks).

The other option, is to statically reserve a fence array, and extend the existing code to deal with N fences, instead of 1.

The driver is still responsible to reserve the correct number of fence slots.

---

v2:

• simplified code
• dropped n_fences
• name changes

---

Signed-off-by: Pierre-Eric Pelloux-Prayer pierre-eric.pelloux-prayer@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 ++-- .../gpu/drm/ttm/tests/ttm_bo_validate_test.c | 11 +++-- drivers/gpu/drm/ttm/tests/ttm_resource_test.c | 5 +- drivers/gpu/drm/ttm/ttm_bo.c | 47 ++++++++++--------- drivers/gpu/drm/ttm/ttm_bo_util.c | 38 ++++++++++++--- drivers/gpu/drm/ttm/ttm_resource.c | 31 +++++++----- include/drm/ttm/ttm_resource.h | 29 ++++++++---- 7 files changed, 109 insertions(+), 60 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 326476089db3..3b46a24a8c48 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -2156,7 +2156,7 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) { struct ttm_resource_manager *man = ttm_manager_type(&adev->mman.bdev, TTM_PL_VRAM); uint64_t size;

• int r;

• int r, i;

if (!adev->mman.initialized || amdgpu_in_reset(adev) || adev->mman.buffer_funcs_enabled == enable || adev->gmc.is_app_apu) @@ -2190,8 +2190,10 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) } else { drm_sched_entity_destroy(&adev->mman.high_pr); drm_sched_entity_destroy(&adev->mman.low_pr);

• dma_fence_put(man->move);
  

• man->move = NULL;
  

• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++) {
  

• 	dma_fence_put(man->eviction_fences[i]);
  

• 	man->eviction_fences[i] = NULL;
  

• }
  

That code should have been a TTM function in the first place.

I suggest to just call ttm_resource_manager_cleanup() here instead and add this as comment:

/* Drop all the old fences since re-creating the scheduler entities will allocate next contexts */

Apart from that looks good to me.

Regards, Christian.

} /* this just adjusts TTM size idea, which sets lpfn to the correct value */ diff --git a/drivers/gpu/drm/ttm/tests/ttm_bo_validate_test.c b/drivers/gpu/drm/ttm/tests/ttm_bo_validate_test.c index 3148f5d3dbd6..8f71906c4238 100644 --- a/drivers/gpu/drm/ttm/tests/ttm_bo_validate_test.c +++ b/drivers/gpu/drm/ttm/tests/ttm_bo_validate_test.c @@ -651,7 +651,7 @@ static void ttm_bo_validate_move_fence_signaled(struct kunit *test) int err; man = ttm_manager_type(priv->ttm_dev, mem_type);

• man->move = dma_fence_get_stub();

• man->eviction_fences[0] = dma_fence_get_stub();

bo = ttm_bo_kunit_init(test, test->priv, size, NULL); bo->type = bo_type; @@ -668,7 +668,7 @@ static void ttm_bo_validate_move_fence_signaled(struct kunit *test) KUNIT_EXPECT_EQ(test, ctx.bytes_moved, size); ttm_bo_put(bo);

• dma_fence_put(man->move);

• dma_fence_put(man->eviction_fences[0]);

} static const struct ttm_bo_validate_test_case ttm_bo_validate_wait_cases[] = { @@ -732,9 +732,9 @@ static void ttm_bo_validate_move_fence_not_signaled(struct kunit *test) spin_lock_init(&fence_lock); man = ttm_manager_type(priv->ttm_dev, fst_mem);

• man->move = alloc_mock_fence(test);

• man->eviction_fences[0] = alloc_mock_fence(test);

• task = kthread_create(threaded_fence_signal, man->move, "move-fence-signal");

• task = kthread_create(threaded_fence_signal, man->eviction_fences[0], "move-fence-signal"); if (IS_ERR(task)) KUNIT_FAIL(test, "Couldn't create move fence signal task\n");

@@ -742,7 +742,8 @@ static void ttm_bo_validate_move_fence_not_signaled(struct kunit *test) err = ttm_bo_validate(bo, placement_val, &ctx_val); dma_resv_unlock(bo->base.resv);

• dma_fence_wait_timeout(man->move, false, MAX_SCHEDULE_TIMEOUT);

• dma_fence_wait_timeout(man->eviction_fences[0], false, MAX_SCHEDULE_TIMEOUT);
• man->eviction_fences[0] = NULL;

KUNIT_EXPECT_EQ(test, err, 0); KUNIT_EXPECT_EQ(test, ctx_val.bytes_moved, size); diff --git a/drivers/gpu/drm/ttm/tests/ttm_resource_test.c b/drivers/gpu/drm/ttm/tests/ttm_resource_test.c index e6ea2bd01f07..c0e4e35e0442 100644 --- a/drivers/gpu/drm/ttm/tests/ttm_resource_test.c +++ b/drivers/gpu/drm/ttm/tests/ttm_resource_test.c @@ -207,6 +207,7 @@ static void ttm_resource_manager_init_basic(struct kunit *test) struct ttm_resource_test_priv *priv = test->priv; struct ttm_resource_manager *man; size_t size = SZ_16K;

• int i;

man = kunit_kzalloc(test, sizeof(*man), GFP_KERNEL); KUNIT_ASSERT_NOT_NULL(test, man); @@ -216,8 +217,8 @@ static void ttm_resource_manager_init_basic(struct kunit *test) KUNIT_ASSERT_PTR_EQ(test, man->bdev, priv->devs->ttm_dev); KUNIT_ASSERT_EQ(test, man->size, size); KUNIT_ASSERT_EQ(test, man->usage, 0);

• KUNIT_ASSERT_NULL(test, man->move);
• KUNIT_ASSERT_NOT_NULL(test, &man->move_lock);

• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++)

• KUNIT_ASSERT_NULL(test, man->eviction_fences[i]);
  

for (int i = 0; i < TTM_MAX_BO_PRIORITY; ++i) KUNIT_ASSERT_TRUE(test, list_empty(&man->lru[i])); diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c index f4d9e68b21e7..0b3732ed6f6c 100644 --- a/drivers/gpu/drm/ttm/ttm_bo.c +++ b/drivers/gpu/drm/ttm/ttm_bo.c @@ -658,34 +658,35 @@ void ttm_bo_unpin(struct ttm_buffer_object *bo) EXPORT_SYMBOL(ttm_bo_unpin); /*

• • Add the last move fence to the BO as kernel dependency and reserve a new
• • fence slot.

