drm/ttm: individualize BO reservation obj when they are freed

Use the BOs reservation object when it is put on the ddelete list. This way we
avoid delaying freeing up the BO because of new fences on the reservation object.

This is used by dma-buf and amdgpu's VM page tables.

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

diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
index 0f5c59f..d3463eb 100644 (file)
--- a/drivers/gpu/drm/ttm/ttm_bo.c
+++ b/drivers/gpu/drm/ttm/ttm_bo.c
@@ -394,14 +394,33 @@ static void ttm_bo_cleanup_memtype_use(struct ttm_buffer_object *bo)
        ww_mutex_unlock (&bo->resv->lock);
 }
 
+static int ttm_bo_individualize_resv(struct ttm_buffer_object *bo)
+{
+       int r;
+
+       if (bo->resv == &bo->ttm_resv)
+               return 0;
+
+       reservation_object_init(&bo->ttm_resv);
+       BUG_ON(reservation_object_lock(&bo->ttm_resv, NULL));
+
+       r = reservation_object_copy_fences(&bo->ttm_resv, bo->resv);
+       if (r) {
+               reservation_object_unlock(&bo->ttm_resv);
+               reservation_object_fini(&bo->ttm_resv);
+       }
+
+       return r;
+}
+
 static void ttm_bo_flush_all_fences(struct ttm_buffer_object *bo)
 {
        struct reservation_object_list *fobj;
        struct dma_fence *fence;
        int i;
 
-       fobj = reservation_object_get_list(bo->resv);
-       fence = reservation_object_get_excl(bo->resv);
+       fobj = reservation_object_get_list(&bo->ttm_resv);
+       fence = reservation_object_get_excl(&bo->ttm_resv);
        if (fence && !fence->ops->signaled)
                dma_fence_enable_sw_signaling(fence);
 
@@ -430,8 +449,19 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
                        ttm_bo_cleanup_memtype_use(bo);
 
                        return;
-               } else
-                       ttm_bo_flush_all_fences(bo);
+               }
+
+               ret = ttm_bo_individualize_resv(bo);
+               if (ret) {
+                       /* Last resort, if we fail to allocate memory for the
+                        * fences block for the BO to become idle and free it.
+                        */
+                       spin_unlock(&glob->lru_lock);
+                       ttm_bo_wait(bo, true, true);
+                       ttm_bo_cleanup_memtype_use(bo);
+                       return;
+               }
+               ttm_bo_flush_all_fences(bo);
 
                /*
                 * Make NO_EVICT bos immediately available to
@@ -443,6 +473,8 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
                        ttm_bo_add_to_lru(bo);
                }
 
+               if (bo->resv != &bo->ttm_resv)
+                       reservation_object_unlock(&bo->ttm_resv);
                __ttm_bo_unreserve(bo);
        }
 
@@ -471,17 +503,25 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
                                          bool no_wait_gpu)
 {
        struct ttm_bo_global *glob = bo->glob;
+       struct reservation_object *resv;
        int ret;
 
-       ret = ttm_bo_wait(bo, false, true);
+       if (unlikely(list_empty(&bo->ddestroy)))
+               resv = bo->resv;
+       else
+               resv = &bo->ttm_resv;
+
+       if (reservation_object_test_signaled_rcu(resv, true))
+               ret = 0;
+       else
+               ret = -EBUSY;
 
        if (ret && !no_wait_gpu) {
                long lret;
                ww_mutex_unlock(&bo->resv->lock);
                spin_unlock(&glob->lru_lock);
 
-               lret = reservation_object_wait_timeout_rcu(bo->resv,
-                                                          true,
+               lret = reservation_object_wait_timeout_rcu(resv, true,
                                                           interruptible,
                                                           30 * HZ);