[i945 tiling] (Needs kernel 2.6.30-rc3) Low performance due to no A17 workaround

Hi llstarks,

Please attach the output of `lspci -vvnn`, and attach your /var/log/Xorg.0.log (and maybe Xorg.0.log.old) file from after reproducing this issue. If you've made any customizations to your /etc/X11/xorg.conf please attach that as well.

[This is an automated message. Apologies if it has reached you inappropriately; please just reply to this message indicating so.]

lspci and Xorg.0.log attached.

Starks, your description says to see the upstream bug for details, however there is no reference to "A17" I can see there. Please explain further.

Also, the patch upstream appears to be a kernel patch to disable GEM. Is your filing of this bug report a request to disable GEM in the kernel? I see you filed it against the 2D intel driver so am curious if you meant differently?

(The upstream patch I assume is being proposed to add to jaunty?)

[I set the Jaunty task, but it is undetermined yet if we actually want to disable GEM like this.]

Maybe Eric Anholt will surprise us with an A17 patch, but if he doesn't, the disable GEM patch is the stop-gap answer.

Well, the GEM disablement option patch has to be taken by the kernel team. It looks safe enough to me.

The A17 patch currently in the upstream bug tracker looks pretty non-trivial, and I'm worried it could carry regressions. I'd be more comfortable to see it in the main tree and tested against a wide array of hardware.

Bryce, I feel you would be in the best position to submit it to the kernel team mailing list. I have no programming background nor can I claim any expertise with regard to the Intel drivers or modules.

This bug is fixed upstream according to https://bugs.freedesktop.org/show_bug.cgi?id=16835 , when will this high importance fix enter Jaunty?

Here's the kernel fix. I've been told this is included in 2.6.30-rc2:

commit 280b713b5b0fd84cf2469098aee88acbb5de859c
Author: Eric Anholt <email address hidden>
Date: Thu Mar 12 16:56:27 2009 -0700

    drm/i915: Allow tiling of objects with bit 17 swizzling by the CPU.

queued to linus

The kernel team has prepared a PPA of 2.6.30-rc2:

PPA: http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc2

<rtg> bryce: 2.6.30-rc2 has a bunch of i915 patches from Dave Arlie. He claims big performance improvements which likely address out issues in Jaunty. Is it possible for you to run that kernel somewhere and test the results? http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc2

@olskar, keep in mind that Jaunty is already in final freeze. Your question should s/when//.

So... Does this fix make jaunty today or jaunty-updates a few weeks from
now?

As Bryce mentioned, it would be great to get feedback regarding testing of http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc2 . I imagine if the feedback is positive this would be considered for a SRU - http://wiki.ubuntu.com/StableReleaseUpdates .

I have now tried the 2.6.30-rc2 and noticed no difference. Glxgears gives ~115 frames where I with intrepid got >1000.

lspci | grep VGA
00:02.0 VGA compatible controller: Intel Corporation Mobile 915GM/GMS/910GML Express Graphics Controller (rev 04)

With this kernel a UXA config freezes almost immediately, even before greeter appears. How's that for feedback. :-)

However, I see there are some updates in Jaunty to mesa packages which I haven't installed yet... but no changelog available so I can't tell what's in there.

Subject says 945 tiling hence my posting... but everyone else here talking about 915?

@ Zack Evans, I think there is a lot of intelcards affected, among others 935 and 915

Apparently Intel's 2.7 graphics driver still has problem
http://www.phoronix.com/scan.php?page=news_item&px=NzIxMQ

@ Zack Evans, I think there is a lot of intelcards affected, among others 945 and 915

Apparently Intel's 2.7 graphics driver still has problem
http://www.phoronix.com/scan.php?page=news_item&px=NzIxMQ

With 2.6.30 rc2 and xserver-xorg-video-intel 2:2.6.99.1+git20090416.b9716b83-0ubuntu0tormod from xorg-edgers ppa I get 1400 fps in glxgears in exa mode with Kwin compositing disabled, and ~700 with compositing enabled. This also helps me work around the annoying suspend-to-ram crashes I get with UXA and still have good performance.

lspci -vnn:
00:02.0 VGA compatible controller [0300]: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller [8086:27a2] (rev 03) (prog-if 00 [VGA controller])
        Subsystem: Dell Device [1028:01d7]
        Flags: bus master, fast devsel, latency 0, IRQ 16
        Memory at eff00000 (32-bit, non-prefetchable) [size=512K]
        I/O ports at eff8 [size=8]
        Memory at d0000000 (32-bit, prefetchable) [size=256M]
        Memory at efec0000 (32-bit, non-prefetchable) [size=256K]
        Capabilities: [90] Message Signalled Interrupts: Mask- 64bit- Queue=0/0 Enable-
        Capabilities: [d0] Power Management version 2
        Kernel modules: intelfb

I can confirm, with linux-image-2.6.30rc2 and xserver-xorg-video-intel-2:2.6.3-0ubuntu9, I get a definite performance boost when using kwin. With other kernels there was stuttering under any compositing, but now performance seems to be back or even better than it was in intrepid (using UXA).

