Re: Bug in ia64 specific down() function??

From: Zoltan Menyhart <Zoltan.Menyhart_at_bull.net>
Date: 2006-01-14 03:25:23
Chen, Kenneth W wrote:
> The memory order semantics for include/asm-ia64/semaphore.h:down()
> doesn't look right.  It is using atomic_dec_return, which eventually
> translate into ia64_fetch_and_add() that uses release semantics.
> Shouldn't it use acquire semantics?

What about this one:

--- linux-2.6.15-test/include/asm-ia64/semaphore.h	2006-01-10 13:54:31.000000000 +0100
+++ linux-2.6.15-test-down/include/asm-ia64/semaphore.h	2006-01-13 16:16:04.000000000 +0100
@@ -61,7 +61,7 @@ down (struct semaphore * sem)
 down (struct semaphore *sem)
 {
 	might_sleep();
-	if (atomic_dec_return(&sem->count) < 0)
+	if (ia64_fetchadd(-1, &sem->count.counter, acq) < 1)
 		__down(sem);
 }
 
@@ -75,7 +75,7 @@ down_interruptible (struct semaphore * sem)
 	int ret = 0;
 
 	might_sleep();
-	if (atomic_dec_return(&sem->count) < 0)
+	if (ia64_fetchadd(-1, &sem->count.counter, acq) < 1)
 		ret = __down_interruptible(sem);
 	return ret;
 }
@@ -85,7 +85,7 @@ down_trylock (struct semaphore *sem)
 {
 	int ret = 0;
 
-	if (atomic_dec_return(&sem->count) < 0)
+	if (ia64_fetchadd(-1, &sem->count.counter, acq) < 1)
 		ret = __down_trylock(sem);
 	return ret;
 }
@@ -93,7 +93,7 @@ up (struct semaphore *sem)
 static inline void
 up (struct semaphore * sem)
 {
-	if (atomic_inc_return(&sem->count) <= 0)
+	if (ia64_fetchadd(1, &sem->count.counter, rel) <= -1)
 		__up(sem);
 }

I do not like the too long chains of the macro definitions which --
as the example shows -- hide how the things work.
This is why I used "ia64_fetchadd()".
("IA64_FETCHADD()" is a bit nasty.)

"__down_interruptible()" and "__down()" could be cleaned up too, but
the atomic operation on the semaphore is followed by a wakeup, that
includes taking a lock, that provides the ".acq" semantics.

"__down_trylock()" is called on failure, the data area protected
by the semaphore will not be touched by the caller.

Zoltan Menyhart
-
To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Received on Sat Jan 14 03:26:08 2006

This archive was generated by hypermail 2.1.8 : 2006-01-14 03:26:16 EST