[Intel-gfx] [PATCH 2/3] drm: Only bump object-reference count when adding first handle

[Ville Syrjälä]

On Mon, Jan 04, 2016 at 10:11:00AM +0000, Chris Wilson wrote:
> We only need a single reference count for all handles (i.e. non-zero
> obj->handle_count) and so can trim a few atomic operations by only
> taking the reference on the first handle and dropping it after the last.
> 
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/drm_gem.c | 17 ++++++++++++-----
>  1 file changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/gpu/drm/drm_gem.c b/drivers/gpu/drm/drm_gem.c
> index a08176debc0e..ad955d7c99fd 100644
> --- a/drivers/gpu/drm/drm_gem.c
> +++ b/drivers/gpu/drm/drm_gem.c
> @@ -220,6 +220,9 @@ static void drm_gem_object_exported_dma_buf_free(struct drm_gem_object *obj)
>  static void
>  drm_gem_object_handle_unreference_unlocked(struct drm_gem_object *obj)
>  {
> +	struct drm_device *dev = obj->dev;
> +	bool final = false;
> +
>  	if (WARN_ON(obj->handle_count == 0))
>  		return;
>  
> @@ -229,14 +232,16 @@ drm_gem_object_handle_unreference_unlocked(struct drm_gem_object *obj)
>  	* checked for a name
>  	*/
>  
> -	mutex_lock(&obj->dev->object_name_lock);
> +	mutex_lock(&dev->object_name_lock);
>  	if (--obj->handle_count == 0) {
>  		drm_gem_object_handle_free(obj);
>  		drm_gem_object_exported_dma_buf_free(obj);
> +		final = true;
>  	}
> -	mutex_unlock(&obj->dev->object_name_lock);
> +	mutex_unlock(&dev->object_name_lock);
>  
> -	drm_gem_object_unreference_unlocked(obj);
> +	if (final)
> +		drm_gem_object_unreference_unlocked(obj);
>  }
>  
>  /**
> @@ -329,6 +334,8 @@ drm_gem_handle_create_tail(struct drm_file *file_priv,
>  	int ret;
>  
>  	WARN_ON(!mutex_is_locked(&dev->object_name_lock));
> +	if (obj->handle_count++ == 0)
> +		drm_gem_object_reference(obj);

Hmm. Yeah, handle_count is protected by object_name_lock, so this looks
better here rather than within the spinlocked section.

Patch looks sane:
Reviewed-by: Ville Syrjälä <ville.syrjala at linux.intel.com>

>  
>  	/*
>  	 * Get the user-visible handle using idr.  Preload and perform
> @@ -338,10 +345,10 @@ drm_gem_handle_create_tail(struct drm_file *file_priv,
>  	spin_lock(&file_priv->table_lock);
>  
>  	ret = idr_alloc(&file_priv->object_idr, obj, 1, 0, GFP_NOWAIT);
> -	drm_gem_object_reference(obj);
> -	obj->handle_count++;
> +
>  	spin_unlock(&file_priv->table_lock);
>  	idr_preload_end();
> +
>  	mutex_unlock(&dev->object_name_lock);
>  	if (ret < 0)
>  		goto err_unref;
> -- 
> 2.6.4
> 
> _______________________________________________
> dri-devel mailing list
> dri-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/dri-devel

-- 
Ville Syrjälä
Intel OTC