[Intel-gfx] [PATCH 3/4] drm/i915: Flush all user surfaces prior to first use

[Chris Wilson]

Quoting Jason Ekstrand (2019-07-19 23:55:03)
> Just to be clear, is this just adding a CLFLUSH or is it actually changing the
> default caching state of buffers from CACHED to NONE?  If it's actually
> changing the default state, that's going to break userspace badly.

What userspace is actually broken? You don't use set-domain, preferring
to do your own cache management and MOCS; where you do default to PTE is
where we expect the buffers to be uncached anyway. In effect, it is only
swap in of the objects that the kernel will insert an extra clflush for
you.

I'm looking for the negative impact of this, and not finding much
repercussion at all. At the end of the day, we have no choice but to
close such a hole; the only question is how much leeway to give and how
to mitigate such issues in future interfaces.
-Chris