Bug 43405

Can you bisect?

git bisect in progress--it goes back to August.
I am gettting different appearances with the different revisions
(snow, blank windows, etc.).

Did you update any other things around the same time?  kernel?  ddx?

I've updated the kernel since, but I know the old version is good because I still have the binaries. I'd switched to fglrx due to thermal reasons (a mix of dust, weather, and bad power management in the driver).
I did not update any other relevant components.

I've gotten a few good results.  It looks like it was one of the Evergreen support fixes that broke it; the first one works, and I've got about 2 revisions left to go.

Git bisect completed.
Here's what I find:

751a6ed893d393eaea266d892402d132f7d15a7d is the first bad commit
commit 751a6ed893d393eaea266d892402d132f7d15a7d
Author: Marek Olšák <maraeo@gmail.com>
Date:   Fri Aug 19 21:48:12 2011 +0200

    r600g: hack around a problem with texture alignment

:040000 040000 a52c6bc437438025f708ff5240db254a0b451f03 4978b1157ec6d23ec96dafe10245fd044d9fac2e M	src

Reverting the commit results in perfect display.

Created attachment 54051 [details] [review]
Revert texture hack

Revert the relevant commit.
This results in good graphics with both glxgears & Google Earth.

I've got a hunch that the texture hack is GPU-dependent. The comment next to the hack suggests it fixes a problem on Evergreen. The problem might have been fixed already by some other commit, not sure.

I am okay with reverting the commit.

Ideally piglit should be tested on Evergreen and if there are regressions, the hack should be applied to Evergreen only, or another solution should be found, but I guess that can be sorted out after the revert.

(In reply to comment #8)

I still get the same issue when using r600g driver from git mesa, no such issue
using mesa-8.0.2.
All 3D demos are affected as well as google earth and GLC_player 3D apps.

Hardware: ATI Radeon HD3200 RS780M/RS780MN

Debian Sid with the following packages updated:
Linux Kernel~3.3.0-1.dmz.1-liquorix-amd64
Latest git version of libdrm, libX11, libxcb, libXi, mesa and xserver.

After doing git bisect, I found this to be another issue.
I filed a new bug: #48747.

I think this issue has been fixed with the new libdrm surface allocator,
since the commit below removes the code in question.

commit 581f7e3101980a4e1068bb75c2eca60bb2071229
Author: Marek Olšák <maraeo@gmail.com>
Date:   Sun Jul 29 18:53:19 2012 +0200

    r600g: drop the old texture allocation code
    
    Made obsolete by the libdrm surface allocator.


Could you confirm that this issue is resolved on mesa git master or mesa git 9.0 branch?

On Mon, 03 Sep 2012 12:22:38 +0000
bugzilla-daemon@freedesktop.org wrote:

> https://bugs.freedesktop.org/show_bug.cgi?id=43405

> Could you confirm that this issue is resolved on mesa git master 
> or mesa git 9.0 branch?
Sorry to take so long replying; school's been keeping me busy.

At present, I'm not getting mesa to build:
#on master
make distclean
git pull
/autogen.sh --prefix=/opt/mesa9 --enable-driglx=direct \
--enable-gallium --enable-texture-float \
--with-dri-driverdir=/opt/mesa9/lib --with-state-trackers=glx,dri \
--without-dri-drivers --with-gallium-drivers=r600 --disable-egl
#succeeds
make -j1
..
make[4]: *** No rule to make target `../../../src/glsl/glsl_lexer.cc', needed by `glsl_lexer.lo'.  Stop.

This is still Debian Squeeze plus several backports; I've updated the kernel to 3.4.11, xcb to 1.8.1, and libdrm/libkms to 2.4.39.

Thanks,
Isaac Dunham <idunham@lavabit.com>

(In reply to comment #12)

Can you try git clean -fdx before autogen?

On Mon, 01 Oct 2012 07:20:49 +0000
bugzilla-daemon@freedesktop.org wrote:

> https://bugs.freedesktop.org/show_bug.cgi?id=43405
> 
> --- Comment #13 from Andreas Boll <andreas.boll.dev@gmail.com> ---
> (In reply to comment #12)
> 
> Can you try git clean -fdx before autogen?
> 
Before I got your message, I ran 
make -C src/glsl glsl_lexer.cc
which generated the file
After that, it built properly.
I'm still getting the same snow display.  I suspect it's an interaction with the Squeeze X stack, since I didn't get this with Precise (which has its own problems...)
I hope to try later when I've updated to the backports.org X server.

I do not see this with Debian Sid/unstable (Mesa 8.0.4) and HD 3200 (RS780).

    01:05.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI RS780 [Radeon HD 3200]

At least `glxgears` works fine. Any special prerequisites to run this with?

With debug message the following is written to `dmesg`.

[ 9350.289921] [drm:drm_mode_addfb], [FB:39]
[ 9350.289975] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.334609] [drm:drm_mode_addfb], [FB:36]
[ 9350.334646] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.349981] [drm:drm_mode_addfb], [FB:39]
[ 9350.350015] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.366670] [drm:drm_mode_addfb], [FB:36]
[ 9350.366701] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.383145] [drm:drm_mode_addfb], [FB:39]
[ 9350.383174] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.400070] [drm:drm_mode_addfb], [FB:36]
[ 9350.400104] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.416752] [drm:drm_mode_addfb], [FB:39]
[ 9350.416779] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.433387] [drm:drm_mode_addfb], [FB:36]
[ 9350.433418] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.449614] [drm:drm_mode_addfb], [FB:39]
[ 9350.449627] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.466732] [drm:drm_mode_addfb], [FB:36]
[ 9350.466766] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.483430] [drm:drm_mode_addfb], [FB:39]
[ 9350.483462] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.500105] [drm:drm_mode_addfb], [FB:36]
[ 9350.500134] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[ 9350.516427] [drm:drm_mode_addfb], [FB:39]
[ 9350.516445] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f6d77c00, cur_bbo = f72df800
[ 9350.533403] [drm:drm_mode_addfb], [FB:36]
[ 9350.533433] [drm:radeon_crtc_page_flip], flip-ioctl() cur_fbo = f72df800, cur_bbo = f6d77c00
[…]

On Mon, 01 Oct 2012 09:58:49 +0000
bugzilla-daemon@freedesktop.org wrote:

> --- Comment #15 from Paul Menzel <paulepanter@users.sourceforge.net> ---
> I do not see this with Debian Sid/unstable (Mesa 8.0.4) and HD 3200 (RS780).
> 
>     01:05.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI
> RS780 [Radeon HD 3200]
> 
> At least `glxgears` works fine. Any special prerequisites to run this with?

At present, I'm suspecting that it's something to do with the Squeeze X stack (I'm using Squeeze with the stock Xorg, and I've updated only the minimum to satisfy configure).

I think this has been fixed since the new surface allocator was added to libdrm. I haven't seen any issues with texture allocation for some time.