Compiz broken on Karmic with NVIDIA driver and CPU without PAT support

Seeing the same here, fresh Karmic build from the daily cd's. Reproduce:

Install the binary driver using the 'Hardware Drivers' app
Reboot as prompted
after you log in (GDM appears to perform fine), screen draws are incomplete, or very very slow.

Screenshot attached is just after login, the menu's etc are all clickable, but draw very slow. Same for starting and navigaing in apps etc.

apt-cache policy nvidia-glx-180
nvidia-glx-180:
  Installed: 185.18.14-0ubuntu3
  Candidate: 185.18.14-0ubuntu3
  Version table:
 *** 185.18.14-0ubuntu3 0
        500 http://gb.archive.ubuntu.com karmic/restricted Packages
        100 /var/lib/dpkg/status

uname -a
Linux yonah-nix 2.6.31-3-generic #19-Ubuntu SMP Tue Jul 14 16:04:41 UTC 2009 i686 GNU/Linux

sudo lshw -class display
  *-display
       description: VGA compatible controller
       product: G72M [GeForce Go 7400]
       vendor: nVidia Corporation
       physical id: 0
       bus info: pci@0000:01:00.0
       version: a1
       width: 64 bits
       clock: 33MHz
       capabilities: pm msi pciexpress bus_master cap_list
       configuration: driver=nvidia latency=0 module=nvidia

Seeing the same here, fresh Karmic build from the daily cd's. Reproduce:

Install the binary driver using the 'Hardware Drivers' app
Reboot as prompted
after you log in (GDM appears to perform fine), screen draws are incomplete, or very very slow.

Screenshot attached is just after login, the menu's etc are all clickable, but draw very slow. Same for starting and navigating in apps etc.

apt-cache policy nvidia-glx-180
nvidia-glx-180:
  Installed: 185.18.14-0ubuntu3
  Candidate: 185.18.14-0ubuntu3
  Version table:
 *** 185.18.14-0ubuntu3 0
        500 http://gb.archive.ubuntu.com karmic/restricted Packages
        100 /var/lib/dpkg/status

uname -a
Linux yonah-nix 2.6.31-3-generic #19-Ubuntu SMP Tue Jul 14 16:04:41 UTC 2009 i686 GNU/Linux

sudo lshw -class display
  *-display
       description: VGA compatible controller
       product: G72M [GeForce Go 7400]
       vendor: nVidia Corporation
       physical id: 0
       bus info: pci@0000:01:00.0
       version: a1
       width: 64 bits
       clock: 33MHz
       capabilities: pm msi pciexpress bus_master cap_list
       configuration: driver=nvidia latency=0 module=nvidia

Should add that setting Visual Effects to 'None' makes the desktop usable, though obviously with no compiz.

Not sure if it's right but I'm assigning to compiz - if it's not right I figure you guys can help me work out where it needs to be.

Ok, steps to reproduce.

System started with binary nvidia module loaded, visual effects off
Log in to desktop
execute compiz.real --replace

Attaching Xorg.log...

.. and a tail -f of .xsession-errors started just after login, and stopped after I kill the compiz process to get the use of my desktop back (the whole file is massive)

And finally the output of executing 'compiz':

Checking for Xgl: not present.
xset q doesn't reveal the location of the log file. Using fallback /var/log/Xorg.0.log
Detected PCI ID for VGA:
Checking for texture_from_pixmap: present.
Checking for non power of two support: present.
Checking for Composite extension: present.
Checking screen 1Comparing resolution (1680x1050) to maximum 3D texture size (4096): Passed.
Checking for Software Rasterizer: Not present.
Checking for nVidia: present.
Checking for FBConfig: present.
Checking for Xgl: not present.

Same story here on a Dell XPS M1210 using Nvidia GO 7400. Made a video of it here:

http://www.youtube.com/watch?v=3iKQbUZv0No

After a chat in #ubuntu+1 and finding someone else on the forum with the same problem - and the same card (GeForce Go 7400), changing this to nvidia-graphics-drivers

http://ubuntuforums.org/showthread.php?t=1187783

@PRGUY85 I get a slightly different effect than that. What software did you use to capture the video, I'll try the same :)

I don't remember now. Itwas available on Synaptic. That isn't the
effect. It gets allslow with dark bars.

On 7/24/09, Darren Worrall <email address hidden> wrote:
> @PRGUY85 I get a slightly different effect than that. What software did
> you use to capture the video, I'll try the same :)
>
> --
> Compiz Slow on Karmic w/ NVIDIA and 2.6.30 Kernel
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>

--
Manuel Mas

The issue is with the Kernel and not the NVIDIA or Compiz version..

I have not yet tried the newest Beta NVIDIA driver (to see if there is a work around in it) or the newest vanilla kernel, or rc from 2.6.31..

But you can log into 2.6.29 and use compiz just fine on the same machine with the same nvidia driver..

so I don't think this belongs with compiz or nvidia..

A kernel regression is less likely I think, you'd see problems all over the shop I'd bet. Same problem on 2.6.31 as well as 2.6.30.

