915gm/945gm/XAA lockup with gray blocks when switching video mode

[Peter Clifton]

On Sat, 2007-10-13 at 17:14 +0800, Keith Packard wrote:
> On Fri, 2007-10-12 at 05:58 -0700, Jesse Barnes wrote:
> 
> >  The latter might be due to the associated pipe being off, 
> > which would disable the registers, possibly causing a GPU crash when 
> > you wrote them (the fact that your crashes go away 100% points to this 
> > possibility).  
> 
> Yeah, the docs say we're not supposed to write the palette when the pipe
> is off.
> 
> I think turning the pipe on to save and restore the palette is a
> reasonable idea; it's not that hard as we don't need any plane connected
> to the pipe, just to have the clock running. That should just be a few
> register writes.

There is a user on the Ubuntu bugzilla who's still seeing a similar
crash, apparently with my patch which helps many other people who see
it. Its _probably_ not the same crash (i've not the HW to test it),
however rebooting without going through power off seems to make a
difference. (It doesn't crash after a reboot, but does if you power on
from scratch).

The logs are attachment:
http://launchpadlibrarian.net/9971789/Xorg.0.log.tar.gz
on bug:
https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-intel/+bug/127101

"""
Shutdown seems to work always, reboot mostly but switching to a console
crashes my system nearly every time (HP compaq nx5000 with 855GM
graphics).
Don't know if it is a coincidence - but everything seem to work (incl.
console switching) every time I did a successful reboot from within
GNOME (without powering off the machine)!?
"""

The only differences I note in the logs, is that dri fires up right
away, not needing to load the i915 kernel module in the log he's
provided after a reboot (which perhaps means its not the first X server
after the reboot?), and that the RENCLK_GATE_D1 register starts at '1'.
Either it survived the reboot, or again, the log is not from the first X
server after the boot.

Any ideas what parts of the HW restore routine might crash on a 855
aside from the Palette? (Assuming a similar failure mode for now).

Best wishes,

-- 
Peter Clifton

Electrical Engineering Division,
Engineering Department,
University of Cambridge,
9, JJ Thomson Avenue,
Cambridge
CB3 0FA

Tel: +44 (0)7729 980173 - (No signal in the lab!)