X segfaults with nvidia driver and quadro

[zovex]

I am running debian unstable with an elsa gloria II video card with uses the quadro chipset. I have an intel pentium III processor and kernel 2.4.18-bf24. I have to use the funny kernel because lilo doesnt like where my linux partition is and I boot from my rescue disk. So needless to say the driver I am using, NVIDIA-Linux-x86-1.0-4349.run doesnt have a precompiled nvidia.o module for my kernel. So I used it to build a custom one, which the installer makes very easy. :) I get the error that there is a cc mismatch between the kernel and compiler; not surprising since i am using a relatively old stock kernel and the newest gcc. I set IGNORE_CC_MISMATCH though and got it to compile ;) . I then modprobed nvidia and lsmod to check that it loaded. (the only warning was that I was tainting my gnu kernel). I then run startx.

X attempts to start, the screen flickers a dark blue and then goes black. Upon flipping back to vtX I see that X has died with this message.

XIO: fatal IO error 104 (connection reset by peer)
...

But there are NOT any other errors. I have ran startx -- -verbose 5 -logverbose 5 and still dont get any errors or warnings.

so I decided to try to use gdb to find the problem, but i didnt want to debug script to I ran X instead of startx, which gave me a better reason why X died. It segfaulted!

using gdb and running step by step untill i get to the seg fault I found the last two commands

0x40067380 in __printf_fp () from /lib/libc.so.6
0x0000002f in ?? ()

I assume that the latter is the memory access violation since its address is so low.

I have tried different resolutions to no avail. and the nv driver works fine. So the only thing I can think of is the gcc/kernel mismatch. Am I out of luck?

Thanks

[Andy Mecham]

Using a kernel compiled with gcc2 and the open files built with gcc3 is known to not work. You should rebuild your kernel with gcc3 or build the open files with gcc2 to see if that's your problem.

--andy

[bwkaz]

To restate what Andy is talking about, you should get the installer to compile your kernel interface without the IGNORE_CC_MISMATCH variable.

I don't know why so many people see the error and immediately try to do the IGNORE thing -- the correct solution to that problem is not to ignore the mismatch, it is to either install the correct compiler, or to recompile the kernel with the current system compiler (in other words, the correct fix is to resolve the mismatch, not ignore it).

[Andy Mecham]

Heh, I wasn't exactly clear, was I? :)

--andy