[Intel-gfx] [PATCH 10/11] drm/i915: PSR Baytrail: Force exit by inactivating it.

[Rodrigo Vivi]

On Sat, Mar 1, 2014 at 5:45 AM, Chris Wilson <chris at chris-wilson.co.uk> wrote:
> On Fri, Feb 28, 2014 at 08:44:45PM -0300, Rodrigo Vivi wrote:
>> Baytrail cannot easily detect screen updates and force PSR exit.
>> So we inactivate it on {busy_ioctl, set_domain, sw_finish and mark_busy}
>> and update to enable it back it later with a delayed workqueue.
>
> Why are we not checking if the object being accessed is indeed being
> used for PSR? In set-domain, you only care about writes.

Ok, so here you are suggesting that I on trigger psr_exit if it is
write domain, right?
I agree.

> sw-finish and
> busy are too late for psr_exit, the damage has already been done and
> presumably the content may already be corrupted?

I also agree but it is better late than never... or than letting
userspace fully control the psr exit.
Most of the cases are coverred by psr_exit at set_domain.
The remaining cases where userspace set domain once, sleep for while
(like 10 seconds) and than write something
was coverred by my test that checked crc and it was different from crc
base already.

> Can you please explain
> that it is safe to do an psr_exit after the damage is already in the scanount
> based on the panel refresh cycle.

The perfect solution is the hardware tracking the changes and doing
psr_exit by itself,
but unfortunatelly se don't have this scenario so we can live with what we have.

If the idea is to allow userspace to let kernel know when to exit psr
I'm in favor of a more generic
ioctl called pre_damage to or a front_buffer_rend something to warn
when some damage will occur
and we can disable psr, fbc and do whatever we want before the damage

if you prefer I can remove this patch and add the patch with ioclts
that let psr_exit full control to userspace.
These patches are ready already. I just really don't like this option
because I don't like the idea to depend
on userspace to get this hardware feature working.

Any other suggestion or ideas or cases, that I might be missing or not
seeing, are welcome!

> -Chris
>
> --
> Chris Wilson, Intel Open Source Technology Centre

Thanks

-- 
Rodrigo Vivi
Blog: http://blog.vivi.eng.br