drm/amdgpu: improve sa_bo->fence by kernel fence

Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian K?nig <christian.koenig@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -441,7 +441,7 @@ int amdgpu_fence_wait_empty(struct amdgpu_ring *ring);
 unsigned amdgpu_fence_count_emitted(struct amdgpu_ring *ring);
 
 signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
-				       struct amdgpu_fence **array,
+				       struct fence **array,
 				       uint32_t count,
 				       bool wait_all,
 				       bool intr,
@@ -654,7 +654,7 @@ struct amdgpu_sa_bo {
 	struct amdgpu_sa_manager	*manager;
 	unsigned			soffset;
 	unsigned			eoffset;
-	struct amdgpu_fence		*fence;
+	struct fence		        *fence;
 };
 
 /*
@@ -696,7 +696,7 @@ bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring,
 				struct amdgpu_semaphore *semaphore);
 void amdgpu_semaphore_free(struct amdgpu_device *adev,
 			   struct amdgpu_semaphore **semaphore,
-			   struct amdgpu_fence *fence);
+			   struct fence *fence);
 
 /*
  * Synchronization
@@ -717,7 +717,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 int amdgpu_sync_rings(struct amdgpu_sync *sync,
 		      struct amdgpu_ring *ring);
 void amdgpu_sync_free(struct amdgpu_device *adev, struct amdgpu_sync *sync,
-		      struct amdgpu_fence *fence);
+		      struct fence *fence);
 
 /*
  * GART structures, functions & helpers

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

@@ -836,30 +836,30 @@ static inline bool amdgpu_test_signaled(struct amdgpu_fence *fence)
 	return test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags);
 }
 
-static bool amdgpu_test_signaled_any(struct amdgpu_fence **fences, uint32_t count)
+static bool amdgpu_test_signaled_any(struct fence **fences, uint32_t count)
 {
 	int idx;
-	struct amdgpu_fence *fence;
+	struct fence *fence;
 
 	for (idx = 0; idx < count; ++idx) {
 		fence = fences[idx];
 		if (fence) {
-			if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags))
+			if (test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->flags))
 				return true;
 		}
 	}
 	return false;
 }
 
-static bool amdgpu_test_signaled_all(struct amdgpu_fence **fences, uint32_t count)
+static bool amdgpu_test_signaled_all(struct fence **fences, uint32_t count)
 {
 	int idx;
-	struct amdgpu_fence *fence;
+	struct fence *fence;
 
 	for (idx = 0; idx < count; ++idx) {
 		fence = fences[idx];
 		if (fence) {
-			if (!test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->base.flags))
+			if (!test_bit(FENCE_FLAG_SIGNALED_BIT, &fence->flags))
 				return false;
 		}
 	}
@@ -885,7 +885,7 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
 	struct amdgpu_fence *fence = to_amdgpu_fence(f);
 	struct amdgpu_device *adev = fence->ring->adev;
 
-	return amdgpu_fence_wait_multiple(adev, &fence, 1, false, intr, t);
+	return amdgpu_fence_wait_multiple(adev, &f, 1, false, intr, t);
 }
 
 /**
@@ -902,7 +902,7 @@ static signed long amdgpu_fence_default_wait(struct fence *f, bool intr,
  * If wait_all is false, it will return when any fence is signaled or timeout.
  */
 signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
-				       struct amdgpu_fence **array,
+				       struct fence **array,
 				       uint32_t count,
 				       bool wait_all,
 				       bool intr,
@@ -910,7 +910,7 @@ signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
 {
 	long idx = 0;
 	struct amdgpu_wait_cb *cb;
-	struct amdgpu_fence *fence;
+	struct fence *fence;
 
 	BUG_ON(!array);
 
