intel driver suspend/resume failure

Jesse Barnes jbarnes at virtuousgeek.org
Fri Jun 8 15:17:59 PDT 2007


On Friday, June 8, 2007 3:08:18 James Bottomley wrote:
> On Fri, 2007-06-08 at 16:32 -0500, James Bottomley wrote:
> > On Fri, 2007-06-08 at 13:29 -0700, Keith Packard wrote:
> > > On Fri, 2007-06-08 at 12:25 -0500, James Bottomley wrote:
> > > > The symptoms are that the X server always crashes on resume (log
> > > > attached).  I've tried the various quirk options to pm-suspend but
> > > > nothing seems to work.
> > >
> > > As usual, the BIOS leaves the chip in a random state on resume, and the
> > > driver isn't going through and carefully resetting everything.
> > >
> > > Can you try using 'vbetool post' between the resume and the switch back
> > > to X to see if that makes it work?
> >
> > Afraid not.  Using a manual version of this (suspending from a vt,
> > issuing the post and switching back to X) causes the X server to crash
> > as soon as it is switched to.
> >
> > Just to try a more cautious test, suspending with X not running, then
> > resuming, then doing vbetool post then starting X also crashes.
> >
> > Curiously enough, If I just start X and no clients at all, it comes up
> > with the X cursor and I can move the mouse.  As soon as I connect a
> > client like an Xterm, it crashes.
>
> Actually, I now have the script that allows me to suspend and resume
> with X: it's this (X is running on vt 7):
>
> [root at mulgrave ~]# chvt 12
> [root at mulgrave ~]# cat /proc/bus/pci/00/02.0 > /var/run/video.state.o
> [root at mulgrave ~]# cat /proc/bus/pci/00/02.1 > /var/run/video.state.1
> [root at mulgrave ~]# echo mem > /sys/power/state
> [root at mulgrave ~]# cat /var/run/video.state.1 > /proc/bus/pci/00/02.1
> [root at mulgrave ~]# cat /var/run/video.state.o > /proc/bus/pci/00/02.0
> [root at mulgrave ~]# chvt 7
>
> I'll update my previous scripts: they were only saving one function of
> the PCI state of the i915 instead of both functions.

Can you send copies of video.state.0 prior to suspend and after?  That way 
we'll know which registers are getting clobbered on resume...

Thanks,
Jesse



More information about the xorg mailing list