[PATCH 1/3] drm/ttm: use try_lock in ttm_bo_delayed_delete again

[Thomas Hellstrom]

This series,

Reviewed-by: Thomas Hellstrom <thellstrom at vmware.com>




On 12/15/2017 07:31 PM, Christian König wrote:
> We only need to wait for the contended lock when the reservation object is
> shared or when we want to remove everything. A trylock should be sufficient
> in all other cases.
>
> Signed-off-by: Christian König <christian.koenig at amd.com>
> ---
>   drivers/gpu/drm/ttm/ttm_bo.c | 13 +++++++++----
>   1 file changed, 9 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/gpu/drm/ttm/ttm_bo.c b/drivers/gpu/drm/ttm/ttm_bo.c
> index ba5b48617bba..122a964cac40 100644
> --- a/drivers/gpu/drm/ttm/ttm_bo.c
> +++ b/drivers/gpu/drm/ttm/ttm_bo.c
> @@ -586,12 +586,17 @@ static bool ttm_bo_delayed_delete(struct ttm_bo_device *bdev, bool remove_all)
>   				      ddestroy);
>   		kref_get(&bo->list_kref);
>   		list_move_tail(&bo->ddestroy, &removed);
> -		spin_unlock(&glob->lru_lock);
>   
> -		reservation_object_lock(bo->resv, NULL);
> +		if (remove_all || bo->resv != &bo->ttm_resv) {
> +			spin_unlock(&glob->lru_lock);
> +			reservation_object_lock(bo->resv, NULL);
>   
> -		spin_lock(&glob->lru_lock);
> -		ttm_bo_cleanup_refs(bo, false, !remove_all, true);
> +			spin_lock(&glob->lru_lock);
> +			ttm_bo_cleanup_refs(bo, false, !remove_all, true);
> +
> +		} else if (reservation_object_trylock(bo->resv)) {
> +			ttm_bo_cleanup_refs(bo, false, !remove_all, true);
> +		}
>   
>   		kref_put(&bo->list_kref, ttm_bo_release_list);
>   		spin_lock(&glob->lru_lock);