Quoting Daniele Ceraolo Spurio (2019-09-12 19:45:27)
> On 9/12/19 6:23 AM, Chris Wilson wrote:
> > We see failures where the context continues executing past a
> > preemption event, eventually leading to situations where a request has
> > executed before we have event submitted it to HW! It seems like tgl is
> AFAIK on TGL the CS can detect tail updates in the image of a running 
> context even if the lrc hasn't been resubmitted via the elsp. Could that 
> be related to the early execution of requests? I haven't looked at the 
> failure logs, so forgive me if I'm completely off-mark :)

We only update the CTX_RING_TAIL just prior to [re]submitting the
context. But fun and games ensue around preemption where we often have
to rewind the RING_TAIL. Still we should never push it behind
RING_HEAD (so long as our semaphore is solid), so it should never be
able to go past any of the previous RING_TAIL updates we made.

At the moment, I'm just focusing on trimming back the features until we
have a solid platform and then we will be better place to be able to look
at what is going wrong.
Intel-gfx mailing list

Reply via email to