[Intel-gfx] Memory corruption on hibernate/thaw with KMS

[Bojan Smojver]

On Mon, 2011-10-10 at 18:15 +1100, Bojan Smojver wrote:
> Will try some more tests, just to confirm whether memory corruption
> thing is still present.

Yes, still present. I did this on my Fedora 15 system with 3.1.0-rc9+
(git pull as of half an hour ago):

echo -n reboot > /sys/power/disk
for (( i=0; i<15; i++)); do pm-hibernate; sleep 2; done

With nomodeset passed into to the kernel, the cycle finished
successfully. There was one program that segfaulted (modem-manager) in
one of the cycles only. I'm guessing this is probably a bug in the
program.

Without nomodeset (i.e. KMS), I got corruption and "unable to handle
paging request", followed by kernel hang on second thaw. It's pretty
much what happens when corruption occurs.

PS. Kernel 3.1.0-rc9+ was patched with my own hibernation patch that
calculates CRC32 of image pages on hibernate/thaw, so the chances that
what was read in was not what was hibernated were minimal. You can find
it here: http://marc.info/?l=linux-kernel&m=131820444524522&w=2. I'm
experiencing the same with Fedora supplied kernels (i.e.
2.6.40.6-0.fc15.x86_64, which is really 3.0.6).

-- 
Bojan