Bug 15481 - Nouveau on nVidia Corporation G71 [GeForce 7950 GT]: works but lockup for 6 min at shutdown
Summary: Nouveau on nVidia Corporation G71 [GeForce 7950 GT]: works but lockup for 6 m...
Status: RESOLVED FIXED
Alias: None
Product: xorg
Classification: Unclassified
Component: Driver/nouveau (show other bugs)
Version: git
Hardware: x86-64 (AMD64) Linux (All)
: high major
Assignee: Nouveau Project
QA Contact: Xorg Project Team
URL:
Whiteboard:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-04-13 00:54 UTC by Eric Valette
Modified: 2011-12-05 09:56 UTC (History)
0 users

See Also:
i915 platform:
i915 features:


Attachments
glxinfo output (6.94 KB, application/octet-stream)
2008-04-13 00:54 UTC, Eric Valette
no flags Details
xorg.conf (3.08 KB, application/octet-stream)
2008-04-13 00:55 UTC, Eric Valette
no flags Details
Xorg.log (113.43 KB, application/octet-stream)
2008-04-13 00:56 UTC, Eric Valette
no flags Details
dmesg output for drm driver (25.73 KB, application/octet-stream)
2008-04-13 00:57 UTC, Eric Valette
no flags Details
trying to run compiz error message (790 bytes, application/octet-stream)
2008-04-13 01:11 UTC, Eric Valette
no flags Details
Updated glxinfo (8.95 KB, text/plain)
2009-05-18 14:44 UTC, Eric Valette
no flags Details
Updated Xorg log (106.59 KB, text/plain)
2009-05-18 14:46 UTC, Eric Valette
no flags Details
updated drm module message (769 bytes, text/plain)
2009-05-18 15:15 UTC, Eric Valette
no flags Details
dmesg containing eveidence of the hang and a strange "DMA Get Lockup" (7.70 KB, text/plain)
2009-05-21 13:53 UTC, Eric Valette
no flags Details

Description Eric Valette 2008-04-13 00:54:45 UTC
Created attachment 15870 [details]
glxinfo output

Yesterday, I've extracted everything related to nouveau from git repositories. That is:
       1) drm
       2) xf86-video-nouveau
       3) mesa (in a second step, see below)

Build everything, made the drm, and nouveau module auto load at startup, then changed my xorg.conf (attached).

Immediately, I did see huge 2D performance improvement, but noticed running glxgears that 3D was not there. Confirmed by the message in Xorg.log.0 about AIGLX. I then started to add

Section "ServerFlags"
        Option "AIGLX" "on"
EndSection

But when restarting X, the machine hang for quite a long time. It is not dead, as I can ping it or event ssh to it. But you do not have any keyboard input (even caps lock led).

I then decided to be adventurous and try the 3D. And managed to install it. I now have some basic 3D acceleration except that if I magnify Glxgears, it gets a sigsegv.

But the "not fatal" hang for 6 min at shutdown remains. The last message I see, before it really starts printing garbled init 0 message again is the "failed to idle channel 1 before destroy. Prepare for strangeness.
Comment 1 Eric Valette 2008-04-13 00:55:56 UTC
Created attachment 15871 [details]
xorg.conf
Comment 2 Eric Valette 2008-04-13 00:56:46 UTC
Created attachment 15872 [details]
Xorg.log
Comment 3 Eric Valette 2008-04-13 00:57:54 UTC
Created attachment 15873 [details]
dmesg output for drm driver
Comment 4 Eric Valette 2008-04-13 01:11:40 UTC
Created attachment 15874 [details]
trying to run compiz error message
Comment 5 Eric Valette 2008-04-27 03:41:06 UTC
The hang does happen even when I have no error message from drm driver.
Comment 6 Eric Valette 2008-11-05 12:37:17 UTC
I update drm and nouveau Xord driver from git yesterday and the roblem has vansihed.
Comment 7 Eric Valette 2009-05-18 14:39:55 UTC
The hang when shuting down is back and solid. The system hangs again for 5 min a shutdown.
Comment 8 Eric Valette 2009-05-18 14:44:22 UTC
Created attachment 25980 [details]
Updated glxinfo
Comment 9 Eric Valette 2009-05-18 14:46:18 UTC
Created attachment 25981 [details]
Updated Xorg log
Comment 10 Eric Valette 2009-05-18 15:15:59 UTC
Created attachment 25982 [details]
updated drm module message
Comment 11 Maarten Maathuis 2009-05-19 05:20:05 UTC
Tried ssh+gdb to figure what exactly it's blocking on? And perhaps modprobe drm debug=1 to get some more verbosity on that side. I'm away from my dev machine so i can't really help, just wanted to give some hints.
Comment 12 Eric Valette 2009-05-19 05:32:40 UTC
Not even sure I still have network. The keyboard is frozen even if I'm back to fb console. The last thing is see is a message about destroying fifo 0.

