[PATCH 12/87] drm/i915: Drain freed objects for mmap space exhaustion

[Tvrtko Ursulin]

On 05/01/2017 23:19, Chris Wilson wrote:
> As we now use a deferred free queue for objects, simply retiring the
> active objects is not enough to immediately free them and recover their
> mmap space - we must now also drain the freed object list.
>
> Fixes: fbbd37b36fa5 ("drm/i915: Move object release to a freelist + worker"
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
> Cc: <drm-intel-fixes at lists.freedesktop.org>
> ---
>  drivers/gpu/drm/i915/i915_gem.c | 24 +++++++++++-------------
>  1 file changed, 11 insertions(+), 13 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
> index c8bc2cd9d479..d2ca6f729f42 100644
> --- a/drivers/gpu/drm/i915/i915_gem.c
> +++ b/drivers/gpu/drm/i915/i915_gem.c
> @@ -2090,23 +2090,21 @@ static int i915_gem_object_create_mmap_offset(struct drm_i915_gem_object *obj)
>  	int err;
>
>  	err = drm_gem_create_mmap_offset(&obj->base);
> -	if (!err)
> +	if (likely(!err))
>  		return 0;
>
> -	/* We can idle the GPU locklessly to flush stale objects, but in order
> -	 * to claim that space for ourselves, we need to take the big
> -	 * struct_mutex to free the requests+objects and allocate our slot.
> -	 */
> -	err = i915_gem_wait_for_idle(dev_priv, I915_WAIT_INTERRUPTIBLE);
> -	if (err)
> -		return err;
> +	/* Attempt to reap some mmap space from dead objects */
> +	do {
> +		err = i915_gem_wait_for_idle(dev_priv, I915_WAIT_INTERRUPTIBLE);
> +		if (err)
> +			break;
>
> -	err = i915_mutex_lock_interruptible(&dev_priv->drm);
> -	if (!err) {
> -		i915_gem_retire_requests(dev_priv);
> +		i915_gem_drain_freed_objects(dev_priv);
>  		err = drm_gem_create_mmap_offset(&obj->base);
> -		mutex_unlock(&dev_priv->drm.struct_mutex);
> -	}
> +		if (!err)
> +			break;
> +
> +	} while (flush_delayed_work(&dev_priv->gt.retire_work));
>
>  	return err;
>  }
>

Hmm.. what about the shrinker and for example the patch I just reviewed 
- "drm/i915: Purge loose pages if we run out of DMA remap space" - we 
would need the same strategy there, no? Maybe move this logic to a 
helper like ..._retire_and_free ?

Regards,

Tvrtko