Anyway, the guys in #ubuntu+1 suggested the driver as culprit, and they're way smarter than I am :D

Do you have the same graphics card as us? (try 'sudo lshw -class display')? Also, posting ~/.xsession-errors and /var/log/Xorg.0.log will be useful to the developers :)

Darren,

The same driver, with the same compiz version with a different kernel (2.6.29.x) works without regressions. I boot with 2.6.30 kernel and I get very poor fps on multiple distributions with multiple versions of the driver (minus the beta since I haven't yet tried it) and multiple version of compiz (w and without the NOMAD patches).

To me that points to a regression in the kernel (even if it's the way the driver builds on the kernel), but again no build failures when using the binary from NVIDIA.

The laptop in question is a Dell D820 (not around me right now but I think it's an 110M) which is equivalent to a GeForce 7600 or 7400 I can't remember.

it is the kernel.. it's the method of PAT forcing back to MTRRs, so this may be per CPU basis and not related to GPU's.

There has been a work around for those who have installed using the binary (this work around is NOT for third party inclusion)
diff -ru usr/src/nv/nv.c usr/src/nv.572433/nv.c
--- usr/src/nv/nv.c 2009-05-27 03:20:53.000000000 -0700
+++ usr/src/nv.572433/nv.c 2009-07-05 11:31:27.635210006 -0700
@@ -888,6 +888,7 @@
     unsigned int pat1, pat2, i;
     U008 PAT_WC_index;

+#if 0 /* disabled */
     if (!test_bit(X86_FEATURE_PAT,
             (volatile unsigned long *)&boot_cpu_data.x86_capability))
     {
@@ -895,6 +896,7 @@
             "NVRM: CPU does not support the PAT, falling back to MTRRs.\n");
         return NV_PAT_MODE_DISABLED;
     }
+#endif

     NV_READ_PAT_ENTRIES(pat1, pat2);
     PAT_WC_index = 0xf;

You can get all the info from nvidia's site:
http://www.nvnews.net/vbulletin/showthread.php?t=133112&page=1

or more directly this post:
http://www.nvnews.net/vbulletin/showpost.php?p=2041755&postcount=29

this may be out of context, but are u guys running with vsync enabled?

They're tracking this bug in Gentoo: http://bugs.gentoo.org/show_bug.cgi?id=274672

Doug Goldstein commented:

http://bugs.gentoo.org/show_bug.cgi?id=274672#c8
"The patch is a copy and paste of the patch posted by Christian from NVIDIA. He
however clearly states in his post that its a hack and not the correct/complete
solution.

I've discussed this issue with NVIDIA in detail and the net result is that
we're really going to recommend users of NVIDIA cards + OpenGL to steer clear
of 2.6.30 until a full work around the regression is completed and tested. It
will be out with the next driver bump."

Question being will that driver bump be done in time for Karmic?

Yes, as linked in my original posting with the code it is from NVIDIA.

Why should it be fixed in the NVIDIA driver when the problem is the kernel? The regression was from the kernel and PAT isn't functioning correctly in general.

I'm not clever enough to know if the change in the kernel constitutes a bug. Just commenting that, according to Doug's discussions, they're planning to fix it in the driver regardless. Afaik, this hasn't been filed in kernel's bugzilla - if you think it needs to, then by all means.

The bug in Gentoo has been marked as FIXED:

http://bugs.gentoo.org/show_bug.cgi?id=274672#c9
"Please upgrade to the in tree 185.x.y series for a resolution."

Has a fix been made available upstream? I didn't spot a patch in gentoo-commits, but I couldn't find a way to search the list either.

Fwiw, the problem still exists (testing Alpha 4 candidate ISO's)

apt-cache policy linux-image-generic
linux-image-generic:
  Installed: 2.6.31.5.16
  Candidate: 2.6.31.5.16
  Version table:
 *** 2.6.31.5.16 0
        500 http://gb.archive.ubuntu.com karmic/main Packages
        100 /var/lib/dpkg/statu

apt-cache policy nvidia-glx-180
nvidia-glx-180:
  Installed: 185.18.14-0ubuntu3
  Candidate: 185.18.14-0ubuntu3
  Version table:
 *** 185.18.14-0ubuntu3 0
        500 http://gb.archive.ubuntu.com karmic/restricted Packages
        100 /var/lib/dpkg/status

uname -a
Linux yonah-nix 2.6.31-5-generic #24-Ubuntu SMP Sat Aug 1 12:48:18 UTC 2009 i686 GNU/Linux

I emailed Doug over at Gentoo to confirm where/how it was fixed - nVidia have fixed the issue in the latest 185 driver. We have a PPA with that latest driver (https://launchpad.net/~nvidia-vdpau/+archive/ppa - 185.18.31) but unfortunately that driver is broken as per https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-180/+bug/408561, so I can't verify.

This is still happening with latest Karmic updates and Nvidia driver 185.

A workaround is available at http://www.nvnews.net/vbulletin/showpost.php?p=2041755&postcount=29

Francesco,

Workaround != Fix

That patch has been around for a while, and as indicated in previous comments it's not considered to be applied to third party installed patched versions, only the vanilla installed NVIDIA Binary.

oops, sorry... I've applied the workaround locally then and wait for a
proper fix.