• • Add the pipelined eviction fencesto the BO as kernel dependency and reserve new
• • fence slots.
  */

-static int ttm_bo_add_move_fence(struct ttm_buffer_object *bo,

• 		 struct ttm_resource_manager *man,
  

• 		 bool no_wait_gpu)
  

+static int ttm_bo_add_pipelined_eviction_fences(struct ttm_buffer_object *bo,

• 				struct ttm_resource_manager *man,
  

• 				bool no_wait_gpu)
  

{ struct dma_fence *fence;

• int ret;

• int i;

• spin_lock(&man->move_lock);
• fence = dma_fence_get(man->move);
• spin_unlock(&man->move_lock);

• spin_lock(&man->eviction_lock);
• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++) {

• fence = man->eviction_fences[i];
  

• if (!fence)
  

• 	continue;
  

• if (!fence)

• return 0;
  

• if (no_wait_gpu) {

• ret = dma_fence_is_signaled(fence) ? 0 : -EBUSY;
  

• dma_fence_put(fence);
  

• return ret;
  

• if (no_wait_gpu) {
  

• 	if (!dma_fence_is_signaled(fence)) {
  

• 		spin_unlock(&man->eviction_lock);
  

• 		return -EBUSY;
  

• 	}
  

• } else {
  

• 	dma_resv_add_fence(bo->base.resv, fence, DMA_RESV_USAGE_KERNEL);
  

• }
  

  }
• spin_unlock(&man->eviction_lock);

• dma_resv_add_fence(bo->base.resv, fence, DMA_RESV_USAGE_KERNEL);
• ret = dma_resv_reserve_fences(bo->base.resv, 1);
• dma_fence_put(fence);
• return ret;

• /* TODO: this call should be removed. */
• return dma_resv_reserve_fences(bo->base.resv, 1);

} /** @@ -718,7 +719,7 @@ static int ttm_bo_alloc_resource(struct ttm_buffer_object *bo, int i, ret; ticket = dma_resv_locking_ctx(bo->base.resv);

• ret = dma_resv_reserve_fences(bo->base.resv, 1);

• ret = dma_resv_reserve_fences(bo->base.resv, TTM_NUM_MOVE_FENCES); if (unlikely(ret)) return ret;

@@ -757,7 +758,7 @@ static int ttm_bo_alloc_resource(struct ttm_buffer_object *bo, return ret; }

• ret = ttm_bo_add_move_fence(bo, man, ctx->no_wait_gpu);
  

• ret = ttm_bo_add_pipelined_eviction_fences(bo, man, ctx->no_wait_gpu);
  

  if (unlikely(ret)) { ttm_resource_free(bo, res); if (ret == -EBUSY)

diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c index acbbca9d5c92..2ff35d55e462 100644 --- a/drivers/gpu/drm/ttm/ttm_bo_util.c +++ b/drivers/gpu/drm/ttm/ttm_bo_util.c @@ -258,7 +258,7 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo, ret = dma_resv_trylock(&fbo->base.base._resv); WARN_ON(!ret);

• ret = dma_resv_reserve_fences(&fbo->base.base._resv, 1);

• ret = dma_resv_reserve_fences(&fbo->base.base._resv, TTM_NUM_MOVE_FENCES); if (ret) { dma_resv_unlock(&fbo->base.base._resv); kfree(fbo);

@@ -646,20 +646,44 @@ static void ttm_bo_move_pipeline_evict(struct ttm_buffer_object *bo, { struct ttm_device *bdev = bo->bdev; struct ttm_resource_manager *from;

• struct dma_fence *tmp;
• int i;

from = ttm_manager_type(bdev, bo->resource->mem_type); /** * BO doesn't have a TTM we need to bind/unbind. Just remember

• * this eviction and free up the allocation
  

• * this eviction and free up the allocation.
  

• * The fence will be saved in the first free slot or in the slot
  

• * already used to store a fence from the same context. Since
  

• * drivers can't use more than TTM_NUM_MOVE_FENCES contexts for
  

• * evictions we should always find a slot to use.
  

  */

• spin_lock(&from->move_lock);
• if (!from->move || dma_fence_is_later(fence, from->move)) {

• dma_fence_put(from->move);
  

• from->move = dma_fence_get(fence);
  

• spin_lock(&from->eviction_lock);
• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++) {

• tmp = from->eviction_fences[i];
  

• if (!tmp)
  

• 	break;
  

• if (fence->context != tmp->context)
  

• 	continue;
  

• if (dma_fence_is_later(fence, tmp)) {
  

• 	dma_fence_put(tmp);
  

• 	break;
  

• }
  

• goto unlock;
  

• }
• if (i < TTM_NUM_MOVE_FENCES) {

• from->eviction_fences[i] = dma_fence_get(fence);
  

• } else {

• WARN(1, "not enough fence slots for all fence contexts");
  

• spin_unlock(&from->eviction_lock);
  

• dma_fence_wait(fence, false);
  

• goto end;
  

  }

• spin_unlock(&from->move_lock);

+unlock:

• spin_unlock(&from->eviction_lock);

+end: ttm_resource_free(bo, &bo->resource); } diff --git a/drivers/gpu/drm/ttm/ttm_resource.c b/drivers/gpu/drm/ttm/ttm_resource.c index e2c82ad07eb4..62c34cafa387 100644 --- a/drivers/gpu/drm/ttm/ttm_resource.c +++ b/drivers/gpu/drm/ttm/ttm_resource.c @@ -523,14 +523,15 @@ void ttm_resource_manager_init(struct ttm_resource_manager *man, { unsigned i;

• spin_lock_init(&man->move_lock); man->bdev = bdev; man->size = size; man->usage = 0;

for (i = 0; i < TTM_MAX_BO_PRIORITY; ++i) INIT_LIST_HEAD(&man->lru[i]);