I don't get any crashes when suspending, but I get a very weirdly rainbow-colored screen right before the computer shuts down and right before X comes back. It looks sort of like when you press on an LCD monitor REALLY hard in the top left and bottom left corners. Doesn't really change anything, just a little weird.

lspci -vnn:
00:02.0 VGA compatible controller [0300]: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller [8086:27a2] (rev 03)
        Subsystem: Fujitsu Limited. Device [10cf:1380]
        Flags: bus master, fast devsel, latency 0, IRQ 16
        Memory at f0300000 (32-bit, non-prefetchable) [size=512K]
        I/O ports at 1800 [size=8]
        Memory at e0000000 (32-bit, prefetchable) [size=256M]
        Memory at f0400000 (32-bit, non-prefetchable) [size=256K]
        Capabilities: [90] Message Signalled Interrupts: Mask- 64bit- Queue=0/0 Enable-
        Capabilities: [d0] Power Management version 2
        Kernel modules: intelfb

Have installed mesa updates now.

I am on
mesa 7.4-0ubuntu2
drm 2.4.9~git20090416.07646002-0ubuntu0tormod
xserver-xorg-video-intel 2:2.6.99.1+git20090416.b9716b83-0ubuntu0tormod

and under .30RC2 with UXA, X freezes even before I have got a greeter.

"bad" Register dump attached - can't think of a way to get a "good" register dump given that X hangs straight away...!

Small comment on my previous post: apparently enabling UXA with tiling causes a lot of the known UXA-related bugs to resurface. Rendering is faster with UXA, but sometimes it seems that the driver decides to stop rendering - I would resize something and suddenly everything will stop and the screen is not updated except for the cursor. Switching to another vt seems to reset it. I switched back to the default EXA and performance is the same as it was in intrepid.

Obvious performance boost with 30rc2. Card is :

00:02.0 VGA compatible controller: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller (rev 03)
 Subsystem: Toshiba America Info Systems Device 0001
 Control: I/O+ Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr- Stepping- SERR- FastB2B- DisINTx-
 Status: Cap+ 66MHz- UDF- FastB2B+ ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR- INTx-
 Latency: 0
 Interrupt: pin A routed to IRQ 16
 Region 0: Memory at ffd80000 (32-bit, non-prefetchable) [size=512K]
 Region 1: I/O ports at cff8 [size=8]
 Region 2: Memory at e0000000 (32-bit, prefetchable) [size=256M]
 Region 3: Memory at ffd40000 (32-bit, non-prefetchable) [size=256K]
 Capabilities: <access denied>
 Kernel modules: intelfb

After switching to 30rc2 glxgears went from ~120fps to over 1000fps. Also performance of Compiz & Flash sites, and especially fullscreen video is much, much better.

I was affected by this bug with my fully updated jaunty i915 machine: compiz was unusable, problems with videos and so on.

Now i installed kernel-image from http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc2/ and everything is perfect.

I can also confirm that using the kernel v2.6.30-rc2 fixes this bug, I would be happy to see the fix in the official jaunty kernel soon.

This really works!!!
Thanks very much.

I've been struggling for a couple months trying to get my FPS to go back to >1000 from the mere 250.

Does anybody have the same kernel image for the LPIA architecture? I'd like to see the same fix on my little Dell but I'm running the LPIA system to get a better battery life.

So for clarity: are we saying this is believed fixed under 2.6.30-RC2 and the plan is to put a .30 kernel into Jaunty updates?

Zack: I think this is simply impossible to happen. Maybe there are some performance/stability patches that can be backported.
btw: this graphic stack works for me and is the one suggested by intel:
http://intellinuxgraphics.org/2009Q1.html

I will have this issue and send to Jaunty SRU

I had the same problem, even switching tabs in firefox took a whole second. I used 2.6.30-rc3 (just for being newer than rc2) from:

http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc3/

... and it fixed the problem without any further changes.

using: Integrated Graphics Chipset: Intel(R) Mobile Intel® GM45 Express Chipset, ubuntu 9.04, acer extensa 5230

I have the same problem with a GMA 4500MHD (ThinkPad X200s). Is it possible to fix it with a stable kernel? If not is there a server version of 2.6.30-rc3? I can't find it under http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30-rc3/.

Best
Christoph

@Christoph -- no we only build -generic kernels generally. What feature of the -server kernel are you needing?

