Bug 30047

Created attachment 38481 [details]
debug output of X server

Created attachment 38504 [details] [review]
Attempt to make radeon_dri2_create_buffer() more robust

Does this X radeon driver patch fix the problem?

Thanks for the patch, I'll tell you as soon as I learn how to apply it, I've never done something like this before, but I'll try. :-)

Now I'm totally confused: hwinfo tells me I have radeon, but the package installed is radeonhd?
Anyhow, neither in the source of radeon nor in the source of radeonhd I can find a file "radeon_dri2.c" where I should apply the patch.
Will it be sufficient to compile just the driver (if I find the appropriate source) , or do I have to compile the whole X stuff (sorry, the only thing I compiled until now were plasmoids)?

(In reply to comment #4)
> Now I'm totally confused: hwinfo tells me I have radeon, but the package
> installed is radeonhd?

radeonhd is a separate driver.  It's not applicable to your chip.  I'm not sure how suse packages the radeon (xf86-video-ati) driver.  It may be part of a larger xorg driver package.

> Anyhow, neither in the source of radeon nor in the source of radeonhd I can
> find a file "radeon_dri2.c" where I should apply the patch.
> Will it be sufficient to compile just the driver (if I find the appropriate
> source) , or do I have to compile the whole X stuff (sorry, the only thing I
> compiled until now were plasmoids)?

All you need to do is recompile the ddx driver.  The upstream source is here:
http://cgit.freedesktop.org/xorg/driver/xf86-video-ati/
and building info is here:
http://wiki.x.org/wiki/radeonBuildHowTo

Good: Thanks to your advice I could patch and compile the ati driver (6.13.1).

Bad: Xorg -configure doesn't generate a working Xorg.conf. When I test it, I get a black screen (usual ATI driver) or a black screen and an X crash (patched driver) with the error log (4 times) "no input driver/identifier attached. ignoring" (in the patched version there's a complaint about xterm, too).
But I don't see anything wrong in Xorg.conf, everything is there, see attachment.

Created attachment 38574 [details]
not working Xorg.conf

(In reply to comment #6)
> Bad: Xorg -configure doesn't generate a working Xorg.conf.

No need to generate a new xorg.conf for testing the patch, just make sure the X server loads the patched radeon_drv.so and try to reproduce the crash.

OK, thanks to your hints I patched, compiled, and installed the new driver, but the crash behaviour is the same.

(In reply to comment #9)
> OK, thanks to your hints I patched, compiled, and installed the new driver, but
> the crash behaviour is the same.

Bummer. Please attach a new GDB backtrace with the patch applied.

Uff, I really try, but I can't get this stuff to work. I can't remember that I did something special last time, but somehow I got the debug output. Now I get only this one:

Backtrace:
[   135.694] 0: X (xorg_backtrace+0x37) [0x80986d7]
[   135.694] 1: X (0x8048000+0x4e3fa) [0x80963fa]
[   135.694] 2: (vdso) (__kernel_rt_sigreturn+0x0) [0xffffe410]
[   135.694] 3: /usr/lib/xorg/modules/extensions/libdri2.so (0xb734f000+0x303c) [0xb735203c]
[   135.694] 4: /usr/lib/xorg/modules/extensions/libdri2.so (0xb734f000+0x363c) [0xb735263c]
[   135.694] 5: X (0x8048000+0x48177) [0x8090177]
[   135.694] 6: X (0x8048000+0x1eb85) [0x8066b85]
[   135.694] 7: /lib/libc.so.6 (__libc_start_main+0xfe) [0xb7465c0e]
[   135.695] 8: X (0x8048000+0x1e771) [0x8066771]
[   135.695] Segmentation fault at address (nil)
[   135.695] 
Fatal server error:
[   135.695] Caught signal 11 (Segmentation fault). Server aborting
[   135.695] 
[   135.695] 
Please consult the The X.Org Foundation support at http://wiki.x.org for help. 

The debug script in the Xorg wiki doesn't work for me (as /usr/X11R6/ is empty, I changed the path to /var/lib/X11/), the X server doesn't start at all with this. I'm sorry, at the moment this surpasses my limited skills.

I'm seeing this with a radeon 9600xt card on opensuse 11.4. Could you update the patch for video-ati-6.14.2 ?

The (old) backtrace is attached.

Created attachment 52634 [details]
Backtrace

I can reproduce this crash reliably with kwin compositing with the attached OO.org document and when I expand the mimetype menu of the kparts-plugin.

Created attachment 52681 [details] [review]
Attempt to make radeon_dri2_create_buffer() more robust

Patch against current Git.

Created attachment 52683 [details] [review]
Attempt to make radeon_dri2_create_buffer() more robust

Updated patch which avoids leaking resources in error paths. Only lightly tested (glxgears works :) for regressions.

No crashes so far. However,
1. from zeile 5 downwards the tooltip becomes corrupted,
2. the mimetype menu in kparts-plugin does not display the menu that would fill the screen, which triggered the crash previously.

Thanks.

(In reply to comment #16)
> No crashes so far.

Great, thanks for testing the fix. I've pushed it to xf86-video-ati Git.


> 1. from zeile 5 downwards the tooltip becomes corrupted,
> 2. the mimetype menu in kparts-plugin does not display the menu that would fill
> the screen, which triggered the crash previously.

I suspect one of two things is happening here:

1. These pixmaps exceed the maximum width/height supported by the 3D engine. That would really need to be handled by the compositing manager earlier on.

2. The X driver fails to allocate GPU accessible memory for these pixmaps. Do you get 'Failed to alloc memory' lines in Xorg.0.log or the X server stderr output? What's the version of the kernel running when this happens?

The tooltips exceed the width of the screen when this occurs. Also, no error message appears either in Xorg.log or dmesg. The kernel is 3.1rc9. So, should I report this to the kwin developers?

(In reply to comment #18)
> Also, no error message appears either in Xorg.log or dmesg. The kernel is
> 3.1rc9. So, should I report this to the kwin developers?

I think so.