Mandrake 10.0 AMD64 2.6.3-6: can't build kernel interface

[bwelch]

I'm trying to install the nVidia 6111 driver on a dual opteron machine running Mandrake 10.0 and currently using the 2.6.3-6 kernel. No matter what, it refuses to build the kernel interface. I've been reading this forum and have tried a number of things that have been suggested, such as running it with --add-this-kernel, adding the minion.de patch, tweaking the nv code (such as removing the syscalls.h line), took the kernel back to 2.6.3-6 when someone said the driver won't build against anything newer than 2.6.3-13, haven't run 'make mrproper', and removed the riva module. It still won't build the kernel interface. If anyone can see from this logfile what is failing I'd hugely appreciate it. This is driving me batty.

nvidia-installer log file '/var/log/nvidia-installer.log'
creation time: Tue Sep 21 11:45:11 2004

option status:
license pre-accepted : false
update : false
force update : false
expert : false
uninstall : false
driver info : false
no precompiled interface: false
no ncurses color : false
query latest driver ver : false
OpenGL header files : false
no questions : false
silent : false
X install prefix : /usr/X11R6
OpenGL install prefix : /usr
Installer install prefix: /usr
kernel source path : (not specified)
kernel install path : (not specified)
proc mount point : /proc
ui : (not specified)
tmpdir : /root/tmp
ftp site : ftp://download.nvidia.com

Using: nvidia-installer ncurses user interface
-> Kernel source path: '/lib/modules/2.6.3-6mdksmp/build'
-> Copying kernel module sources to temporary directory.
-> Building kernel interface:
executing: 'cd /root/tmp/nvidia-3842; make nv-linux.o SYSSRC=/lib/modules/2.
6.3-6mdksmp/build'...

NVIDIA: calling KBUILD...
make CC=cc KBUILD_VERBOSE=1 -C /lib/modules/2.6.3-6mdksmp/build SUBDIRS=/ro
ot/tmp/nvidia-3842 modules
make -f scripts/Makefile.build obj=scripts
make -f scripts/Makefile.build obj=scripts/genksyms
*** Warning: Overriding SUBDIRS on the command line can cause
*** inconsistencies
mkdir -p .tmp_versions
make -f scripts/Makefile.build obj=arch/x86_64/kernel arch/x86_64/kernel/asm
-offsets.s
make[3]: `arch/x86_64/kernel/asm-offsets.s' is up to date.
CHK include/asm-x86_64/offset.h
make -f scripts/Makefile.build obj=/root/tmp/nvidia-3842
cc -Wp,-MD,/root/tmp/nvidia-3842/.nv.o.d -nostdinc -iwithprefix include -D
__KERNEL__ -Iinclude -D__KERNEL__ -Iinclude -Wall -Wstrict-prototypes -Wno
-trigraphs -fno-strict-aliasing -fno-common -mno-red-zone -mcmodel=kernel -
pipe -fno-reorder-blocks -Wno-sign-compare -fno-asynchronous-unwind-tables
-funit-at-a-time -O2 -I/root/tmp/nvidia-3842 -Wall -Wimplicit -Wreturn-type
-Wswitch -Wformat -Wchar-subscripts -Wparentheses -Wpointer-ar
ith -Wno-multichar -Werror -O -fno-common -mno-red-zone -MD -Wno-cast-qu
al -Wno-error -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE -mcmodel=kernel - DNTRM -D_GNU_SOURCE -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE -DNV_MAJOR_
VERSION=1 -DNV_MINOR_VERSION=0 -DNV_PATCHLEVEL=6111 -DNV_UNIX -DNV_LINUX
-DNV_INT64_OK -DNVCPU_X86_64 -DNV_64_BITS -UDEBUG -U_DEBUG -DNDEBUG -DN
V_CHANGE_PAGE_ATTR_PRESENT -DNV_CLASS_SIMPLE_PRESENT -DMODULE -DKBUILD_BASE
NAME=nv -DKBUILD_MODNAME=nvidia -c -o /root/tmp/nvidia-3842/.tmp_nv.o /root/
tmp/nvidia-3842/nv.c
In file included from include/asm/dma-mapping.h:4,
from include/linux/dma-mapping.h:13,
from include/asm/pci.h:305,
from include/linux/pci.h:724,
from /root/tmp/nvidia-3842/nv-linux.h:75,
from /root/tmp/nvidia-3842/nv.c:14:
include/asm-generic/dma-mapping.h: In function `dma_map_page':
include/asm-generic/dma-mapping.h:75: warning: pointer of type `void *' used
in arithmetic
/root/tmp/nvidia-3842/nv.c: In function `nv_acpi_event':
/root/tmp/nvidia-3842/nv.c:3700: error: parse error before "nv_printf"
/root/tmp/nvidia-3842/nv.c:3701: warning: unreachable code at beginning of s
witch statement
make[3]: *** [/root/tmp/nvidia-3842/nv.o] Error 1
make[2]: *** [/root/tmp/nvidia-3842] Error 2
NVIDIA: left KBUILD.
nvidia.ko failed to build!
make[1]: *** [module] Error 1
make: *** [nv-linux.o] Error 2
-> Error.
ERROR: Unable to build the NVIDIA kernel module interface.
ERROR: Unable to add a precompiled kernel interface for the running kernel.

