Re: [patch 2.4.11-pre5] atomic_dec_and_lock() for alpha

[Richard Henderson]

On Tue, Oct 09, 2001 at 02:30:13PM +0400, Ivan Kokshaysky wrote:
> OK. I prefer the latter - rewriting in assembly won't allow DEBUG_SPINLOCK
> stuff in this function. OTOH, moving the return outside an asm statement
> adds only one instruction - conditional branch that falls through in the
> fast path.

Hmm.  What about a mixture:

  asm (".text					\n\
	.global atomic_dec_and_lock		\n\
	.ent atomic_dec_and_lock		\n\
atomic_dec_and_lock:				\n\
	.prologue 0				\n\
1:	ldl_l	$1, 0($16)			\n\
	subl	$1, 1, $1			\n\
	beq	2f				\n\
	stl_c	$1, 0($16)			\n\
	beq	$1, 3f				\n\
	mb					\n\
	ret					\n\
	.align	4				\n\
3:	br	1b				\n\
2:	lda	$27, atomic_dec_and_lock_1");

	/* FALLTHRU */
	
static int
atomic_dec_and_lock_1(atomic_t *atomic, spinlock_t *lock)
{
	/* Slow path */
	spin_lock(lock);
	if (atomic_dec_and_test(atomic))
		return 1;
	spin_unlock(lock);
	return 0;
}

r~