[Intel-gfx] [PATCH] drm/i915: Add private api for power well usage -- alignment between graphic team and audio team

[Wang, Xingchao]

Hi Takashi,


> -----Original Message-----
> From: Takashi Iwai [mailto:tiwai at suse.de]
> Sent: Friday, April 26, 2013 11:13 PM
> To: Daniel Vetter
> Cc: Li, Jocelyn; Daniel Vetter; Wang, Xingchao; Zanoni, Paulo R;
> ville.syrjala at linux.intel.com; Lin, Mengdong; Girdwood, Liam R;
> intel-gfx at lists.freedesktop.org; alsa-devel at alsa-project.org; Wang Xingchao;
> Barnes, Jesse; Wysocki, Rafael J; Hindman, Gavin
> Subject: Re: [PATCH] drm/i915: Add private api for power well usage --
> alignment between graphic team and audio team
> 
> At Fri, 26 Apr 2013 16:57:08 +0200,
> Daniel Vetter wrote:
> >
> > On Fri, Apr 26, 2013 at 07:53:41AM +0000, Li, Jocelyn wrote:
> > > -----Original Message-----
> > > From: Daniel Vetter [mailto:daniel.vetter at ffwll.ch]
> > > Sent: Friday, April 26, 2013 3:25 PM
> > > To: Li, Jocelyn
> > > Cc: Wang, Xingchao; Zanoni, Paulo R; ville.syrjala at linux.intel.com;
> > > Lin, Mengdong; Girdwood, Liam R; intel-gfx at lists.freedesktop.org;
> > > alsa-devel at alsa-project.org; Wang Xingchao; Takashi Iwai; Barnes,
> > > Jesse; Wysocki, Rafael J; Hindman, Gavin
> > > Subject: Re: [PATCH] drm/i915: Add private api for power well usage
> > > -- alignment between graphic team and audio team Once we've figured
> > > out what needs to be changed in the audio driver we can look at the entire
> patch series and the interface to i915.ko.
> > > That's also why I didn't comment on Xingchao's patch right away, but
> > > only once he specifically asked for feedback, since doing a real
> > > review of the interface doesn't make sense until we have all the
> > > pieces (and a working audio driver).
> > >
> > > [Jocelyn] I think you have made constructive comments on Xingchao's
> > > patch yesterday. Next, shall we have Xingchao improve his patch? Or
> > > we just have Xingchao wait till you have completed your pieces.
> > > Sorry, I am a little confused :)
> >
> > I think the next step is to use Xinchao's patch as-is and get the
> > audio side going. Once we have that fixed up, we can revisit the
> > interface and make it solid. But for now trying to polish this
> > relatively simple part seems like wasted time.
> >
> > Also, reviewing an interface is much easier if we have both the i915
> > pieces (already here) and the audio pieces (which I haven't seen yet)
> > avaialble.
> 
> I haven't checked the patch properly yet, but the patch pasted in the post looks
> like i915 driver exports the functions to control power well, and let audio driver
> calling them.  If so, the big mess in such a case is the dependency between
> driver modules.
> 
> A simple workaround would be to split this function and the relevant instance
> out of i915 and snd-hda-intel and put into an individual module (e.g.
> i915-powerwell-ctl).

I agree this a single independent module could solve the dependency issue. I tested the patch
without i915 module loaded, there's symbol error in audio drvier side. 

> 
> Also, it would be feasible to implement some PM governor over both graphics
> and audio, that is, it gives the proper serialization of power up for audio
> controller, for example.
> 

This could avoid more private APIs between gfx and audio driver as there're really some dependencies on that now.
Besides that could the governor help serialize the resume sequence for gfx and audio? We have some good cases
to start for the implementations now. i.e. audio driver resume too early while gfx doesnot enable relative transcoders 
which cause codec pin issues. If you have more guide on this, that would be appreciated.

Thanks
--xingchao