drm/amdgpu: deal with foreign fences in amdgpu_sync

This also requires some error handling from the callers of that function.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

5 files changed, 45 insertions, 29 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 069cc28941ba..70e783a849ed 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -699,8 +699,8 @@ struct amdgpu_sync {
 };
 
 void amdgpu_sync_create(struct amdgpu_sync *sync);
-void amdgpu_sync_fence(struct amdgpu_sync *sync,
-                       struct amdgpu_fence *fence);
+int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
+                      struct fence *f);
 int amdgpu_sync_resv(struct amdgpu_device *adev,
                      struct amdgpu_sync *sync,
                      struct reservation_object *resv,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 1f040d85ac47..53e6a10fe9f9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -482,6 +482,8 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
 
         if (p->bo_list) {
                 for (i = 0; i < p->bo_list->num_entries; i++) {
+                        struct fence *f;
+
                         /* ignore duplicates */
                         bo = p->bo_list->array[i].robj;
                         if (!bo)
@@ -495,7 +497,10 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
                         if (r)
                                 return r;
 
-                        amdgpu_sync_fence(&p->ibs[0].sync, bo_va->last_pt_update);
+                        f = &bo_va->last_pt_update->base;
+                        r = amdgpu_sync_fence(adev, &p->ibs[0].sync, f);
+                        if (r)
+                                return r;
                 }
         }
 
@@ -715,9 +720,12 @@ static int amdgpu_cs_dependencies(struct amdgpu_device *adev,
                                 return r;
                         }
 
-                        amdgpu_sync_fence(&ib->sync, fence);
+                        r = amdgpu_sync_fence(adev, &ib->sync, &fence->base);
                         amdgpu_fence_unref(&fence);
                         amdgpu_ctx_put(ctx);
+
+                        if (r)
+                                return r;
                 }
         }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index bc0fac618a3f..2722815eddbb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -167,7 +167,11 @@ int amdgpu_ib_schedule(struct amdgpu_device *adev, unsigned num_ibs,
                 /* grab a vm id if necessary */
                 struct amdgpu_fence *vm_id_fence = NULL;
                 vm_id_fence = amdgpu_vm_grab_id(ibs->ring, ibs->vm);
-                amdgpu_sync_fence(&ibs->sync, vm_id_fence);
+                r = amdgpu_sync_fence(adev, &ibs->sync, &vm_id_fence->base);
+                if (r) {
+                        amdgpu_ring_unlock_undo(ring);
+                        return r;
+                }
         }
 
         r = amdgpu_sync_rings(&ibs->sync, ring);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
index 21accbdd0a1a..9c292cf770f4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c
@@ -53,20 +53,24 @@ void amdgpu_sync_create(struct amdgpu_sync *sync)
 }
 
 /**
- * amdgpu_sync_fence - use the semaphore to sync to a fence
+ * amdgpu_sync_fence - remember to sync to this fence
  *
  * @sync: sync object to add fence to
  * @fence: fence to sync to
  *
- * Sync to the fence using the semaphore objects
  */
-void amdgpu_sync_fence(struct amdgpu_sync *sync,
-                       struct amdgpu_fence *fence)
+int amdgpu_sync_fence(struct amdgpu_device *adev, struct amdgpu_sync *sync,
+                      struct fence *f)
 {
+        struct amdgpu_fence *fence;
         struct amdgpu_fence *other;
 
-        if (!fence)
-                return;
+        if (!f)
+                return 0;
+
+        fence = to_amdgpu_fence(f);
+        if (!fence || fence->ring->adev != adev)
+                return fence_wait(f, true);
 
         other = sync->sync_to[fence->ring->idx];
         sync->sync_to[fence->ring->idx] = amdgpu_fence_ref(
@@ -79,6 +83,8 @@ void amdgpu_sync_fence(struct amdgpu_sync *sync,
                         amdgpu_fence_later(fence, other));
                 amdgpu_fence_unref(&other);
         }
+
+        return 0;
 }
 
 /**
@@ -106,11 +112,7 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
 
         /* always sync to the exclusive fence */
         f = reservation_object_get_excl(resv);
-        fence = f ? to_amdgpu_fence(f) : NULL;
-        if (fence && fence->ring->adev == adev)
-                amdgpu_sync_fence(sync, fence);
-        else if (f)
-                r = fence_wait(f, true);
+        r = amdgpu_sync_fence(adev, sync, f);
 
         flist = reservation_object_get_list(resv);
         if (!flist || r)
@@ -120,15 +122,14 @@ int amdgpu_sync_resv(struct amdgpu_device *adev,
                 f = rcu_dereference_protected(flist->shared[i],
                                               reservation_object_held(resv));
                 fence = f ? to_amdgpu_fence(f) : NULL;
-                if (fence && fence->ring->adev == adev) {
-                        if (fence->owner != owner ||
-                            fence->owner == AMDGPU_FENCE_OWNER_UNDEFINED)
-                                amdgpu_sync_fence(sync, fence);
-                } else if (f) {
-                        r = fence_wait(f, true);
-                        if (r)
-                                break;
-                }
+                if (fence && fence->ring->adev == adev &&
+                    fence->owner == owner &&
+                    fence->owner != AMDGPU_FENCE_OWNER_UNDEFINED)
+                                continue;
+
+                r = amdgpu_sync_fence(adev, sync, f);
+                if (r)
+                        break;
         }
         return r;
 }
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 9a4e3b63f1cb..0c8c9904d880 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -732,7 +732,9 @@ static int amdgpu_vm_bo_update_mapping(struct amdgpu_device *adev,
 
                 for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
                         struct amdgpu_fence *f = vm->ids[i].last_id_use;
-                        amdgpu_sync_fence(&ib.sync, f);
+                        r = amdgpu_sync_fence(adev, &ib.sync, &f->base);
+                        if (r)
+                                return r;
                 }
         }
 
@@ -861,7 +863,7 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
                              struct amdgpu_vm *vm, struct amdgpu_sync *sync)
 {
         struct amdgpu_bo_va *bo_va = NULL;
-        int r;
+        int r = 0;
 
         spin_lock(&vm->status_lock);
         while (!list_empty(&vm->invalidated)) {
@@ -878,8 +880,9 @@ int amdgpu_vm_clear_invalids(struct amdgpu_device *adev,
         spin_unlock(&vm->status_lock);
 
         if (bo_va)
-                amdgpu_sync_fence(sync, bo_va->last_pt_update);
-        return 0;
+                r = amdgpu_sync_fence(adev, sync, &bo_va->last_pt_update->base);
+
+        return r;
 }
 
 /**