2009/8/31 EclipseAgent <email address hidden>:
> Francesco,
>
> Workaround != Fix
>
> That patch has been around for a while, and as indicated in previous
> comments it's not considered to be applied to third party installed
> patched versions, only the vanilla installed NVIDIA Binary.
>
> --
> Compiz Slow on Karmic w/ NVIDIA and 2.6.30 Kernel
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>
> Status in “nvidia-graphics-drivers-180” package in Ubuntu: Confirmed
>
> Bug description:
> Compiz is extremely slow when using the NVIDIA Proprietary driver and Kernel 2.6.30 which ships with Karmic.
>
> I'm running Karmic (Ubutnu 9.10) and the 180 package from non-free repositories (and not installed from nvidia binary).
>
> I am running kernel 2.6.30-10 generic.. the most recent one released..
>
> Thank you
>

Any update on this issue?

Still a problem on 185.18.36.

Related discussion at http://www.nvnews.net/vbulletin/showthread.php?p=2041755#post2041755.

See post #50 in that forum thread for current status.

In the meantime, the post at http://www.nvnews.net/vbulletin/showpost.php?p=2041996&postcount=37 shows how to add the workaround patch to DKMS.

I should note - I just applied the workaround patch with no effect, 2.6.31-9-generic and 185.18.36.

Gabriel,

If I understand the regression correctly it should be fixed when you go to a 2.6.31 kernel since the PAT code was again changed.

I'll verify and try to find the commit.

Will it be solved for Karmic before release time?

On Sun, Sep 6, 2009 at 10:12 AM, EclipseAgent <email address hidden> wrote:

> Gabriel,
>
> If I understand the regression correctly it should be fixed when you go
> to a 2.6.31 kernel since the PAT code was again changed.
>
> I'll verify and try to find the commit.
>
> --
> Compiz Slow on Karmic w/ NVIDIA and 2.6.30 Kernel
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>

--
Manuel Mas

I see identical behaviour on both the .31 and .30 kernels. The workaround patch has no effect that I can see.

I can confirm this as well. Just updated and rebooted.

Interesting tidbits from kern.log:

Sep 10 18:12:53 Phoenix kernel: [ 5.867229] nvidia: module license 'NVIDIA' taints kernel.
Sep 10 18:12:53 Phoenix kernel: [ 5.867235] Disabling lock debugging due to kernel taint
Sep 10 18:12:53 Phoenix kernel: [ 5.876511] dcdbas dcdbas: Dell Systems Management Base Driver (version 5.6.0-3.2)
Sep 10 18:12:53 Phoenix kernel: [ 5.877300] intel_rng: FWH not detected
Sep 10 18:12:53 Phoenix kernel: [ 6.126223] nvidia 0000:01:00.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
Sep 10 18:12:53 Phoenix kernel: [ 6.126232] nvidia 0000:01:00.0: setting latency timer to 64
Sep 10 18:12:53 Phoenix kernel: [ 6.126383] NVRM: loading NVIDIA UNIX x86 Kernel Module 185.18.36 Fri Aug 14 17:18:04 PDT 2009
Sep 10 18:12:53 Phoenix kernel: [ 6.126456] NVRM: CPU does not support the PAT, falling back to MTRRs.

This is from /var/log/Xorg.0.log

(II) LoadModule: "glx"
(II) Loading /usr/lib/xorg/modules/extensions//libglx.so
dlopen: /usr/lib/xorg/modules/extensions//libglx.so: undefined symbol: _nv000028gl
(EE) Failed to load /usr/lib/xorg/modules/extensions//libglx.so
(II) UnloadModule: "glx"
(EE) Failed to load module "glx" (loader failed, 7)

I have also attached a copy of my nvidia bug report log

I am by no means an expert, but part of the problem seems to be that the GLX Module is not being loaded. I hope that this info helps resolve the problem.

BUMP.

Can someone please take a look at this bug? This will be a showstopper for people with NVIDIA cards using Karmic!!

Currently, I can only get to a CLI Interface...

This still does not work on Ubuntu Karmic Alpha 6. Any ideas when this will be fixed?

I can confirm, this is also broken on Alpha 6.

My output when I run Compiz --replace in the terminal:

diggs@Phoenix:~$ compiz --replace
Checking for Xgl: not present.
xset q doesn't reveal the location of the log file. Using fallback /var/log/Xorg.0.log
Detected PCI ID for VGA:
Checking for texture_from_pixmap: present.
Checking for non power of two support: present.
Checking for Composite extension: present.
Checking screen 1Comparing resolution (1280x800) to maximum 3D texture size (4096): Passed.
Checking for Software Rasterizer: Not present.
Checking for nVidia: present.
Checking for FBConfig: present.
Checking for Xgl: not present.
Starting gtk-window-decorator
/usr/bin/compiz.real (video) - Warn: No 8 bit GLX pixmap format, disabling YV12 image format
diggs@Phoenix:~$

NVIDIA Restricted Drivers: 185.18.36
Kernel: 2.6.31-10-generic

Please let me know what I can add or try to get this bug squashed....

