> If you can reliably reproduce it, can you try adjusting 
> downwards the desiredvnodes multiplier at the beginning of 
> gnu/sys/ufs/ffs/ffs_softdep.c:softdep_initialize() 
> and see if that causes the problem to go away?

I'll check that, but I'm not sure I can reliably reproduce it.

What I've done botht times the panic happened was:

 - sup new sources
 - build new kernel
 - reboot
 - make includes, build usr/src/lib and install it
 - make && make install in /usr/src

That was OK. But then I do the regression test: make cleandir && make in 
/usr/src.

Somewhere during the make the system paniced.

Sometimes I reboot after the make install (before the cleandir and second build)
and I've never seen a panic then.

Has the softdep code some problem with still busy vnodes not referenced any
more from the filesystem (/sbin/init, removed and replaced by the first build)?

BTW, the backtrace I posted showed Xintr5 (the ISDN card) near the bottom, a
lot of stuff above it and then Xintr14 (the IDE controller), but the ISDN 
driver never touches a VNODE nor knows anything about softdeps or fs-related
locks, so I guess it's just a coincidence (an ISDN packet arriving, interrupted
by a disk activity, although there were no active ISDN connections at that 
time, but there may be D channel backround stuff going on).


Martin