Re: [PATCH] MIPS: add smp_call_function_single()

[Stephane Eranian <eranian@hpl.hp.com>]

Hello,

Thanks for catching this. We will resubmit an updated version.

PHil, Peter, would you be able/willing to do this?

thanks.

On Sat, Jul 28, 2007 at 11:19:50AM +0200, Heiko Carstens wrote:
> On Fri, Jul 27, 2007 at 06:53:23AM -0700, Stephane Eranian wrote:
> > Ralf,
> > 
> > Here is take 2.
> > 
> > [MIPS] add smp_call_function_single (take 2)
> > 
> > signed-off-by: Stephane Eranian <eranian@hpl.hp.com>
> > signed-off-by: Phil Mucci <mucci@cs.utk.edu>
> > 
> > diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c
> > index be7362b..d47234c 100644
> > --- a/arch/mips/kernel/smp.c
> > +++ b/arch/mips/kernel/smp.c
> > @@ -193,6 +193,53 @@ void smp_call_function_interrupt(void)
> >  	}
> >  }
> > 
> > +int smp_call_function_single(int cpu, void (*func) (void *info), void *info, int retry,
> > +			     int wait)
> > +{
> > +	struct call_data_struct data;
> > +	int me = smp_processor_id();
> > +
> > +	/*
> > +	 * Can die spectacularly if this CPU isn't yet marked online
> > +	 */
> > +	BUG_ON(!cpu_online(me));
> > +	if (cpu == me) {
> > +		WARN_ON(1);
> > +		return -EBUSY;
> > +	}
> > +
> > +	/* Can deadlock when called with interrupts disabled */
> > +	WARN_ON(irqs_disabled());
> > +
> > +	data.func = func;
> > +	data.info = info;
> > +	atomic_set(&data.started, 0);
> > +	data.wait = wait;
> > +	if (wait)
> > +		atomic_set(&data.finished, 0);
> > +
> > +	spin_lock(&smp_call_lock);
> > +	call_data = &data;
> > +	mb();
> > +
> > +	/* Send a message to the other CPU */
> > +	core_send_ipi(cpu, SMP_CALL_FUNCTION);
> > +
> > +	/* Wait for response */
> > +	/* FIXME: lock-up detection, backtrace on lock-up */
> > +	while (atomic_read(&data.started) != 1)
> > +		barrier();
> > +
> > +	if (wait)
> > +		while (atomic_read(&data.finished) != 1)
> > +			barrier();
> > +	call_data = NULL;
> > +	spin_unlock(&smp_call_lock);
> > +
> > +	return 0;
> > +}
> > +EXPORT_SYMBOL(smp_call_function_single);
> > +
> >  static void stop_this_cpu(void *dummy)
> >  {
> 
> This will not do the right thing. Semantics of smp_call_function_single()
> changed recently. It now is supposed to call func() locally with irqs
> disabled if cpu == smp_processor_id(). See i386/x86_64 and powerpc.
> Unfortunately ia64 hasn't been changed yet, so it will behave differently.

-- 

-Stephane