On Tue, 2004-04-27 at 12:52, Steve Lord wrote:
> Craig Tierney wrote:
> > I have enabled debugging and tracing features in XFS
> > using the 2.4.26 kernel. When I load the xfs module,
> > it reports that tracing and debugging are enabled.
> > (From dmesg)
> > SGI XFS with tracing, debug enabled
> > SGI XFS Quota Management subsystem
> > To enable tracing, the Configure.help file says to set
> > /proc/sys/vm/pagebuf/debug. However, I could not find this
> > sysctl, or anything that was similar in the /proc/fs/xfs or
> > /proc/sys/fs/xfs directories.
> > How do I turn this on?
> > Also, with debugging enabled, do all debug messages go to the
> > kernel log?
> The tracing is to internal kernel buffers and is not much use
> unless you have a debugger capable of looking at them, which
> means kdb and the xfs support for kdb.
> Any reason for turning on debug? It adds thousands of checks to
> the code base, any one of them will oops if it triggers. Never
> run this way with data you care about.
I am exporting my XFS filesystems over NFS. Under load we are
seeing corruption on writes. Corruption is in multiples of 4KB,
and are always paged aligned.
I sent a post to the list last week:
Besides that post, it happens when kernel is booted as "nosmp,noapic",
and corruptions seems to happen more as the performance of the
server goes up. I get more corruption for a given number of runs
on a Dual 2.6Ghz Xeon than on a Dual 933Mhz PIII.
I was able to generate corrupted output with debugging enabled, but
no oopses occurred.