Hard hangs on AMD64 with mpsafenet enabled

[Robert Watson]

Thanks for the report -- I have some questions below that it would be
helpful if you could answer.

On Tue, 26 Oct 2004, [iso-8859-2] S³awek ¯ak wrote:

>     I've got a Sun V20z 2 cpu Opteron box. I experience hard hangs when
>     accessing NFS simulatneously from 2 processes (tested with parallel port
>     builds with /usr/ports mounted over NFS with
>     nosuid,nodev,soft,bg,intr). rpc.lockd and rpc.statd are both enabled for
>     NFS.

From the above, can I assume that this is a problem on the NFS client, and
that the NFS server is on another system reachable via a local area
network?

When "hung", can the machine be pinged from another machine?

From your subject line, it looks like you mean "when debug.mpsafenet=0,
this doesn't happen".  Is that a correct reading?

Could you try running with WITNESS and INVARIANTS enabled, and see if you
get any specific warnings or assertion failures?  A hard hang could imply
a deadlock, which WITNESS would be able to report on.  Other sources of
hard hangs may be easier to debug with INVARIANTS and WITNESS enabled.

If possible, getting access to a serial console might make this problem
significantly easier to debug.

>     I cannot also enter the debugger with C-M-ESC (no serial console at this
>     moment, sorry). When the system is running and I try to enter the debugger
>     on video console I get garbage on the screen and a reboot immediately
>     after. Scary stuff. I can't play with MP watchdog now (4 CPU box arrives in
>     two weeks).

So when there isn't a problem and you try to enter the debugger on the
video console, you get the garbage, or only when this problem is
manifesting? 

Thanks,

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research