Even the caps lock led does not work. Will see if network is still up when the lock occur (hang is during shutdown but I guess it is before the network stop script).

Will try to see if I can still log into the machine and report.
Comment 13 Eric Valette 2009-05-21 13:51:49 UTC
I did a ssh session before ending a kde4 session. Ending the kde4 session itself worked. I been back to kdm prompt. Then I did login to tty console from kdm mennu and I got the hang. I have been able to do some ps and some command for a while (< 1mn) and then the ssh was stuck. Ping did continue to work.

Part of dmesg after hang is added. NB: the console get back after the 6 mins but the charcater displayed are totally garbled. If I restart kdm, then I got a normal screen.
Comment 14 Eric Valette 2009-05-21 13:53:27 UTC
Created attachment 26079 [details]
dmesg containing eveidence of the hang and a strange "DMA Get Lockup"

If only there was a real in kernle debugger...
Comment 15 Eric Valette 2009-05-21 14:14:14 UTC
./drivers/video/nvidia/nv_accel.c:137:                                         printk("DMA Get lockup\n");

 in function NVDmaWait()

So apparently nouveau and the in kernel nv frame buffer are doing incompatible things with video registers.

Comment 16 Ben Skeggs 2009-05-21 15:04:07 UTC
Disable nvidiafb and retry.  Nouveau and nvidiafb can *not* work together.
Comment 17 Eric Valette 2009-05-22 00:23:23 UTC
Why the hell? If I use nouveau I cannot use a framebuffer. What is supposed to happend when KMS will be generalized?

Will do the test but still...
Comment 18 Ben Skeggs 2009-05-22 00:31:22 UTC
You can't reliably have two drivers thinking they have direct control of the hw, nvidiafb will blindly destroy a lot of setup nouveau does to the card, it can't possibly work.  The binary driver will flat-out refuse to load if nvidiafb is running, nv only works by sheer accident (a combination of the very basic way it and nvidiafb use the hw, and because they use the card almost identically).

If everything works fine without nvidiafb, I'll close this as NOTABUG.  You can use vesafb if you really want a framebuffer console..  With KMS, nouveau provides the console, so there isn't an issue.
Comment 19 Eric Valette 2009-05-22 10:38:30 UTC
I say that because it basically works. I can switch from X11 to vt back and forth without problem. I only hangs when I exit X.

As soon as nouveau use KMS, and provide a framebuffer console, I would be happy.
Comment 20 Eric Valette 2009-05-23 05:14:23 UTC
Works without nvidia framebuffer. But I cannot stand having 80x25 tty on a 1600x1050 23 inch display...
Comment 21 Ben Skeggs 2009-05-23 06:21:02 UTC
There's vesafb if you really must have a framebuffer console.  Until KMS (probably soon, it'll be there once the nouveau kernel tree is sorted out, the code's basically written), that's your only option.
Comment 22 Jinghua Luo 2009-05-24 17:14:00 UTC
I get KMS working on my C51 and Go 7300. My work is available in my drm tree[1] at f.d.o.
It may be unstable then upstream master, if you're brave enough, you can try it:)

[1] git://anongit.freedesktop.org/~jinghua/drm modesetting-gem branch
Comment 23 Eric Valette 2009-05-27 01:06:15 UTC
The bug happened yesterday with vesa framebuffer. So we only chnaged the bug frequency using vesa.

Sorry for this annoying reopen...
Comment 24 Lucas Stach 2011-02-15 01:13:07 UTC
Is this still an issue with recent nouveau, or could we close this bug?
Comment 25 Marcin Slusarz 2011-12-05 09:56:28 UTC
Should work fine now.
Closing as RESOLVED FIXED.


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.