[PATCH] drm/ttm: Always and only destroy bo->ttm_resv in ttm_bo_release_list

Chunming Zhou zhoucm1 at amd.com
Thu Nov 2 02:30:21 UTC 2017


I think so, it should be enough we free it in the last place.

We also find many issues for bo->resv and bo->ttm_resv when debugging 
PER-VM-BO feature, after ttm_bo_individualize_resv, we should set 
bo->resv = &bo->ttm_rev before adding to destroy list, and the root resv 
is preventing eviction and swapout. Roger will give a summary and send 
those draft patches.

For this patch, Reviewed-by: Chunming Zhou <david1.zhou at amd.com>


On 2017年11月02日 00:15, Michel Dänzer wrote:
> From: Michel Dänzer <michel.daenzer at amd.com>
>
> Fixes a use-after-free due to a race condition in
> ttm_bo_cleanup_refs_and_unlock, which allows one task to reserve a BO
> and destroy its ttm_resv while another task is waiting for it to signal
> in reservation_object_wait_timeout_rcu.
>
> Fixes: 0d2bd2ae045d "drm/ttm: fix memory leak while individualizing BOs"
> Signed-off-by: Michel Dänzer <michel.daenzer at amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c | 13 +++----------
>   1 file changed, 3 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index 379ec41d2c69..a19a0ebf32ac 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -150,8 +150,7 @@ static void ttm_bo_release_list(struct kref *list_kref)
>   	ttm_tt_destroy(bo->ttm);
>   	atomic_dec(&bo->glob->bo_count);
>   	dma_fence_put(bo->moving);
> -	if (bo->resv == &bo->ttm_resv)
> -		reservation_object_fini(&bo->ttm_resv);
> +	reservation_object_fini(&bo->ttm_resv);
>   	mutex_destroy(&bo->wu_mutex);
>   	if (bo->destroy)
>   		bo->destroy(bo);
> @@ -406,10 +405,8 @@ static int ttm_bo_individualize_resv(struct ttm_buffer_object *bo)
>   	BUG_ON(!reservation_object_trylock(&bo->ttm_resv));
>   
>   	r = reservation_object_copy_fences(&bo->ttm_resv, bo->resv);
> -	if (r) {
> +	if (r)
>   		reservation_object_unlock(&bo->ttm_resv);
> -		reservation_object_fini(&bo->ttm_resv);
> -	}
>   
>   	return r;
>   }
> @@ -457,10 +454,8 @@ static void ttm_bo_cleanup_refs_or_queue(struct ttm_buffer_object *bo)
>   		if (reservation_object_test_signaled_rcu(&bo->ttm_resv, true)) {
>   			ttm_bo_del_from_lru(bo);
>   			spin_unlock(&glob->lru_lock);
> -			if (bo->resv != &bo->ttm_resv) {
> +			if (bo->resv != &bo->ttm_resv)
>   				reservation_object_unlock(&bo->ttm_resv);
> -				reservation_object_fini(&bo->ttm_resv);
> -			}
>   
>   			ttm_bo_cleanup_memtype_use(bo);
>   			return;
> @@ -560,8 +555,6 @@ static int ttm_bo_cleanup_refs_and_unlock(struct ttm_buffer_object *bo,
>   	}
>   
>   	ttm_bo_del_from_lru(bo);
> -	if (!list_empty(&bo->ddestroy) && (bo->resv != &bo->ttm_resv))
> -		reservation_object_fini(&bo->ttm_resv);
>   	list_del_init(&bo->ddestroy);
>   	kref_put(&bo->list_kref, ttm_bo_ref_bug);
>   



More information about the amd-gfx mailing list