SCHED_ULE / NetGraph interaction broken somwhere between r227874 and r229818

[Коньков Евгений]

Здравствуйте, Lev.

Вы писали 12 января 2012 г., 13:07:15:

LS> Hello, Lev.
LS> You wrote 12 января 2012 г., 15:00:20:

>>> But what mav says makes sense.
>>  It is it -- stack size. Setting KSTACK_PAGES=6 fixes situation.
LS>   OOOPS. Not. After another 5 minutes ng_queue again consumes 100% CPU
LS> :(

FreeBSD r229881   10-CURRENT

IF Load http://piccy.info/view3/2474120/f4a4eb06a8af54434e9643024737ae30/orig/
CPU load http://piccy.info/view3/2474125/0d4474f8f3f4491eb49a11cdf3a301f4/orig/

if traffic goes normal: there are {irq256: re0}, ng_queue  and netisr threads in top
when there is fall: {irq256: re0}, ng_queue and no intr{swi1: netisr 3} in top,
when intr{swi1: netisr 3} appear in 'top' then traffic starts to flow
normally until next fall: 16:05; 16:28, 18:05: 18:30, 19:05: 19:30
it is like 30min interval.

19:45 is reboot

Also freebsd stops to work on igb after couple of hours.
There is no such problem on re

-- 
С уважением,
 Коньков                          mailto:kes-kes at yandex.ru