[bwelch]

I should add, this is the code it seems to be complaining about. It was one of the tweaks to nv.c:

switch (state)
{
case PM_SUSPEND_MEM
nv_printf(NV_DBG_INFO, "NVRM: ACPI: received standby event\n");
status = rm_power_management(nv, 0, NV_PM_ACPI_STANDBY);
break;

case PM_RESUME:
nv_printf(NV_DBG_INFO, "NVRM: ACPI: received resume event\n");
status = rm_power_management(nv, 0, NV_PM_ACPI_RESUME);
break;

default:
nv_printf(NV_DBG_WARNINGS, "NVRM: ACPI: unsupported event: %d\n", state);
return -1;
}

if (status != RM_OK)
nv_printf(NV_DBG_ERRORS, "NVRM: ACPI: failed event: %d\n", state);

return status;
}

[zander]

You're missing a colon after PM_SUSPEND_MEM.

[bwelch]

Thanks Zander. I caught that typo after taking a break for a bit I had yesterday off work and have all new problems today.

So the interface is built now, but I still can't install the driver. The nvidia-installer.log says:
kernel module load error: insmod: error inserting './usr/src/nv/nvidia.ko': -1 invalid module format.

while /var/log/messages says:
kernel: nvidia: disagrees about version of symbol struct_module

I saw a message in this forum about making links for everything that kernel.h calls in the include directory, then accept it failing, reboot, maybe it'll fail again, maybe reinstall... That doesn't sound very stable to me - and this is my boss' machine, not a home machine.

The source is the correct version for the running kernel, I haven't done anything to the compiler, and as to setting the BIOS to "assign IRQ to AGP" that isn't an option in this BIOS. It only has "IRQ to PCI VGA".

Any other ideas out there?

Thanks.

[zander]

The NVIDIA kernel module obtains version information from the kernel source tree, so chances are it doesn't fully match the running kernel. As to the IRQ assignment option, I'd expect either of the options you named to perform basically the same operation.

[bwelch]

The source and the kernel are the same versions. Both are 2.6.3-6mdk, with the kernel being the smp form (of course the source doesn't come in smp or any other form). Is there something else that could cause the driver to not believe that? Or anything else that could be involved?

[zander]

Please note that the NVIDIA doesn't actively refuse to be loaded, the kernel rejects it based on information statically linked into the module at build time; in order for the module to build, link and load successfully, the kernel source tree needs to be configured correctly, i.e. its configuration needs to match that of the target kernel. If in doubt, clean the source tree and reconfigure it with the correct configuration file (likely installed in /boot).