Problem with multi-head and compiz

[dura91]

Hi,

I have a multihead setup with an LCD screen and an AV receiver with HDMI audio enabled wich are connected to a GT240 card. If I start one X server with multihead enabled and compiz and then a second, the screen go off and VT-swithching stop working for a few tens of seconds. No problem with two multihead X-server without compiz or two X-server with compiz without multihead. 256.53 and 256.35 drivers are equally affected and there is no problem with 195.36.24 driver.

Note that if using kwin from kde 4.5.1 with composite effects activated instead of compiz the problem does not occur. The compiz version is 0.8.4.

See nvidia-bug-report.log.gz for details about the multihead setup.

Thanks

[dura91]

I've tested the new beta 260.19.04 and it got even worse: VT-switching has completely stopped working after X server with compiz and multihead was started.

I also tried the 195.36.31driver and it's got the problem too. So the regression was introduced between 195.36.24 and 195.36.31, it should be easier to find it.

[dura91]

Problem remain with latest beta 260.19.06.

I can see this in syslog which is correlated to nvidia module when I'm trying to VT-switch:

Sep 16 00:52:26 dura2 kernel: [   89.003482] NVRM: os_schedule: Attempted to yield the CPU while in atomic or interrupt context
Sep 16 00:52:38 dura2 kernel: [  101.101444] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:41 dura2 kernel: [  104.104026] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:44 dura2 kernel: [  107.137555] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:47 dura2 kernel: [  109.813009] NVRM: Xid (0001:00): 16, Head 00000000 Count 00000bef
Sep 16 00:52:47 dura2 kernel: [  110.139081] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:47 dura2 kernel: [  110.139151] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:47 dura2 kernel: [  110.142213] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:50 dura2 kernel: [  113.144263] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:53 dura2 kernel: [  116.185371] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:53 dura2 kernel: [  116.185447] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:54 dura2 kernel: [  117.009592] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:54 dura2 kernel: [  117.009647] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:54 dura2 kernel: [  117.109733] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:54 dura2 kernel: [  117.109792] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005
Sep 16 00:52:54 dura2 kernel: [  117.162940] NVRM: Xid (0001:00): 53, CMDre 00000000 00000080 00000000 00000005 00000005

I think that there was the same messages with the prvious beta (at least the first one).

Now, I'm switching back to 195.36.24...

[ajw1980]

I seem to have similar problems with VT switching and compiz with 256.53. The X server is slow to respond and is just kind of acting weird. This is also a gt240 card. For me, 195.36.31 seems to be ok, though.

I'll attach my bug report.

[dura91]

I've not tested the 260.19.12 just after its release because there was no indication that the problem would be solved, and I was right, the problem still there with the 260.19.12 release. Switching back again to the 195.36.24.

Nvidia guys, if you need more informations to resolve the issue, please ask!

[dura91]

I've tried the 260.19.21 release yesterday and the problem is now solved (after nine months and two major releases...). All things are working fine including VDPAU which seems to be problematic with previous releases.
However, there are two problems:
- Overclock in nvidia-settings does not work (was working in 195.36.24 at least for core). Already reported here: http://www.nvnews.net/vbulletin/showthread.php?t=156164
- MSI interrupt does not work with kernel 2.6.35.x with x > 4 (might be 5, 6 or 7, I don't remember) and 2.6.36. 195.36.24 have this problem too. Already reported here: http://www.nvnews.net/vbulletin/showthread.php?t=156231

Perhaps in the future, I will try to make a git bisect to isolate kernel commit that trigger the MSI problem. I hope NVidia devs already done it and could post the id of the first bad commit here or in the other thread.