[Intel-gfx] [PATCH 4/4] drm/i915: skip useless vblank wait on Haswell audio sequence

[Daniel Vetter]

On Thu, Oct 10, 2013 at 8:33 PM, Paulo Zanoni <przanoni at gmail.com> wrote:
> 2013/9/22 Lin, Mengdong <mengdong.lin at intel.com>:
>> Hi Daniel and Paulo,
>>
>> I think we need a confirmation from HW owner whether vblank wait can be skipped in audio enabling and disabling. I'll ping HW owner and involve you.
>> The original code just follows b-spec but there is no explanation why.
>
> Hi, any news on this?
>
> Please notice that even though the spec says we need to wait for a
> vblank, this code runs on a place where the pipe is disabled, so
> there's no vblank to wait. Bspec also suggests Audio should only be
> enabled at the very end of the mode set sequence, so the "proper" fix
> would probably be to move the place where haswell_write_eld is called.
> While this is not really done, I think we could probably merge this
> patch because the 50ms delay here seems pointless.

I think updating the ELD is one of those things we can do any time, as
long as it's early enough (similar to setting new pipe timings). Maybe
we need to move the updating of the eld_valid bits (which is what will
generate the unsolictid interrupt event on the audio driver side
afaik). But we have a sparate bit (plus interrupt type on the audio
side) for "audio is now enabled" and I think that's the really
important part. And that one is enabled at the right spot in the
modeset sequence afaics.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch