Christos Zoulas wrote:
[snip]
> I have no idea. I am running current here on two machines and it seems to
> work. But that is i386, not amd64.

Updating to -current as of just now (November 11, 1pm PST) shows the 
same problem.

Breaking into a debugger on boot yields:

Starting named.
Stopped in pid 632.1 (syslogd) at       netbsd:cpu_Debugger+0x5: 
leave
db> step
Stopped in pid 632.1 (syslogd) at       netbsd:cpu_Debugger+0x6:        ret
db>

I can keep stepping though things, and it looks like syslogd rarely 
yields (occasionally pflogd will pop up, but mostly I'm stuck running 
syslogd).

Disabling syslogd via rc.conf allows the system to boot and be 
(seemingly) fully functional.

In retrospect, this correlates with a problem I was seeing originally 
with the October 24th(ish) build, where syslogd would consume 100% cpu 
(the system was still otherwise usable).

So, now I suspect it's something with the syslog interface, rather than 
libc in general (which was my prior assumption).  There were a LOT of 
syslog related changes (that whole _ss/_r debacle) lately, so it's 
possible something was overlooked there.  I've done a complete rm -rf of 
my source tree, and re-cvs checkout, and the problem persists, so I 
don't think it's an out-of-sync database.

Any ideas what to look at next?  This is 100% repeatable for me. :-(

	ScottE