undefined symbol: pciNumBuses (in 9.1 driver version)

[asfalatus]

Hi,

I reported this bug: http://ati.cchtml.com/show_bug.cgi?id=1423 , but maybe someone here can help me too.

Bug content:

Greetings,

I have this issue with 9.1 driver:

(II) LoadModule: "fglrx"
(II) Loading /usr/lib/xorg/modules/drivers//fglrx_drv.so
dlopen: /usr/lib/xorg/modules/drivers//fglrx_drv.so: undefined symbol:
pciNumBuses
(EE) Failed to load /usr/lib/xorg/modules/drivers//fglrx_drv.so
(II) UnloadModule: "fglrx"
(EE) Failed to load module "fglrx" (loader failed, 7)

After this 'No screens found' msg follows and X Server doesn't start.

dmesg short fglrx related output (nothing strange here):
[   26.585797] fglrx: module license 'Proprietary. (C) 2002 - ATI Technologies,
Starnberg, GERMANY' taints kernel.
[   26.624107] [fglrx] Maximum main memory to use for locked dma buffers: 3525
MBytes.
[   26.624271] [fglrx]   vendor: 1002 device: 9612 count: 1
[   26.624865] [fglrx]   vendor: 1002 device: 95c4 count: 2
[   26.625215] [fglrx] ioport: bar 1, base 0x9000, size: 0x100
[   26.625296] pci 0000:01:05.0: enabling device (0006 -> 0007)
[   26.625362] pci 0000:01:05.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
[   26.625425] pci 0000:01:05.0: setting latency timer to 64
[   26.625994] [fglrx] ioport: bar 1, base 0xa000, size: 0x100
[   26.626073] pci 0000:02:00.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
[   26.626139] pci 0000:02:00.0: setting latency timer to 64
[   26.626361] [fglrx] Kernel PAT support detected, disabling driver built-in
PAT support
[   26.626469] [fglrx] module loaded - fglrx 8.57.2 [Jan 14 2009] with 2 minors

Last working version: none of 9.x branch
Kernel version: 2.6.28.1.mobileworkshop (custom, mobileworkshop is hostname of
machine)
X server: 
dpkg -l | grep xorg
ii  xserver-xorg                                 1:7.4~5     the X.Org X server
ii  xserver-xorg-core                            2:1.5.99.901-2     Xorg X
server - core server

Anyone who have this issue?
PS: I have Fujitsu-Siemens laptop, model Amilo Pa3553 with HybridCrossfire (HD3470 discrete, HD3200 integrated).

[bridgman]

Which distro version are you running with ? Looks like an upgraded kernel from what came with the distro ?

[asfalatus]

Which distro version are you running with ? Looks like an upgraded kernel from what came with the distro ?

I am using self-compiled kernel 2.6.28.1 with this patch: http://bugzilla.kernel.org/show_bug.cgi?id=12514

Distribution is Debian sid/experimental (mostly experimental )

[pflynn]

This is not a exactly a kernel problem. It is a XOrg one: the fglrx_drv.so module (used by the XOrg server) is complaining it cannot find the pciNumBuses internal symbol. I wish I had enough knowledge of XOrg's internals so I could help you...anyway, this problem seems to be related to the server (XOrg) version.

[asfalatus]

This is not a exactly a kernel problem. It is a XOrg one: the fglrx_drv.so module (used by the XOrg server) is complaining it cannot find the pciNumBuses internal symbol. I wish I had enough knowledge of XOrg's internals so I could help you...anyway, this problem seems to be related to the server (XOrg) version.

X Server 1.6.0 RC1 is not supported by this recent driver?

[pflynn]

X Server 1.6.0 RC1 is not supported by this recent driver?

Not sure, but in the xorg-server 1.5.99.1 announce (xorg-server 1.6 beta) -
http://lists.freedesktop.org/archive...er/000696.html ? they mention that the pciNumBuses internal variable was removed from this version:

Drop the never-called SkippedRequestCallback list.
const cleanup
Remove the Must_have_memory hack.
Remove unused pciNumBuses
loader: sort the mi symbol list
Remove a bunch of unused API from the mi export list.

This could explain why you can't compile and install Catalyst 9.1 within your system.

[asfalatus]

Not sure, but in the xorg-server 1.5.99.1 announce (xorg-server 1.6 beta) -
http://lists.freedesktop.org/archive...er/000696.html ? they mention that the pciNumBuses internal variable was removed from this version:




This could explain why you can't compile and install Catalyst 9.1 within your system.

I can compile and install it, but when X server core tries load fglrx module then it gets error.

So do you think there is a possible workaround for this issue? Downgrading X server isn't solution for me.

Thank you for helpful informations you gave me.

[asfalatus]

So, I decided to report bug to xorg-team. I thought that restructuralization of fglrx driver is harder than re-adding "unused pciNumBuses".

Link: http://bugs.freedesktop.org/show_bug.cgi?id=19878

Someone wrote me it isn't their bug. So there is no solution for me for 1-N months.

Many thanks for people who tried to help me.

[pflynn]

I can compile and install it, but when X server core tries load fglrx module then it gets error.

Yes: the code will compile successfully and everything will be installed properly. But it is a link time question, so the problem will raise whenever X server core tries to load the module.

Someone wrote me it isn't their bug. So there is no solution for me for 1-N months.

Yes, they are right: it's not a XOrg bug. It's just some change to the server's code. Catalyst driver maintainers are expected to find this and apply the corresponding changes to their code (maybe in the form of a patch).

So do you think there is a possible workaround for this issue? Downgrading X server isn't solution for me.

A workaround could be a patch to the Catalyst code. But it'd be better first know if someone could successfully run the drivers using the same X server version as yours.

[Kano]

That symbol error is an ABI change. Of course ATI has to take care of it. But the driver passed out at 2 other stages (the error is later in the log - and when you workaround the first you will see a 2nd one) where the ABI checks are done. But even if you bypass those checks it does not work. The xserver would just lock up, so erroring out it maybe not the worst *g*