[RV200] [XAA] garbled screen with compiz but no KMS on ATI Radeon Mobility 7500

[Jamie Strandboge]

Binary package hint: xorg

In trying workarounds for bug #507148 I disabled KMS by booting with radeon.modeset=0. When I logged in compiz was enabled, but opening applications resulted in garbled parts of the screen. See attached screenshot.

ProblemType: Bug
Architecture: i386
Date: Thu Jan 28 13:52:56 2010
DistroRelease: Ubuntu 10.04
DkmsStatus: Error: [Errno 2] No such file or directory
EcryptfsInUse: Yes
Lsusb:
 Bus 004 Device 002: ID 0483:2016 SGS Thomson Microelectronics Fingerprint Reader
 Bus 004 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 003 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 002 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
MachineType: IBM 2378RRU
Package: xorg 1:7.5+1ubuntu2
PccardctlIdent:
 Socket 0:
   no product info available
 Socket 1:
   no product info available
PccardctlStatus:
 Socket 0:
   no card
 Socket 1:
   no card
ProcCmdLine: BOOT_IMAGE=/boot/vmlinuz-2.6.32-11-generic root=UUID=0c1cec18-6655-4ef5-8c03-2009d66c50e2 ro splash quiet radeon.modeset=0
ProcEnviron:
 PATH=(custom, user)
 LANG=en_US.UTF-8
 SHELL=/bin/bash
ProcVersionSignature: Ubuntu 2.6.32-11.15-generic
RelatedPackageVersions:
 xserver-xorg 1:7.5+1ubuntu2
 libgl1-mesa-glx 7.7-0ubuntu8
 libdrm2 2.4.17-0ubuntu2
 xserver-xorg-video-intel 2:2.9.1-1ubuntu1
SourcePackage: xorg
Uname: Linux 2.6.32-11-generic i686
dmi.bios.date: 06/18/2007
dmi.bios.vendor: IBM
dmi.bios.version: 1RETDRWW (3.23 )
dmi.board.name: 2378RRU
dmi.board.vendor: IBM
dmi.board.version: Not Available
dmi.chassis.asset.tag: No Asset Information
dmi.chassis.type: 10
dmi.chassis.vendor: IBM
dmi.chassis.version: Not Available
dmi.modalias: dmi:bvnIBM:bvr1RETDRWW(3.23):bd06/18/2007:svnIBM:pn2378RRU:pvrThinkPadT42:rvnIBM:rn2378RRU:rvrNotAvailable:cvnIBM:ct10:cvrNotAvailable:
dmi.product.name: 2378RRU
dmi.product.version: ThinkPad T42
dmi.sys.vendor: IBM
system:
 distro: Ubuntu
 architecture: i686kernel: 2.6.32-11-generic

[lspci]
01:00.0 VGA compatible controller [0300]: ATI Technologies Inc Radeon Mobility M7 LW [Radeon Mobility 7500] [1002:4c57]
     Subsystem: IBM Device [1014:0530]

[Bryce Harrington (bryce)]

I'm fairly certain this is the well-known "screen corruption with XAA". Indeed, EXA was made to replace XAA specifically because of bugs like this.

So the real bug here is why is your system using XAA rather than EXA, and that I believe is because in certain circumstances EXA has severe problems, and so EXA is being quirked off in these cases. One situation is where the video memory is too low. You can try forcing EXA via the AccelMethod option. If it turns out that EXA works fine on your hardware, perhaps the quirk can be adjusted to not be employed in that case. If EXA does not work right, than probably we should focus efforts on *that* issue, since it is extraordinarily unlikely anyone will be fixing this problem with XAA at this point.

This quirk is a fairly recent thing. In Jaunty we (Ubuntu) forced all ATI hardware to use EXA. I don't recall whether the quirk to use XAA appeared in Karmic or only on Lucid, but this may explain why you see this as a regression.

Alternatively, compiz could be quirked to not work for this graphics card. I don't care for that option since it really just sweeps the bug under the carpet, but it is something which could be considered if we get down to release and the underlying issues haven't been resolved.

[Jamie Strandboge (jdstrand)]

Verified that with no xorg.conf file it loaded with XAA.

Enabled compiz and adjusted xorg.conf to have (with still using radeon.modeset=0):
Section "Device"
        Identifier "Configured Video Device"
        #Option "RenderAccel" "off"
        Option "AccelMethod" "EXA"
EndSection

Section "Monitor"
        Identifier "Configured Monitor"
EndSection

Section "Screen"
        Identifier "Default Screen"
        Monitor "Configured Monitor"
        Device "Configured Video Device"
EndSection

This does fix the screen corruption, but X is painfully slow. Eg, resizing a firefox window with nothing else open takes seconds.

[Jamie Strandboge (jdstrand)]

I should also mention that I verified that X did indeed start with 'EXA' after doing the above. For giggles, I later disabled compiz and X was still painfully slow.

