F14 kernel process loop, possibly in ext4 filesystem code

[Eric Sandeen]

Michał Piotrowski wrote:
> CC'ing kernel list
> 
> Regards,
> Michal
> 
> 2010/10/4 Anne & Lynn Wheeler <lynn at garlic.com>:
>> EXT4 filesystem directory with 100,000+ small files, 4-core intel
>> processor, batch process that runs parallel on all four processors
>> ... adding/deleting several hundred files.
> 
> Let me guess - fsfuzzer? :)
> 
>> No problem on F13 but fairly consistently happens since recent
>> "preupgrade" to F14. If I serialize the processing, problem
>> doesn't occur.
>>
>> Initial symptom is that batch process doesn't complete and gkrellm
>> shows one of the processors at 100% cpu in the kernel. Eventually
>> whole system locks up and requires reboot (with manual reset switch).
>> Directory will then frequently show numerous newly added files
>> as zero bytes (i.e. metadata had been logged, but actual data
>> never written to disk).

Is there any actual information about the kernel version, the testcase,
or the failure available?

sysrq-w or sysrq-t would at least show which processes are stuck (former)
and/or where they all are in the callchains (latter)

-Eric