[SOLVED] Gfx corruption with Radeon HD 6950 and 3.4.0-gentoo

[mangobrain]

I upgraded to kernel 3.4.0-gentoo today and noticed immediately that GDM didn't look right. It was pretty obvious, since over the course of a couple of reboots, it either showed a completely corrupt screen, or - strangely - a perfectly good background image, but absolutely no controls (no user list, no top bar, etc.). The pointer was fine, and GDM was obviously running and responding because I could log in, though I only knew this because the background image changed and the hard drive activity was audible. GNOME Shell didn't render any better than GDM.

Thankfully, I could still switch VTs, and log in on a good old fashioned text console, allowing me to reboot and go back to 3.3.5-gentoo. All I know so far is the problem doesn't seem to be related to the choice of PCIe ASPM profile; I tried all three, since the option was presented as new when doing "make oldconfig", but none seemed to work.

Anyone else having similar problems?

[BillWho]

mangobrain,

If you're using the ati closed source drivers, there is a problem with it.

https://bugs.gentoo.org/show_bug.cgi?id=413569

I added my two cents to it yesterday and then decided to migrate to the radeon driver.

Good luck
_________________
Good luck

Since installing gentoo, my life has become one long emerge

[mangobrain]

I'm using the open source driver - have been for a while now, and this is the first time I've had any sort of big regression.

[i92guboj]

Did you try some other DM the where mutter compositor is not involved?

Also, are you sure you are using the gallium opengl backend?
_________________
Gentoo Handbook | My website

[mangobrain]

I didn't, because I don't have any other DMs installed at the moment. glxinfo (after booting 3.3.5) states the following:

OpenGL renderer string: Gallium 0.4 on AMD CAYMAN
OpenGL version string: 2.1 Mesa 8.0.2

As noted, plain-text consoles appeared to work fine under 3.4.0, but I didn't do much other than switch VTs and reboot. GNOME 3 is the only desktop I have installed at the moment, and I'd rather stay on 3.3.5 until this is figured out rather than go through the pain of installing and configuring another environment.

[i92guboj]

You don't have to go thru the pain of anything. All you need to do is to emerge xdm, set it to be your DM by editing /etc/conf.d/xdm, and do /etc/init.d/xdm restart, and see if that works. If it does, then you know it's some kind of weirdness between the radeon driver and the mutter compositor, which is the most likely candidate.

If you prefer to wait then it's fine, your choice. But it's not like you have to install kde to check this

You could also try setting USE="-gnome-shell -plymouth" for gdm, and recompiling it, to see if that helps.
_________________
Gentoo Handbook | My website

[mangobrain]

Fair point, XDM is pretty lightweight. Will try that later when I get home.

[mangobrain]

OK... booting into 3.4.0-gentoo, killing GDM then just doing plain-old "startx", it seems that X itself is mostly OK, the problem appears to be restricted to OpenGL-related things. Running glxgears from xterm produces the following output, repeated over and over:

radeon: The kernel rejected CS, see dmesg for more information.

dmesg says:

radeon 0000:01:00.0: evergreen_packet3_check:2055 invalid cmd stream 994
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
radeon 0000:01:00.0: evergreen_surface_value_conv_check:325 cb invalid array mode 6
radeon 0000:01:00.0: evergreen_packet3_check:2055 invalid cmd stream 582
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
radeon 0000:01:00.0: evergreen_cs_track_validate_texture:796 texture bo too small (layer size 7526400, offset 0, max layer 1, depth 1, bo size 7299072) (1792 1050)
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
radeon 0000:01:00.0: evergreen_surface_value_conv_check:325 cb invalid array mode 6
radeon 0000:01:00.0: evergreen_packet3_check:2055 invalid cmd stream 515
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
radeon 0000:01:00.0: evergreen_cs_track_validate_depth:668 depth read bo too small (layer size 30277632, offset 0, max layer 1, bo size 11468800)
radeon 0000:01:00.0: evergreen_packet3_check:2055 invalid cmd stream 561
[drm:radeon_cs_ib_chunk] *ERROR* Invalid command stream !
radeon 0000:01:00.0: evergreen_surface_value_conv_check:325 cb invalid array mode 6

and so on and so forth, repeated over and over.

Should I log this on the kernel bug tracker, or over on freedesktop? I assume there's definitely something there worth reporting.

[mangobrain]

Found the solution here: http://comments.gmane.org/gmane.comp.freedesktop.xorg.drivers.ati/21981

Opened up xorg.conf and commented out the ColorTiling2D option, restarted GDM, and things seem to be working again.

[i92guboj]

I am glad you managed to solve it, well, more or less. From what I read there, the definitive solution will come with the next release of mesa. If you lack the patience you could try the x11 overlay. It's in the layman repo. You can usually find the latest and greatest mesa in it.
_________________
Gentoo Handbook | My website

[mangobrain]

I've played with that overlay in the past, when the open-source Radeon driver was newer and had more issues. I'm in no hurry to play with it again when things are largely working, even if I have had to disable a feature which ought to improve performance. This is mostly my machine to play with, but sometimes I need to use it to work from home, so I can't risk introducing too much instability.

This solution will suit me for the time being.