• man->move = NULL;

• spin_lock_init(&man->eviction_lock);
• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++)

• man->eviction_fences[i] = NULL;
  

} EXPORT_SYMBOL(ttm_resource_manager_init); @@ -551,7 +552,7 @@ int ttm_resource_manager_evict_all(struct ttm_device *bdev, .no_wait_gpu = false, }; struct dma_fence *fence;

• int ret;

• int ret, i;

do { ret = ttm_bo_evict_first(bdev, man, &ctx); @@ -561,18 +562,24 @@ int ttm_resource_manager_evict_all(struct ttm_device *bdev, if (ret && ret != -ENOENT) return ret;

• spin_lock(&man->move_lock);
• fence = dma_fence_get(man->move);
• spin_unlock(&man->move_lock);

• ret = 0;

• if (fence) {

• ret = dma_fence_wait(fence, false);
  

• dma_fence_put(fence);
  

• if (ret)
  

• 	return ret;
  

• spin_lock(&man->eviction_lock);
• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++) {

• fence = man->eviction_fences[i];
  

• if (fence && !dma_fence_is_signaled(fence)) {
  

• 	dma_fence_get(fence);
  

• 	spin_unlock(&man->eviction_lock);
  

• 	ret = dma_fence_wait(fence, false);
  

• 	dma_fence_put(fence);
  

• 	if (ret)
  

• 		return ret;
  

• 	spin_lock(&man->eviction_lock);
  

• }
  

  }
• spin_unlock(&man->eviction_lock);

• return 0;

• return ret;

} EXPORT_SYMBOL(ttm_resource_manager_evict_all); diff --git a/include/drm/ttm/ttm_resource.h b/include/drm/ttm/ttm_resource.h index f49daa504c36..50e6added509 100644 --- a/include/drm/ttm/ttm_resource.h +++ b/include/drm/ttm/ttm_resource.h @@ -50,6 +50,15 @@ struct io_mapping; struct sg_table; struct scatterlist; +/**

• • define TTM_NUM_MOVE_FENCES - How many entities can be used for evictions
• • Pipelined evictions can be spread on multiple entities. This
• • is the max number of entities that can be used by the driver
• • for that purpose.
• */

+#define TTM_NUM_MOVE_FENCES 8

/**

• enum ttm_lru_item_type - enumerate ttm_lru_item subclasses

*/ @@ -180,8 +189,8 @@ struct ttm_resource_manager_func {

• @size: Size of the managed region.
• @bdev: ttm device this manager belongs to
• @func: structure pointer implementing the range manager. See above

• • @move_lock: lock for move fence
• • @move: The fence of the last pipelined move operation.

• • @eviction_lock: lock for eviction fences
• • @eviction_fences: The fences of the last pipelined move operation.
  • @lru: The lru list for this memory type.
  • This structure is used to identify and manage memory types for a device.

@@ -195,12 +204,12 @@ struct ttm_resource_manager { struct ttm_device *bdev; uint64_t size; const struct ttm_resource_manager_func *func;

• spinlock_t move_lock;

• /*

• * Protected by @move_lock.
  

• /* This is very similar to a dma_resv object, but locking rules make

• * it difficult to use one in this context.
  

  */

• struct dma_fence *move;

• spinlock_t eviction_lock;
• struct dma_fence *eviction_fences[TTM_NUM_MOVE_FENCES];

/* * Protected by the bdev->lru_lock. @@ -421,8 +430,12 @@ static inline bool ttm_resource_manager_used(struct ttm_resource_manager *man) static inline void ttm_resource_manager_cleanup(struct ttm_resource_manager *man) {

• dma_fence_put(man->move);
• man->move = NULL;

• int i;
• for (i = 0; i < TTM_NUM_MOVE_FENCES; i++) {

• dma_fence_put(man->eviction_fences[i]);
  

• man->eviction_fences[i] = NULL;
  

• }

} void ttm_lru_bulk_move_init(struct ttm_lru_bulk_move *bulk);

On 11/13/25 17:05, Pierre-Eric Pelloux-Prayer wrote:

It was always false.

Signed-off-by: Pierre-Eric Pelloux-Prayer pierre-eric.pelloux-prayer@amd.com Reviewed-by: Christian König christian.koenig@amd.com

Please push to amd-staging-drm-next.

---

drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c | 2 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 +++++++------------ drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 2 +- drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 2 +- 4 files changed, 10 insertions(+), 16 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c index 199693369c7c..02c2479a8840 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c @@ -39,7 +39,7 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size, for (i = 0; i < n; i++) { struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,

• 		       false, false, 0);
  

• 		       false, 0);
  

  if (r) goto exit_do_move; r = dma_fence_wait(fence, false);

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 3b46a24a8c48..c985f57fa227 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -354,7 +354,7 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, } r = amdgpu_copy_buffer(ring, from, to, cur_size, resv,

• 		       &next, false, true, copy_flags);
  

• 		       &next, true, copy_flags);
  

  if (r) goto error;

@@ -2211,16 +2211,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) } static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,

• 		  bool direct_submit,
  	  unsigned int num_dw,
  	  struct dma_resv *resv,
  	  bool vm_needs_flush,
  	  struct amdgpu_job **job,
  	  bool delayed, u64 k_job_id)
  

{

• enum amdgpu_ib_pool_type pool = direct_submit ?

