[Intel-gfx] [PATCH 4/8] drm/i915: avoid waking up from PC8 on DP AUX operations

[Chris Wilson]

On Wed, Jul 31, 2013 at 11:31:42AM -0300, Paulo Zanoni wrote:
> 2013/7/30 Chris Wilson <chris at chris-wilson.co.uk>:
> > On Mon, Jul 29, 2013 at 05:48:23PM -0300, Paulo Zanoni wrote:
> >> From: Paulo Zanoni <paulo.r.zanoni at intel.com>
> >>
> >> If we're already allowing PC8, just don't use the IRQs, so we won't
> >> need to wake from PC8. Waking up from PC8 is a slow thing, so avoid it
> >> when we can.
> >
> > This raises the question of why we where holding the power well wake
> > lock if we weren't using IRQs in the first place.
> 
> I can't understand the sentence above. Also notice that the power well
> is kinda orthogonal to allowing/disallowing PC8. We need the power
> well disabled to allow PC8, but we won't allow PC8 whenever the power
> well is disabled.

This is the key sentence that is missing that explains why this patch is
relevant:

> And the GMBUS/DP AUX registers are on the PCH, so
> they don't get affected by the state of the power well.

That is we can use the gpio pins irrespective of whether either the package
state or display power well is enabled. So if the package is asleep, we
can use the slightly slower, higher CPU, method to communicate with the
display device rather than incur the overhead of powering up the package
to monitor IRQs.

I have not looked at what is covered by the power wells and by the
package cstate - so I am relying on the patch series to teach me and be
consistent with the story it tells. (You explain the rationale of the
patch in the changelog, and then I can check that the implementation
follows suit. And can then also go back and check the rationale against
the power management guides, etc.)
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre