dealing with interupt storms .....

[John Baldwin]

On Friday 12 November 2004 08:49 am, Michal Mertl wrote:
> Alext Wilkinson wrote:
> > Hi all,
> >
> > How does one stop an interrupt storm in RELENG_5 ?
> >
> >> >From top(1)
> >
> > CPU states: 10.9% user,  0.0% nice,  3.9% system, 85.3% interrupt,  0.0%
> > idle
> >
> > >From vmstat(1)
> >
> >                                  total          rate
> >                                  -----          ----
> > irq18: uhci2 bt0++               9187313        426
>
> Answer to your question might be to modify sysctl hw.intr_storm_threshold.
>
> I can confirm, there is a problem in bktr(4) generating interrupt storms.
> For several months I was seeing messages about interrupt storms and system
> run fine (I haven't been using bktr for long time so I didn't care). I
> tried to roll the bktr sources as far back as possible (that is to the time
> before phk changed dev_t to struct cdev *) to no avail, so I suspect the
> problem is caused by something outside of the driver. After latest changes
> to interrupt throttling (rev. 1.117 of src/sys/kern_intr.c by jhb) the
> system performance is pretty affected until I unload bktr driver because
> bktr now seems to be doing some 120000 interrupts/second and not getting
> throttled).

This sounds like the bktr(4) driver doesn't properly program the device to 
turn off interrupts when it detaches if I've parsed this correctly.

-- 
John Baldwin <jhb at FreeBSD.org>  <><  http://www.FreeBSD.org/~jhb/
"Power Users Use the Power to Serve"  =  http://www.FreeBSD.org