[Intel-gfx] [PATCH 18/39] drm/i915: Differentiate between ggtt->mutex and ppgtt->mutex

Chris Wilson chris at chris-wilson.co.uk
Thu Jan 3 13:55:50 UTC 2019


Quoting Tvrtko Ursulin (2019-01-03 13:43:42)
> 
> Mika's reply here reminded me that the patch looks familiar and indeed I 
> r-b-ed it back in September.
> 
> Can keep it after the blank line adjustment and ordering shuffles of course.
> 
> Not sure on the is_ggtt vs the subclass id - perhaps even it would make 
> more sense to set is_ggtt before the call to i915_address_space_init and 
> then inside derive the lock subclass from it. Although even that is not 
> fully elegant. Maybe is_ggtt gneneric flag to i915_address_space_init to 
> do both tasks?

I'd rather keep .is_ggtt distinct; that's a special flag regarding the
list ordering. The lock_subclass we many want to extend later, if we
have some fancy notion of nesting address_spaces or something. Or maybe
better explained, let's try to keep the notion of nesting locks explicit
:)
-Chris


More information about the Intel-gfx mailing list