Weird system cpu usage

Mike Tancsa mike at sentex.net
Wed Mar 19 13:34:17 PDT 2008


At 02:06 PM 3/19/2008, Jeremy Chadwick wrote:
>On Wed, Mar 19, 2008 at 06:53:36PM +0100, Oliver Fromme wrote:
> > Charlie Root <root at ha-web1.hockeyarena.net> wrote:
> >  > root@[ha-web1 ~]# vmstat -i
> >  > interrupt                          total       rate
> >  > irq1: atkbd0                          12          0
> >  > irq16: ohci0                           1          0
> >  > irq17: ohci1 ohci3                     1          0
> >  > irq18: ohci2 ohci4                     1          0
> >  > irq20: em0                      86255835       1361
> >  > irq22: em1 atapci0           18611379049     293795
> >
> > Now that looks unusual indeed.  Do you get that rate
> > on irq22 right after boot, before the services have
> > started?  It looks like either hardware or driver
> > problems.  Do you have polling enabled on em1?
>
>Also, I believe there was a report from another user who saw similar
>issues with em(4), and found that disabling MSI fixed the storm in
>question.  I believe you can disable MSI/MSIX by placing the following
>in /boot/loader.conf, then reboot:
>
>hw.pci.enable_msi="0"
>hw.pci.enable_msix="0"

When MSI is enabled, the irq will be a strangely high number.  e.g.

% vmstat -i
interrupt                          total       rate
irq4: sio0                            76          0
irq17: em3                           360          0
irq19: atapci1                      2901          0
cpu0: timer                     33719800       1999
irq257: em1                        56571          3
irq258: em2                            4          0
cpu1: timer                     33717664       1999
Total                           67497376       4003

If anything, I found enabling MSI helped matters where I saw strange 
IRQ issues.  However, not sure if the original poster's hardware 
supports it.   One thing it does remind me of is some strange IRQ 
issues I had on an AMD board where a USB setting for "legacy handoff" 
(something like that) would really slow down the machine with an in 
inordinate amount of IRQs firing.  I forget if I had to enable it or 
disable it to fix the problem.  If anything, I would try disabling 
USB all together if its not being used even though its not figuring 
in the above really high rate of IRQs.


         ---Mike 



More information about the freebsd-stable mailing list