Compiz crash with recent nvidia drivers

Hi,

Since approximately driver version 337.25 compiz crashes when playing video with mplayer and repeatedly using alt+tab or toggling full screen mode. My multi screen is somewhat special: I want to have compiz and all 3D stuff on the first screen and a second screen fully independent from the first one for HDMI audio and HD video. This goal is achieve with my customized xorg.conf (see attachement).

With the latest driver (349.12), I’ve got these messages in the kernel log :

NVRM: Xid (PCI:0000:03:00): 31, Ch 00000005, engmask 00000101, intr 10000000
compiz[884]: segfault at fffff8 ip 00007feb4d89b944 sp 00007fffcbf42a08 error 4 in libc-2.20.so[7feb4d81f000+199000

Xorg does not crash and compiz can be restarted (and the bug reproduced again)

With driver 337.25 it’s even worse. I got:

NVRM: Xid (0000:03:00): 31, Ch 00000003, engmask 00000101, intr 10000000
NVRM: os_pci_init_handle: invalid context!
compiz[9003]: segfault at 0 ip 00007fcb56265049 sp 00007ffea621bc40 error 6 in libnvidia-glcore.so.337.25[7fcb54a63000+1e10000

I can’t test 337.19 because of Xorg ABI incompatibility. But I’m pretty sure that it was working and at the same 337.25 was not.

Furthermore there is some valgrind complaint when compiz is launched with it and the bug is reproduced, none when using a driver which does not exhibit the bug. Since the problem is not reproduced with old nvidia drivers and nouveau driver, I suppose that the culprit is recent nvidia drivers, but it might be a compiz bug. Note that the bug is reproduced with the same xorg.conf when xorg is started single head (HDMI is unplugged).

My configuration:
GTX 570 with :
DVI =⇒ standard monitor (compiz and 3D stuff) mplayer is launched on this screen when the bug is triggered
HDMI =⇒ AV receiver

I’m now stuck with 331.113 due to kernel/Xorg incompatibilities with 337.x series.

Please find attached: nvidia-bug-report.log.gz for 331.113 driver and 349.12 driver after the bug. For xorg.conf, please look into nvidia-bug-report.log.gz

If you need more informations, please contact me. If relevant I can provide valgrind logs.
Best Regards
nvidia-bug-report-349.12.log.gz (115 KB)
nvidia-bug-report-331.113.log.gz (177 KB)

Experiencing the exact same issue on my Laptop.
Is any bug with Linux Kernel?

The bug stills there with drivers version 352.09 and a GTX750ti

Until this is fixed, temporary workaround might be going through Compiz settings and disabling every Mipmap option you can find, eg. I use Shift Switcher plugin fot alt-tabbing etc. and disabling Mipmaps option under Appearance in its settings fixed crashing for me. Same for Thumbnail Window Previews, disabling Generate Mipmaps option fixed crashing when moving mouse pointer over some window button in taskbar.

And again, no improvement with latest driver 352.21.

No change with driver 352.30. Nvidia guys, are you looking at the problem? Did you try to reproduce it?
Even if the bug is a compiz bug, there should not be xid error in nvidia driver. Am I wrong?
If you need more infos, please tell me!

I had a little hope when I saw the 355.11 release changelog that one of the free bugs corrected might be mine. But no, the problem is still there.

And again the problem stills there with 358.09, but this time there is a little change: No compiz crah but the X server hang immediately. The computer can been reach though network, but the only way to make it usable again is a hard reset.
dmesg after the problem occurs show this:

[   95.211903] NVRM: GPU at PCI:0000:04:00: GPU-e7f43557-15c8-9535-9850-fc91f8ad3c8e
[   95.211907] NVRM: Xid (PCI:0000:04:00): 31, Ch 00000010, engmask 00000101, intr 10000000
[  203.272145] nvidia-modeset: WARNING: GPU:0: Lost display notification; continuing.
[  205.283843] nvidia-modeset: ERROR: GPU:0: Idling EVO timed out: 0x0000947d:0:0:0x00000020
[  207.494338] nvidia-modeset: ERROR: GPU:0: Idling EVO timed out: 0x0000947d:0:0:0x00000040

Again, when I saw the 361.16 changelog I had a little hope that the problem might be resolved, but no it is still there and is reproducible as easily as with previous driver versions.

GTX 970, Arch Linux, VLC, any currently available driver as of 2016, may 24.

NVRM: Xid (PCI:0000:01:00): 31, Ch 00000041, engmask 00008100, intr 10000000

Happens whenever VLC goes to next title on the DVD, completely locking up the UI for minutes and sometimes hours before VLC goes to next track.

It seems the GPU state is somehow corrupted after it happens and can take a complete poweroff (not just a reset) to get it to work well again without random lockups.

Workaround in my OS:

rm /usr/lib/libvdpau.so*

rm -r /usr/lib/vdpau

/Allan