Re: ksoftirqd/n permanently eating 60% of a CPU

[Louis-David Mitterrand <vindex+lists-linux-kernel@apartia.org>]

On Mon, Sep 13, 2010 at 09:52:37AM +0200, Louis-David Mitterrand wrote:
> On Sun, Sep 12, 2010 at 08:57:15PM +0200, Andreas Mohr wrote:
> > > Examining /proc/interrupts shows that "Local timer interrupts" are
> > > mainly firing.
> > 
> > > Strangely enough, the interrupt storm starts only 5 minutes or so after
> > > bootup.
> > 
> > And have you tried running powertop or manually (perhaps via
> > sophisticated shell script) keeping a watch on
> > /proc/timer[list|stats]?
> 
> powertop seems to give a strong hint:
> 
> 	54.5% (250.2)   [kernel core] dsp_cmx_send (dsp_cmx_send)
> 	25.8% (118.6)   usbhid-ups
> 	5.8% ( 26.7)   [ahci] <interrupt>
> 	2.6% ( 12.0)   [kernel core] usb_hcd_poll_rh_status (rh_timer_func)
> 
> Apparently dsp_cmx_send is in the mISDN stack which, in my kernel, is
> compiled statically. This happens regardless of the box having isdn
> hardware.

I can confirm that removing mISDN support from the kernel (or compiling
it as module without loading it) fixes the irq storm. I sent a note to
isdn4linux@listserv.isdn4linux.de about it.