Radeon Screen Corruption since Kernel (maybe) 3.10.5

Logiarch · 2013-09-05 20:03:48

Hi!

I own a board: 880GMH/U3S3 with radeon chipset HD 4250.
I update very frequently to get current state of archlinux and to minimize migration efforts.

Last time I updated my system and encountered screen corruptions after the kernel switches to kms.
After some downgrading some packages (linux-firmware, mesa, ...) I found out, that the kernel package (at the time 3.10.5) was the bad guy.
Now that kernel 3.11 is entering archlinux, I tried that today - with the same screen corruption result.
When I reverted the kernel package to 3.10.2 the system works as expected - no screen corruptions.

I made some <real> screenshots while booting on console:
Boot Screen

and KDE:
KDE Screen

If someone would like to look at the kernel output:
dmesg output

Now I'm at a point where I'm glue less how to solve this issue...

Lone_Wolf · 2013-09-06 12:54:49

[    0.000000] Your BIOS doesn't leave a aperture memory hole
[    0.000000] Please enable the IOMMU option in the BIOS setup

If you're running a 32-bit kernel, this can be ignored.
For a 64-bit kernel on linux however, having IOMMU disabled tends to have very bad results.

Logiarch · 2013-09-06 18:14:40

Thanks for your reply.

For a 64-bit kernel on linux however, having IOMMU disabled tends to have very bad results.

I don't have the ability to turn on the IOMMU inside the BIOS :-(
I tried the kernel parameters iommu=noaperture, iommu=force, iommu=allowdac with no luck (Same result). Maybe you can suggest some other things?

Last edited by Logiarch (2013-09-06 18:33:28)

xeekei · 2013-09-07 03:15:24

I have a similar problem. Though it only occurs when I load the GPU, like hardware acceleration in VLC, or games. Could be the exact same problem. Last time I was in Arch was with Linux 3.9.x, and I came back during 3.10.9.

Lone_Wolf · 2013-09-07 10:34:08

Logiarch, i've looked a bit deeper into the dmesg, and may have been wrong about the IOMMU message causing your problems.

[    2.057464] [drm] initializing kernel modesetting (RS880 0x1002:0x9715 0x1849:0x9715).
<snip>
[    2.059736] [drm] Loading RS780 Microcode

Your videocard is detected as a RS880/HD4200 type, but the firmware it loads is for the RS780/HD3200 cards.
This seems weird.

I suggest you compare the IOMMU & RSx80 messages in dmesg from the corruted graphics kernel AND dmesg from the working kernel.

another thing, probably unrelated to the graphics corruption :
in the lines from systemd , there are several services that are failing to start (dcronie service is one of them) , you should check your setup.

Xeekei : logiarch has very different symptoms as his problems appear way before X is started.
Imo you should start a new thread and include xorg0.log .

xeekei · 2013-09-07 17:31:02

Lone_Wolf wrote:

Xeekei : logiarch has very different symptoms as his problems appear way before X is started.
Imo you should start a new thread and include xorg0.log .

I posted this: https://bbs.archlinux.org/viewtopic.php?id=169312 Probably in the wrong section though. I'll look for that xorg0.log and add that.

Logiarch · 2013-09-07 18:01:23

Hi Lone_Wolf!

Thanks for taken time for my individual problem!

Lone_Wolf wrote:

Logiarch, i've looked a bit deeper into the dmesg, and may have been wrong about the IOMMU message causing your problems.
[    2.057464] [drm] initializing kernel modesetting (RS880 0x1002:0x9715 0x1849:0x9715).
<snip>
[    2.059736] [drm] Loading RS780 Microcode
Your videocard is detected as a RS880/HD4200 type, but the firmware it loads is for the RS780/HD3200 cards.
This seems weird.
I suggest you compare the IOMMU & RSx80 messages in dmesg from the corruted graphics kernel AND dmesg from the working kernel.

Both kernels suggests to enable iommu on BIOS.

Your BIOS doesn't leave a aperture memory hole
Please enable the IOMMU option in the BIOS setup

Both kernel loads the RS780 Microcode

Loading RS780 Microcode

I can't see real differences between the drm init routines, only the version changes (as expected ):

[drm] Initialized radeon 2.34.0 (not working) vs.[drm] Initialized radeon 2.33.0 (working)

I did some git bisect between kernel version v3.10.2 and v3.10.5 today and found the commit which causes this behavior:

6f8bbaf568c7f2c497558bfd04654c0b9841ad57 is the first bad commit
commit 6f8bbaf568c7f2c497558bfd04654c0b9841ad57
Author: Alex Deucher <alexander.deucher@amd.com>
Date:   Tue Jul 30 00:22:53 2013 -0400

    drm/radeon/atom: initialize more atom interpretor elements to 0
    
    commit 42a21826dc54583cdb79cc8477732e911ac9c376 upstream.
    
    The ProcessAuxChannel table on some rv635 boards assumes
    the divmul members are initialized to 0 otherwise we get
    an invalid fb offset since it has a bad mask set when
    setting the fb base.  While here initialize all the
    atom interpretor elements to 0.
    
    Fixes:
    https://bugzilla.kernel.org/show_bug.cgi?id=60639
    
    Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
    Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

:040000 040000 d2bb057047f71419a89def40e6e21dc948c5784c 7e49987ae73078e644723f0cb6c791e15e102ab0 M      drivers

Hm, I try to build the linux3.11 kernel without this patch and check out if this would work.
The patch doesn't seems complicated, but fails (on my system).
Edit: It works.

Last edited by Logiarch (2013-09-07 20:05:44)

Lone_Wolf · 2013-09-08 10:59:42

nice catch, Logiarch .

It seems that patch not only doesn't fix the problem in #60639, but zeroes out something that's needed for your system to work correctly.

I suggest you open a new kernel bug, point to #60639 and add alex deucher & greg Koah-Hartmann to the cc list for the new bug.

supplying xorg0.log , dmesg for both working / non working kernel would also be good.

Xeekei, i've reported your thread for moving it to this subboard.

Logiarch · 2013-09-10 20:29:41

Seems that issue is fixed in case the suggested patch from bug
#60791 is applied on the mainline kernel.

Arch Linux

#1 2013-09-05 20:03:48

Radeon Screen Corruption since Kernel (maybe) 3.10.5

#2 2013-09-06 12:54:49

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#3 2013-09-06 18:14:40

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#4 2013-09-07 03:15:24

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#5 2013-09-07 10:34:08

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#6 2013-09-07 17:31:02

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#7 2013-09-07 18:01:23

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#8 2013-09-08 10:59:42

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

#9 2013-09-10 20:29:41

Re: Radeon Screen Corruption since Kernel (maybe) 3.10.5

Board footer