[Intel-gfx] Breaking suspend/resume by the Pipe A quirk

[Daniel Vetter]

On Mon, Jun 02, 2014 at 12:41:36PM +0200, Thomas Richter wrote:
> Am 02.06.2014 10:27, schrieb Daniel Vetter:
> >
> >
> >Can you go right ahead and please submit this as a patch?
> 
> Certainly, but I would prefer to get more information on this. Even though
> the R31 *also* works without the pipe A quirk, I am not sure it does work on
> all other hardware configurations.
> 
> There is, however, an important difference between the R31 and the S6010:
> The R31 uses two independent display pipes for the generating the display,
> LVDS for the internal and VGA for the external display. As a result, frame
> rates and resolutions can be different between the two outputs.
> 
> The S6010, however, seems to use a single pipe design, with the internal
> display connected via DVI (not LVDS!) and the external by VGA. This has the
> unfortunate side effect that I cannot set the resolutions of internal and
> external display independently. Any attempt to modify the external
> resolution while using the internal screen results in an "no crtc found for
> output VGA1" when using xrandr. (Not quite sure what this means, but I
> believe that the VGA output is simply a duplicate of the DVI output, and the
> two are probably connected through a bios-switchable bridge chip).
> 
> Thus, I would *prefer* to be conservative and only disable the pipe_A quirk
> only in situations where there is a single display pipe (as in the S6010)
> and, just to be on the safe side, keep it enabled in dual-pipe (as in R31)
> configurations.

We've put a crtc restriction on VGA (it needs to be crtc 0) to work around
some issues. DVI/LVDS should work on crtc 1. You can set this with the
--crtc knob for xrandr.

> Now I wonder how I could possibly distinguish between the two. Could you
> please provide some pointers?

You're probably the last real user of this hw left. You're needs win,
especially if you know that it fixes stuff on other platforms, too.

So holesale removal of the pipe quirk for i830M seems like the right thing
to do here. Especially since Chris also complained that it makes stuff
worse for his i845.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch