FS#19971 - [xf86-video-ati] KDE session will be reset shortly after login; kernel: [drm] Resetting GPU

Attached to Project: Arch Linux
Opened by Kay Abendroth (kay.abendroth) - Friday, 25 June 2010, 11:30 GMT
Last edited by Andreas Radke (AndyRTR) - Friday, 20 August 2010, 10:29 GMT
Task Type Bug Report
Category Packages: Extra
Status Closed
Assigned To Jan de Groot (JGC)
Andreas Radke (AndyRTR)
Architecture All
Severity High
Priority Normal
Reported Version
Due in Version Undecided
Due Date Undecided
Percent Complete 100%
Votes 7
Private No

Details

Description: After upgrading xorg-server from version 1.7.6 to version 1.8.1.902 I experienced the following problem. Shortly after logging into a KDE session, the session got "reset" and I was looking at the KDE login screen again.

/var/log/messages.log says:

Jun 25 13:07:40 kay kernel: mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Jun 25 13:07:40 kay kernel: mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Jun 25 13:07:40 kay kernel: mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Jun 25 13:07:40 kay kernel: mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Jun 25 13:07:40 kay kernel: mtrr: type mismatch for d0000000,10000000 old: write-back new: write-combining
Jun 25 13:07:40 kay kernel: [drm] Setting GART location based on new memory map
Jun 25 13:07:40 kay kernel: [drm] Loading RV610 CP Microcode
Jun 25 13:07:40 kay kernel: platform r600_cp.0: firmware: requesting radeon/RV610_pfp.bin
Jun 25 13:07:40 kay kernel: platform r600_cp.0: firmware: requesting radeon/RV610_me.bin
Jun 25 13:07:40 kay kernel: [drm] Resetting GPU
Jun 25 13:07:40 kay kernel: [drm] writeback test succeeded in 1 usecs


Additional info:
* xorg-server-1.7.6-3-x86_64


Steps to reproduce:
1. Upgrade xorg-server to version 1.8.1.902
2. Login to KDE session.
3. Do something and after a few minutes you will see the login screen again.
   xorg.conf (1.6 KiB)
This task depends upon

Closed by  Andreas Radke (AndyRTR)
Friday, 20 August 2010, 10:29 GMT
Reason for closing:  Fixed
Additional comments about closing:  problem doesn't occur with KMS enabled
Comment by Kay Abendroth (kay.abendroth) - Friday, 25 June 2010, 11:32 GMT
Additional info above should be: xorg-server-1.8.1.902-1-x86_64 ! My mistake.

I reinstalled xorg-server-1.7.6-3-x86_64 to get rid of the problem.
Comment by Torsten Flammiger (moneo) - Saturday, 26 June 2010, 05:56 GMT
I had the same problem here on an x86. To reproduce try minimize and restore a window.
If X crashes see my post here: http://bugs.archlinux.org/task/19661
Comment by Lionel Poiblanc (Laserpithiumm) - Sunday, 27 June 2010, 07:18 GMT
Same issue here, but doesn't look for me related with time after loging but to KDE activity.
For me crash happens only when I try to restore a window after having minimized it.
Problem in UMS mode, but everything is fine in KMS mode (using ati oss driver from extra repo).
x86_64 system, up to date on core and extra. Testing not used.
ATI HD4650 Mobility (rv730). Not using any xorg.conf
Comment by Chris Roche (sylentdogg) - Sunday, 27 June 2010, 19:39 GMT
I too am having a similar problem. I have an ati x1550 rv516 and I am using the xf86-video-ati 6.13.0-1 driver and xorg-server 1.8.1.902-1, as well as KDE 4.4. If I run with KMS, X does not restart but my desktop is extremely slow. And without KMS certain things like making a video full screen will cause X to restart. These all happen when compositing is enabled. When compositing is disabled, KMS is a little faster but not by much. And without KMS, it does not restart X as far as I have noticed. I amost forgot, I am using an x86 machine. I have also tried it with and without xorg.conf and the same things happen whether or not there is an xorg.conf file. I was thinking the KMS thing might be related to this, but it may not. Hopefully some of this info will be of use.
Comment by Kay Abendroth (kay.abendroth) - Monday, 28 June 2010, 09:34 GMT
My configuration (that also lead to the error):
- KMS is disabled.
- Card: ATI Technologies Inc RV610 [Radeon HD 2400 XT]
- Driver: xf86-video-ati
Comment by xCrucialDudex (xCrucialDudex) - Monday, 28 June 2010, 12:55 GMT
Setup:
- Samsung R-20 notebook with ATI Radeon XPress 1250 IGP
- kernel26 2.6.34-2
- xorg-server 1.8.1.902-1
- xf86-video-ati 6.13.0-1
- kde****-4.4.4-x

- Compositing is ON
- KMS is turned OFF

Steps to reproduce:
Minimizing a window in KDE results in X crashing.

Reproducible:
Almost always. Sometimes it won't crash but often times it does rather than not.

Details:
If KMS is turned ON heavy3D performance degradation is observed (glxgears yields ~80FPS against usual >200FPS) with random 1-2 seconds complete system freezes (input devices stop responding). Overall system performance is stable though.

