[Intel-gfx] [RFC 1/4] drm/i915: Reduce recursive mutex locking from the shrinker

[Tvrtko Ursulin]

On 03/01/2019 12:29, Chris Wilson wrote:
> Quoting Tvrtko Ursulin (2019-01-03 12:23:26)
>> From: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
>>
>> In two codepaths internal to the shrinker we know we will end up taking
>> the resursive mutex path.
>>
>> It instead feels more elegant to avoid this altogether and not call
>> mutex_trylock_recursive in those cases.
>>
>> We achieve this by adding a new I915_SHRINK_LOCKED flag which gets passed
>> to i915_gem_shrink by the internal callers.
> 
> Since we reach here from many paths with struct_mutex either locked or
> unlocked, special casing one such seems silly. Especially when removing
> struct_mutex from here entirely is within our grasp.

Hm, for me it is silly that we rely on mutex_trylock_recursive to handle 
what the code clearly knows. Would you be happy with the 
__i915_gem_shrink which does no locking, and i915_gem_shrink wraps with 
locking, approach instead?

Regards,

Tvrtko