I want to add that I also have this bug. Card is: nVidia Corporation G72M [Quadro NVS 110M/GeForce Go 7300] (rev a1). Computer is a Dell latitude D620.

Karmic working ok if compiz is not enabled.

I am constantly updating Ubuntu Karmic and have to report I am still getting
this bug with Compiz. Driver works fine if effects are turned off (compiz).

As there is a fix available, I'm wondering why it is not included yet. Because of this bug, I have to install the drivers manually which is quite annoying after wach kernel upgrade,

@Ernst
NVIDIA made it very clear that "patch is not fit for inclusion in third-party NVIDIA driver packages":
http://www.nvnews.net/vbulletin/showpost.php?p=2041755&postcount=29

That's true, but in that thread everyone who experienced this problem has been running the patch without problems. So, if we can predict which systems are affected (it seems a combination of certain mobile processors and gpu's), Ubuntu can offer this 'experimental' driver for which we expect it works better than the unmodified driver.

Can you attach the output of the following command, please?
grep flags /proc/cpuinfo

$ grep flags /proc/cpuinfo
flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx constant_tsc arch_perfmon bts pni monitor est tm2 xtpr pdcm
flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx constant_tsc arch_perfmon bts pni monitor est tm2 xtpr pdcm

Just wanted to say Nvidia works fine for me under Karmic with compiz on my MacBook Pro 5,1:

lspci -v:
2:00.0 VGA compatible controller: nVidia Corporation G96 [GeForce 9600M GT] (rev a1)
 Subsystem: Apple Computer Inc. Device 00a9
 Flags: bus master, fast devsel, latency 0, IRQ 23
 Memory at a4000000 (32-bit, non-prefetchable) [size=16M]
 Memory at 80000000 (64-bit, prefetchable) [size=256M]
 Memory at a2000000 (64-bit, non-prefetchable) [size=32M]
 I/O ports at 3000 [size=128]
 [virtual] Expansion ROM at a5000000 [disabled] [size=512K]
 Capabilities: <access denied>
 Kernel driver in use: nvidia
 Kernel modules: nvidia, nvidiafb

$ grep flags /proc/cpuinfo
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc arch_perfmon pebs bts rep_good pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 lahf_lm tpr_shadow vnmi flexpriority
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm constant_tsc arch_perfmon pebs bts rep_good pni dtes64 monitor ds_cpl vmx smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 lahf_lm tpr_shadow vnmi flexpriority

So this clearly doesn't affect all Nvidia users under Karmic.

The GPU is largely irrelevant (other than being an nVidia card), the bug
stems from an erratum, at least affecting the Intel T2500. The nVidia driver
was relying on certain behaviours in the kernel pertaining to the CPU which
changed in 2.6.30.

2009/9/29 Alex Murray <email address hidden>

> Just wanted to say Nvidia works fine for me under Karmic with compiz on
> my MacBook Pro 5,1:
>
> lspci -v:
> 2:00.0 VGA compatible controller: nVidia Corporation G96 [GeForce 9600M GT]
> (rev a1)
> Subsystem: Apple Computer Inc. Device 00a9
> Flags: bus master, fast devsel, latency 0, IRQ 23
> Memory at a4000000 (32-bit, non-prefetchable) [size=16M]
> Memory at 80000000 (64-bit, prefetchable) [size=256M]
> Memory at a2000000 (64-bit, non-prefetchable) [size=32M]
> I/O ports at 3000 [size=128]
> [virtual] Expansion ROM at a5000000 [disabled] [size=512K]
> Capabilities: <access denied>
> Kernel driver in use: nvidia
> Kernel modules: nvidia, nvidiafb
>
> $ grep flags /proc/cpuinfo
> flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
> cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm
> constant_tsc arch_perfmon pebs bts rep_good pni dtes64 monitor ds_cpl vmx
> smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 lahf_lm tpr_shadow vnmi flexpriority
> flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca
> cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx lm
> constant_tsc arch_perfmon pebs bts rep_good pni dtes64 monitor ds_cpl vmx
> smx est tm2 ssse3 cx16 xtpr pdcm sse4_1 lahf_lm tpr_shadow vnmi flexpriority
>
> So this clearly doesn't affect all Nvidia users under Karmic.
>
> --
> Compiz Slow on Karmic w/ NVIDIA and 2.6.30 Kernel
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>

lspci -v:
01:00.0 VGA compatible controller: nVidia Corporation G72M [Quadro NVS 110M/GeForce Go 7300] (rev a1)
 Subsystem: Dell Device 01cc
 Flags: bus master, fast devsel, latency 0, IRQ 16
 Memory at ed000000 (32-bit, non-prefetchable) [size=16M]
 Memory at d0000000 (64-bit, prefetchable) [size=256M]
 Memory at ee000000 (64-bit, non-prefetchable) [size=16M]
 [virtual] Expansion ROM at ef000000 [disabled] [size=128K]
 Capabilities: <access denied>
 Kernel driver in use: nvidia
 Kernel modules: nvidia, nvidiafb

grep flags /proc/cpuinfo:

flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx constant_tsc arch_perfmon bts pni monitor vmx est tm2 xtpr pdcm
flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx constant_tsc arch_perfmon bts pni monitor vmx est tm2 xtpr pdcm

ok, so the problem seems to affect CPUs which do not support "pat" (mine does it).

