Re: [PATCH][2.5] Protect smp_call_function_data w/ spinlocks on Alpha

[Ivan Kokshaysky <ink@jurassic.park.msu.ru>]

On Mon, Feb 17, 2003 at 03:32:09AM -0500, Zwane Mwaikambo wrote:
> Assigns whatever the pointer happens to be at the time, be it NULL or the 
> next incoming message call.

No, the pointer is guaranteed to be valid.

> Therefore we'd need a lock to protect both the variable and critical 
> section.

But smp_call_function_data pointer itself is exactly such a lock -
other CPUs can't enter the section between 'if (pointer_lock())' and
'smp_call_function_data = 0', so there is no need for extra lock
variable. Additionally, pointer_lock() with retry = 0 acts as spin_trylock.

> > I happen to like the pointer_lock a lot, and think we should
> > make more use of it on systems known to have cmpxchg.  It
> > saves on the number of cache lines that have to get bounced
> > between processors.
> 
> I have to agree there, it would save on locked operations per 
> 'acquisition' which can be a win on a lot of systems.

Here's cmpxchg version for illustration (untested).

Ivan.

--- linux/arch/alpha/kernel/smp.c.orig	Mon Feb 10 21:38:15 2003
+++ linux/arch/alpha/kernel/smp.c	Mon Feb 17 17:05:47 2003
@@ -680,17 +680,7 @@ pointer_lock (void *lock, void *data, in
 	mb();
  again:
 	/* Compare and swap with zero.  */
-	asm volatile (
-	"1:	ldq_l	%0,%1\n"
-	"	mov	%3,%2\n"
-	"	bne	%0,2f\n"
-	"	stq_c	%2,%1\n"
-	"	beq	%2,1b\n"
-	"2:"
-	: "=&r"(old), "=m"(*(void **)lock), "=&r"(tmp)
-	: "r"(data)
-	: "memory");
-
+	old = cmpxchg(lock, 0, data);
 	if (old == 0)
 		return 0;
 	if (! retry)