Bug 20104 – media-video/ati-drivers-2.5.1-r1 doesn't work with sys-kernel/gentoo-sources-2.4.20-r3

• Product: Gentoo Linux
• Status: RESOLVED FIXED

[Description Ian Abbott 2003-04-28 10:22:45 UTC]

Emerging ati-drivers-2.5.1-r1 for gentoo-sources-2.4.20-r3 and xfree-4.2.1-r2 appears to 
build and install okay, but the X server fails to load the fglrx.o module. 

Reproducible: Always
Steps to Reproduce:
1. emerge gentoo-sources 
2. set /usr/src/linux symlink, build and install kernel 
3. reboot into new kernel 
4. emerge =x11-base/xfree-4.2.1-r2 
5. emerge ati-drivers (this should change the opengl configuration to use the AT 
6. Configure X server using /opt/ati/bin/fglrxconfig 
7. Run startx 
8. In an xterm, run /opt/ati/bin/fgl_glxgears 
Actual Results:  
Error message about fbconfig 

Expected Results:  
Display whizzy-fast rotating cube with rotating gears! 

It's the back-ported rmap patch that does it.  The open-source bits of the ati driver calls the 
pte_offset macro for 2.4 kernels, but this macro is no longer defined.  For 2.5 kernels it calls 
the pte_offset_kernel macro.  (There is also some conditional stuff that calls 
pte_offset_atomic instead, if it is defined, but it isn't.) 
 
I made a patch that calls pte_offset_kernel instead of pte_offset, and defines 
pte_offset_kernel if it isn't already defined.  I haven't tested it yet (I'll try it at home tonight), 
but I think it should work.

[Comment 1 Ian Abbott 2003-04-28 10:24:47 UTC]

Created attachment 11247 [details, diff]
Proposed fix

I haven't tested the patch yet.  I'll test it tonight and report back tomorrow.

[Comment 2 Ian Abbott 2003-04-29 05:17:51 UTC]

I have now tested the patch by building ati-drivers for a gentoo-sources-2.4.20-r3 kernel and it worked fine.  I haven't tested it for a non-rmap kernel yet. 

[Comment 3 Luca Barbato 2003-04-29 08:42:30 UTC]

Please try the lastest driver in portage.

[Comment 4 Ian Abbott 2003-04-29 11:10:55 UTC]

I'll try it tonight (also means building xfree and qt). 
 
I just want to say that my patch was probably bogus, as the updated version of 
ati-drivers effectively calls pte_offset_map in the places where my patch called 
pte_offset_kernel. 

[Comment 5 Luca Barbato 2003-04-29 23:46:43 UTC]

I couldn't look at the issue yet (due an exams quite near in time) but I had for ages rmap kernel on my box and the 2.5.1 was working fine, please tell me if I should close the bug or just keep that open and fully investigate on it.
(Thanks for you time anyway)

[Comment 6 Ian Abbott 2003-04-30 06:08:45 UTC]

Created attachment 11322 [details, diff]
Different (probably better) patch, based on ati-drivers-2.9.8

In my original patch, for the 2.4 kernel, I effectivel called the
pte_offset_kernel macro (if defined by the kernel rmap patch) instead of
pte_offset, as that is what was called for the 2.5 kernel.  However, the
ati-drivers-2.9.8 sources effectively call the pte_offset_map macro instead of
pte_offset for the 2.4 kernel (if pte_offset is left undefined by the rmap
patch).

This new patch just pulls in the relevant bit from the ati-drivers-2.9.8
sources and applies it to ati-drivers-2.5.1.

I don't understand why pte_offset_kernel is used for the 2.5 kernel and
pte_offset_map is used for the 2.4 kernel, but who am I to judge?

[Comment 7 Ian Abbott 2003-04-30 06:20:33 UTC]

I tested ati-drivers-2.9.8 on my system with gentoo-sources-2.4.20-r3 and xfree-4.3.1-r2 
and it seems to work fine.  The fgl_glxgears program looks a lot different now, which I 
thought was a bug, but everything else using OpenGL looks the same as before. (The 
fgl_glxgears program has been updated, so I guess that explains the difference in 
appearance.) 
 
As far as I am concerned, the bug is now a non-issue, except perhaps if someone 
really needs to keep on using XFree 4.2.  For me, it was a PITA to keep xfree pinned 
to this older version (requiring several injections!) just for ati-drivers compatibility. 

[Comment 8 Arkadi Shishlov 2003-05-10 06:56:11 UTC]

Can you please fix 2.5.1 in portage? 2.9.8 have severe 2D rendering problems on my Radeon8500 - it is unusable.

[Comment 9 Luca Barbato 2003-05-10 21:47:27 UTC]

Updated the 2.5.1-r1 ebuild, please reopen the bug if something is wrong