Re: 2.6.16-rc6: known regressions (v2)

[Nathan Scott <nathans@sgi.com>]

On Sat, Mar 18, 2006 at 08:57:09PM +0000, Claudio Martins wrote:
>   Hi Adrian, Nathan and all,
> 
>  If think I might have hit this one! 

OK - any hints that might lead us toward a test case?

> I managed to get an oops which showed xfs related functions on the backtrace. 
> The process involved was "rm" and the specific stress test was some 32 
> paralell kernel builds (each one with "make -j8") on a quad Opteron box with 
> a 1 TB xfs filesystem. Preemption was disabled.

>  After that the machine was still alive, but an fsck.xfs after a reboot showed 
> corruption that I was able to repair with xfs_repair. This was also with an 

Hmm, fsck.xfs wont report corruption - did you mean xfs_check?

> almost empty filesystem, hence the similarity with the above bug report.

Well, not sure its the same yet - what was your stack trace & did
repair report inodes with nlink==0?

> I didn't record the oops and output from xfs_repair. I'll update my git tree 

Ah, doh.

> tonight, rebuild and retest in hopes to find that oops again.

Great, thanks!

cheers.

-- 
Nathan