Bug 7767 - No hardware rendering for "SuperSavage/IXC 64" and "Xorg 7.2 RC3"
Summary: No hardware rendering for "SuperSavage/IXC 64" and "Xorg 7.2 RC3"
Status: CLOSED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: Video(DRI - non Intel) (show other bugs)
Hardware: i386 Linux
: P2 normal
Assignee: drivers_video-dri
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2007-01-04 00:23 UTC by Joachim Frieben
Modified: 2007-06-02 04:01 UTC (History)
2 users (show)

See Also:
Kernel Version: 2.6.20-rc3-git3
Subsystem:
Regression: ---
Bisected commit-id:


Attachments
Xorg.0.log for FC kernel 2.6.19-1.2917.fc7 [2.6.20rc7] (62.55 KB, text/plain)
2007-02-02 08:29 UTC, Joachim Frieben
Details
Xorg.0.log for FC kernel 2.6.19-1.2917.fc7 [2.6.20rc7] using DRM modules from the git tree (62.64 KB, text/plain)
2007-02-02 08:39 UTC, Joachim Frieben
Details
/var/log/Xorg.0.log (65.06 KB, text/plain)
2007-02-03 20:41 UTC, Sérgio M Basto
Details
Output of "dmesg" for 2.6.19-1.2917.fc7 stock kernel "DRM" modules with "debug=1" (28.18 KB, text/plain)
2007-02-04 02:13 UTC, Joachim Frieben
Details
Output of "dmesg" for 2.6.19-1.2917.fc7 "git" kernel "DRM" modules with "debug=1" (23.30 KB, text/plain)
2007-02-04 02:14 UTC, Joachim Frieben
Details
Output of "dmesg" for 2.6.20-1.2922.fc7 stock kernel "DRM" modules with "modprobe drm debug=1" (29.16 KB, text/plain)
2007-02-08 05:15 UTC, Joachim Frieben
Details
Output of "dmesg" for 2.6.20-1.2922.fc7 "git" kernel "DRM" modules with "modprobe drm debug=1" (34.68 KB, text/plain)
2007-02-09 01:21 UTC, Joachim Frieben
Details
Output of "dmesg" for 2.6.19-1.2911.6.3.fc6 [2.6.19.5rc1] "git" kernel tree "DRM" modules with "modprobe drm debug=1" (27.53 KB, text/plain)
2007-02-28 12:29 UTC, Joachim Frieben
Details
Output of "dmesg" for 2.6.20-1.2953.fc7 [2.6.21rc1] "git" kernel tree "DRM" modules with "modprobe drm debug=1" (30.69 KB, text/plain)
2007-03-02 11:08 UTC, Joachim Frieben
Details

Description Joachim Frieben 2007-01-04 00:23:30 UTC
Most recent kernel where this bug did *NOT* occur: none
Distribution: "Fedora Core [Development]"
Hardware Environment: "IBM ThinkPad T23"
Software Environment: n/a
Problem Description: Obsolete kernel "DRM" modules in kernel "2.6.20-rc3-git3"
inhibit hardware rendering for "SuperSavage/IXC 64". For the new upstream
version 2.1.12 of the "savage" driver in "Xorg 7.2 RC3", "glxinfo" reports:

  "libGL error: drmMap of framebuffer failed (Invalid argument)"
  "libGL error: reverting to (slow) indirect rendering"

This error was absent for driver version 2.1.11.

Steps to reproduce:
1. Install "savage" driver version 2.1.12.
2. Restart "X".
3. Log in to "X" session and execute "glxinfo".

Additional comment: After compiling the latest "drm.ko" and "savage.ko" from
[http://anongit.freedesktop.org/git/mesa/drm] and loading them prior to starting
the "X" server, hardware rendering is back and working as before.
Comment 1 Joachim Frieben 2007-01-04 00:27:30 UTC
Sorry, the correct link is [git://anongit.freedesktop.org/git/mesa/drm].
Comment 2 Andrew Morton 2007-01-31 02:03:07 UTC
hm, so the freedesktop.org guys have better drivers that Linus does?

Dave, will that stuff get synced up sometime?
Comment 3 Dave Airlie 2007-01-31 14:42:15 UTC
it should all be synced up for ages, I've no outstanding differences between the
2.6.20 savage and drm git tree that I can see with a quick look.

