5336 driver causes panic under load

[jbaietto]

I'm running stock Red Hat Enterprise Linux 3.0 + all updates on a Dell 650:

kernel-2.4.21-4.EL
XFree86-4.3.0-55.EL
gcc-3.2.3-24
glibc-2.3.2-95.6

I start a few clients to create some X and OpenGL noise (ico, glxgears, chromium) and then proceed to do a kernel build (using the "rpm" target in the kernel Makefile) and the system only survives a few minutes after that. I see similar problems running 2.6.3 too. This is very reproducible.

Attached is my XF86Config file. Note that I'm running a TwinView dual monitor setup. Following is an example panic message. Any help or advice will be greatly appreciated.

Unable to handle kernel paging request at virtual address 0b0b0b1f
printing eip:
f8eb55b0
*pde = 00000000
Oops: 0000
agpgart nvidia i810_audio ac97_codec soundcore parport_pc lp parport autofs nfs lockd sunrpc e1000 8139too mii crc32 sr_mod cdrom floppy sg microcode keybdev
CPU: 1
EIP: 0060:[<f8eb55b0>] Tainted: P
EFLAGS: 00210087

EIP is at _nv003605rm [nvidia] 0x34 (2.4.21-9.0.1.ELsmp/i686)
eax: f8faf000 ebx: 000001dd ecx: 00000000 edx: 001002ed
esi: 0b0b0b0b edi: 0b0b0b0b ebp: d166bd10 esp: d166bd04
ds: 0068 es: 0068 ss: 0068
Process glxgears (pid: 14583, stackpage=d166b000)
Stack: f6488000 d6d4f200 f6488000 d166bd2c f8eb57e8 f6488000 00000005 00000200
00000000 00000001 d166bd70 f8eb5ebe f6488000 00001000 f6488000 dedf9c00
d748f000 dedf9c00 00001c9e d6e06800 00000000 00000000 00000000 00000000
Call Trace: [<f8eb57e8>] _nv003615rm [nvidia] 0x58 (0xd166bd14)
[<f8eb5ebe>] _nv002012rm [nvidia] 0x19e (0xd166bd30)
[<f8eb6078>] _nv002090rm [nvidia] 0x90 (0xd166bd74)
[<f8eb93c1>] _nv002051rm [nvidia] 0xc1 (0xd166bda4)
[<f8ea24e3>] _nv003564rm [nvidia] 0xd3 (0xd166bde8)
[<f8ddcfff>] _nv005301rm [nvidia] 0x103 (0xd166be3c)
[<f8dc6142>] _nv001223rm [nvidia] 0x12 (0xd166be4c)
[<f8ddd647>] _nv004046rm [nvidia] 0x3a3 (0xd166be58)
[<f8edeee3>] _nv001476rm [nvidia] 0x1d3 (0xd166be98)
[<f8dc777a>] _nv000896rm [nvidia] 0x4a (0xd166bf20)
[<f8f8402c>] nv_linux_devices [nvidia] 0x8c (0xd166bf2c)
[<f8dc8f94>] rm_isr_bh [nvidia] 0xc (0xd166bf3c)
[<f8da8858>] nv_kern_isr_bh [nvidia] 0xe (0xd166bf48)
[<f8da881c>] nv_kern_isr [nvidia] 0x1d (0xd166bf58)
[<c012ebb7>] tasklet_action [kernel] 0x67 (0xd166bf68)
[<c012ea15>] do_softirq [kernel] 0xd5 (0xd166bf7c)
[<c010db48>] do_IRQ [kernel] 0x148 (0xd166bf98)
[<c0124584>] sys_sched_yield [kernel] 0xa4 (0xd166bfa8)
[<c010da00>] do_IRQ [kernel] 0x0 (0xd166bfbc)

Code: 8b 46 14 8b 4d 14 51 68 b4 0b 40 00 8b 56 18 52 8b 40 24 ff

Kernel panic: Fatal exception
In interrupt handler - not syncing

[jbaietto]

It appears that I can reproduce the same problem with the 5328 driver version, so it's not just 5336.

At this point I think it's time to test my machine's memory...

[jbaietto]

Looks like memtest-86 v3.0 passes with flying colors (though I'll let it run overnight just for good measure).

So, I don't think it is bad memory. That narrows it down to one of the following:

- A bug that has been in the nvidia driver for multiple releases
- A bug in the Red Hat kernel
- Some intermittent hardware failure other than memory

After the memory tests complete tomorrow I'll go back some further revs of the nvidia driver and see if I can make those fail under load too.