Regression 2.6.27 -> 2.6.28

[Michel Dänzer]

On Thu, 2009-01-29 at 20:04 +0100, Jörg-Volker Peetz wrote:
> Michel Dänzer wrote:
> > On Tue, 2009-01-27 at 21:38 +0100, Jörg-Volker Peetz wrote:
> >> Michel Dänzer wrote:
> >>> On Sun, 2009-01-25 at 20:03 +0100, Jörg-Volker Peetz wrote:
> >>>> with kernel 2.6.28 and 2.6.28.1 while in X with active DPMS screen off ("xset
> >>>> dpms force off") the CPU does not go into C3 state when idle.
> >>>> With 2.6.27.10 and 2.6.27.12 the CPU does fall into C3 state when idle as
> >>>> observed with powertop 1.11.
> >>> Was the DRI already enabled with the 2.6.27 kernels? If yes, you may
> >>> need to use git bisect to find the change that introduced the problem.
> >>>
> >> DRI was enabled both on 2.6.27 and 2.6.28.
> >>
> >> Today I also checked another hardware, a laptop with Pentium M CPU, Intel
> >> chipset ICH6, and ATI M22 [Mobility Radeon X300]. Similar behavior: while in X
> >> with active DPMS screen off ("xset dpms force off") there are nearly 60
> >> wakeups/s instead of 3 wakeups/s with DPMS disabled.
> >>
> >> I've not done a git bisect before. It will take me a while.
> > 
> > FWIW, I think it's most likely related to commit
> > 0a3e67a4caac273a3bfc4ced3da364830b1ab241 ('drm: Rework vblank-wait
> > handling to allow interrupt reduction.') and friends. Can you try
> > reverting that and seeing if the problem still happens?
> > 
> 
> You were right. I did the "git bisect good 2.6.27 bad 2.6.28" and the outcome is
> 
> 
> 0a3e67a4caac273a3bfc4ced3da364830b1ab241 is first bad commit

Thanks for taking the time to confirm this!


I think this may actually be an X driver issue though; apparently it's
telling the kernel that the CRTC is disabled, but the CRTC is obviously
still generating vertical blank interrupts.

> xserver-xorg-video-radeon    1:6.9.0-1+lenny4

Can you try if the problem persists with current xf86-video-ati Git or
at least the 6.10.0 release? If so, does disabling the
radeon_crtc_modeset_ioctl() calls in legacy_crtc_dpms() in
xf86-video-ati/src/legacy_crtc.c avoid the problem?

(Alex, do you know or can you find out under what circumstances exactly
the CRTC frame counter registers reset to 0?)


-- 
Earthling Michel Dänzer           |                http://www.vmware.com
Libre software enthusiast         |          Debian, X and DRI developer