Bug 47519

Created attachment 58692 [details]
2nd trace

Created attachment 58693 [details]
Xorg.log

Please attach your dmesg output.

Some notes (but it may be fate or something, but it feels like it does have some influence):
Using this config - it does still hang from time to time (e.g. one or two times a day):

Section "Device"
        Identifier  "Card0"
	Option	"EnablePageFlip"        "off"  #supported on all R/RV/RS4xx and older hardware and set on by default
	Option	"ColorTiling"           "off"  #enabled by default on RV300 and later radeon cards.
	Option	"EXAPixmaps"            "off"  #when on icreases 2D performance, but may also cause artifacts on some old cards
	Option	"AccelDFS"              "off"  #default is on, read the radeon manpage for more information
EndSection

If i remove those config or set all to "on" it happens really frequently (enough to make me report this one here to seek help :-) ).

Created attachment 58694 [details]
dmesg

Is this a regression?  If so what components did you change and what was the last working version?

I suffer this one since i'll changed from lucid to oneiric, where gnome2 was removed and i did switch to gnome3.
Because it did freeze very often, i switched to lxde where is does not seem to happen - but i am back to gnome3 now to see if the error is gone (seems not ;-) ).

So the combination of gnome-shell and xorg does have this problem for me since the beginning.
Maybe its a regression, maybe not - don't know. Any other help or information i can provide?

> Restarting X the complete X server does hang at the main loop (looks like 47512
> - but i don't know if they are related at all).

Might well be one and the same problem.


(In reply to comment #4)
> If i remove those config or set all to "on" it happens really frequently
> (enough to make me report this one here to seek help :-) ).

Option "AccelDFS" has no effect with KMS (the log file incorrectly claims otherwise). It might be interesting to narrow down which of the other three options really make a difference for you. My first guess would be page flipping.


(In reply to comment #5)
> dmesg

Did you capture this after a hang occurred? If so, it looks like we can rule out the hangs being due to GPU lockups.

Does booting with radeon.msi=0 on the kernel command line work around the problem?


(In reply to comment #7)
> I suffer this one since i'll changed from lucid to oneiric, where gnome2 was
> removed and i did switch to gnome3.

Was lucid even using KMS yet?


> Because it did freeze very often, i switched to lxde where is does not seem to
> happen

Not even for OpenGL apps? Is the GNOME3 fallback mode affected now?

> (In reply to comment #4)
> > If i remove those config or set all to "on" it happens really frequently
> > (enough to make me report this one here to seek help :-) ).
> 
> Option "AccelDFS" has no effect with KMS (the log file incorrectly claims
> otherwise). It might be interesting to narrow down which of the other three
> options really make a difference for you. My first guess would be page
> flipping.

Ok - it happens too with all of them off. But yes, it feels like page flipping.

> 
> 
> (In reply to comment #5)
> > dmesg
> 
> Did you capture this after a hang occurred? If so, it looks like we can rule
> out the hangs being due to GPU lockups.

Yes i'll take it after the hang. It did hang too today again with all options off. dmesg is similar. And the trace if got after hanging is the same as in 47512 - so yes it looks like the same.
Should other one be marked as duplicate?

> 
> Does booting with radeon.msi=0 on the kernel command line work around the
> problem?

I'll try it (with all options on again) and will report.

> 
> 
> (In reply to comment #7)
> > I suffer this one since i'll changed from lucid to oneiric, where gnome2 was
> > removed and i did switch to gnome3.
> 
> Was lucid even using KMS yet?

With backported kernel if was possible afaik. But this comment was about that i don't know if its a regression. So it should not matter at all.

> 
> 
> > Because it did freeze very often, i switched to lxde where is does not seem to
> > happen
> 
> Not even for OpenGL apps? Is the GNOME3 fallback mode affected now?

Hm it might happen for OpenGL apps - any hint here what app i can try to provoke it? Some hints for me?
Gnome3 fallback mode i'll did not try - but i will test it too (with msi=1 and msi=0).
All test cycles will take some time - guess i get a result in the next few days.

First feedback:

After running 3 days with radeon.msi=0 it seems the hang is gone. Did not suffer this problem until now.
I am using this config now:

/etc/X11/xorg.conf.d/20-radeon:

Section "Device"
        Identifier  "Card0"
        Option  "EnablePageFlip"        "on"
        Option  "ColorTiling"           "on"
        Option  "EXAPixmaps"            "off"
        Option  "AccelDFS"              "on"
        Option  "MigrationHeuristic" "greedy"
EndSection


"EXAPixmaps" is "off" because i see many artifacts sometimes on the screen otherwise - maybe its the greedy option, did not test both exclusive. See the screen for artifacts i am seeing, maybe someone can tell me if i can try other options here than pixmap = off.

I'll test gnome3 fallback + lxde with openGl in the next days and provide additional feedback.

Created attachment 58857 [details]
screenshot with artifacts

(In reply to comment #10)
> After running 3 days with radeon.msi=0 it seems the hang is gone. Did not
> suffer this problem until now.

Okay, so I'm reassigning this to the kernel driver for now, though this might just be a duplicate of another MSI related report here or in the kernel bugzilla.


> "EXAPixmaps" is "off" because i see many artifacts sometimes on the screen
> otherwise - maybe its the greedy option, did not test both exclusive.

Option "MigrationHeuristic" has no effect with KMS. Disabling EXAPixmaps probably just avoids the problem because it prevents hardware acceleration in most cases. The corruption looks like bug 47266, would be nice if somebody could find a working driver snapshot and bisect.

*** Bug 47512 has been marked as a duplicate of this bug. ***

Current Linux kernels disable MSI by default on these cards.