drm/radeon: fence BO_VAs manually

This allows us to finally remove the VM fence and
so allow concurrent use of it from different engines.

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

3 files changed, 14 insertions, 16 deletions

diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 79f5f5bf4c0c..3207bb60715e 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -456,6 +456,7 @@ struct radeon_bo_va {
         struct list_head                bo_list;
         uint32_t                        flags;
         uint64_t                        addr;
+        struct radeon_fence             *last_pt_update;
         unsigned                        ref_count;
 
         /* protected by vm mutex */
@@ -915,6 +916,8 @@ struct radeon_vm_id {
 };
 
 struct radeon_vm {
+        struct mutex            mutex;
+
         struct rb_root          va;
 
         /* BOs moved, but not yet updated in the PT */
@@ -932,10 +935,6 @@ struct radeon_vm {
 
         struct radeon_bo_va     *ib_bo_va;
 
-        struct mutex            mutex;
-        /* last fence for cs using this vm */
-        struct radeon_fence     *fence;
-
         /* for id and flush management per ring */
         struct radeon_vm_id     ids[RADEON_NUM_RINGS];
 };
diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 30437aa00014..75f22e5e999f 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -505,6 +505,9 @@ static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
         if (r)
                 return r;
 
+        radeon_sync_resv(p->rdev, &p->ib.sync, vm->page_directory->tbo.resv,
+                         true);
+
         r = radeon_vm_clear_freed(rdev, vm);
         if (r)
                 return r;
@@ -536,6 +539,8 @@ static int radeon_bo_vm_update_pte(struct radeon_cs_parser *p,
                 r = radeon_vm_bo_update(rdev, bo_va, &bo->tbo.mem);
                 if (r)
                         return r;
+
+                radeon_sync_fence(&p->ib.sync, bo_va->last_pt_update);
         }
 
         return radeon_vm_clear_invalids(rdev, vm);
@@ -580,7 +585,6 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
                         DRM_ERROR("Failed to sync rings: %i\n", r);
                 goto out;
         }
-        radeon_sync_fence(&parser->ib.sync, vm->fence);
 
         if ((rdev->family >= CHIP_TAHITI) &&
             (parser->chunk_const_ib_idx != -1)) {
diff --git a/drivers/gpu/drm/radeon/radeon_vm.c b/drivers/gpu/drm/radeon/radeon_vm.c
index e38efe4962f3..f45761469e95 100644
--- a/drivers/gpu/drm/radeon/radeon_vm.c
+++ b/drivers/gpu/drm/radeon/radeon_vm.c
@@ -275,9 +275,6 @@ void radeon_vm_fence(struct radeon_device *rdev,
 {
         unsigned vm_id = vm->ids[fence->ring].id;
 
-        radeon_fence_unref(&vm->fence);
-        vm->fence = radeon_fence_ref(fence);
-
         radeon_fence_unref(&rdev->vm_manager.active[vm_id]);
         rdev->vm_manager.active[vm_id] = radeon_fence_ref(fence);
 
@@ -707,8 +704,6 @@ int radeon_vm_update_page_directory(struct radeon_device *rdev,
                 }
                 ib.fence->is_vm_update = true;
                 radeon_bo_fence(pd, ib.fence, false);
-                radeon_fence_unref(&vm->fence);
-                vm->fence = radeon_fence_ref(ib.fence);
         }
         radeon_ib_free(rdev, &ib);
 
@@ -999,8 +994,8 @@ int radeon_vm_bo_update(struct radeon_device *rdev,
         }
         ib.fence->is_vm_update = true;
         radeon_vm_fence_pts(vm, bo_va->it.start, bo_va->it.last + 1, ib.fence);
-        radeon_fence_unref(&vm->fence);
-        vm->fence = radeon_fence_ref(ib.fence);
+        radeon_fence_unref(&bo_va->last_pt_update);
+        bo_va->last_pt_update = radeon_fence_ref(ib.fence);
         radeon_ib_free(rdev, &ib);
 
         return 0;
@@ -1026,6 +1021,7 @@ int radeon_vm_clear_freed(struct radeon_device *rdev,
         list_for_each_entry_safe(bo_va, tmp, &vm->freed, vm_status) {
                 r = radeon_vm_bo_update(rdev, bo_va, NULL);
                 radeon_bo_unref(&bo_va->bo);
+                radeon_fence_unref(&bo_va->last_pt_update);
                 kfree(bo_va);
                 if (r)
                         return r;
@@ -1084,6 +1080,7 @@ void radeon_vm_bo_rmv(struct radeon_device *rdev,
                 bo_va->bo = radeon_bo_ref(bo_va->bo);
                 list_add(&bo_va->vm_status, &vm->freed);
         } else {
+                radeon_fence_unref(&bo_va->last_pt_update);
                 kfree(bo_va);
         }
 
@@ -1130,8 +1127,6 @@ int radeon_vm_init(struct radeon_device *rdev, struct radeon_vm *vm)
         int i, r;
 
         vm->ib_bo_va = NULL;
-        vm->fence = NULL;
-
         for (i = 0; i < RADEON_NUM_RINGS; ++i) {
                 vm->ids[i].id = 0;
                 vm->ids[i].flushed_updates = NULL;
@@ -1192,11 +1187,13 @@ void radeon_vm_fini(struct radeon_device *rdev, struct radeon_vm *vm)
                 if (!r) {
                         list_del_init(&bo_va->bo_list);
                         radeon_bo_unreserve(bo_va->bo);
+                        radeon_fence_unref(&bo_va->last_pt_update);
                         kfree(bo_va);
                 }
         }
         list_for_each_entry_safe(bo_va, tmp, &vm->freed, vm_status) {
                 radeon_bo_unref(&bo_va->bo);
+                radeon_fence_unref(&bo_va->last_pt_update);
                 kfree(bo_va);
         }
 
@@ -1206,8 +1203,6 @@ void radeon_vm_fini(struct radeon_device *rdev, struct radeon_vm *vm)
 
         radeon_bo_unref(&vm->page_directory);
 
-        radeon_fence_unref(&vm->fence);
-
         for (i = 0; i < RADEON_NUM_RINGS; ++i) {
                 radeon_fence_unref(&vm->ids[i].flushed_updates);
                 radeon_fence_unref(&vm->ids[i].last_id_use);