[PATCH] ARM: update show_interrupts for online cpu's only

[Russell King - ARM Linux]

On Sat, Jul 31, 2010 at 08:18:05PM +0530, Shilimkar, Santosh wrote:
> > -----Original Message-----
> > From: Russell King - ARM Linux [mailto:linux at arm.linux.org.uk]
> > Sent: Saturday, July 31, 2010 8:06 PM
> > To: Shilimkar, Santosh
> > Cc: linux-arm-kernel at lists.infradead.org
> > Subject: Re: [PATCH] ARM: update show_interrupts for online cpu's only
> > 
> > On Sat, Jul 31, 2010 at 07:38:53PM +0530, Santosh Shilimkar wrote:
> > > Even after CPUx is logically offline, it's interrupt are displayed
> > > in proc entry
> > >
> > > This patch fixes the same by use of 'for_each_online_cpu' instead
> > > of 'for_each_present_cpu' in 'show_interrupts' function
> > 
> > What's the synchronisation mechanism between CPUs being brought on or
> > offline and this file being read?
> >
> Isn't that taken care by "cpu_online_mask" mask ?

The problem comes when you have:

	for_each_online_cpu(cpu)
		do_something(cpu);

	/* cpu goes offline */

	for_each_online_cpu(cpu)
		do_something_else(cpu);

This causes first iterates over more CPUs than the second.  When you
consider we iterate over the online CPU mask for each interrupt, which
could cross a read() call, it's possible that a CPU can go offline
in the middle of reading /proc/interrupts.

The real question is whether this matters.  If you're reading this
file as part of a daemon which is controlling the hotplug CPUs based
on interrupt load, then probably not.

If you're reading this as part of a daemon involved with balancing IRQs
across several CPUs, and CPU2 of 4 CPUs goes offline, you could get
confused (but hopefully your parser is good enough to spot the format
change and signal an exception.)

I feel rather uneasy about files which can spontaneously change format
on a reader at any time.