Can you send me a Xorg.0.log from the old and new modules?
Comment 4 Joachim Frieben 2007-02-02 08:29:37 UTC
Created attachment 10258 [details]
Xorg.0.log for FC kernel 2.6.19-1.2917.fc7 [2.6.20rc7]

The kernel is the Fedora flavour of 2.6.20rc7:

* Wed Jan 31 2007 Dave Jones <davej@redhat.com>
- 2.6.20rc7

The "X" log file contains an "AIGLX" related error message:

(EE) AIGLX error: drmMap of framebuffer failed (Invalid argument)(EE) AIGLX:
reverting to software rendering
Comment 5 Joachim Frieben 2007-02-02 08:39:52 UTC
Created attachment 10259 [details]
Xorg.0.log for FC kernel 2.6.19-1.2917.fc7 [2.6.20rc7] using DRM modules from the git tree

Xorg.0.log for FC kernel 2.6.19-1.2917.fc7 [2.6.20rc7]

The kernel is the Fedora flavour of 2.6.20rc7:

* Wed Jan 31 2007 Dave Jones <davej@redhat.com>
- 2.6.20rc7

This time, "drm.ko" and "savage.ko" have been built from the current "git" tree
and are loaded at the end of "rc.sysinit". Hardware rendering is up and running
albeit slow. When I run "glxgears", the fps are only half the value obtained
when running identical "Xorg 7.2 RC3" on an "FC5" system.
Comment 6 Joachim Frieben 2007-02-03 03:16:32 UTC
I cannot confirm the claim of comment #3. By merely looking at the handful of
"savage" related source files, it becomes clear that there actually -are-
differences between the "2.6.20rc7" kernel "DRM" and the one from the "git" tree.
Most are simple formatting issues but at least in "savage_drv.c", more
substantial things have changed, namely the "probe" function declared at line 36.
It looks like they are "oustanding" enough to break "DRI", so they are
apparently not that trivial.
Comment 7 Dave Airlie 2007-02-03 13:06:30 UTC
I think it is a change elsewhere in the drm that is affecting your system, just
comparing the files isn't valid around the savage in this case as the out of the
kernel drm does thing differently than the in-kernel one with respect to probing
the driver....
Comment 8 Joachim Frieben 2007-02-03 13:49:00 UTC
Ok, I never claimed anything different. However, I do continue to insist, that
albeit "2.6.20rc7" being a recent kernel, it is not up to date regarding the
"git" development tree. That was my one and only point.
Comment 9 Dave Airlie 2007-02-03 13:54:25 UTC
Okay I looked at the logs, you do have hardware rendering, dri works fine it is
in fact AIGLX that is broken, that is indeed strange, there must be something in
the drm map that is broken...
Comment 10 Dave Airlie 2007-02-03 14:10:55 UTC
And that is true at all stages, the git tree is a "development" tree, it isn't
always suitable for inclusion in the upstream kernel...  you can check if an -mm
kernel fixes it or not as that has a lot of patches queued, but I can't think if
anything is in that area...
Comment 11 Dave Airlie 2007-02-03 14:56:50 UTC
can you load the drm modules in both cases with debug=1 option and attach those
logs...

the drm buffer stuff is quite different between both trees at the moment due to
the ttm merge pending..
Comment 12 Sérgio M Basto 2007-02-03 20:41:02 UTC
Created attachment 10276 [details]
/var/log/Xorg.0.log

Hi, I tried to disable AIGLX
http://fedoraproject.org/wiki/RenderingProject/aiglx#head-936632a0253a7a1e6c04e697932546dde44175c1


but same error appears :
Error -12
Failed to initialize batch pool - possible incorrect agpgart installed
libGL warning: 3D driver returned no fbconfigs.
libGL error: InitDriver failed
libGL error: reverting to (slow) indirect rendering
on a i915 with kernel-2.6.19-1.2917.fc7 and git drm 

(WW) I810(0): Extended BIOS function 0x5f05 failed.
maybe this message say something.
Comment 13 Dave Airlie 2007-02-03 22:25:36 UTC
please don't attached other problems to this bug... that isn't the same problem
you are seeing, you have i915 reporter has a savage...
Comment 14 Joachim Frieben 2007-02-04 02:13:42 UTC
Created attachment 10277 [details]
Output of "dmesg" for 2.6.19-1.2917.fc7 stock kernel "DRM" modules with "debug=1"
Comment 15 Joachim Frieben 2007-02-04 02:14:42 UTC
Created attachment 10278 [details]
Output of "dmesg" for 2.6.19-1.2917.fc7 "git" kernel "DRM" modules with "debug=1"
Comment 16 Joachim Frieben 2007-02-04 02:48:51 UTC
Attachment http://bugzilla.kernel.org/attachment.cgi?id=10278&action=view turns
out to be not useful as turning on the "debug" flag inhibits direct rendering
completely: "glxinfo" now returns "direct rendering: No", and "dmesg" reports
many complaints about unknown symbols. After rebooting the system without the
"debug" flag, "glxinfo" again reports "direct rendering: Yes".
Relevant package info for the affected system is:

- kernel-2.6.19-1.2917.fc7 [2.6.20rc7]
- xorg-x11-server-Xorg-1.2.0-4.fc7 [X Window System Version 7.2.0]
- xorg-x11-drv-savage-2.1.2-1.fc7 [version 2.1.2]
- mesa-libGL[U]-6.5.2-4.fc7 [1.5 Mesa 6.5.2]
Comment 17 Sérgio M Basto 2007-02-05 20:51:04 UTC
Joachim had you try disable AIGLX ?