• AMDGPU_IB_POOL_DIRECT :
  

• AMDGPU_IB_POOL_DELAYED;
  

• enum amdgpu_ib_pool_type pool = AMDGPU_IB_POOL_DELAYED; int r; struct drm_sched_entity *entity = delayed ? &adev->mman.low_pr : &adev->mman.high_pr;

@@ -2246,7 +2243,7 @@ static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev, int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, uint64_t dst_offset, uint32_t byte_count, struct dma_resv *resv,

•        struct dma_fence **fence, bool direct_submit,
  

•        struct dma_fence **fence,
    bool vm_needs_flush, uint32_t copy_flags)
  

{ struct amdgpu_device *adev = ring->adev; @@ -2256,7 +2253,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, unsigned int i; int r;

• if (!direct_submit && !ring->sched.ready) {

• if (!ring->sched.ready) { dev_err(adev->dev, "Trying to move memory with ring turned off.\n"); return -EINVAL;

@@ -2265,7 +2262,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, max_bytes = adev->mman.buffer_funcs->copy_max_bytes; num_loops = DIV_ROUND_UP(byte_count, max_bytes); num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8);

• r = amdgpu_ttm_prepare_job(adev, direct_submit, num_dw,

• r = amdgpu_ttm_prepare_job(adev, num_dw, resv, vm_needs_flush, &job, false, AMDGPU_KERNEL_JOB_ID_TTM_COPY_BUFFER); if (r)

@@ -2283,10 +2280,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, amdgpu_ring_pad_ib(ring, &job->ibs[0]); WARN_ON(job->ibs[0].length_dw > num_dw);

• if (direct_submit)

• r = amdgpu_job_submit_direct(job, ring, fence);
  

• else

• *fence = amdgpu_job_submit(job);
  

• *fence = amdgpu_job_submit(job); if (r) goto error_free;

@@ -2315,7 +2309,7 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data, max_bytes = adev->mman.buffer_funcs->fill_max_bytes; num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes); num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8);

• r = amdgpu_ttm_prepare_job(adev, false, num_dw, resv, vm_needs_flush,

• r = amdgpu_ttm_prepare_job(adev, num_dw, resv, vm_needs_flush, &job, delayed, k_job_id); if (r) return r;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h index 577ee04ce0bf..50e40380fe95 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h @@ -166,7 +166,7 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, uint64_t dst_offset, uint32_t byte_count, struct dma_resv *resv,

•        struct dma_fence **fence, bool direct_submit,
  

•        struct dma_fence **fence,
    bool vm_needs_flush, uint32_t copy_flags);
  

int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo, struct dma_resv *resv, diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index 46c84fc60af1..378af0b2aaa9 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -153,7 +153,7 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, } r = amdgpu_copy_buffer(ring, gart_s, gart_d, size * PAGE_SIZE,

• 		       NULL, &next, false, true, 0);
  

• 		       NULL, &next, true, 0);
  

  if (r) { dev_err(adev->dev, "fail %d to copy memory\n", r); goto out_unlock;

On 11/13/25 17:05, Pierre-Eric Pelloux-Prayer wrote:

This way the caller can select the one it wants to use.

Signed-off-by: Pierre-Eric Pelloux-Prayer pierre-eric.pelloux-prayer@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c | 3 +- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 75 +++++++++++-------- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 16 ++-- drivers/gpu/drm/amd/amdkfd/kfd_migrate.c | 3 +- 5 files changed, 60 insertions(+), 41 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c index 02c2479a8840..b59040a8771f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c @@ -38,7 +38,8 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size, stime = ktime_get(); for (i = 0; i < n; i++) { struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;

• r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,
  

• r = amdgpu_copy_buffer(ring, &adev->mman.default_entity.base,
  

• 		       saddr, daddr, size, NULL, &fence,
  	       false, 0);
  

  if (r) goto exit_do_move;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c index e08f58de4b17..c06c132a753c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -1321,8 +1321,8 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo) if (r) goto out;

• r = amdgpu_fill_buffer(abo, 0, &bo->base._resv, &fence, true,

• 	       AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
  

• r = amdgpu_fill_buffer(&adev->mman.clear_entity, abo, 0, &bo->base._resv,

• 	       &fence, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);
  

  if (WARN_ON(r)) goto out;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 42d448cd6a6d..c8d59ca2b3bd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -164,6 +164,7 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo, /**

• amdgpu_ttm_map_buffer - Map memory into the GART windows

• • @entity: entity to run the window setup job
  • @bo: buffer object to map
  • @mem: memory object to map
  • @mm_cur: range to map

@@ -176,7 +177,8 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,

• Setup one of the GART windows to access a specific piece of memory or return
• the physical address for local memory.

*/ -static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, +static int amdgpu_ttm_map_buffer(struct drm_sched_entity *entity,

• 		 struct ttm_buffer_object *bo,
  

Probably better to split this patch into multiple patches.

One which changes amdgpu_ttm_map_buffer() and then another one or two for the higher level copy_buffer and fill_buffer functions.

		 struct ttm_resource *mem,
		 struct amdgpu_res_cursor *mm_cur,
		 unsigned int window, struct amdgpu_ring *ring,

@@ -224,7 +226,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8); num_bytes = num_pages * 8 * AMDGPU_GPU_PAGES_IN_CPU_PAGE;

• r = amdgpu_job_alloc_with_ib(adev, &adev->mman.default_entity.base,

• r = amdgpu_job_alloc_with_ib(adev, entity, AMDGPU_FENCE_OWNER_UNDEFINED, num_dw * 4 + num_bytes, AMDGPU_IB_POOL_DELAYED, &job,

@@ -274,6 +276,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, /**

• amdgpu_ttm_copy_mem_to_mem - Helper function for copy
• @adev: amdgpu device

• • @entity: entity to run the jobs
  • @src: buffer/address where to read from
  • @dst: buffer/address where to write to
  • @size: number of bytes to copy

@@ -288,6 +291,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, */ __attribute__((nonnull)) static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,

• 		      struct drm_sched_entity *entity,
  	      const struct amdgpu_copy_mem *src,
  	      const struct amdgpu_copy_mem *dst,
  	      uint64_t size, bool tmz,
  

@@ -320,12 +324,14 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, cur_size = min3(src_mm.size, dst_mm.size, 256ULL << 20); /* Map src to window 0 and dst to window 1. */

• r = amdgpu_ttm_map_buffer(src->bo, src->mem, &src_mm,
  

• r = amdgpu_ttm_map_buffer(entity,
  

• 			  src->bo, src->mem, &src_mm,
  		  0, ring, tmz, &cur_size, &from);
  

  if (r) goto error;