[Jamie Strandboge (jdstrand)]

Is low the correct priority for this? If KMS ends up being blacklisted for radeon due to bug #507148 it may be very important to also blacklist compiz, which is the current default. The only known stable configuration for the radeon as of today is XAA/metacity/RenderAccel off/no KMS. If the compiz and KMS bugs are not fixed in time for Lucid, we should quirk to this stable configuration (I hope we don't have to resort to this!).

[Jamie Strandboge (jdstrand)]

This is still an issue with 2.6.32-14.20-generic and the latest xorg-edgers drivers. I maintain that this is a higher priority than 'Low' because this bug will affect people if KMS is blacklisted for the radeon (as mentioned in my last comment, the only known stable configuration for the radeon 7500 is still XAA/metacity/RenderAccel off/no KMS).

[Jamie Strandboge (jdstrand)]

On up to date Lucid as of 2010-03-09, this is still a problem with no xorg.conf file (X defaults to XAA).

$ cat /proc/version_signature
Ubuntu 2.6.32-16.24-generic

[Jamie Strandboge (jdstrand)]

I forgot to mention, VT switching does not help.

[Bryce Harrington (bryce)]

Forwarding this bug from Ubuntu reporter Jamie Strandboge:
http://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-ati/+bug/513956

[Problem]
Due to bug #26302, KMS seems to be broken on this hardware, so we're considering blacklisting this pci id to use UMS. However, with UMS instead of KMS, the screen suffers corruption when compositing is used.

This might be the old "Random screen corruption with 3D on XAA", however it appears the driver is opting for XAA rather than EXA as the default in this circumstance so that seems suboptimal. The user also tried forcing the hardware to use EXA but found it resulted in unacceptably slow performance.

The only known stable configuration for the radeon 7500 is XAA/metacity/RenderAccel off/no KMS).

This has been tested against Ubuntu 2.6.32-16.24-generic which contains the 2.6.33 drm backported, as airlied recommended.

[Original Description]
In trying workarounds for LP bug #507148 (fdo #26302) I disabled KMS by booting with radeon.modeset=0. When I logged in compiz was enabled, but opening applications resulted in garbled parts of the screen. See attached screenshot.

Architecture: i386
Date: Thu Jan 28 13:52:56 2010
DistroRelease: Ubuntu 10.04
DkmsStatus: Error: [Errno 2] No such file or directory
EcryptfsInUse: Yes
Lsusb:
 Bus 004 Device 002: ID 0483:2016 SGS Thomson Microelectronics Fingerprint Reader
 Bus 004 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 003 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 002 Device 001: ID 1d6b:0001 Linux Foundation 1.1 root hub
 Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub
MachineType: IBM 2378RRU
Package: xorg 1:7.5+1ubuntu2
PccardctlIdent:
 Socket 0:
   no product info available
 Socket 1:
   no product info available
PccardctlStatus:
 Socket 0:
   no card
 Socket 1:
   no card
ProcCmdLine: BOOT_IMAGE=/boot/vmlinuz-2.6.32-11-generic root=UUID=0c1cec18-6655-4ef5-8c03-2009d66c50e2 ro splash quiet radeon.modeset=0
ProcEnviron:
 PATH=(custom, user)
 LANG=en_US.UTF-8
 ProcVersionSignature: Ubuntu 2.6.32-11.15-generic
RelatedPackageVersions:
 xserver-xorg 1:7.5+1ubuntu2
 libgl1-mesa-glx 7.7-0ubuntu8
 libdrm2 2.4.17-0ubuntu2
 xserver-xorg-video-intel 2:2.9.1-1ubuntu1
SourcePackage: xorg
Uname: Linux 2.6.32-11-generic i686
dmi.bios.date: 06/18/2007
dmi.bios.vendor: IBM
dmi.bios.version: 1RETDRWW (3.23 )
dmi.board.name: 2378RRU
dmi.board.vendor: IBM
dmi.board.version: Not Available
dmi.chassis.asset.tag: No Asset Information
dmi.chassis.type: 10
dmi.chassis.vendor: IBM
dmi.chassis.version: Not Available
dmi.modalias: dmi:bvnIBM:bvr1RETDRWW(3.23):bd06/18/2007:svnIBM:pn2378RRU:pvrThinkPadT42:rvnIBM:rn2378RRU:rvrNotAvailable:cvnIBM:ct10:cvrNotAvailable:
dmi.product.name: 2378RRU
dmi.product.version: ThinkPad T42
dmi.sys.vendor: IBM
system:
  architecture: i686kernel: 2.6.32-11-generic

[lspci]
01:00.0 VGA compatible controller [0300]: ATI Technologies Inc Radeon Mobility M7 LW [Radeon Mobility 7500] [1002:4c57]
     Subsystem: IBM Device [1014:0530]

[Bryce Harrington (bryce)]

Created an attachment (id=34181)
compiz_and_no_kms_garbled_screen.png

[Bryce Harrington (bryce)]

Created an attachment (id=34182)
XorgLog.txt

[Bryce Harrington (bryce)]

Created an attachment (id=34183)
XorgLogOld.txt

[Bryce Harrington (bryce)]

Created an attachment (id=34184)
BootDmesg.txt

[Bryce Harrington (bryce)]

Created an attachment (id=34185)
CurrentDmesg.txt

[Airlied-freedesktop (airlied-freedesktop)]

isn't KMS + EXA + metacity stable?