Thanks Andy for asking! I use PAE.

2.6.30 kernel is now in karmic.

linux-generic:
  Installed: (none)
  Candidate: 2.6.30.2.1
  Version table:
     2.6.30.2.1 0
        500 http://archive.ubuntu.com karmic/restricted Packages

Probably I'm not aware of every practice in fixing bugs. But this is a bug in Jaunty---how does it help Jaunty users if a fix for upcoming Karmic is released?

I wouldn't worry Christoph, the bug is marked as high priority and it affects a very big userbase. Ubuntu is used in a lot of netbooks, they even release a netbook version of Ubuntu. Most notebooks use intelcards and not to fix this in Jaunty would be very, very strange. PR suicide. I think a fix is coming in time :)

@Christoph -- it helps to track the bug in Karmic as well as you do not want the bug back when karmic releases.

I try to cherry-pick the bit 17 swizzling patch, it looks like a proper solution for this issue, but lots of conflict when I merge to current Jaunty tree, I check for the upstream and find out that I need to pull lots of commit to solve the conflict. It is against the policy of SRU.

Fortunately we have the disable-gem patch which will force intel driver in the traditional mode and have some improvement with the benchmark.

I merge the patch and build the kernel image. I put those packages at http://people.ubuntu.com/~ikepanhc/lp349992/

If anyone suffers with this issue, could you have a test with the http://people.ubuntu.com/~ikepanhc/lp349992/linux-image-2.6.28-13-ikepanhc-lp349992_1_i386.deb, and add the following line to "/etc/modprobe.d/options"
options i915 enable_gem=0

Please let me know if it works fine or not.

hi,

i tried the 2.6.28-13-ikepanhc-lp349992 kernel and setting the options but it was as slow as with vmlinuz-2.6.28-11-generic. i use an acer extensa 5230 with intel gma 4500. attached is a diff between Xorg.0.log with the ike-kernel with options and 2.6.30rc3 without options, hope it helps.

Hi broe,

Thanks for your information, I will see if there is anyway to merge the patch.

Any news on this?

Can this information be used to fix this?

http://www.ubuntugeek.com/new-intel-graphics-drivers-for-ubuntu-9-04-jaunty.html

Unfortunately no, that's just rehashing stuff already in place for karmic.

On Sat, Jun 13, 2009 at 03:19:29PM -0000, Åskar wrote:
> Can this information be used to fix this?
>
> http://www.ubuntugeek.com/new-intel-graphics-drivers-for-
> ubuntu-9-04-jaunty.html
>
> --
> [i945 tiling] (Needs kernel 2.6.30-rc3) Low performance due to no A17 workaround
> https://bugs.launchpad.net/bugs/349992
> You received this bug notification because you are a bug assignee.
>
> Status in The Linux Kernel: New
> Status in X.org xf86-video-intel: Fix Released
> Status in “linux” source package in Ubuntu: Triaged
> Status in “xserver-xorg-video-intel” source package in Ubuntu: Fix Released
> Status in linux in Ubuntu Jaunty: Triaged
> Status in xserver-xorg-video-intel in Ubuntu Jaunty: Won't Fix
>
> Bug description:
> See upstream bug for technical details.
>
> [lspci]
> 00:00.0 Host bridge [0600]: Intel Corporation Mobile 945GM/PM/GMS, 943/940GML and 945GT Express Memory Controller Hub [8086:27a0] (rev 03)
> Subsystem: Dell Device [1028:01d8]
> 00:02.0 VGA compatible controller [0300]: Intel Corporation Mobile 945GM/GMS, 943/940GML Express Integrated Graphics Controller [8086:27a2] (rev 03)
> Subsystem: Dell Device [1028:01d8]

Hi,

Thanks for the informations, it seems the upstream commit 280b713b5b0fd84cf2469098aee88acbb5de859c is the only solution. I am sorry to say after cherry-pick, the patch is too large to be a SRU patches.

If you suffer with low performance, could you use mainline kernel instead, it contains with the commit and we have reports that it works fine. You can easily find the mainline kernel at http://kernel.ubuntu.com/~kernel-ppa/mainline/v2.6.30/

I will leave this bug and am sorry that I can not help anything.

I'm setting Jaunty nomination to Won't Fix per Ike's previous comment. This patch is however in the current kernel for Karmic so I'm setting the actively developed "linux (Ubuntu)" task status to Fix Released. As Ike mentioned, the patch should also be available in the mainline kernel builds. Thanks.

Also setting the upstream bug watch to Invalid.

Since it is still a legitimate issue on jaunty, and a fix is known, probably it should be left open, just unassigned with the ct-rev tag.

Jaunty has reached end-of-life, so I'll close this report. The bug is marked as being fixed in later versions of Ubuntu