> I you find this problem interesting and need some info, I can try to
> provide it. I've still got the kernel running (after a reboot), and my
> guess is 3 days until the first signs begin again... If I should pay
> special attention to something, please say so.
> Sorry for the long post and thanks for some (otherwise) great kernels: I
> also run 2.0.32pre6 on my 200MHz K6 with 48Megs and *NO* problems!!!
> This box is not running all the time so max uptime is less than 2 days.
> Anyway, an 8 Meg box is faster to find a memleak...
> Nov 23 02:02:21 hobbes kernel: EIP: 0010:[__iget+60/544]
> Nov 23 02:02:21 hobbes kernel: EFLAGS: 00010206
> Nov 23 02:02:21 hobbes kernel: eax: 00000301 ebx: 00000100 ecx: 00000901 edx: 001b1839
All exept close_fd OOPSes were caused by dereferencing 0x00000100 value in a
context well protected against NULL (0x00000000) pointers. It looks like a
single bit flip - bad memory/cache/whatever. How that machine was working
before, with different kernels? (note that other kernel may just put its data
structures in some other place and localized memory defect may not be seen)
Try to compile the kernel on it - it is the best memory tester.
BTW. [__iget+60/544] is fs/inode.c:613
[__iget+454/544] is fs/inode.c:662
Greetings,
-- Tomasz Motylewski