[Intel-gfx] [PATCH 03/20] drm/i915/gem: Don't drop the timeline lock during execbuf

[Chris Wilson]

Quoting Tvrtko Ursulin (2020-07-09 11:52:19)
> 
> On 08/07/2020 19:08, Chris Wilson wrote:
> > Quoting Tvrtko Ursulin (2020-07-08 17:54:51)
> >>
> >> On 06/07/2020 07:19, Chris Wilson wrote:
> >>> +static struct i915_request *
> >>> +nested_request_create(struct intel_context *ce)
> >>> +{
> >>> +     struct i915_request *rq;
> >>> +
> >>> +     /* XXX This only works once; replace with shared timeline */
> >>
> >> Once as in attempt to use the same local intel_context from another eb
> >> would upset lockdep? It's not a problem I think.
> > 
> > "Once" as in this is the only time we can do this nested locking between
> > engines of the same context in the whole driver, or else lockdep would
> > have been right to complain. [i.e. if we ever do the reserve nesting, we
> > are screwed.]
> > 
> > Fwiw, I have posted patches that will eliminate the need for a nested
> > timeline here :)
> 
> In this series or just on the mailing list?

It's the implement a ring-scheduler for gen6 series, which is currently
sitting at the end of the vdeadline scheduler series.
-Chris