• r = amdgpu_ttm_map_buffer(dst->bo, dst->mem, &dst_mm,
  

• r = amdgpu_ttm_map_buffer(entity,
  

• 			  dst->bo, dst->mem, &dst_mm,
  		  1, ring, tmz, &cur_size, &to);
  

  if (r) goto error;

@@ -353,7 +359,7 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, write_compress_disable)); }

• r = amdgpu_copy_buffer(ring, from, to, cur_size, resv,
  

• r = amdgpu_copy_buffer(ring, entity, from, to, cur_size, resv,
  	       &next, true, copy_flags);
  

  if (r) goto error;

@@ -394,7 +400,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo, src.offset = 0; dst.offset = 0;

• r = amdgpu_ttm_copy_mem_to_mem(adev, &src, &dst,

• r = amdgpu_ttm_copy_mem_to_mem(adev,

• 		       &adev->mman.move_entity.base,
  

• 		       &src, &dst,
  	       new_mem->size,
  	       amdgpu_bo_encrypted(abo),
  	       bo->base.resv, &fence);
  

@@ -406,8 +414,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo, (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) { struct dma_fence *wipe_fence = NULL;

• r = amdgpu_fill_buffer(abo, 0, NULL, &wipe_fence,
  

• 		       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
  

• r = amdgpu_fill_buffer(&adev->mman.move_entity,
  

• 		       abo, 0, NULL, &wipe_fence,
  

• 		       AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);
  

  if (r) { goto error; } else if (wipe_fence) {

@@ -2223,16 +2232,15 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) } static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,

• 		  struct drm_sched_entity *entity,
  	  unsigned int num_dw,
  	  struct dma_resv *resv,
  	  bool vm_needs_flush,
  	  struct amdgpu_job **job,
  

• 		  bool delayed, u64 k_job_id)
  

• 		  u64 k_job_id)
  

{ enum amdgpu_ib_pool_type pool = AMDGPU_IB_POOL_DELAYED; int r;

• struct drm_sched_entity *entity = delayed ? &adev->mman.clear_entity.base :

• 				    &adev->mman.move_entity.base;
  

  r = amdgpu_job_alloc_with_ib(adev, entity, AMDGPU_FENCE_OWNER_UNDEFINED, num_dw * 4, pool, job, k_job_id);

@@ -2252,7 +2260,9 @@ static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev, DMA_RESV_USAGE_BOOKKEEP); } -int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, +int amdgpu_copy_buffer(struct amdgpu_ring *ring,

•        struct drm_sched_entity *entity,
  

•        uint64_t src_offset,
    uint64_t dst_offset, uint32_t byte_count,
    struct dma_resv *resv,
    struct dma_fence **fence,
  

@@ -2274,8 +2284,8 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, max_bytes = adev->mman.buffer_funcs->copy_max_bytes; num_loops = DIV_ROUND_UP(byte_count, max_bytes); num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8);

• r = amdgpu_ttm_prepare_job(adev, num_dw,

• 		   resv, vm_needs_flush, &job, false,
  

• r = amdgpu_ttm_prepare_job(adev, entity, num_dw,

• 		   resv, vm_needs_flush, &job,
  	   AMDGPU_KERNEL_JOB_ID_TTM_COPY_BUFFER);
  

  if (r) return r;

@@ -2304,11 +2314,13 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, return r; } -static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data, +static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring,

• 	       struct drm_sched_entity *entity,
  

• 	       uint32_t src_data,
         uint64_t dst_addr, uint32_t byte_count,
         struct dma_resv *resv,
         struct dma_fence **fence,
  

• 	       bool vm_needs_flush, bool delayed,
  

• 	       bool vm_needs_flush,
         u64 k_job_id)
  

{ struct amdgpu_device *adev = ring->adev; @@ -2321,8 +2333,8 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data, max_bytes = adev->mman.buffer_funcs->fill_max_bytes; num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes); num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8);

• r = amdgpu_ttm_prepare_job(adev, num_dw, resv, vm_needs_flush,

• 		   &job, delayed, k_job_id);
  

• r = amdgpu_ttm_prepare_job(adev, entity, num_dw, resv,

• 		   vm_needs_flush, &job, k_job_id);
  

  if (r) return r;

@@ -2386,13 +2398,14 @@ int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo, /* Never clear more than 256MiB at once to avoid timeouts */ size = min(cursor.size, 256ULL << 20);

• r = amdgpu_ttm_map_buffer(&bo->tbo, bo->tbo.resource, &cursor,
  

• r = amdgpu_ttm_map_buffer(&adev->mman.clear_entity.base,
  

• 			  &bo->tbo, bo->tbo.resource, &cursor,
  		  1, ring, false, &size, &addr);
  

  if (r) goto err;

• r = amdgpu_ttm_fill_mem(ring, 0, addr, size, resv,
  

• 			&next, true, true,
  

• r = amdgpu_ttm_fill_mem(ring, &adev->mman.clear_entity.base, 0, addr, size, resv,
  

• 			&next, true,
  		AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);
  

  if (r) goto err;

@@ -2408,12 +2421,12 @@ int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo, return r; } -int amdgpu_fill_buffer(struct amdgpu_bo *bo,

• 	uint32_t src_data,
  

• 	struct dma_resv *resv,
  

• 	struct dma_fence **f,
  

• 	bool delayed,
  

• 	u64 k_job_id)
  

+int amdgpu_fill_buffer(struct amdgpu_ttm_entity *entity,

•        struct amdgpu_bo *bo,
  

•        uint32_t src_data,
  

•        struct dma_resv *resv,
  

•        struct dma_fence **f,
  

•        u64 k_job_id)
  

{ struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; @@ -2437,13 +2450,15 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo, /* Never fill more than 256MiB at once to avoid timeouts */ cur_size = min(dst.size, 256ULL << 20);

• r = amdgpu_ttm_map_buffer(&bo->tbo, bo->tbo.resource, &dst,
  

• r = amdgpu_ttm_map_buffer(&entity->base,
  

• 			  &bo->tbo, bo->tbo.resource, &dst,
  		  1, ring, false, &cur_size, &to);
  

  if (r) goto error;

• r = amdgpu_ttm_fill_mem(ring, src_data, to, cur_size, resv,
  

• 			&next, true, delayed, k_job_id);
  

• r = amdgpu_ttm_fill_mem(ring, &entity->base,
  

• 			src_data, to, cur_size, resv,
  

• 			&next, true, k_job_id);
  

  if (r) goto error;

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h index d2295d6c2b67..e1655f86a016 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h @@ -167,7 +167,9 @@ int amdgpu_ttm_init(struct amdgpu_device *adev); void amdgpu_ttm_fini(struct amdgpu_device *adev); void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable); -int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, +int amdgpu_copy_buffer(struct amdgpu_ring *ring,

•        struct drm_sched_entity *entity,
  

If I'm not completely mistaken you should be able to drop the ring argument since that can be determined from the entity.