@@ -924,7 +924,7 @@ signed long amdgpu_fence_wait_multiple(struct amdgpu_device *adev,
 		fence = array[idx];
 		if (fence) {
 			cb[idx].task = current;
-			if (fence_add_callback(&fence->base,
+			if (fence_add_callback(fence,
 					&cb[idx].base, amdgpu_fence_wait_cb)) {
 				/* The fence is already signaled */
 				if (wait_all)
@@ -967,7 +967,7 @@ fence_rm_cb:
 	for (idx = 0; idx < count; ++idx) {
 		fence = array[idx];
 		if (fence)
-			fence_remove_callback(&fence->base, &cb[idx].base);
+			fence_remove_callback(fence, &cb[idx].base);
 	}
 
 err_free_cb:

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

@@ -93,8 +93,8 @@ int amdgpu_ib_get(struct amdgpu_ring *ring, struct amdgpu_vm *vm,
  */
 void amdgpu_ib_free(struct amdgpu_device *adev, struct amdgpu_ib *ib)
 {
-	amdgpu_sync_free(adev, &ib->sync, ib->fence);
-	amdgpu_sa_bo_free(adev, &ib->sa_bo, ib->fence);
+	amdgpu_sync_free(adev, &ib->sync, &ib->fence->base);
+	amdgpu_sa_bo_free(adev, &ib->sa_bo, &ib->fence->base);
 	amdgpu_fence_unref(&ib->fence);
 }
 

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

@@ -193,7 +193,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 			    unsigned size, unsigned align);
 void amdgpu_sa_bo_free(struct amdgpu_device *adev,
 			      struct amdgpu_sa_bo **sa_bo,
-			      struct amdgpu_fence *fence);
+			      struct fence *fence);
 #if defined(CONFIG_DEBUG_FS)
 void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager,
 					 struct seq_file *m);

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

@@ -139,6 +139,20 @@ int amdgpu_sa_bo_manager_suspend(struct amdgpu_device *adev,
 	return r;
 }
 
+static uint32_t amdgpu_sa_get_ring_from_fence(struct fence *f)
+{
+	struct amdgpu_fence *a_fence;
+	struct amd_sched_fence *s_fence;
+
+	s_fence = to_amd_sched_fence(f);
+	if (s_fence)
+		return s_fence->entity->scheduler->ring_id;
+	a_fence = to_amdgpu_fence(f);
+	if (a_fence)
+		return a_fence->ring->idx;
+	return 0;
+}
+
 static void amdgpu_sa_bo_remove_locked(struct amdgpu_sa_bo *sa_bo)
 {
 	struct amdgpu_sa_manager *sa_manager = sa_bo->manager;
@@ -147,7 +161,7 @@ static void amdgpu_sa_bo_remove_locked(struct amdgpu_sa_bo *sa_bo)
 	}
 	list_del_init(&sa_bo->olist);
 	list_del_init(&sa_bo->flist);
-	amdgpu_fence_unref(&sa_bo->fence);
+	fence_put(sa_bo->fence);
 	kfree(sa_bo);
 }
 
@@ -161,7 +175,7 @@ static void amdgpu_sa_bo_try_free(struct amdgpu_sa_manager *sa_manager)
 	sa_bo = list_entry(sa_manager->hole->next, struct amdgpu_sa_bo, olist);
 	list_for_each_entry_safe_from(sa_bo, tmp, &sa_manager->olist, olist) {
 		if (sa_bo->fence == NULL ||
-		    !fence_is_signaled(&sa_bo->fence->base)) {
+		    !fence_is_signaled(sa_bo->fence)) {
 			return;
 		}
 		amdgpu_sa_bo_remove_locked(sa_bo);
@@ -246,7 +260,7 @@ static bool amdgpu_sa_event(struct amdgpu_sa_manager *sa_manager,
 }
 
 static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
-				   struct amdgpu_fence **fences,
+				   struct fence **fences,
 				   unsigned *tries)
 {
 	struct amdgpu_sa_bo *best_bo = NULL;
@@ -275,7 +289,7 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
 		sa_bo = list_first_entry(&sa_manager->flist[i],
 					 struct amdgpu_sa_bo, flist);
 
