Quoting Chris Wilson (2018-02-26 14:38:20)
> Quoting Chris Wilson (2018-02-26 14:28:36)
> > Quoting Michał Winiarski (2018-02-26 13:59:59)
> > > + for_each_engine(engine, dev_priv, id) {
> > struct intel_engine *ce = &client->owner->engine[id];
> >
> > /* The preempt context must be pinned on each engine);
> > GEM_BUG_ON(!ce->pin_count);
> >
> > /*
> > * We rely on the context image *not* being saved after
> > * preemption. This ensures that the RING_HEAD / RING_TAIL
> > * do not advance and they remain pointing at this command
> > * sequence forever.
> > */
>
> Hmm, this is not true. See intel_lr_context_resume().
Continuing this chain of thought, that doesn't matter. The reg state is
reset to 0, which is what we expect with context-save-inhibit anyway.
What it does do is reset ring->tail to 0 as well. That doesn't play well
with my idea to use ring->tail.
-Chris
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