Apart from that looks rather good to me.

Regards, Christian.

•        uint64_t src_offset,
    uint64_t dst_offset, uint32_t byte_count,
    struct dma_resv *resv,
    struct dma_fence **fence,
  

@@ -175,12 +177,12 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo, struct dma_resv *resv, struct dma_fence **fence); -int amdgpu_fill_buffer(struct amdgpu_bo *bo,

• 	uint32_t src_data,
  

• 	struct dma_resv *resv,
  

• 	struct dma_fence **fence,
  

• 	bool delayed,
  

• 	u64 k_job_id);
  

+int amdgpu_fill_buffer(struct amdgpu_ttm_entity *entity,

•        struct amdgpu_bo *bo,
  

•        uint32_t src_data,
  

•        struct dma_resv *resv,
  

•        struct dma_fence **f,
  

•        u64 k_job_id);
  

int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo); void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c index d74ff6e90590..09756132fa1b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -157,7 +157,8 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys, goto out_unlock; }

• r = amdgpu_copy_buffer(ring, gart_s, gart_d, size * PAGE_SIZE,
  

• r = amdgpu_copy_buffer(ring, &entity->base,
  

• 		       gart_s, gart_d, size * PAGE_SIZE,
  	       NULL, &next, true, 0);
  

  if (r) { dev_err(adev->dev, "fail %d to copy memory\n", r);

Le 14/11/2025 à 15:41, Pierre-Eric Pelloux-Prayer a écrit :

Le 14/11/2025 à 14:07, Christian König a écrit :

...

On 11/13/25 17:05, Pierre-Eric Pelloux-Prayer wrote:

...

This way the caller can select the one it wants to use.

Signed-off-by: Pierre-Eric Pelloux-Prayer pierre-eric.pelloux-prayer@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c |  3 +-   drivers/gpu/drm/amd/amdgpu/amdgpu_object.c    |  4 +-   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c       | 75 +++++++++++--------   drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h       | 16 ++--   drivers/gpu/drm/amd/amdkfd/kfd_migrate.c      |  3 +-   5 files changed, 60 insertions(+), 41 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c b/drivers/gpu/drm/ amd/amdgpu/amdgpu_benchmark.c index 02c2479a8840..b59040a8771f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_benchmark.c @@ -38,7 +38,8 @@ static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,       stime = ktime_get();       for (i = 0; i < n; i++) {           struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; -        r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence, +        r = amdgpu_copy_buffer(ring, &adev->mman.default_entity.base, +                       saddr, daddr, size, NULL, &fence,                          false, 0);           if (r)               goto exit_do_move; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/ amd/amdgpu/amdgpu_object.c index e08f58de4b17..c06c132a753c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c @@ -1321,8 +1321,8 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)       if (r)           goto out; -    r = amdgpu_fill_buffer(abo, 0, &bo->base._resv, &fence, true, -                   AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE); +    r = amdgpu_fill_buffer(&adev->mman.clear_entity, abo, 0, &bo->base._resv, +                   &fence, AMDGPU_KERNEL_JOB_ID_CLEAR_ON_RELEASE);       if (WARN_ON(r))           goto out; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/ amdgpu/amdgpu_ttm.c index 42d448cd6a6d..c8d59ca2b3bd 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -164,6 +164,7 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,   /**    * amdgpu_ttm_map_buffer - Map memory into the GART windows

• • @entity: entity to run the window setup job

* @bo: buffer object to map    * @mem: memory object to map    * @mm_cur: range to map @@ -176,7 +177,8 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,    * Setup one of the GART windows to access a specific piece of memory or return    * the physical address for local memory.    */ -static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo, +static int amdgpu_ttm_map_buffer(struct drm_sched_entity *entity, +                 struct ttm_buffer_object *bo,

Probably better to split this patch into multiple patches.

One which changes amdgpu_ttm_map_buffer() and then another one or two for the higher level copy_buffer and fill_buffer functions.

OK.

...

...

struct ttm_resource *mem,                    struct amdgpu_res_cursor *mm_cur,                    unsigned int window, struct amdgpu_ring *ring, @@ -224,7 +226,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,       num_dw = ALIGN(adev->mman.buffer_funcs->copy_num_dw, 8);       num_bytes = num_pages * 8 * AMDGPU_GPU_PAGES_IN_CPU_PAGE; -    r = amdgpu_job_alloc_with_ib(adev, &adev->mman.default_entity.base, +    r = amdgpu_job_alloc_with_ib(adev, entity,                        AMDGPU_FENCE_OWNER_UNDEFINED,                        num_dw * 4 + num_bytes,                        AMDGPU_IB_POOL_DELAYED, &job, @@ -274,6 +276,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,   /**    * amdgpu_ttm_copy_mem_to_mem - Helper function for copy    * @adev: amdgpu device

