[Pkg-fglrx-devel] fglrx driver locks machine

[Matteo Semplice]

Dear everybody,
 I am having troubles with my new ATI Sapphire HD2600PRO. 
Also, part (or all?) of the problem is that the fglrx kernel module does not 
compile correctly, so I am Cc-ing to the debian-x and pkg-fglrx-devel lists. 
Apologies if you get multiple copies of this message.

My system is based on etch.
The card is brand new and it works at low resolution with the VESA X11 driver, 
so I guess it's not broken...

The fglrx driver in etch do not recognize the card (since it's too old): X 
won't even start.

The fglrx driver in etch-backports recognizes the card and initializes it 
correctly. However the DRI functionality depends on the fglrx kernel module 
and without this module even moving a window or scrolling in a terminal is 
very slow and not smooth. (I mean _seriously not smooth_, it's almost 
unusable)

When compiling (with module-assistant) the fglrx kernel module, I always get 
errors or at least warnings. This happens in various degrees if I compile 
against the 2.6.18-k7 kernel in etch, the 2.6.24-etchnhalf kernel, the 
2.6.26~bpo kernel in etch-backports. I am not reporting on these, but I have 
the logs, if anyone wants to see them.

I got the best results getting kernel 2.6.26 and fglrx source packages from 
the lenny repository and compiling them on my machine. With this setup, m-a 
creates the package fglrx-kernel-2.6.26-1-686_8-7-2+2.6.26-5_i386.deb
but the logfile 
/var/cache/modass/fglrx-source.buildlog.2.6.26-1-686.1222445844
contains several warnings about misused pointers and kernel symbol. Here are 
just a few:

make[2]: Entering directory `/usr/src/linux-headers-2.6.26-1-686'
  CC [M]  /usr/src/modules/fglrx/firegl_public.o
/usr/src/modules/fglrx/firegl_public.c: In function '__ke_vm_phys_addr_str':
/usr/src/modules/fglrx/firegl_public.c:3996: warning: return makes pointer 
from integer without a cast
/usr/src/modules/fglrx/firegl_public.c: In function 'ip_vm_nopage':
/usr/src/modules/fglrx/firegl_public.c:4031: warning: passing argument 2 
of 'vm_nopage' makes pointer from integer without a cast
/usr/src/modules/fglrx/firegl_public.c:4031: warning: assignment makes pointer 
from integer without a cast
/usr/src/modules/fglrx/firegl_public.c: At top level:
/usr/src/modules/fglrx/firegl_public.c:4091: warning: initialization from 
incompatible pointer type
WARNING: modpost: module fglrx.ko uses symbol 'init_mm' marked UNUSED

If I install such .deb package, the fglrx kernel module gets inserted at 
startup and the graphics card works at full resolution and with DRI enabled. 
However logging out from the X server locks the machine, instead of giving me 
back the kdm prompt. Also repeatedly switching VT1 to/from VT7 can lock it. 
With locked, I mean that the machine does not respond any more to mouse, 
keyboard, Ctrl+Alt+Backspace, Ctrl+Alt+Del and I cannot switch it off from 
another pc connected via ssh (even if I establish the connection before the 
lock happens).

Finally I performed the following tests with the 2.6.26 kernel and fglrx 
compiled from the lenny source packages.
1) boot machine, log into kdm as user, switch to VT1, log in as root and 
issue "halt". The last line of the Xorg.log is
   (II) AIGLX: Suspending AIGLX clients for VT switch
Note that if I instead switch back to VT7, the machine would lock.

2) boot machine, log into kdm as user, open root terminal and issue "halt". 
The last lines of Xorg.log are
(II) fglrx(0): Shutdown CMMQS
(II) fglrx(0): [drm] removed 1 reserved context for kernel
(II) fglrx(0): [drm] unmapping 8192 bytes of SAREA 0x2000 at 0xb7f3d000

3) uninstall the fglrx kernel module. Boot machine, log into kdm as user, 
switch to VT1, back to VT7, back to VT1, back to VT7. Machine locks, leaving 
this Xorg.log

(II) Configured Mouse: ps2EnableDataReporting: succeeded
Could not init font path element /usr/X11R6/lib/X11/fonts/cyrillic, removing 
from list!
Could not init font path element /usr/X11R6/lib/X11/fonts/Type1, removing from 
list!
(EE) fglrx(0): firegl_SetSuspendResumeState FAILED -1003.
(WW) Open ACPI failed (/var/run/acpid.socket) (No such file or directory)
(II) No APM support in BIOS or kernel
(EE) fglrx(0): firegl_GetSuspendResumeState FAILED -1003.
(II) Configured Mouse: ps2EnableDataReporting: succeeded
(EE) fglrx(0): firegl_SetSuspendResumeState FAILED -1003.
FreeFontPath: FPE "/usr/share/fonts/X11/misc" refcount is 2, should be 1; 
fixing.

Now, if you had the patience to read this far... do you have any suggestion?

For the maintainers or anyone willing to look deeply into this issue: I saved 
the log files I mentioned above and I am willing to send them to you and/or 
do more testing if needed. Also, if you need me to file a bug about this, 
please state so and maybe give me some directives, as I am no professional 
programmer.

Thanks in advance to anyone who has any suggestion.

Best regards,
 Matteo Semplice