Alberto, that appears to be correct. My CPU does not support PAT and I have this problem.

While the workaround patch may not be "fit" for general release, it is effective for the time being and it's pretty trivial. Any chance of including it until nvidia decides to make their driver behave with 2.6.3x?

This is a pretty big regression from Jaunty, and the nvidia folks don't seem to consider it a priority.

+1: Compiz is also not working for me since the upgrade to Karmic. Jaunty was fine though. I attached the contents of /proc/cpuinfo in case this helps.

Otherwise, Karmic rocks :-)

I applied the patch provided by EclipseAgent (comment #15) and Compiz is now working again on my T2300 CPU with nVidia 7400 Go GPU. To install the patch, I used the DKMS approach: just put the patch file in the folder /usr/src/nvidia-185.18.36/patches/ and name it "pat.patch". Then add the following two lines to the bottom of /usr/src/nvidia-185.18.36/dkms.conf:

PATCH[1]="pat.patch"
PATCH_MATCH[1]="2.6.31"

After that, you have to recompile the kernel module nvidia and reboot afterwards:

$ sudo dkms remove -m nvidia -v 185.18.36 --all
$ sudo dkms add -m nvidia -v 185.18.36
$ sudo dkms build -m nvidia -v 185.18.36

In case Ubuntu releases a new version of nvidia-glx-*, you will most likely have to repeat the above procedure, but of course with different version numbers - that is, if the bug is still alive then ;-)

I have tried with kubuntu and I get a black screen during the transition between the splash and the desktop, obviously makes the desktop unusable

Using a T2300 CPU with nVidia 7400 Go GPU

No problems with ubuntu though

Hope it's solved soon, I can't wait to move to karmic!

Oh, and this is the patch that worked for me:

--- nv/nv.c 2009-05-27 03:20:53.000000000 -0700
+++ nv/nv.572433.c 2009-07-05 11:31:27.635210006 -0700
@@ -888,6 +888,7 @@
     unsigned int pat1, pat2, i;
     U008 PAT_WC_index;

+#if 0 /* disabled */
     if (!test_bit(X86_FEATURE_PAT,
             (volatile unsigned long *)&boot_cpu_data.x86_capability))
     {
@@ -895,6 +896,7 @@
             "NVRM: CPU does not support the PAT, falling back to MTRRs.\n");
         return NV_PAT_MODE_DISABLED;
     }
+#endif

     NV_READ_PAT_ENTRIES(pat1, pat2);
     PAT_WC_index = 0xf;

The above patch worked for me as well. I now have Compiz running.

I keep getting an error when dkms tries to apply the patch. The weird thing is that I tried this in Ubuntu and it worked fine and now I'm just trying it in Kubuntu. The error:

Kernel preparation unnecessary for this kernel. Skipping...
applying patch rt_preempt_31.patch...patching file nv-linux.h

applying patch pat.patch...patching file nv.c
patch: **** malformed patch at line 10: @@ -895,6 +896,7 @@

(bad exit status: 2)

Error! Application of patch pat.patch failed.
Check /var/lib/dkms/nvidia/185.18.36/build/ for more information.

Any ideas?

Hi
My laptop is an HP Pavilion 6000, with a G72M (GeForce Go 7400) chipset and the following CPU profile :

processor : 1
vendor_id : GenuineIntel
cpu family : 6
model : 14
model name : Genuine Intel(R) CPU T2250 @ 1.73GHz
stepping : 8
cpu MHz : 800.000
cache size : 2048 KB
.....

cpuid level : 10
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx constant_tsc arch_perfmon bts pni monitor est tm2 xtpr pdcm

And objousvly I exprerienced the same problem.

The latest driver 185.18.36 runs in the same way you described WITH and WITHOUT the workaround above (that looking at the code seems to be a test exclusion but does not solve the problem).
I tested everything with kernels x.31-12 and x.31-13.

I tested a new beta driver frome the NVIDIA site (the 190 beta) by a manual installation and it does not solve the problem so I reversed back to Ubuntu version (185).

However I think this is a strange behavior because the glx module is properly installed and "glxinfo" returns a properly running extension :

name of display: :0.0
display: :0 screen: 0
direct rendering: Yes
server glx vendor string: NVIDIA Corporation
....

glxgears runs properly and Google earth it's extremely fast, but there are problem with COMPIZ and with KDE also that does not use COMPIZ.

I don't know why but probably this is due to an OPENGL feature that fall back to a software implementation if the CPU does not have PAT .

I am in doubt about a kernel patch solution or a simple test removal, I think this is an architectural problem into the driver that dorce PAT instead of MTRR , and the new 190 driver still does not work.

Bye.

This patch also don't solve my problems, well, the video is not so slow, to be honest it is fast for most things but not fast as 9.04 and is really hard to watch videos with Composite enabled.

