Bug 69982

Summary: [NV86] GPU+system crash with nouveau and kernel 3.12-rc3
Product: xorg Reporter: Mike C <mike.cloaked>
Component: Driver/nouveauAssignee: Nouveau Project <nouveau>
Status: RESOLVED INVALID QA Contact: Xorg Project Team <xorg-team>
Severity: major    
Priority: medium CC: rcoe
Version: unspecified   
Hardware: x86-64 (AMD64)   
OS: Linux (All)   
See Also: https://bugs.freedesktop.org/show_bug.cgi?id=69928
Whiteboard:
i915 platform: i915 features:
Attachments:
Description Flags
kernel log file edited for relevant lines with nouveau problems
none
errors.log edited for lines relevant to the crash
none
xorg log file at the point the system crashed
none
full kernel log file without editing down
none
full errors log file without editing down
none
kernel log with nouveau.config=NvMSI=0 at boot none

Description Mike C 2013-09-30 20:13:39 UTC
Created attachment 86871 [details]
kernel log file edited for relevant lines with nouveau problems

Boot linux kernel mainline 3.12-rc3 in arch linux. Graphics hardware is:

01:00.0 VGA compatible controller: NVIDIA Corporation G86GLM [Quadro FX 360M] (rev a1)

KDM greeter allows login to KDE.

Within a few seconds of logging in to KDE the DE freezes, and the mouse and all controls are frozen. It is not possible to get to a text console and reboot is necessary to unfreeze the system - using stable 3.11.2 kernel to continue working.

Nouveau packages installed are: 

$ sudo pacman -Ss nouveau
extra/nouveau-dri 9.2.0-2 [installed]
    Mesa drivers for Nouveau
extra/xf86-video-nouveau 1.0.9-1 (xorg-drivers xorg) [installed]
    Open Source 2D acceleration driver for nVidia cards

Attaching log file segments relevant to the crash as follows:
kernel-sep30-crash.log
error.sep30-crash.log
Xorg.0.log.sep30.log
Comment 1 Mike C 2013-09-30 20:14:50 UTC
Created attachment 86872 [details]
errors.log edited for lines relevant to the crash
Comment 2 Mike C 2013-09-30 20:15:25 UTC
Created attachment 86873 [details]
xorg log file at the point the system crashed
Comment 3 Ilia Mirkin 2013-09-30 20:17:34 UTC
Can you see if booting with nouveau.config=NvMSI=0 helps?

(In the future, unless you're on some super-mega-secret machine whose full configuration you can't reveal, editing logs down is undesirable.)
Comment 4 Mike C 2013-09-30 20:25:23 UTC
OK I will try adding the kernel line you suggest and I will attach the full version of the log files for you shortly.
Comment 5 Mike C 2013-09-30 20:26:30 UTC
Created attachment 86874 [details]
full kernel log file without editing down
Comment 6 Mike C 2013-09-30 20:27:07 UTC
Created attachment 86875 [details]
full errors log file without editing down
Comment 7 Mike C 2013-09-30 20:39:16 UTC
I added nouveau.config=NvMSI=0 to the kernel boot line and the system booted fine so far, KDM gave a normal offer of login, and KDE login worked fine. I tested the system for several minutes, running konsole, and the chrome browser and so far there is no crash at all. I will attach the new kernel log that is current for this configuration.
Comment 8 Mike C 2013-09-30 20:40:14 UTC
Created attachment 86876 [details]
kernel log with nouveau.config=NvMSI=0 at boot
Comment 9 Mike C 2013-09-30 20:47:05 UTC
Ilia's suggestion is spot on - and armed with that information I now see that there is a related bug at https://bugs.freedesktop.org/show_bug.cgi?id=69928
Comment 10 Mike C 2013-10-01 13:02:27 UTC
Having now also seen that there was a long running thread "drm/nouveau: use MSI interrupts" in the nouveau devel list, and that Ilia is a key developer, I would appreciate knowing if a patch to fix this issue will get into the upstream code? Thank you for any new info about this.
Comment 11 Mike C 2013-10-01 13:33:14 UTC
As a temporary workaround I made a file /etc/modprobe.d/nouveau.conf containing one line:

option nouveau config:NvMSI=0

Now booting the 3.12 rc3 kernel with this in place gives a working system until the nouveau driver has a fix.
Comment 12 Mike C 2013-10-08 09:19:18 UTC
I have now tested kernel 3.12 rc4 with my arch linux system which had updates to nouveau-dri to version 9.2.1-1, and to mesa-libgl version 9.2.1-1

The original problem of the system freezing within a minute or so of logging in to KDE has come back, despite using NvMSI=0 so the issue remains, but I don't know if this is due to the mesa/nouveau updates or to changes in the rc4 kernel.
Comment 13 Mike C 2013-10-14 18:09:34 UTC
With the release of 3.12 rc5 yesterday I can finally run 3.12 rc5 with stable operation and no freeze.  

I believe that it was this patch in the kernel release patchset in rc5 that resulted in a fix for the moment:

Ben Skeggs (1):
drm/nouveau/mc: disable msi support by default, it's busted in
tons of places

I will continue to test as further 3.12-rcX releases happen, so in the meantime it would be best to leave this bug not resolved and not closed, since whether the bug re-surfaces or not will depend on whether there are more problems if msi is set by default in the future and whether the associated code is buggy or not. Therefore it will need to wait until 3.12 reaches final release and then see if the kernel remains stable before closing this bug.
Comment 14 Tobias Klausmann 2015-01-17 00:27:34 UTC
Do the freezes still occur with a recent kernel (3.17, 3.18) and newer Mesa, Xserver while not specifying "nouveau.config=NvMSI=0"?
Comment 15 Ilia Mirkin 2015-10-22 07:10:13 UTC
(In reply to Mike C from comment #11)
> As a temporary workaround I made a file /etc/modprobe.d/nouveau.conf
> containing one line:
> 
> option nouveau config:NvMSI=0
> 
> Now booting the 3.12 rc3 kernel with this in place gives a working system
> until the nouveau driver has a fix.

MSI was enabled again in 3.13, hopefully in a more reliable fashion. As there hasn't been a response to a test request almost a year ago, closing as invalid.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.