Re: [PATCH powerpc ] Avoid debug_smp_processor_id() check in arch_spin_unlock_wait()

[Benjamin Herrenschmidt <benh@kernel.crashing.org>]

On Thu, 2013-01-10 at 17:00 +0800, Li Zhong wrote:
> Use local_paca directly in arch_spin_unlock_wait(), as all processors have the
> same value for the field shared_proc, so we don't need care racy here.

Of course that won't build if CONFIG_PPC_SPLPAR isn't defined...

Maybe you could change the definition of the SHARED_PROCESSOR
macro itself. The only possible "risk" would be a stale lppaca
if we preempt & hot unplug the CPU at the wrong time (provided
we no longer stop_machine either), I suppose if that's a real
concern we could delay freeing of lppaca's via RCU or such.

Ben.

> Reported-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> Signed-off-by: Li Zhong <zhong@linux.vnet.ibm.com>
> ---
>  arch/powerpc/lib/locks.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/powerpc/lib/locks.c b/arch/powerpc/lib/locks.c
> index bb7cfec..850bea6 100644
> --- a/arch/powerpc/lib/locks.c
> +++ b/arch/powerpc/lib/locks.c
> @@ -72,7 +72,7 @@ void arch_spin_unlock_wait(arch_spinlock_t *lock)
>  {
>  	while (lock->slock) {
>  		HMT_low();
> -		if (SHARED_PROCESSOR)
> +		if (local_paca->lppaca_ptr->shared_proc)
>  			__spin_yield(lock);
>  	}
>  	HMT_medium();