Re: [V3 PATCH] x86: avoid calling arch_trigger_all_cpu_backtrace() at the same time on SMP

[Don Zickus <dzickus@redhat.com>]

On Thu, Nov 11, 2010 at 12:17:02PM +0100, Ingo Molnar wrote:
> 
> Hm, another thing i noticed is that there's two of these:

I guess it depends how we want to deprecate the old nmi watchdog.
Dongdong was just patching the old and new code.

There are 3 ways we can go here.

- rip out the 'arch_trigger_all_cpu_backtrace' from the old nmi_watchdog,
  rework the Makefile and have both watchdogs use common code in hw_nmi.c

- patch both places

- rely on the fact that the old watchdog doesn't really work any more and
  only patch the new watchdog

I can put together a patch to do the first one, then have Dongdong put his
patch on top.  Selfishly I wouldn't mind just doing the third option. ;-)

Thoughts?

Cheers,
Don

> 
> >  #ifdef ARCH_HAS_NMI_WATCHDOG
> > +/* "in progress" flag of arch_trigger_all_cpu_backtrace */
> > +static unsigned long backtrace_flag;
> > +
> >  void arch_trigger_all_cpu_backtrace(void)
> >  {
> >  	int i;
> > +	unsigned long flags;
> > +
> > +	/*
> > +	 * Have to disable irq here, as the
> > +	 * arch_trigger_all_cpu_backtrace() could be
> > +	 * triggered by "spin_lock()" with irqs on.
> > +	 */
> > +	local_irq_save(flags);
> 
> > +/* "in progress" flag of arch_trigger_all_cpu_backtrace */
> > +static unsigned long backtrace_flag;
> > +
> >  void arch_trigger_all_cpu_backtrace(void)
> >  {
> >  	int i;
> > +	unsigned long flags;
> > +
> > +	/*
> > +	 * Have to disable irq here, as the
> > +	 * arch_trigger_all_cpu_backtrace() could be
> > +	 * triggered by "spin_lock()" with irqs on.
> > +	 */
> > +	local_irq_save(flags);
> > +
> > +	if (test_and_set_bit(0, &backtrace_flag))
> 
> A fair amount of code is being duplicated in two places - which is not nice. Lets 
> try to create a shared facility instead?
> 
> Thanks,
> 
> 	Ingo