• • @entity: entity to run the jobs

* @src: buffer/address where to read from    * @dst: buffer/address where to write to    * @size: number of bytes to copy @@ -288,6 +291,7 @@ static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,    */   __attribute__((nonnull))   static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, +                      struct drm_sched_entity *entity,                         const struct amdgpu_copy_mem *src,                         const struct amdgpu_copy_mem *dst,                         uint64_t size, bool tmz, @@ -320,12 +324,14 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,           cur_size = min3(src_mm.size, dst_mm.size, 256ULL << 20);           /* Map src to window 0 and dst to window 1. */ -        r = amdgpu_ttm_map_buffer(src->bo, src->mem, &src_mm, +        r = amdgpu_ttm_map_buffer(entity, +                      src->bo, src->mem, &src_mm,                         0, ring, tmz, &cur_size, &from);           if (r)               goto error; -        r = amdgpu_ttm_map_buffer(dst->bo, dst->mem, &dst_mm, +        r = amdgpu_ttm_map_buffer(entity, +                      dst->bo, dst->mem, &dst_mm,                         1, ring, tmz, &cur_size, &to);           if (r)               goto error; @@ -353,7 +359,7 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,                                    write_compress_disable));           } -        r = amdgpu_copy_buffer(ring, from, to, cur_size, resv, +        r = amdgpu_copy_buffer(ring, entity, from, to, cur_size, resv,                          &next, true, copy_flags);           if (r)               goto error; @@ -394,7 +400,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,       src.offset = 0;       dst.offset = 0; -    r = amdgpu_ttm_copy_mem_to_mem(adev, &src, &dst, +    r = amdgpu_ttm_copy_mem_to_mem(adev, +                       &adev->mman.move_entity.base, +                       &src, &dst,                          new_mem->size,                          amdgpu_bo_encrypted(abo),                          bo->base.resv, &fence); @@ -406,8 +414,9 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,           (abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {           struct dma_fence *wipe_fence = NULL; -        r = amdgpu_fill_buffer(abo, 0, NULL, &wipe_fence, -                       false, AMDGPU_KERNEL_JOB_ID_MOVE_BLIT); +        r = amdgpu_fill_buffer(&adev->mman.move_entity, +                       abo, 0, NULL, &wipe_fence, +                       AMDGPU_KERNEL_JOB_ID_MOVE_BLIT);           if (r) {               goto error;           } else if (wipe_fence) { @@ -2223,16 +2232,15 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)   }   static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev, +                  struct drm_sched_entity *entity,                     unsigned int num_dw,                     struct dma_resv *resv,                     bool vm_needs_flush,                     struct amdgpu_job **job, -                  bool delayed, u64 k_job_id) +                  u64 k_job_id)   {       enum amdgpu_ib_pool_type pool = AMDGPU_IB_POOL_DELAYED;       int r; -    struct drm_sched_entity *entity = delayed ? &adev->mman.clear_entity.base : -                            &adev->mman.move_entity.base;       r = amdgpu_job_alloc_with_ib(adev, entity,                        AMDGPU_FENCE_OWNER_UNDEFINED,                        num_dw * 4, pool, job, k_job_id); @@ -2252,7 +2260,9 @@ static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,                              DMA_RESV_USAGE_BOOKKEEP);   } -int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, +int amdgpu_copy_buffer(struct amdgpu_ring *ring, +               struct drm_sched_entity *entity, +               uint64_t src_offset,                  uint64_t dst_offset, uint32_t byte_count,                  struct dma_resv *resv,                  struct dma_fence **fence, @@ -2274,8 +2284,8 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,       max_bytes = adev->mman.buffer_funcs->copy_max_bytes;       num_loops = DIV_ROUND_UP(byte_count, max_bytes);       num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8); -    r = amdgpu_ttm_prepare_job(adev, num_dw, -                   resv, vm_needs_flush, &job, false, +    r = amdgpu_ttm_prepare_job(adev, entity, num_dw, +                   resv, vm_needs_flush, &job,                      AMDGPU_KERNEL_JOB_ID_TTM_COPY_BUFFER);       if (r)           return r; @@ -2304,11 +2314,13 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,       return r;   } -static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data, +static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, +                   struct drm_sched_entity *entity, +                   uint32_t src_data,                      uint64_t dst_addr, uint32_t byte_count,                      struct dma_resv *resv,                      struct dma_fence **fence, -                   bool vm_needs_flush, bool delayed, +                   bool vm_needs_flush,                      u64 k_job_id)   {       struct amdgpu_device *adev = ring->adev; @@ -2321,8 +2333,8 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data,       max_bytes = adev->mman.buffer_funcs->fill_max_bytes;       num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes);       num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8); -    r = amdgpu_ttm_prepare_job(adev, num_dw, resv, vm_needs_flush, -                   &job, delayed, k_job_id); +    r = amdgpu_ttm_prepare_job(adev, entity, num_dw, resv, +                   vm_needs_flush, &job, k_job_id);       if (r)           return r; @@ -2386,13 +2398,14 @@ int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,           /* Never clear more than 256MiB at once to avoid timeouts */           size = min(cursor.size, 256ULL << 20); -        r = amdgpu_ttm_map_buffer(&bo->tbo, bo->tbo.resource, &cursor, +        r = amdgpu_ttm_map_buffer(&adev->mman.clear_entity.base, +                      &bo->tbo, bo->tbo.resource, &cursor,                         1, ring, false, &size, &addr);           if (r)               goto err; -        r = amdgpu_ttm_fill_mem(ring, 0, addr, size, resv, -                    &next, true, true, +        r = amdgpu_ttm_fill_mem(ring, &adev->mman.clear_entity.base, 0, addr, size, resv, +                    &next, true,                       AMDGPU_KERNEL_JOB_ID_TTM_CLEAR_BUFFER);           if (r)               goto err; @@ -2408,12 +2421,12 @@ int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,       return r;   } -int amdgpu_fill_buffer(struct amdgpu_bo *bo, -            uint32_t src_data, -            struct dma_resv *resv, -            struct dma_fence **f, -            bool delayed, -            u64 k_job_id) +int amdgpu_fill_buffer(struct amdgpu_ttm_entity *entity, +               struct amdgpu_bo *bo, +               uint32_t src_data, +               struct dma_resv *resv, +               struct dma_fence **f, +               u64 k_job_id)   {       struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);       struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring; @@ -2437,13 +2450,15 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,           /* Never fill more than 256MiB at once to avoid timeouts */           cur_size = min(dst.size, 256ULL << 20); -        r = amdgpu_ttm_map_buffer(&bo->tbo, bo->tbo.resource, &dst, +        r = amdgpu_ttm_map_buffer(&entity->base, +                      &bo->tbo, bo->tbo.resource, &dst,                         1, ring, false, &cur_size, &to);           if (r)               goto error; -        r = amdgpu_ttm_fill_mem(ring, src_data, to, cur_size, resv, -                    &next, true, delayed, k_job_id); +        r = amdgpu_ttm_fill_mem(ring, &entity->base, +                    src_data, to, cur_size, resv, +                    &next, true, k_job_id);           if (r)               goto error; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/ amdgpu/amdgpu_ttm.h index d2295d6c2b67..e1655f86a016 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h @@ -167,7 +167,9 @@ int amdgpu_ttm_init(struct amdgpu_device *adev);   void amdgpu_ttm_fini(struct amdgpu_device *adev);   void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev,                       bool enable); -int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset, +int amdgpu_copy_buffer(struct amdgpu_ring *ring, +               struct drm_sched_entity *entity,

