[Intel-gfx] [PATCH 1/2] drm/i915/gem: Excise the per-batch whitelist from the context

[Chris Wilson]

Quoting Mika Kuoppala (2019-11-26 11:13:27)
> Chris Wilson <chris at chris-wilson.co.uk> writes:
> 
> > One does not lightly add a new hidden struct_mutex dependency deep within
> > the execbuf bowels! The immediate suspicion in seeing the whitelist
> > cached on the context, is that it is intended to be preserved between
> > batches, as the kernel is quite adept at caching small allocations
> > itself. But no, it's sole purpose is to serialise command submission in
> > order to save a kmalloc on a slow, slow path!
> 
> I needed n+1 cups of coffee for bouncing between this, code and
> the codebase. No harm done unless I start to spam irc in chattyness
> due to overdoze.
> 
> ...when the real tip-off is the patch subject!
> 
> Thanks for explaining it through in irc.
> 
> It is obvious now when one does get it but it wasn't
> before. So please add few lines to the commit message of what
> the excise gives us wrt to future improvements in here.
> 
> something like
> "By removing the whitelest dependancy from context our freedom
> to chop the big mutex is greatly augmented"

Ok.
 
> ..in proper english.

Hah, that was far more eloquent than my usual drivel.
 
> What are your thoughts on using nonatomic bitops?

Fine. Even a hint of concurrent use of the whitelist is a recipe for a
nightmare, so s/set_bit/__set_bit/ is ok.
-Chris