Re: heartbeat panic: swapping?

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

> From: Thomas Klausner <wiz%netbsd.org@localhost>
> Date: Tue, 17 Feb 2026 08:34:57 +0100
> 
> savecore: reboot after panic: [ 312649.0082489] panic: cpu0: softints stuck for 16 seconds
> [...]
> #3  0xffffffff80e7736a in panic (fmt=fmt@entry=0xffffffff816f1d30 "%s: softints stuck for %u seconds") at /usr/src/sys/kern/subr_prf.c:209
> #4  0xffffffff80e0e03c in heartbeat () at /usr/src/sys/kern/kern_heartbeat.c:702
> #5  0xffffffff80df34ad in hardclock (frame=0xffffa72469e83ce8) at /usr/src/sys/kern/kern_clock.c:342
> #6  0xffffffff8021075e in Xresume_lapic_ltimer ()
> 
> and since I had been putting pressure on the system to use much swap,
> and this was the result, I was wondering if there might be a relation.
> 
> I'm bulk building in tmpfs, and at times when there are a lot of
> pkg_add/pkg_delete and the system kind of hangs already (which would
> be nice to avoid). Can this be related?

It's unlikely to be directly related to swapping because softints are
forbidden to do anything that sleeps waiting for memory, and you would
be more likely hit an ASSERT_SLEEPABLE() panic immediately if such a
path were even attempted.

> Anyway, core dump available if someone has ideas what to check.

In crash(8), can you get the output of the following commands?

ps
ps/w
show all tstiles

In the ps output, you'll see various softXXX/N lines for the softint
threads (softnet, softclock, ...) on cpuN.  Can you also get the
output of

bt/a 0xffff0123abcd3210

for the lwp addr in each of those lines?

And, can you file a PR with all this information?

(We should maybe convince the heartbeat panic to automatically show ps
output, or perhaps just the softint-related ps output, since deadlocks
like PR 59963 that happen to involve softints will always manifest
with the same panic and stack trace you saw even if they are otherwise
unrelated.)