If Compositing is turned OFF (via KDE's System Settings) no degradation of system performance can be seen.
Comment by Eric (eric2) - Thursday, 08 July 2010, 18:23 GMT
The latest xorg-server that works without this issue is 1.8.0.
I've compiled different versions of xorg-server myself and found out that all versions after 1.8.0 has this issue.

My current (working) setup is:
- xorg-server: 1.8.0 (self compiled);
- xf86-video-ati: 6.13.0 (self compiled, but Archlinux' version probably also works).
The rest (libdrm, dri2proto, mesa, libgl, ati-dri, xf86-input-* and other xf86-video-*) are the same as the current version from Archlinux.
Comment by Warren Huard (Warren) - Friday, 09 July 2010, 00:49 GMT
Same problem...
- KMS is disabled.
- Compositing is on.
- xorg-server 1.8.1.902-1
- xf86-video-ati 6.13.0-1
- kdelibs 4.4.5-1
- Thinkpad T400 with ATI Mobility Radeon HD 3400
- kernel 2.6.34.1-1 (x86_64)

There is no problem when I turn compositing off or switch to Openbox. I can't try it with KMS enabled for other reasons.
Comment by Brad Fanella (cesura) - Tuesday, 03 August 2010, 04:36 GMT
Because I have KMS issues with 2.6.34, I booted using 2.6.32-lts with KMS enabled and this seemed to solve the issue, at least temporarily. Using an older version of xf86-video-ati seemed to work, but there were a couple of problems so I simply downgraded my kernel.
Comment by Andreas Radke (AndyRTR) - Wednesday, 04 August 2010, 22:16 GMT
Please everybody check where the problem is. Check various kernels (.32LTS, .34 and .35) for different KMS/drm modules. KMS is the only supported way from .34 on. That should be working or we need to tell this upstream. Always check dmesg for the drm messages. Once the kmd/drm radeon module is working well you can look out for issues/bugs in the ddx driver.
Comment by Andreas Radke (AndyRTR) - Saturday, 14 August 2010, 22:03 GMT
state?
Comment by Brad Fanella (cesura) - Sunday, 15 August 2010, 03:01 GMT
Doesn't seem to happen in 2.6.34.3-1. The screen occasionally goes blank for a couple of seconds and returns to normal, but I believe that is a separate issue.
Comment by Panos (nameerf) - Tuesday, 17 August 2010, 18:17 GMT
Same issue here as well.
It happens when minimizing and restoring a window and also when restoring a video from full screen.
This happens when I'm using the latest packages from extra (xorg-server 1.8.1.902, xf86-video-ati 6.13.1...).
So I'm stuck with xorg-server 1.7.6 and its relevant dependencies.
I have an x86 system with an ATI Radeon X1950XT card with compositing ON and KMS OFF.

PS: I also tried turning KMS ON and the problem indeed disappears BUT perfomance(3D mostly) is way downgraded!
Comment by Brad Fanella (cesura) - Tuesday, 17 August 2010, 19:08 GMT
Could you please post your dmesg output from right after the problem occurs?
Comment by Panos (nameerf) - Tuesday, 17 August 2010, 20:40 GMT
Here are the last lines of dmesg just right after the crash occurs:

[drm] Num pipes: 4
agpgart-intel 0000:00:00.0: AGP 3.0 bridge
agpgart-intel 0000:00:00.0: putting AGP V3 device into 4x mode
pci 0000:01:00.0: putting AGP V3 device into 4x mode
[drm] Setting GART location based on new memory map
[drm] Loading R500 Microcode
platform radeon_cp.0: firmware: requesting radeon/R520_cp.bin
[drm] Num pipes: 4
[drm] writeback test succeeded in 2 usecs
ata1.00: configured for UDMA/100
ata1.01: configured for UDMA/33
ata1: EH complete
ata3.00: configured for UDMA/133
ata3: EH complete
ata4.00: configured for UDMA/133
ata4: EH complete
Comment by Andreas Radke (AndyRTR) - Wednesday, 18 August 2010, 04:49 GMT
KMS is the recommended upstream way. Please use it even if it slows down 3D stuff a bit. That can be improved later. KMS is the only supported way for a while now!

So is there any wrong in KMS mode expect speed?
Comment by Panos (nameerf) - Wednesday, 18 August 2010, 10:54 GMT
As I said in my earlier post I currently use xorg-server 1.7.6 with KMS OFF.
I haven't used my system with KMS ON for too long because of the low performance.
As far as I remember when I turned KMS ON, KDE turned compositing off because of the low performance.
On the other hand, I could test my system with KMS ON for a while to see if any other problem occurs (except the slowness in speed) if this would be any helpful.

I hope however that in the coming kernels the perfomance of KMS improves a lot because now its unbearable..
Comment by Andreas Radke (AndyRTR) - Wednesday, 18 August 2010, 16:22 GMT
If nobody can confirm the bug still being present in .35 kernel in testing in KMS mode I will close this bug. UMS won't receive bugs fixes and is no more the recommended way to use your card. If you prefer it for other reasons you won't get support.
Comment by Kay Abendroth (kay.abendroth) - Friday, 20 August 2010, 10:18 GMT
With KMS enabled, kernel26 2.6.35.2-1 and xorg-server 1.8.1.902-1 the problem does not occur.

Upgrading KDEmod from 4.4 to 4.5 however results in KDM not starting because the user kdm won't be created. But that's a different issue.

Loading...