Severity level of disable IRQ messages?

[Josh Boyer]

On Thu, Mar 08, 2012 at 09:37:14AM -0500, Adam Jackson wrote:
> On Wed, 2012-03-07 at 16:35 -0600, Bruno Wolff III wrote:
> > I am getting syslog warnings sent out to my terminal sessions when
> > IRQs get disabled. Is this really something severe enough to deserve
> > that kind of notice? If so, should I be filing bugs?
> > 
> > Sample message:
> > Message from syslogd at wolff at Mar  7 16:29:51 ...
> >  kernel:[63120.139549] Disabling IRQ 15
> > 
> > This is with 3.3.0-0.rc6.git0.2.fc17.i686.PAE.
> 
> KERN_EMERG is a bit extreme but they're often real bugs.  That message
> only fires if we've received a ton of interrupts with nothing handling
> them, which means there's a device somewhere that we're either not
> handling properly or not ignoring properly.
> 
> I think we'd need the content of /proc/interrupts and probably the
> output of lspci -v to be able to do anything with the bug report.

Actually, I think this is fallout from a patch we added to automatically
fall back to irqpoll mode when this happens.  A few others have
mentioned it's now spamming their logs.  See bug 797369 for a similar
comment.

The patch is supposed to help some machines that have a balky pci bridge
that doesn't seem to do interrupt acks correctly.  It just seems a bit
verbose at the moment.

josh