If I'm not completely mistaken you should be able to drop the ring argument since that can be determined from the entity.

OK will do.

AFAIU the only way to get the ring from the entity is to get it from the drm_gpu_scheduler pointer. This would require adding a new function:

struct drm_gpu_scheduler * drm_sched_entity_get_scheduler(struct drm_sched_entity *entity) { struct drm_gpu_scheduler *sched; spin_lock(&entity->lock); if (entity->rq) sched = entity->rq->sched; spin_unlock(&entity->lock); return sched; }

Alternatively, I can access the ring from the buffer_funcs_ring / buffer_funcs_sched stored in amdgpu_mman.

What do you think?

Thanks, Pierre-Eric

...

Apart from that looks rather good to me.

Regards, Christian.

...

+               uint64_t src_offset,                  uint64_t dst_offset, uint32_t byte_count,                  struct dma_resv *resv,                  struct dma_fence **fence, @@ -175,12 +177,12 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,   int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo,                   struct dma_resv *resv,                   struct dma_fence **fence); -int amdgpu_fill_buffer(struct amdgpu_bo *bo, -            uint32_t src_data, -            struct dma_resv *resv, -            struct dma_fence **fence, -            bool delayed, -            u64 k_job_id); +int amdgpu_fill_buffer(struct amdgpu_ttm_entity *entity, +               struct amdgpu_bo *bo, +               uint32_t src_data, +               struct dma_resv *resv, +               struct dma_fence **f, +               u64 k_job_id);   int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);   void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo); diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c b/drivers/gpu/drm/amd/ amdkfd/kfd_migrate.c index d74ff6e90590..09756132fa1b 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_migrate.c @@ -157,7 +157,8 @@ svm_migrate_copy_memory_gart(struct amdgpu_device *adev, dma_addr_t *sys,               goto out_unlock;           } -        r = amdgpu_copy_buffer(ring, gart_s, gart_d, size * PAGE_SIZE, +        r = amdgpu_copy_buffer(ring, &entity->base, +                       gart_s, gart_d, size * PAGE_SIZE,                          NULL, &next, true, 0);           if (r) {               dev_err(adev->dev, "fail %d to copy memory\n", r);

Le 17/11/2025 à 09:44, Christian König a écrit :

On 11/13/25 17:05, Pierre-Eric Pelloux-Prayer wrote:

...

If a resv object is passed, its fences are treated as a dependency for the amdgpu_ttm_map_buffer operation.

This will be used by amdgpu_bo_release_notify through amdgpu_fill_buffer.

Why should updating the GART window depend on fences in a resv object?

You're right, this is not needed. I'll drop the patch.

Pierre-Eric

Regards, Christian.

...

Signed-off-by: Pierre-Eric Pelloux-Prayer pierre-eric.pelloux-prayer@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index b13f0993dbf1..411997db70eb 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -184,7 +184,8 @@ static int amdgpu_ttm_map_buffer(struct drm_sched_entity *entity, struct amdgpu_res_cursor *mm_cur, unsigned int window, struct amdgpu_ring *ring, bool tmz, uint64_t *size, uint64_t *addr,

• 		 struct dma_fence *dep)
  

• 		 struct dma_fence *dep,
  

• 		 struct dma_resv *resv)
  

  { struct amdgpu_device *adev = ring->adev; unsigned int offset, num_pages, num_dw, num_bytes;

@@ -239,6 +240,10 @@ static int amdgpu_ttm_map_buffer(struct drm_sched_entity *entity, if (dep) drm_sched_job_add_dependency(&job->base, dma_fence_get(dep));

• if (resv)

• drm_sched_job_add_resv_dependencies(&job->base, resv,
  

• 				    DMA_RESV_USAGE_BOOKKEEP);
  

• src_addr = num_dw * 4; src_addr += job->ibs[0].gpu_addr;

@@ -332,14 +337,14 @@ static int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev, r = amdgpu_ttm_map_buffer(&entity->base, src->bo, src->mem, &src_mm, entity->gart_window_id0, ring, tmz, &cur_size, &from,

• 			  NULL);
  

• 			  NULL, NULL);
  

  if (r) goto error;

r = amdgpu_ttm_map_buffer(&entity->base, dst->bo, dst->mem, &dst_mm, entity->gart_window_id1, ring, tmz, &cur_size, &to,

• 			  NULL);
  

• 			  NULL, NULL);
  

  if (r) goto error;

@@ -2451,7 +2456,7 @@ int amdgpu_ttm_clear_buffer(struct amdgpu_bo *bo, r = amdgpu_ttm_map_buffer(&entity->base, &bo->tbo, bo->tbo.resource, &cursor, entity->gart_window_id1, ring, false, &size, &addr,

• 			  NULL);
  

• 			  NULL, NULL);
  

  if (r) goto err;

@@ -2506,7 +2511,8 @@ int amdgpu_fill_buffer(struct amdgpu_ttm_buffer_entity *entity, &bo->tbo, bo->tbo.resource, &dst, entity->gart_window_id1, ring, false, &cur_size, &to,

• 			  dependency);
  

• 			  dependency,
  

• 			  resv);
  

  if (r) goto error;