-		if (!fence_is_signaled(&sa_bo->fence->base)) {
+		if (!fence_is_signaled(sa_bo->fence)) {
 			fences[i] = sa_bo->fence;
 			continue;
 		}
@@ -299,7 +313,8 @@ static bool amdgpu_sa_bo_next_hole(struct amdgpu_sa_manager *sa_manager,
 	}
 
 	if (best_bo) {
-		++tries[best_bo->fence->ring->idx];
+		uint32_t idx = amdgpu_sa_get_ring_from_fence(best_bo->fence);
+		++tries[idx];
 		sa_manager->hole = best_bo->olist.prev;
 
 		/* we knew that this one is signaled,
@@ -315,7 +330,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 		     struct amdgpu_sa_bo **sa_bo,
 		     unsigned size, unsigned align)
 {
-	struct amdgpu_fence *fences[AMDGPU_MAX_RINGS];
+	struct fence *fences[AMDGPU_MAX_RINGS];
 	unsigned tries[AMDGPU_MAX_RINGS];
 	int i, r;
 	signed long t;
@@ -373,7 +388,7 @@ int amdgpu_sa_bo_new(struct amdgpu_device *adev,
 }
 
 void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo,
-		       struct amdgpu_fence *fence)
+		       struct fence *fence)
 {
 	struct amdgpu_sa_manager *sa_manager;
 
@@ -383,10 +398,11 @@ void amdgpu_sa_bo_free(struct amdgpu_device *adev, struct amdgpu_sa_bo **sa_bo,
 
 	sa_manager = (*sa_bo)->manager;
 	spin_lock(&sa_manager->wq.lock);
-	if (fence && !fence_is_signaled(&fence->base)) {
-		(*sa_bo)->fence = amdgpu_fence_ref(fence);
-		list_add_tail(&(*sa_bo)->flist,
-			      &sa_manager->flist[fence->ring->idx]);
+	if (fence && !fence_is_signaled(fence)) {
+		uint32_t idx;
+		(*sa_bo)->fence = fence_get(fence);
+		idx = amdgpu_sa_get_ring_from_fence(fence);
+		list_add_tail(&(*sa_bo)->flist, &sa_manager->flist[idx]);
 	} else {
 		amdgpu_sa_bo_remove_locked(*sa_bo);
 	}
@@ -413,8 +429,16 @@ void amdgpu_sa_bo_dump_debug_info(struct amdgpu_sa_manager *sa_manager,
 		seq_printf(m, "[0x%010llx 0x%010llx] size %8lld",
 			   soffset, eoffset, eoffset - soffset);
 		if (i->fence) {
-			seq_printf(m, " protected by 0x%016llx on ring %d",
-				   i->fence->seq, i->fence->ring->idx);
+			struct amdgpu_fence *a_fence = to_amdgpu_fence(i->fence);
+			struct amd_sched_fence *s_fence = to_amd_sched_fence(i->fence);
+			if (a_fence)
+				seq_printf(m, " protected by 0x%016llx on ring %d",
+					   a_fence->seq, a_fence->ring->idx);
+			if (s_fence)
+				seq_printf(m, " protected by 0x%016llx on ring %d",
+					   s_fence->v_seq,
+					   s_fence->entity->scheduler->ring_id);
+
 		}
 		seq_printf(m, "\n");
 	}

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

@@ -87,7 +87,7 @@ bool amdgpu_semaphore_emit_wait(struct amdgpu_ring *ring,
 
 void amdgpu_semaphore_free(struct amdgpu_device *adev,
 			   struct amdgpu_semaphore **semaphore,
-			   struct amdgpu_fence *fence)
+			   struct fence *fence)
 {
 	if (semaphore == NULL || *semaphore == NULL) {
 		return;

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

@@ -234,7 +234,7 @@ int amdgpu_sync_rings(struct amdgpu_sync *sync,
  */
 void amdgpu_sync_free(struct amdgpu_device *adev,
 		      struct amdgpu_sync *sync,
-		      struct amdgpu_fence *fence)
+		      struct fence *fence)
 {
 	unsigned i;
 

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

@@ -1042,7 +1042,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring,
 	}
 
 	amdgpu_ring_unlock_commit(ring);
-	amdgpu_sync_free(adev, &sync, *fence);
+	amdgpu_sync_free(adev, &sync, &(*fence)->base);
 
 	return 0;
 }