Memory corruption from suspend-to-disk with Intel graphics and KMS

amunkres · 2010-02-17 10:34:10

Apparently there is a known bug with the kernel's driver "i915.ko" affecting suspend-to-disk:

http://bugzilla.kernel.org/show_bug.cgi?id=13811

I've experienced problems similar to the ones reported there. (My video chipset is reported by xorg as a 945GM, and I'm running an x86_64 kernel.) Sometimes the system has locked up (was using uswsusp when this happened), another time (also using uswsusp) it would fail to start any new processes, and another time (this time using kernel swsusp) it caused filesystem corruption. (The directory /lib/udev vanished and fsck reported errors upon boot. When I ran fsck in repair mode, it put what appeared to be the contents of /lib/udev into /lost+found. I'm not certain that this bug is what caused the filesystem corruption, but it seems likely.)

This is really annoying since xf86-video-intel version 2.10.0 requires KMS; this means that those who have Intel graphics don't have working suspend-to-disk if they're using the current xf86-video-intel. For now, I'm going to try to go back to version 2.9.1.

Last edited by amunkres (2010-02-17 10:35:58)

Ramses de Norre · 2010-02-17 10:46:57

I have the same problem. It started around the upgrade of the kernel (2.6.32.7 -> 2.6.32.8) and the intel driver (2.9.1 -> 2.10.0). I tried downgrading the intel driver but that didn't help...

The strange thing is, the bug you mention was already there in older kernels but I have been using KMS for quite a while now and this bug only popped up this week.

And I have to agree, this is _very_ annoying! I hope the intel guys manage to fix this soon.

xduugu · 2010-02-17 12:19:27

Apparently, I'm also affected by this bug since some months (kernel suspend and uswsusp). However, I recently switched to tuxonice which works much better than the in-kernel hibernation ever did.

ccc1 · 2010-02-17 19:12:15

Ramses de Norre wrote:

The strange thing is, the bug you mention was already there in older kernels but I have been using KMS for quite a while now and this bug only popped up this week.

same here on a thinkpad x200 with an intel gma x4500 graphic card. for me the problem also started a few days ago. i had kms enabled since it was available and never had a problem with hibernate until now. has anyone reported this issue at https://bugs.freedesktop.org? see also:
http://intellinuxgraphics.org/how_to_report_bug.html & http://intellinuxgraphics.org/suspend-resume.html

ccc1

anrxc · 2010-02-17 20:20:56

I reported this bug some days ago to the Arch bug tracker. Uswsusp or direct kernel makes no difference, memory gets corrupted and process segfault is common. Everything pointed to KMS, and I'm glad to know for sure that indeed it is: http://bugs.archlinux.org/task/18334

Unfortunately in 2010 suspend is still like playing lottery with your data and hardware. You never know when it will break, and it does break in cycles of every N months. It simply can not be trusted so you can't hibernate, even in periods when it does work, because of it.

kokoko3k · 2010-02-17 20:33:15

I had this problem even before kms with 2.6.27 on a notebook:
00:02.1 Display controller: Intel Corporation Mobile 945GM/GMS/GME, 943/940GML Express Integrated Graphics Controller (rev 03)
..and xf86-video-intel 2.3.2-1

On a new netbook, eeepc1005ha, hibernation totally fails with kernel method and uswsusp, (screen goes black for a while and desktop is restored)
http://bbs.archlinux.org/viewtopic.php?id=90570
didn't tried tuxonice.

Last edited by kokoko3k (2010-02-17 20:34:15)

anrxc · 2010-02-17 20:55:09

hibernation totally fails with kernel method and uswsusp

This is not about hibernation failing. Hibernation is working, but the memory gets corrupted and system processes crash (and all kind of other side effects)

screen goes black for a while and desktop is restored

...as I was saying, I don't think this is the same problem. This doesn't have anything to do with blank desktop or whatever, desktop is restored (on those ocassions X11 doesn't segfault).

I hope others will not start dumping all kinds of hibernation failures here. Especially if "it happened even before KMS".

Last edited by anrxc (2010-02-17 21:00:55)

kokoko3k · 2010-02-18 08:54:54

anrcx wrote:

Especially if "it happened even before KMS".

Am i missing something or you say that the problem is so clearly related to kms, that the memory corruption issue i had in the past (and still have) with *no* kms can't be related to your?
Maybe you isolated the problem by hibernating the same system (i mean same hardware and same packages/software-version) multiple times without kms and no segfaults?

Probably you missed my point and/or it is not clear to me why you think that memory corruption problem comes for sure from kms:

anrcx wrote:

kokoko3k wrote:
screen goes black for a while and desktop is restored
...as I was saying, I don't think this is the same problem.

