Re: 2.1.56-SMP: still no problems for me

Richard Gooch (rgooch@atnf.CSIRO.AU)
Sun, 21 Sep 1997 16:41:27 +1000


I wrote:
> H. Peter Anvin writes:
> > I am still unable to in any form reproduce the hanging automounter
> > bug, running on a 2.1.56 stock SMP kernel (dual PPro/200), Red Hat
> > libc 5.3.12-17, and automounter 0.3.14-pre5. I'm completely
> > dumbfounded.
>
> Well, I just found something interesting while trying to tickle a
> kernel oops again: I manually mounted an NFS filesystem. File access
> worked fine. When I went to manually unmount it, my load average goes
> up to 1! And a snippet from ps:
> 234 p3 SW 0:00 rpciod
> 235 p3 DW 0:00 lockd
>
> I can do nothing to kill these processes. The NFS mounted FS is indeed
> unmounted, and there is no zombie umount process. Using top gives no
> indication of who is chewing the CPU. A compute-bound process manages
> to get over 98% of the CPU, so it doesn't appear that the phantom is
> using much CPU.
> But I've found a cute way of increasing my load average: sucessive
> mounts and unmounts of an NFS filesystem. Every time I mount, I get a
> new lockd with state "DW". Every time I unmount, I get loadavg++.
> Even more interesting: even if I don't unmount, after a few minutes I
> still get loadavg++. Hmmm...

I've just rebooted and tried this again: I get loadavg++ as soon as I
mount an NFS filesystem. The load avarage starts creeping up as soon
as I mount, not a few minutes later as I first reported. Can get a
little difficult to see when xload rescales for higher load averages
:-(

I'm running 2.1.56-SMP with mount 2.7e, libc 5.4.33, gcc 2.7.2.f.1 on
a dual PPro. I haven't even started the automounter yet and I get the
above behaviour. The NFS client is compiled as a module. The NFS
server is a 2.0.31-pre9-SMP uniprocessor Pentium running the userspace
NFS server (I last compiled my rpc.mountd and rpc.nfsd binaries
Jul-1995).

The problems I orignally reported with 2.1.56, autofs-0.3.14-pre5
(before going back home and testing further) were on a similarly
configured dual PPro, but the NFS server was a Sparc running Solaris
2.5.1.

Regards,

Richard....