Bug 7844

Summary: ext3 with data=journal hangs when running fsx-linux
Product: File System Reporter: Randy Dunlap (rdunlap)
Component: ext3Assignee: Andrew Morton (akpm)
Severity: normal CC: bunk
Priority: P2    
Hardware: i386   
OS: Linux   
Kernel Version: 2.6.20-rc2 thru -rc5 Subsystem:
Regression: --- Bisected commit-id:

Description Randy Dunlap 2007-01-17 09:48:17 UTC
Most recent kernel where this bug did *NOT* occur:  2.6.20-rc1
Distribution:  openSUSE 10.2
Hardware Environment:  x86_64 AMD64 single-CPU, 1 GB RAM
Software Environment:  SMP kernel
Problem Description:  ext3 data=journal hangs, other journal modes run OK.

Steps to reproduce:  mkfs.ext3, mount with -o data=journal, run fsx-linux

config:  http://oss.oracle.com/~rdunlap/kerneltest/configs/config-2620-rc4-hangs
msg log:  http://oss.oracle.com/~rdunlap/kerneltest/logs/fsx-capture.txt
Comment 1 Adrian Bunk 2007-01-18 01:09:31 UTC
Does it also happen with

There weren't any ext3 changes between 2.6.20-rc1 and 2.6.20-rc2, and the most
prominent filesystem related change was the file corruption fix.

Your log also contains ACPI error messages and
[ 3603.375828] warning: many lost ticks.
[ 3603.375830] Your time source seems to be instable or some driver is hogging
[ 3603.387303] rip __do_softirq+0x4b/0xc7

Does a diff between the 2.6.20-rc1 and the 2.6.20-rc2 dmesg tell that any of
them were new in -rc2?
Comment 2 Randy Dunlap 2007-01-18 10:09:38 UTC runs with no problems.

I would be suspecting changes in mm/ rather than in fs/ext3/.

The lost ticks are not new in 2.6.20-rc2.  I see them in all of these:
Comment 3 Randy Dunlap 2007-01-25 08:59:05 UTC
fsx-linux options:  -l 100M -N 500000 -S 0 /mountpoint/fsxtextfile
Comment 4 Adrian Bunk 2007-01-27 04:41:01 UTC
Fix is in Linus' tree.