Of course not,
in other words, what i said is that from my experience, problems coming from hibernation which are kms related are simply... others.

Last edited by kokoko3k (2010-02-18 08:56:53)

amunkres · 2010-02-19 02:36:18

kokoko3k wrote:

Maybe you isolated the problem by hibernating the same system (i mean same hardware and same packages/software-version) multiple times without kms and no segfaults?

Yes, exactly. I haven't experienced this problem when KMS is disabled, and neither did the person who reported the bug at bugzilla.kernel.org:

Evgeni Golov wrote:

As soon I disable KMS, I can't reproduce this behavior anymore.

ccc1 · 2010-02-19 03:14:55

i just tried what was suggested on the Suspend/Resume Bug Report Guide from intel (http://intellinuxgraphics.org/suspend-resume.html):

S4 hibernation (a.k.a Suspend To Disk):
Check if hibernation under console mode has the same failure as in X mode.
1. boot the system into init level 3 (text mode).
2. run: echo disk > /sys/power/state
3. press the power button to see if the system can be resumed correctly.
4. If it can be resumed normally, the issue may be related to the graphics driver.
5. If it can't be resumed normally, the issue may be related to the Linux kernel.

i can resume fine from init level 3. a bug report should be filed at https://bugs.freedesktop.org following the guidelines given in:
http://intellinuxgraphics.org/how_to_report_bug.html and http://intellinuxgraphics.org/suspend-resume.html

ccc1

amunkres · 2010-02-19 06:19:15

ccc1 wrote:

i can resume fine from init level 3. a bug report should be filed at https://bugs.freedesktop.org following the guidelines given in:
http://intellinuxgraphics.org/how_to_report_bug.html and http://intellinuxgraphics.org/suspend-resume.html
ccc1

It already was, here:
https://bugs.freedesktop.org/show_bug.cgi?id=22886
Also, the user who reported it had the same problem when X wasn't running:

Evgeni Golov wrote:

Now I disabled X, rebooted, and just did a plain
echo disk > /sys/power/state
and got the very same segfaults after resume.

kokoko3k · 2010-02-19 08:28:40

amunkres wrote:

I haven't experienced this problem when KMS is disabled, and neither did the person who reported the bug at bugzilla.kernel.org:
Evgeni Golov wrote:
As soon I disable KMS, I can't reproduce this behavior anymore.

Understood, yesterday bugzilla.kernel.org was down for me, so i couldn't check.

Evgeni Golov 2009-07-22 01:06:35 PST wrote:

After some more testing, it looks more like a kernel issue here.

Still, the fact that memory corruption occours on some systems without kms points me to think that the real culprit could be not kms itself, don't you think? Of course, it could be a pure coincidence...

But if i ask myself what's the common thing to both system segfaulting (with and without kms), maybe the answer just lies in a framebuffer console.
Think about it, adding nomodeset to grub line, not only disables kms, but probably gives the user a text mode only console because he has already disabled the "vga=xxx" parameter in the past.
inferences, with some sense...

Last edited by kokoko3k (2010-02-19 08:33:32)

ccc1 · 2010-02-22 12:04:02

ccc1 wrote:

i can resume fine from init level 3.

my mistake. i tried a few times to resume from runlevel 3 and i get segmentation faults without X as well ...

bdjnk · 2010-02-22 18:30:10

I too have experienced this same issue of every new process dying immediately with a segmentation fault after resuming from hibernation.

I downgraded my Intel video driver from 2.9.1-1 to 2.8.1-1 and the problem is gone. I have changed nothing else, KMS is still fully enabled, the latest kernel 2.6.32.8-1, etc.

anrxc · 2010-02-24 19:27:57

I upgraded to 2.6.32.9 and just resumed from hibernation. The system is working so far, nothing crashed. Can anyone confirm?

Acheron · 2010-02-25 04:52:55

I upgraded the kernel to 2.6.32.9, but once I got resume to work in the first place (I had to add the resume hook to have the computer load the hibernated image instead of booting normally whereas this wasn't necessary in the past, though I doubt this is related in any way), the segmentation faults persist for me.

bdjnk · 2010-02-25 07:40:27

Acheron wrote:

I upgraded the kernel to 2.6.32.9, but once I got resume to work in the first place (I had to add the resume hook to have the computer load the hibernated image instead of booting normally whereas this wasn't necessary in the past, though I doubt this is related in any way), the segmentation faults persist for me.

This is almost my situation but I can't even tell if the segmentation faults persist because mouse and keyboard input fails.

