Lockups on Dual Xeon with i7505 chipset and Kernel 2.6.x

[dancres]

Suspect most of this has been reported before but I wanted to put it all in one place.

Two postings from linux-kernel - in both cases, waggling the mouse is sufficient to cause the crash. I've seen this problem with all versions of the NVIDIA drivers since about 4496.

First posting:

Setup: Dual Xeon 2.4s (hyperthreading on) on i7505 chipset based
motherboard. Running the nvidia binary driver (from vanilla debian
nvidia-kernel-installer).

ksymoops 2.4.9 on i686 2.6.5. Options used
-V (default)
-k /proc/kallsyms (specified)
-l /proc/modules (default)
-o /lib/modules/2.6.5/ (default)
-m /boot/System.map-2.6.5 (default)

Warning (read_ksyms): no kernel symbols in ksyms, is /proc/kallsyms a
valid ksyms file?
No modules in ksyms, skipping objects
No ksyms, skipping lsmod
NMI Watchdog detected LOCKUP on CPU2, eip c01475ca, registers:
CPU: 2
EIP: 0060:[<c01475ca>] Tainted: P
Using defaults from ksymoops -t elf32-i386 -a i386
EFLAGS: 00000046 (2.6.5)
eax: 00000000 ebx: f7fd6400 ecx: f7ff0248 edx: f738e000
esi: f5e0d000 edi: 00000007 ebp: f738ff04 esp: f738fed8
ds: 007b es: 007b ss: 0068
Stack: f738fef0 c016f3b3 c2795b1c f5e0d018 f7ff02a8 f7ff0270 f7fd6410
f7ff0278
00000296 00000000 ffffffe9 f738ff18 c0147a65 f7ff0248 000000d0
00000000
f738ff30 c01610c2 f7ff0248 000000d0 f7fdb380 00000000 f738ff50
c015f34e
Call Trace:
[<c016f3b3>] permission+0x2f/0x49
[<c0147a65>] kmem_cache_alloc+0x4a/0x4c
[<c01610c2>] get_empty_filp+0x48/0xe9
[<c015f34e>] dentry_open+0x16/0x221
[<c01476d9>] cache_alloc_refill+0x1c1/0x2a4
[<c015f336>] filp_open+0x5d/0x5f
[<c015f8a8>] sys_open+0x55/0x85
[<c010924f>] syscall_call+0x7/0xb
Code: 83 6a 14 01 a8 08 0f 85 5d 01 00 00 8b 55 08 8b 42 58 39 46



>>>>EIP; c01475ca <cache_alloc_refill+b2/2a4> <=====



>>>>ebx; f7fd6400 <__crc_pci_find_subsys+12fdb4/17053c>
>>>>ecx; f7ff0248 <__crc_pci_find_subsys+149bfc/17053c>
>>>>edx; f738e000 <__crc_redraw_screen+15c1f/1b1afa>
>>>>esi; f5e0d000 <__crc_cap_bprm_compute_creds+145e81/17660f>
>>>>ebp; f738ff04 <__crc_redraw_screen+17b23/1b1afa>
>>>>esp; f738fed8 <__crc_redraw_screen+17af7/1b1afa>


Trace; c016f3b3 <permission+2f/49>
Trace; c0147a65 <kmem_cache_alloc+4a/4c>
Trace; c01610c2 <get_empty_filp+48/e9>
Trace; c015f34e <dentry_open+16/221>
Trace; c01476d9 <cache_alloc_refill+1c1/2a4>
Trace; c015f336 <filp_open+5d/5f>
Trace; c015f8a8 <sys_open+55/85>
Trace; c010924f <syscall_call+7/b>

Code; c01475ca <cache_alloc_refill+b2/2a4>
00000000 <_EIP>:
Code; c01475ca <cache_alloc_refill+b2/2a4> <=====
0: 83 6a 14 01 subl $0x1,0x14(%edx) <=====
Code; c01475ce <cache_alloc_refill+b6/2a4>
4: a8 08 test $0x8,%al
Code; c01475d0 <cache_alloc_refill+b8/2a4>
6: 0f 85 5d 01 00 00 jne 169 <_EIP+0x169>
Code; c01475d6 <cache_alloc_refill+be/2a4>
c: 8b 55 08 mov 0x8(%ebp),%edx
Code; c01475d9 <cache_alloc_refill+c1/2a4>
f: 8b 42 58 mov 0x58(%edx),%eax
Code; c01475dc <cache_alloc_refill+c4/2a4>
12: 39 46 00 cmp %eax,0x0(%esi)


1 warning issued. Results may not be reliable.


Second can be found at: http://marc.theaimsgroup.com/?l=linu...6771821495&w=2

And there have been further mentionings of this problem:

http://marc.theaimsgroup.com/?l=linu...6176219093&w=2


This problem seems to have been around since last July, any feedback on what's happening with this would be great.

Thanks,

Dan.

[SuLinUX]

I believe nvidia are looking into this and a fix in the next driver release, does it still lockup with Hyperthreading off?

[tjw]

I don't see anything that in that trace that points a finger at the nvidia kernel module.

I use a I7505 board with dual 2.4 xeons with a GeForce 4 Ti 4600 card on 2.6.5 and have never had a lockup. Have you tried memtest86 to verify your main memory? Also make sure you're not using Fast Writes or Side Band Addressing:

#cat /proc/driver/nvidia/agp/status
Status: Enabled
Driver: NVIDIA
AGP Rate: 4x
Fast Writes: Disabled
SBA: Disabled

[dancres]

Originally Posted by tjw

I don't see anything that in that trace that points a finger at the nvidia kernel module.

I use a I7505 board with dual 2.4 xeons with a GeForce 4 Ti 4600 card on 2.6.5 and have never had a lockup. Have you tried memtest86 to verify your main memory? Also make sure you're not using Fast Writes or Side Band Addressing:

#cat /proc/driver/nvidia/agp/status
Status: Enabled
Driver: NVIDIA
AGP Rate: 4x
Fast Writes: Disabled
SBA: Disabled

I'll try those out, thanks.

Re: no evidence of nvidia, no, there isn't in the first trace but if you look at the other postings you will see very similar traces which blatantly *are* associated with the nvidia driver.

Lastly, just wondering what chipset you have on your MB?

P.S. I've run memtest with no problems and, whatever's going on doesn't happen with a 2.4 kernel so I'm guessing it's possibly interrupt/acpi related.

[dancres]

Originally Posted by SuLinUX

I believe nvidia are looking into this and a fix in the next driver release, does it still lockup with Hyperthreading off?

That'd be great if they were and yes, with HT off, it still dies a death.