Re: Interrupt storm

[christos%zoulas.com@localhost (Christos Zoulas)]

On Oct 25, 12:35pm, kab00m%lich.phys.spbu.ru@localhost (Dima Veselov) wrote:
-- Subject: Re: Interrupt storm

| On Wed, Oct 24, 2018 at 06:01:53PM +0000, Christos Zoulas wrote:
| > >> 
| > >> I have two identical Dell R220 servers running NetBSD 8-STABLE
| > >> and they are working fine, but I noticed permanent high CPU
| > >> usage.
| > >> 
| > >> I think this is kind of a driver problem, but how can I identify
| > >> which hardware cause that load?
| > 
| > If you switch to the "threads" display by typing "t" you'll see the thread
| > that is using all the cpu. My guess is it is "ioflush".
| 
| No, it is sysmon. Everything down those 3 take 0%.
| 
|     0    15 root      96 RUN/3     86.5H 73.00% 73.00% sysmon    [system]
|     0    88 root     221 raidio/2   1:01  2.64%  2.64% raidio3   [system]
|  6799     7 named     85 kqueue/0   4:50  1.12%  1.12% -         named

Ok, I lose :-) So it must be some driver looping in sysmon events...

Perhaps you can use:

# crash
crash> ps
...
0       15 3   1       200   fffffe8c106f44e0             sysmon smtaskq
...
crash> t /a fffffe8c106f44e0
trace: pid 0 lid 15 at 0xfffffe8157910e90
sleepq_block() at sleepq_block+0xa0
cv_wait() at cv_wait+0xfb
sysmon_task_queue_thread() at sysmon_task_queue_thread+0x81


To see if you can catch a stacktrace that has the driver involved
in one of its frames.

christos