sparc64/142102: FreeBSD 8.0 kernel panics on sparc64 when
accessing NFS
Marius Strobl
marius at alchemy.franken.de
Tue Dec 29 01:00:14 UTC 2009
The following reply was made to PR sparc64/142102; it has been noted by GNATS.
From: Marius Strobl <marius at alchemy.franken.de>
To: Manuel Tobias Schiller <mala at hinterbergen.de>
Cc: Mark Linimon <linimon at lonesome.com>, FreeBSD-gnats-submit at FreeBSD.org
Subject: Re: sparc64/142102: FreeBSD 8.0 kernel panics on sparc64 when accessing NFS
Date: Tue, 29 Dec 2009 01:58:29 +0100
On Mon, Dec 28, 2009 at 10:06:02PM +0100, Manuel Tobias Schiller wrote:
> On Mon, 28 Dec 2009 14:26:50 -0600
> Mark Linimon <linimon at lonesome.com> wrote:
>
> > Just to add a data point, the Netra T1s in the package build cluster do
> > not show this problem, so I'm guessing that you are hitting an edge
> > on large file transfers. We run them pretty hard.
I think you guys are talking about different things; AFAIK the
package build cluster nodes only act as NFS clients but this
problem is about when using machines with strict alignment
requirements as an NFS server.
> >
> > mcl
> >
> Hi Mark, Marius,
>
> thanks for the data point. In the meantime, I managed to test a kernel
> with the "official" fix from PR 140797, and I still get the crash when
> trying to write a moderately sized file (20 Megabytes) to NFS. According
> to the backtrace below, it crashes in line 1355 of
> /usr/src/sys/nfsserver/nfs_srvsubs.c.
>
> Since I'm using the fix from the original problem report and RELENG_8_0
> sources cvsupped on December 20, I guess that means we are either using
> different source trees (i.e. there is something in the sources for the
> kernels used on your machines that helps), or there is some other
> difference which I have not been able to pinpoint yet. Maybe you could
> clarify before I try hacking away or finding other differences...
>
> Thanks,
>
> Manuel
>
>
> ---- backtrack of panic follows ----
> panic: trap: memory address not aligned
> cpuid = 0
> KDB: stack backtrace:
> panic() at panic+0x1c8
> trap() at trap+0x4d0
> -- memory address not aligned sfar=0xfffff800016d08aa sfsr=0x40029
> %o7=0xc0528934 -- nfsm_srvmtofh_xx() at nfsm_srvmtofh_xx+0x24
> fha_assign() at fha_assign+0x12c
> svc_run_internal() at svc_run_internal+0x71c
> svc_thread_start() at svc_thread_start+0x8
> fork_exit() at fork_exit+0x80
> fork_trampoline() at fork_trampoline+0x8
> Uptime: 2m1s
> Dumping 512 MB (2 chunks)
> chunk at 0: 268435456 bytes |
>
I'm using a more or less current HEAD but the NFS code hasn't
changed that much since 8.0, at least it doesn't contain any
other alignment fixes I'm aware of.
I think I got what the problem is but I still haven't managed
to reproduce it so far. Could you please test whether the
following patch makes a difference?
http://people.freebsd.org/~marius/fha_extract_info_realign.diff
Marius
More information about the freebsd-sparc64
mailing list