7.0R panic in nfs_readdirrpc()

[Hiroki Sato]

Marcel Moolenaar <xcllnt at mac.com> wrote
  in <BDB0D269-FA6E-4017-B1B7-12872207DB56 at mac.com>:

xc>
xc> On Jul 25, 2008, at 12:10 AM, Hiroki Sato wrote:
xc>
xc> > Almost the same panic occurred on sparc64, too.  Memory alignment
xc> > issue?
xc>
xc> Yes. The exception is for a 4-byte load on a 1-byte aligned
xc> (i.e. odd) address. This is a code bug.
xc>
xc> I can't remember having these issues with NFS. Can you tell
xc> me more about your configuration?

 Yes, it was 7.0R/i386 (server)<->7.0R/ia64 (client), mounted over
 TCP.  Doing "cvs co src" from the server triggered the panic.  I
 tried NFS over UDP (default) and one over TCP, the former has no
 problem under the same load.  The mount_nfs options were
 "-T,ro,bg,intr,noinet6" and the difference was the transport protocol
 only.  Maybe it is because TCP mount uses a larger block size than
 UDP by default.  7.0R/sparc64 can trigger it with the same
 configuration as I reported.  I am not sure if 7.0R/ppc does, though.

 In i386<->i386 and i386<->amd64 case, TCP mount seems to work fine,
 but it is probably because i386 does not throw exception due to
 misaligned access.

--
| Hiroki SATO
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 195 bytes
Desc: not available
Url : http://lists.freebsd.org/pipermail/freebsd-sparc64/attachments/20080726/0f207498/attachment.pgp