The problem occurs not only with Compiz enabled, it also happens when enabled composite on Metacity.

Also, I don't think that it is only a NVidia problem because I have a laptop with Intel 965 and it is also slow when composite enabled. When I say "slow" I mean slow compared as example to other distros.

@Everaldo: I think you suffer from a different bug: Compiz is something else as composite. Furthermore, this bug is solely targeted at certain nvidia gpu's in a mobile configuration: that has been confirmed by the nvidia devs. So, I think you should fill a new bug report.

Those nvidia devs have issued a new patch (http://www.nvnews.net/vbulletin/showpost.php?p=2100096&postcount=59), which will be included in new releases. (This new patch is different from the old one!) The current version is 190.36, which should be the last driver without patch included. I think the solution for this bug is to include this new driver release (release date is not known)? Maybe Karmic can offer this new driver as nvidia-190, and offer it as default closed source driver for cpu's without pat?

@Craig: Maybe you can get some help in the thread I just posted (http://www.nvnews.net/vbulletin/showthread.php?t=133112)? I think the nvidia devs are interested in cases the patch does not work correctly.

@Ernst: Thanks, but I managed to figure out with some (ok most ;) help from #ubuntu+1 . The patching process still doesn't work, so I just inserted the two small pieces of code manually.

Just a side note...I just ran updates today (first time since Wednesday...its now Saturday), and I had to re-run the patching procedure. Not a big deal. Is there some way to make script for this? My scripts tend to break things really bad and I would rather not do THAT to everyone else. What needs to happen is this (taken from #52): Keep in mind...This ONLY works if you have successfully applied the patch PREVIOUSLY:

(1. ) add the following two lines to the bottom of /usr/src/nvidia-185.18.36/dkms.conf:

PATCH[1]="pat.patch"
PATCH_MATCH[1]="2.6.31"

(2. ) After that, you have to recompile the kernel module nvidia and reboot afterwards:

$ sudo dkms remove -m nvidia -v 185.18.36 --all
$ sudo dkms add -m nvidia -v 185.18.36
$ sudo dkms build -m nvidia -v 185.18.36

Look here: http://ubuntuforums.org/showthread.php?p=8034561#post8034561

On Sun, Oct 11, 2009 at 01:41, Diggs808 <email address hidden> wrote:

> Just a side note...I just ran updates today (first time since
> Wednesday...its now Saturday), and I had to re-run the patching
> procedure. Not a big deal. Is there some way to make script for this?
> My scripts tend to break things really bad and I would rather not do
> THAT to everyone else. What needs to happen is this (taken from #52):
> Keep in mind...This ONLY works if you have successfully applied the
> patch PREVIOUSLY:
>
> (1. ) add the following two lines to the bottom of
> /usr/src/nvidia-185.18.36/dkms.conf:
>
> PATCH[1]="pat.patch"
> PATCH_MATCH[1]="2.6.31"
>
> (2. ) After that, you have to recompile the kernel module nvidia and
> reboot afterwards:
>
> $ sudo dkms remove -m nvidia -v 185.18.36 --all
> $ sudo dkms add -m nvidia -v 185.18.36
> $ sudo dkms build -m nvidia -v 185.18.36
>
> --
> Compiz broken on Karmic with NVIDIA driver and CPU without PAT support
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>

Ok, sorry, my fault.
The reinstallation of the Nvidia Sources and the complete procedure showed by Diggs808, solve any problem.
I noticed a slight decrease of performances in GL but everything seems to run properly now, my card has understood that I want to use accelerations.
I have not jet understood, why someone includes that test into the nvidia.ko kernel module, and if this patch does not produces any problem in future, but my sistem goes right now and this is ok.
Thanks !

Alberto/Mario,

I think maybe it is time to get the patch included now that more testing has taken place:
http://www.nvnews.net/vbulletin/showpost.php?p=2100983&postcount=63
10-08-09, 06:52 PM
Nvidia: "Ok, thanks for giving the patch a try and providing feedback. It should be included in future NVIDIA Linux graphics driver releases."

It would be nice to get it merged soon so we can have working+tested compiz+nvidia on the GA release of 9.10.

Thanks

I discussed this with Nvidia and I'll include the following patch ASAP:
http://www.nvnews.net/vbulletin/showpost.php?p=2100096&postcount=59

This bug was fixed in the package nvidia-graphics-drivers-180 - 185.18.36-0ubuntu7

---------------
nvidia-graphics-drivers-180 (185.18.36-0ubuntu7) karmic; urgency=low

  * debian.binary/patches, dkms.conf.in:
    - fall_back_on_mtrr_if_no_pat.patch: Fall back on MTRRs if the
      CPU doesn't support the PAT (LP: #391461).

