Kernel 3.7.9 crashes (nouveau?)

[tomtom69]

Hi,

since the last kernel update from 3.6.11 to 3.7.9 I get frequent freezes which seem to be graphics related.
It happens for example when using googleearh, mplayer or some other applications which use graphics with more bandwidth. glxgears is also a candidate to cause the lockup.
When the kernel crashes I only get stripes on the screen. Sometimes the picture freezes for some seconds until the stripes appear. No blinking LEDs, no chance to change terminal, no ssh, only reset is possible.
Unfortunately also no output in dmesg before the crash.
Chipset is Nvidia GeForce 7025 / nForce 630a.
Reverting to kernel 3.6.11 makes the crashes disappear.
Is this a known issue?

Kernel config is here:
http://pastebin.com/cv53m85H

tom

[Maitreya]

For stripes to appear after a crash would seem like a memory issue on the videocard. This could be heat. Did any powersaving/fan options changed during the kernel upgrade or could you monitor temperatures?

[Chris W]

I've been seeing hard locks on NVidia hardware with 3.7.10 Gentoo sources (a MythTV box). Have not found the cause yet: they're intermittent and do not seem to have a consistent trigger. I'm rolling back to 3.6 and I'll see how it goes.
_________________
Cheers,
Chris W
"Common sense: The collection of prejudices acquired by age 18." -- Einstein

[tomtom69]

Hi,

The chipset is onboard and has no specific cooling or dedicated memory. A manual temperature check of the chipset does not show any "hot spots" so I assume temperature should not be an issue. The crashes appear reproducible at the start of a graphics intensive application (kernel 3.7), where I would not assume any temperature rise, but never with kernel 3.6.11. 3D accel or such things are not used - no gaming etc.
At the moment I hope that kernel 3.8 cures this - till then I will get stuck at 3.6.11, which works without any problems since getting stable.

tom

[TomWij]

There has been refactoring in Nouveau in the 3.7 branch, you will definitely want to avoid it and prefer an earlier or the latest kernel.

[ian.au]

Irritating, I keep away from nVidia cards in general for my Gentoo machines.

I have an old Toshiba Satellite Laptop P-20 Laptop using NV34M [GeForce FX Go5200 64M] which is totally broken under 3.7.10 (loses graphics altogether as soon as the module comes up.

Another machine here runs GT218 [GeForce 210] graphics and seems to be fine so far under 3.7.10.

I.

[TomWij]

[ian.au]

[TomWij]

[ian.au]

[tomtom69]

Update:
Problem stays with kernel 3.8.13
Looks like I need to drop nouveau and use the nvidia blob again.

[wcg]

Does enabling the MTRR sanitizer by default help with the nforce630a
chipset? I have an nforce430 machine with that enabled, and it works
on that machine (no dmesg complaints about chunk size, etc). The gpu
is Ge6150SE.

In your kernel .config, I noticed that you have

[wcg]

Actually I just started seeing this on the nforce430-Ge6150SE board with
4gb of ram and scads of swap when firefox-17.0.{5,6} starts, on kernel
3.7.10 (I know I tested this after the kernel was installed, and it worked
after re-emerging the xorg nouveau driver; apparently that was only luck)
and kernel 3.8.13. No problem with kernel 3.5.7, and no problem
with 3.7.10 on another box that has a 990fx chipset and a gt218 gpu
in a PCIe slot.

I ran memtest86, guessing it might be a memory problem (that's what firefox
does differently than most other processes that run in xfce4 in X,
allocate lots of memory when it starts), but the dimms tested with
0 errors (actually I suspected a dimm socket rather than the dimms
themselves, but neither produced any errors). Does not happen when
I start evince and load a .pdf, gimp, emacs, etc.

Hangs the kernel (cannot ssh in and kill the process, because the kernel
is no longer running, so the network is not responding). No messages
in /var/log/*, no xorg messages, etc.

I have a spare nvidia PCIe x16 card I can try in the nforce430 box and
see if it happens with a 3.7.10 or 3.8.x kernel. (The Ge6150SE onboard
video will reserve some of main memory for a video memory buffer;
a PCIe card will not, yet they will use the same video driver.)

But it could be other things than the video driver (something to do with chipset
setup in the BIOS). The system hangs before firefox ever displays the page.
_________________
TIA

[_______0]

upgrade card's bios, many seem to shipped with questionable stability.

Nouveau had fixed an important bug in 3.9, can't recall the details but might be worth trying. Bear in mind that nouvau lags behind in sophistication due to nvidias attitued. On the other hand radeon's are rock solid.

I experienced similar issues, search my posts. What I did is to purposely stress the card with many things until triggering the darn bug. It's fun because it's predictable. Without stressing it's all aight though.

let me know about 3.9.

[tomtom69]

Hi,

I tried the following:
(1) CONFIG_MTRR_SANITIZER_ENABLE_DEFAULT=1
(2) Disabling MTRR support completely
(3) increase the shared memory size from 64MB to 256MB
but unfortunately everything without success. The time to crash varies, but it was always less than some minutes. Reminds me to Win9x which caused me to move to linux a long time ago.

Hardware defect should not be the case because I can see the fault on 2 systems with identical chipset. And I have a system with a different chipset (GeForce 6150SE) but nearly same kernel .config which works without problems using nouveau.

I'll give kernel 3.9 a try as soon as it hits stable and report the results.

tom

[wcg]

Yes, I don't think it is really hardware, or at least not the Ge 6150SE video
hardware. That hardware works fine with nouveau on kernel 3.5.7. firefox-17.0.5
and firefox-17.0.6 load on that kernel and run with no problems. That gpu-driver
combination worked fine on 3.3.8, 3.2.12, 3.1.6, and so on.

It is a new bug (new for me with 3.7.x kernels, anyway), and it is intermittent.
It seems to depend on one's memory allocation pattern before the offending
process runs whether the kernel hangs or not (reference the time firefox
worked on 3.7.10 right after I re-emerged the xorg nouveau driver and
restarted xorg).

So, I guess one would have to step up kernel versions one version at a time
until seeing the hang, then git bisect it to find the actual kernel patch that
allows it to happen. (Unless one has more money than time and can just
replace the mb and/or gpu with products that don't trigger the hang.)

If it is fixed in kernel 3.9.x, that would be cool (someone else already found
it and fixed it).
_________________
TIA

[tomtom69]

GE6150SE works for me on one machine. What does not work is GE7025.
But I do not know how to get and apply all the patches from 3.6.11 to 3.7.10 in order to see when the problem apperared first.
For filing a bug I think some more information would be necessary than just "hangs with striped screen". However the crash is so "quick and heavy" that things like sysrq key or log files are all empty.

[wcg]

I would have said kernel.org to get the source to any kernel version,
but connecting to http://www.kernel.org/ seems to be broken for
this. One gets a front page with a filtered list instead of a directory
listing that one can navigate to a directory with base 3.x kernel source
trees and then a long list of 3.x.x patches.

(I am *not* reconfiguring my router to let in remotely initiated
connect attempts so that ftp works.)

However, connecting to https://www.kernel.org/pub/linux/kernel/v3.x/
gets one to the traditional directory listing of kernel source trees
and version patches.

edit:
These would not be managed by portage, of course, and I don't know
if genkernel will work with kernel.org source trees. Old school.
_________________
TIA

[TomWij]

Please file this at https://bugs.gentoo.org as we can't actively track the forums to fix things, thanks in advance.

[tomtom69]

OK. Bug 472200 Submitted. Hopefully with all information needed.