64-bit atomic ops on 32-bit CPU -- was: ZFS .. on PowerPC ?

[Marcel Moolenaar]

On Sep 24, 2008, at 2:07 PM, Marcel Moolenaar wrote:

> Ok. Let's assume we lose the reservation and we're
> forced to loop and try again. Step 2 then reads W2,
> which on a retry is the word written in step 4. As
> such, if we have to loop and retry X times, then
> the atomic increment would increment by X and not
> 1.
>
> I don't think an unconditional write to W2 will
> work if we don't preserve the origional value of
> W2.

What about the following algorithm:

/* Atomically read 64-bit entity. */
1:
	lwarx	P[0]
	ldw	P[1]
	stwcx	P[0]
	b.fail	1

/* Perform operation on 64-bit entity. */
	{Q[0],Q[1]} = operation({P[0],P[1]})

/*
  * Pseudo-atomically write 64-bit value.
  * The 64-bit entity may have been clobbered
  * by this time.
  */
	lwarx	t[0]
	ldw	t[1]
	cmp	t[0],P[0]
	b.ne	1
	cmp	t[1],P[1]
	b.ne	1
	stw	Q[1]
	stwcx	Q[0]
	b.ok	2
	stw	P[1]
	b	1
2:
	sync
	ret

Interrupts should be disabled to minimize the time
the 64-bit entity is inconsistent WRT non-atomic
reads.

Thoughts?

-- 
Marcel Moolenaar
xcllnt at mac.com