Section "ServerFlags"
    Option  "AIGLX" "off"
EndSection

Section "Extensions"
    Option "Composite" "Disable"
EndSection
Comment 18 Joachim Frieben 2007-02-08 02:41:36 UTC
In reply to comment #17:
After disabling "AIGLX" and extension "Composite" altogether, entries:

  (**) Option "AIGLX" "off"
  (**) Extension "Composite" is disabled

appear in "Xorg.0.log". The server still spports direct rendering:

  (II) SAVAGE(0): Direct rendering enabled

However, "glxinfo" still complains:

  libGL error: drmMap of framebuffer failed (Invalid argument)
  libGL error: reverting to (slow) indirect rendering
Comment 19 Dave Airlie 2007-02-08 02:46:37 UTC
please send the drm logs from the failure with glxinfo, please note you only add
debug=1 to the drm module not the savage module, which is  why it broke in the
other place...
Comment 20 Joachim Frieben 2007-02-08 05:15:09 UTC
Created attachment 10342 [details]
Output of "dmesg" for 2.6.20-1.2922.fc7 stock kernel "DRM" modules with "modprobe drm debug=1"

The section below the last entry [drm:drm_stub_open] got added after executing
"env LIBGL_DEBUG=verbose glxinfo"
Comment 21 Dave Airlie 2007-02-08 14:39:08 UTC
Okay can you try the git modules with debugging again and log that?

I'm not really sure what could be causing this,

make sure you rmmod the kernel drm and savage ones before installing the git
tree ones with debug=1 passed to drm..

the log you posted didn't do remove the old modules..
Comment 22 Joachim Frieben 2007-02-09 01:21:15 UTC
Created attachment 10364 [details]
Output of "dmesg" for 2.6.20-1.2922.fc7 "git" kernel "DRM" modules with "modprobe drm debug=1"

The section below the 2nd last entry [drm:drm_stub_open] got added after
executing
"env LIBGL_DEBUG=verbose glxinfo". Hardware rendering is working correctly,
albeit very slow. When I had installed "FC5", the fps of "glxgears" for both
"Xorg 7.0" and "Xorg 7.2 RC3" was about 190. Now, for "pre-FC7" and a "2.6.20"
kernel, even with active "DRI", the value drops to a value of about 80 fps.
Comment 23 Dave Airlie 2007-02-17 22:20:51 UTC
Okay I've added some changes to the drm-2.6 git tree to see if they fix this
problem..

Can you test the next -mm kernel when Andrew releases it or grab the tree from
drm-mm branch from
git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git

It may take a while to mirror out correctly..
497db0d3027679e7830c490c917d08f43aa9ee7b is the git head..
Comment 24 Joachim Frieben 2007-02-24 04:25:17 UTC
I do not see any difference between the "savage" source files ["savage_bci.c",
"savage_drm.h", "savage_drv.c", "savage_drv.h", "savage_state.c"] from kernel
"2.6.20-1.2940.fc7" [which still fails] and the "git" tree from
"www.kernel.org". However, they both differ from the corresponding files of the
"Mesa" tree at "www.freedesktop.org" which I use to build working modules
"drm.ko" and "savage.ko" against the kernel headers installed on my system.
Comment 25 Dave Airlie 2007-02-24 15:00:18 UTC
did you test it though?

As I said further up this bug, there are no "functional" difference between the
savage in the drm tree and the one in the kernel tree, they are just either
whitespace or a difference caused by the device binding methods used... so what
I did was bring over a bunch of drm core code that could affect the area you are
seeing the problem in..
Comment 26 Anonymous Emailer 2007-02-25 11:35:49 UTC
Reply-To: alanh@fairlite.demon.co.uk

I know what the problem is here. And bugzilla #5714 will cure it. So I'm
getting inclined to pull the frontbuffer-removal branch into master and
fix up the drivers on the fly.

How does this sound to others.

Alan.

