Ext3 users - Oct 2001 - Hi all - Kernel Opps on PowerPC

Hello all,

We recently switched to kernel 2.4.10 from kernel 2.4.4, and in doing
so also added the EXT3 patch 2.4-0.9.10-2410.  

Everything seems to work great, however I have been able to reproduce
a crash over and over again.

Here it is:

#!/bin/sh

mke2fs -j /dev/hda2
mount /dev/hda2 /mnt

About 75% of the time, mount will result in a kernel Oops, the crash
seems to be about something in journal_update_superblock()...

If I put a small delay, of any kind, in bewteen the mke2fs and
the mount, all things are OK.

Is this another imfamous bug created by the separation of the
Page and Buffer cache in 2.4.10?

Just wanted to report it...

Here is the PowerPC Opps, don't know if it'll tell you any
more or not...

Oops: Exception in kernel mode, sig: 4
NIP: C0080574 XER: 20000000 LR: C0080240 SP: CDF39D70 REGS: cdf39cc0 TRAP: 0700
MSR: 00089032 EE: 1 PR: 0 FP: 0 ME: 1 IR/DR: 11
TASK = cdf38000[104] 'mount' Last syscall: 21 
last math cdf38000 last altivec 00000000
GPR00: 00000000 CDF39D70 CDF38000 C0810C00 00000001 00000000 CDF39D88 00000002 
GPR08: 00000001 C0210000 00002000 00000001 22002022 0185FAF8 00000000 00000000 
GPR16: 00000000 00000000 00000000 00000000 00000000 00000001 00000302 CDED2800 
GPR24: CDED4420 00000400 00000002 00000000 CDED2800 00000001 CC5B3800 C0810C00 
Call backtrace: 
C003E7E8 C0080240 C0080840 C00765EC C0075DFC C00449D8 C0045B2C 
C0045D0C C004615C C004621C C0003DFC 0181F43C 0181F740 0181FC2C 
01834280 01833EE8 016DBA08 00000000 
Illegal instruction

If you need anything else from me, please let me know....

Paul W.

On Tue, Oct 09, 2001 at 06:22:32PM -0700, Paul White
wrote:
> 
> Here is the PowerPC Opps, don't know if it'll tell you any
> more or not...
> 
As with all Oopses, you'll need to run it through ksymoops...

Hi,

On Tue, Oct 09, 2001 at 06:22:32PM -0700, Paul White wrote:
 
> We recently switched to kernel 2.4.10 from kernel 2.4.4, and in doing
> so also added the EXT3 patch 2.4-0.9.10-2410.  
> 
> Everything seems to work great, however I have been able to reproduce
> a crash over and over again.
ext3 on 2.4.10 is known to have problems --- far too much changed in
2.4.10's VFS internals.
> mke2fs -j /dev/hda2
> mount /dev/hda2 /mnt
> 
> About 75% of the time, mount will result in a kernel Oops, the crash
> seems to be about something in journal_update_superblock()...
Almost certainly the 2.4.10 changes --- it's precisely the
relationship between buffer-cache operations such as mke2fs and fs
operations such as mount which has changed the most in 2.4.10.

We're doing primary testing for ext3 on 2.4.10-ac right now, rather
than the Linus tree, until we have a bit more confidence in the
Linus-tree ext3 interactions.  

But to be sure of what's happening here, you'll have to send us a
ksymoops parsing of the kernel message.

Cheers,
 Stephen