9 hangs with idletick = 0

[Dag-Erling Smørgrav]

Alexander Motin <mav at FreeBSD.org> writes:
> If short freezes you've descrived happens often enough, you may try to
> log them down with enabling KTR_SPARE2 ktr event type and disabling
> logging within few seconds after such freeze happened.

I've been working with adri to try to isolate it.  We've eliminated nfs
and zfs as possible sources.  I definitely think it's network-related,
because I can trigger it by downloading a large file to /dev/null.

Interestingly, it looks like serial console activity wakes it up - not
immediately, but when I hooked up the console, it woke up within seconds
after being frozen for almost ten minutes.

I've just built a kernel with KTR support, and with KTR_SPARE2, KTR_INTR
and KTR_SCHED enabled by default.  I'll see what turns up.  I'm also
going to try machdep.idle=hlt with kern.eventtimer.idletick=0, and using
a PCI re(4) instead of the on-board msk(4) while running with default
settings.

BTW, can I suggest appropriating one of KTR_SPARE[234] and renaming it
to KTR_CLOCK?  I don't see why cxgb should use them, let alone all
three; it should use KTR_DEV or KTR_NET instead.

DES
-- 
Dag-Erling Smørgrav - des at des.no