 -- Alberto Milone <email address hidden> Wed, 14 Oct 2009 14:30:13 +0200

Patch fall_back_on_~_no_pat.patch does not solve problem for my G72M [Quadro NVS 110M/GeForce Go 7300] card on ASUS A6 notebook. CPU info attached. Xorg.log stil says:

(II) LoadModule: "glx"
(II) Loading /usr/lib/xorg/modules/extensions//libglx.so
dlopen: /usr/lib/xorg/modules/extensions//libglx.so: undefined symbol: _nv000028gl
(EE) Failed to load /usr/lib/xorg/modules/extensions//libglx.so
(II) UnloadModule: "glx"
(EE) Failed to load module "glx" (loader failed, 7)

Kernel and NVIDIA drivers updates 2009-10-15,

2.6.31-14-generic #46-Ubuntu SMP Tue Oct 13 16:47:59 UTC 2009 i686 GNU/Linux

Forget to mention that switching to text console now works. When I turn on visual effects in KDE, some effects (i.e. transparency) work, but after some use CPU load goes 100%.

Oleksiy,

what's the output of the following command?
apt-cache policy nvidia-glx-185

I can confirm that with latest updates on Karmic, this error has been fixed.

On Mon, Oct 5, 2009 at 9:09 AM, ixothym <email address hidden> wrote:

> I applied the patch provided by EclipseAgent (comment #15) and Compiz is
> now working again on my T2300 CPU with nVidia 7400 Go GPU. To install
> the patch, I used the DKMS approach: just put the patch file in the
> folder /usr/src/nvidia-185.18.36/patches/ and name it "pat.patch". Then
> add the following two lines to the bottom of
> /usr/src/nvidia-185.18.36/dkms.conf:
>
> PATCH[1]="pat.patch"
> PATCH_MATCH[1]="2.6.31"
>
> After that, you have to recompile the kernel module nvidia and reboot
> afterwards:
>
> $ sudo dkms remove -m nvidia -v 185.18.36 --all
> $ sudo dkms add -m nvidia -v 185.18.36
> $ sudo dkms build -m nvidia -v 185.18.36
>
> In case Ubuntu releases a new version of nvidia-glx-*, you will most
> likely have to repeat the above procedure, but of course with different
> version numbers - that is, if the bug is still alive then ;-)
>
> --
> Compiz Slow on Karmic w/ NVIDIA and 2.6.30 Kernel
> https://bugs.launchpad.net/bugs/391461
> You received this bug notification because you are a direct subscriber
> of the bug.
>

--
Manuel Mas

>Oleksiy,
>what's the output of the following command?
>apt-cache policy nvidia-glx-185
It says:
root@lukin:~# apt-cache policy nvidia-glx-185
nvidia-glx-185:
  Installed: 185.18.36-0ubuntu7
  Candidate: 185.18.36-0ubuntu7
  Version table:
 *** 185.18.36-0ubuntu7 0
        500 http://ua.archive.ubuntu.com karmic/restricted Packages
        100 /var/lib/dpkg/status

This bug was fixed in the package nvidia-graphics-drivers-96 - 96.43.13-0ubuntu4

---------------
nvidia-graphics-drivers-96 (96.43.13-0ubuntu4) karmic; urgency=low

  * debian.binary/patches, dkms.conf.in:
    - fall_back_on_mtrr_if_no_pat.patch: Fall back on MTRRs if the
      CPU doesn't support the PAT (LP: #391461).

 -- Alberto Milone <email address hidden> Fri, 16 Oct 2009 11:36:09 +0200

This bug was fixed in the package nvidia-graphics-drivers-173 - 173.14.20-0ubuntu3

---------------
nvidia-graphics-drivers-173 (173.14.20-0ubuntu3) karmic; urgency=low

  * debian.binary/patches, dkms.conf.in:
    - fall_back_on_mtrr_if_no_pat.patch: Fall back on MTRRs if the
      CPU doesn't support the PAT (LP: #391461).

 -- Alberto Milone <email address hidden> Fri, 16 Oct 2009 11:30:55 +0200

@64 Albero Milione

That patch has been included in today update of the 185.18.36 drivers as

fall_back_on_mtrr_if_no_pat.patch

with the specific processors profile.

Everthing seems to be ok now.

What to look on and what to check if it does not work with latest updates?
I'm stil getting in Xorg.log:

 "Failed to load /usr/lib/xorg/modules/extensions//libglx.so"

dmesg says:

[ 19.300347] nvidia: module license 'NVIDIA' taints kernel.
[ 19.300355] Disabling lock debugging due to kernel taint
[ 19.566799] intel_rng: FWH not detected
[ 19.574817] nvidia 0000:01:00.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
[ 19.574831] nvidia 0000:01:00.0: setting latency timer to 64
[ 19.575039] NVRM: loading NVIDIA UNIX x86 Kernel Module 185.18.36 Fri Aug 14 17:18:04 PDT 2009

There's no message defined in patch:

 "NVRM: CPU does not support the PAT, falling back to MTRRs.\n"

So I suspect that cdriver thinks taht PAT available bit it is not because /proc/cpuinfo says:

flags : fpu vme de pse tsc msr mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss tm pbe up bts est tm2

Works for me with the latest driver (185.18.36-0ubuntu7) ubuntu and kubuntu

Stil does not work on my Asus A6 notebook.

