[PATCH 21/22] ARM: CPU hotplug: ensure correct ordering of unplug

[Catalin Marinas]

On Mon, 2010-12-06 at 18:46 +0000, Russell King - ARM Linux wrote:
> On Mon, Dec 06, 2010 at 06:13:37PM +0000, Catalin Marinas wrote:
> > On 3 December 2010 20:26, Russell King - ARM Linux
> > <linux at arm.linux.org.uk> wrote:
> > > Don't call idle_task_exit() with interrupts disabled, and ensure
> > > that we have a memory barrier after interrupts are disabled but
> > > before signalling that this CPU has shut down.
> > >
> > > Signed-off-by: Russell King <rmk+kernel at arm.linux.org.uk>
> > > ---
> > >  arch/arm/kernel/smp.c |    4 +++-
> > >  1 files changed, 3 insertions(+), 1 deletions(-)
> > >
> > > diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
> > > index bebe3bd..8cc9c03 100644
> > > --- a/arch/arm/kernel/smp.c
> > > +++ b/arch/arm/kernel/smp.c
> > > @@ -269,9 +269,11 @@ void __ref cpu_die(void)
> > >  {
> > >        unsigned int cpu = smp_processor_id();
> > >
> > > -       local_irq_disable();
> > >        idle_task_exit();
> > >
> > > +       local_irq_disable();
> > > +       mb();
> > > +
> > >        /* Tell __cpu_die() that this CPU is now safe to dispose of */
> > >        complete(&cpu_died);
> >
> > I don't fully understand the point of the mb() here. If you want to
> > drain the write buffer that would be a dsb().
> 
> It's there to ensure that writes prior to the IRQ disable do not
> cross the complete() call, that is all.  local_irq_disable() does
> not have any built-in barrier semantics.

The DMB only ensures relative ordering between memory accesses and not
function calls. What other memory access you try to prevent getting
reordered? The complete() function takes a spinlock anyway so it has a
memory barrier.

-- 
Catalin