Edit:
This is going to sound stark raving mad, but I went ahead and tried to hibernate and resume three more times. The first time X didn't finish coming back. The second time everything worked fine, including input, with no segfaults nor other oddities. The third time input failed again. I am at a loss.

Last edited by bdjnk (2010-02-25 08:17:09)

anrxc · 2010-02-27 15:58:28

Still broken with 2.6.33 in [testing].

steff · 2010-03-02 17:37:24

Maybe same problem here with newest kernel 2.6.32.9-1, but no difference downgrade to 26-2.6.32.7-1 .
It just do not resume and boots normal.

dmesg
PM: Checking image partition /dev/sda6
PM: Resume from disk failed.

@bdjnk: Where did you find intel-video-2.8.1-1? I cleaned my archives...

EDIT: solved with adding "resume" to hooks in mkinitcpio.conf

Last edited by steff (2010-03-02 20:45:55)

bdjnk · 2010-03-03 05:52:43

steff wrote:

@bdjnk: Where did you find intel-video-2.8.1-1? I cleaned my archives...

Yeah, I haven't cleaned out my package archive since installing Arch almost a year ago. I figure, why should I? I never know when I might need an older version of something. Modern computers have way too much disk space anyways.

I was going to advise you to check the Arch Rollback Machine as mentioned in the Downgrading Packages page of the Wiki, but then I did a search and found out they haven't got it. So I really don't know what to say other than, I still have it and could send it your way if you want it.

steff · 2010-03-03 08:03:41

Thank you a lot for the offer, but as you see the problem was not the driver. It' did not resume even in init3, and now it works at 100% as just before with the hook "resume". No idea why it got necessary...

bdjnk · 2010-03-10 00:50:44

Does anyone know if this issue is resolved in the new 2.6.34-rc1. I looked through the Changelog but didn't see anything decisive, I can't find anyone talking about it, and my only machine with this issue needs to remain stable(ish).

lyman · 2010-03-10 03:45:52

In my case, which is tuning backlight on GM965 will randomly freeze the system and make hibernation not stable, KMS is the reason.

The workaround is to disable KMS.

1, downgrade xf86-video-intel from 2.10.0 to 2.9.1 'cause 2.10.0 does not support non-KMS anymore.
Find your old package at http://arm.konnichi.com/, mine is http://arm.konnichi.com/2009/11/01/extr … pkg.tar.gz
2, disable KMS according to http://wiki.archlinux.org/index.php/Int … Setting.29

Now everything should be fine.

marten_kv · 2010-03-11 08:44:05

Problem gone for me when I deleted KMS stuff from initramfs (but kept kernel and xf86-video-intel as is from repo). My hardware is 945GME / Atom N280

Last edited by marten_kv (2010-03-11 11:56:26)

Kirurgs · 2010-03-12 08:19:30

Hi!

I would like to add that I have T500 with "Intel Corporation Mobile 4 Series Chipset Integrated Graphics Controller (rev 07)". I'm using KMS with TuxOnIce (for superior speed of hibernation) and I have not encountered a single fault. I have 2.6.32.9 x86_64 + TOI. I hibernate my system every day sometimes even twice a day and not a single problem.
Probably it's related to TOI which I doubt but couple of weeks ago I tried Fedora 12 x86_64 and I encountered FS corruption. I'm not sure what are they using for standart hibernation but resume from hibernation was kinda 50% success of all tries I made.

regards
Kirurgs

Arch Linux

#1 2010-02-17 10:34:10

Memory corruption from suspend-to-disk with Intel graphics and KMS

#2 2010-02-17 10:46:57

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#3 2010-02-17 12:19:27

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#4 2010-02-17 19:12:15

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#5 2010-02-17 20:20:56

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#6 2010-02-17 20:33:15

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#7 2010-02-17 20:55:09

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#8 2010-02-18 08:54:54

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#9 2010-02-19 02:36:18

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#10 2010-02-19 03:14:55

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#11 2010-02-19 06:19:15

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#12 2010-02-19 08:28:40

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#13 2010-02-22 12:04:02

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#14 2010-02-22 18:30:10

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#15 2010-02-24 19:27:57

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#16 2010-02-25 04:52:55

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#17 2010-02-25 07:40:27

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#18 2010-02-27 15:58:28

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#19 2010-03-02 17:37:24

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#20 2010-03-03 05:52:43

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#21 2010-03-03 08:03:41

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#22 2010-03-10 00:50:44

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#23 2010-03-10 03:45:52

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#24 2010-03-11 08:44:05

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

#25 2010-03-12 08:19:30

Re: Memory corruption from suspend-to-disk with Intel graphics and KMS

Board footer