Re: [PATCH v3 0/3] implement GPU reset API

[Daniel Almeida]

Hi Onur,

I went through this and Sami's downstream implementation as asked. I think we
can strike a middle ground here, because it looks like the two trees need
different things at the moment.

Upstream tyr today has no scheduler and no jobs. It seems like everywhere we
might want to issue resets can get a hold of &Device<Bound>: probe,
ScopedQueue, IRQ handlers and etc. For this, I think your 'bound design fits,
and I'd like v4 to keep it.

I think the problem starts when the job queue lands, because now we might find
ourselves in places that only have access to the DRM device data which has a
'static bound and thus a hard requirement on Devres, AFAIU. I think we can keep
that in tyr-dev for now and land on top of your code later, if we shape a few
things accordingly. I will leave comments on 3/3.

In any case, there are changes in !74 which I think are improvements regardless
of the 'bound vs devres question, and I also think v4 should come with a user.
How about adding the GLB_PING watchdog? It would exercise schedule(), give
ScopedQueue its first user and lets us test this on the board.

Lastly, can you use b4 to manage the dependencies for this series, if you
haven't yet? It would make it easier to apply.


-- Daniel