Xorg.log:

(II) LoadModule: "glx"
(II) Loading /usr/lib/xorg/modules/extensions//libglx.so
dlopen: /usr/lib/xorg/modules/extensions//libglx.so: undefined symbol: _nv000028gl
(EE) Failed to load /usr/lib/xorg/modules/extensions//libglx.so
(II) UnloadModule: "glx"
(EE) Failed to load module "glx" (loader failed, 7)

dmesg:

[ 9.418992] nvidia: module license 'NVIDIA' taints kernel.
[ 9.418999] Disabling lock debugging due to kernel taint
[ 9.677530] nvidia 0000:01:00.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
[ 9.677544] nvidia 0000:01:00.0: setting latency timer to 64
[ 9.677738] NVRM: loading NVIDIA UNIX x86 Kernel Module 185.18.36 Fri Aug 14 17:18:04 PDT 2009
[ 9.677821] NVRM: CPU does not support the PAT, falling back to MTRRs.

As it can be seen, patch discovers absence of PAT support, but MTRR code probably is still broken.

I have the same problem (even with the first patch) until the version 185.18.36-0ubuntu7 solves it. (Karmic rocks!) Here is the output of "x86info -f". Strangely it reports "pat" as a feature flag.

Found 2 CPUs
--------------------------------------------------------------------------
CPU #1
EFamily: 0 EModel: 0 Family: 6 Model: 14 Stepping: 8
CPU Model: Unknown model.
Processor name string: Genuine Intel(R) CPU T2400 @ 1.83GHz
Type: 0 (Original OEM) Brand: 0 (Unsupported)
Number of cores per physical package=2
Number of logical processors per socket=2
Number of logical processors per core=1
APIC ID: 0x0 Package: 0 Core: 0 SMT ID 0
Feature flags:
 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat clflsh ds acpi mmx fxsr sse sse2 ss ht tm pbe
Extended feature flags:
 sse3 monitor vmx est tm2 xTPR [15]

--------------------------------------------------------------------------
CPU #2
EFamily: 0 EModel: 0 Family: 6 Model: 14 Stepping: 8
CPU Model: Unknown model.
Processor name string: Genuine Intel(R) CPU T2400 @ 1.83GHz
Type: 0 (Original OEM) Brand: 0 (Unsupported)
Number of cores per physical package=2
Number of logical processors per socket=2
Number of logical processors per core=1
APIC ID: 0x1 Package: 0 Core: 0 SMT ID 1
Feature flags:
 fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat clflsh ds acpi mmx fxsr sse sse2 ss ht tm pbe
Extended feature flags:
 sse3 monitor vmx est tm2 xTPR [15]

--------------------------------------------------------------------------

On my Asus A6 some strange things are happening. Kernel says CPU does not support PAT, but xorg sees it.

 x86info -f
x86info v1.24. Dave Jones 2001-2009
Feedback to <email address hidden>.

Found 1 CPU
--------------------------------------------------------------------------
EFamily: 0 EModel: 0 Family: 6 Model: 13 Stepping: 6
CPU Model: Pentium M (Dothan) [B-1]
Processor name string: Intel(R) Pentium(R) M processor 1.80GHz
Type: 0 (Original OEM) Brand: 6 (Mobile Intel® Pentium® III processor)
Number of cores per physical package=1
Number of logical processors per socket=1
Number of logical processors per core=1
APIC ID: 0x0 Package: 0 Core: 0 SMT ID 0
Feature flags:
 fpu vme de pse tsc msr mce cx8 apic sep mtrr pge mca cmov pat clflsh ds acpi mmx fxsr sse sse2 ss tm pbe
Extended feature flags:
 est tm2

Kernel definitely says that

[ 0.000000] PAT not supported by CPU.

/proc/cpuinfo flags are:
flags : fpu vme de pse tsc msr mce cx8 apic mtrr pge mca cmov clflush dts acpi mmx fxsr sse sse2 ss tm pbe up bts est tm2

Certainly something is broken either in kernel or in xorg.

Forget to say that "nouveau" driver does not work at all on my Asus A6 causing kernel memory oops but somehow worked in Fedora 11. On Karnic after re-installing nvidia-glx-185 glx extension module of xorg loaded but GL support itself does not work correctly in composite mode. The same situation with 190.x.x beta driver. The same situation with Fedora 11 last updates I had on notebook week ago. I tried almost every variant available but composite opengl mode does not work.

So it is common problem, I guess, and the roots of problem is kernel. My notebook video worked perfectly until last kernel updates until May of this year. At least in Ubuntu Jaunty it worked perfect.

Finally found solution and compisite is fully functional.
Solution is to disable PAT support in kernel command line.

Edit file /etc/default/grub and add:

GRUB_CMDLINE_LINUX="nopat"

Then run update-grub as root and reboot.

Well, as I said, problem is in kernel PAT support for my Intel Pentium M (Dothan) [B-1]

This patch seems to break the 185 drivers for the RT kernel, causing X to fail to start. I removed it and things are working again. See this bug for more info and another patch we are using for the RT kernel:
https://bugs.launchpad.net/ubuntu/+source/nvidia-graphics-drivers-180/+bug/413296