[agd5f (agd5f)]

I've gone ahead and disabled XAA render accel, it's been broken in the xserver for a while now.

5c256808cb5fea955eea96ffe9196473715156aa

[Bryce Harrington (bryce)]

> isn't KMS + EXA + metacity stable?

Jamie was saying that this particular configuration led to gpu lockups after a few hours of use. However I do not know if this configuration got re-tested with the 2.6.33 drm backport.

Jamie, you indicated the lockup issue still exists with KMS/XAA/compiz, and the compiz crash still occurs for KMS/EXA/compiz. Can you confirm that the KMS/EXA/metacity combo works, or does it still freeze after a few hours?

[Bryce Harrington (bryce)]

I've forwarded this bug upstream to https://bugs.freedesktop.org/show_bug.cgi?id=27151 and subscribed you to it.

I'm pretty sure this bug is considered unsolvable upstream, since they're preferring to move to EXA, but I'm interested in hearing what they say about it. If this hardware can't be fixed with KMS then this seems like the next best configuration.

[Jamie Strandboge (jdstrand)]

XAA/metacity/RenderAccel off/no KMS has been the only stable configuration for my system. KMS/EXA/metacity locks up, though not as often as with compiz. When I retested everything with the recent .33 backported DRM, I did not specifically test KMS/EXA/metacity since KMS/EXA/compiz was still broken.

I lent my laptop to a member of Canonical's kernel team for testing fixes/reproducing. I can follow up with him to test this configuration if required.

[agd5f (agd5f)]

(In reply to comment #9)
> XAA/metacity/RenderAccel off/no KMS has been the only stable configuration for
> my system. KMS/EXA/metacity locks up, though not as often as with compiz. When
> I retested everything with the recent .33 backported DRM, I did not
> specifically test KMS/EXA/metacity since KMS/EXA/compiz was still broken.

Do the patches in bug 26302 help with kms?

[Launchpad Janitor (janitor)]

This bug was fixed in the package xserver-xorg-video-ati - 1:6.12.192-2ubuntu1

---------------
xserver-xorg-video-ati (1:6.12.192-2ubuntu1) lucid; urgency=low

  * Merge new upstream version from Debian.
    - r6xx/r7xx: fix domain handling in accel code. Improves GetImage
      performace by a factor of ~10. (LP: #544496 ?)
    - radeon: add support for pal on legacy IGP chips (LP: #161893)
    - radeon: disable frac fb div with new pll code
    - Allocate Xv buffers to GTT. This is nice performance boost for Xv
      under KMS.
    - XAA: disable render accel. It's been reported broken for a while.
      (LP: #513956)
    - radeon: avoid using DRI1 init path on DRI2 driver. Improves behavior
      with multi-seat
    - kms: fix ums naming compat for DisplayPort
  * Remaining changes:
    - Add quilt to Build-Depends and call patch/unpatch target.
    - 100_radeon-6.9.0-bgnr-enable.patch: Turn on option to allow xserver to
      skip drawing the root background window for the radeon driver. This
      will make the boot up sequence smoother by eliminating one flicker.
  * Drop 101_support_hd4290.patch - included upstream
 -- Bryce Harrington <email address hidden> Tue, 30 Mar 2010 19:24:28 -0700

[Jamie Strandboge (jdstrand)]

Alex,

The patches from 26302 *do* help. Quoting me from that bug:

"These patches resolve this issue completely. After a lot of automated testing
and many hours of real world use with up to date packages that include these
fixes, it no longer crashes. Thanks to everyone who helped fix this. :)

From our kernel changelog:
  [ Upstream Kernel Changes ]
  ...
  * drm/radeon/bo: add some fallback placements for VRAM only objects.
    - LP: #507148
  * drm/radeon/kms: don't print error on -ERESTARTSYS.
    - LP: #507148
  ...
"

I've not specifically testing KMS with XAA, but KMS with EXA/compiz works well now.

[agd5f (agd5f)]

(In reply to comment #11)
>
> I've not specifically testing KMS with XAA, but KMS with EXA/compiz works well
> now.

XAA doesn't work with kms, so there's nothing to test :)