On Sat, 2007-02-24 at 15:00 -0800, bugme-daemon@bugzilla.kernel.org
wrote:
> http://bugzilla.kernel.org/show_bug.cgi?id=7767
> 
> 
> 
> 
> 
> ------- Additional Comments From airlied@linux.ie  2007-02-24 15:00 -------
> did you test it though?
> 
> As I said further up this bug, there are no "functional" difference between the
> savage in the drm tree and the one in the kernel tree, they are just either
> whitespace or a difference caused by the device binding methods used... so what
> I did was bring over a bunch of drm core code that could affect the area you are
> seeing the problem in..
> 
> ------- You are receiving this mail because: -------
> You are the assignee for the bug, or are watching the assignee.
> 
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share your
> opinions on IT & business topics through brief surveys-and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> --
> _______________________________________________
> Dri-devel mailing list
> Dri-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/dri-devel

Comment 27 Joachim Frieben 2007-02-28 12:19:29 UTC
Reply to comment #25:

In the meantime, I have reverted my system from "FC rawhide" to "FC6", and so I
have pulled in your "DRM" tree from "kernel.org" into the
"kernel-2.6.19-1.2911.6.3.fc6" source package which, a priori, is only
"2.6.19.5rc1" based. Nevertheless, modules "drm.ko" and "savage.ko" got built
without complaint, and it is also possible to load them. However, I still do
obtain the message "libGL error: drmMap of framebuffer failed (Invalid argument)
.." when launching "glxinfo", whereas the modules built from the
"freedesktop.org" tree work correctly [The next time, I try "FC test" or "FC
rawhide" again, I will be able to report results for version "2.6.20" and later].
Comment 28 Joachim Frieben 2007-02-28 12:29:17 UTC
Created attachment 10558 [details]
Output of "dmesg" for 2.6.19-1.2911.6.3.fc6 [2.6.19.5rc1] "git" kernel tree "DRM" modules with "modprobe drm debug=1"
Comment 29 Joachim Frieben 2007-03-02 11:08:57 UTC
Created attachment 10584 [details]
Output of "dmesg" for 2.6.20-1.2953.fc7 [2.6.21rc1] "git" kernel tree "DRM" modules with "modprobe drm debug=1"

I have merged the "DRM" source from:

  git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6.git

as of 2007-03-02 with the "2.6.20-1.2953.fc7" [2.6.21rc1] kernel package from
the "FC development" tree. Unfortunately still no success: after building and
loading modules "drm.ko" and "savage.ko" via "insmod" after booting into text
mode and launching the "X" server afterwards, I still obtain the familiar error
message: "libGL error: drmMap of framebuffer failed (Invalid argument) .." when
I execute "glxinfo". In "Xorg.0.log" appears the error message:
"(EE) AIGLX error: drmMap of framebuffer failed (Invalid argument)
 (EE) AIGLX: reverting to software rendering"
Comment 30 Anonymous Emailer 2007-03-09 12:53:36 UTC
Reply-To: alanh@fairlite.demon.co.uk

I know what the problem is here. And bugzilla #5714 will cure it. So I'm
getting inclined to pull the frontbuffer-removal branch into master and
fix up the drivers on the fly.

How does this sound to others.

Alan.

On Sat, 2007-02-24 at 15:00 -0800, bugme-daemon@bugzilla.kernel.org
wrote:
> http://bugzilla.kernel.org/show_bug.cgi?id=7767
> 
> 
> 
> 
> 
> ------- Additional Comments From airlied@linux.ie  2007-02-24 15:00 -------
> did you test it though?
> 
> As I said further up this bug, there are no "functional" difference between the
> savage in the drm tree and the one in the kernel tree, they are just either
> whitespace or a difference caused by the device binding methods used... so what
> I did was bring over a bunch of drm core code that could affect the area you are
> seeing the problem in..
> 
> ------- You are receiving this mail because: -------
> You are the assignee for the bug, or are watching the assignee.
> 
> -------------------------------------------------------------------------
> Take Surveys. Earn Cash. Influence the Future of IT
> Join SourceForge.net's Techsay panel and you'll get the chance to share your
> opinions on IT & business topics through brief surveys-and earn cash
> http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
> --
> _______________________________________________
> Dri-devel mailing list
> Dri-devel@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/dri-devel


-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
--
_______________________________________________
Dri-devel mailing list
Dri-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/dri-devel

Comment 31 Joachim Frieben 2007-04-10 03:09:46 UTC
Issue still open for kernel "2.6.21-rc6-git1". Anything I can do here?
Comment 32 Joachim Frieben 2007-05-16 00:39:05 UTC
"DRM" still broken in "2.6.21.1". Building "drm.ko" and "savage.ko" from the
"freedesktop.org" development tree works again after some recent breakage, and
the resulting modules restore working hardware rendering.
Comment 33 Joachim Frieben 2007-06-02 04:01:22 UTC
Solved for me in kernel "2.6.22-rc3-git6". Thanks!

Note You need to log in before you can comment on or make changes to this bug.