Re: [PATCH 2/6] BC: beancounters core (API)

[Oleg Nesterov <oleg@tv-sign.ru>]

On 08/23, Andrew Morton wrote:
>
> On Wed, 23 Aug 2006 15:03:07 +0400
> Kirill Korotaev <dev@sw.ru> wrote:
> 
> > +void __put_beancounter(struct beancounter *bc)
> > +{
> > +	unsigned long flags;
> > +
> > +	/* equivalent to atomic_dec_and_lock_irqsave() */
> > +	local_irq_save(flags);
> > +	if (likely(!atomic_dec_and_lock(&bc->bc_refcount, &bc_hash_lock))) {
> > +		local_irq_restore(flags);
> > +		if (unlikely(atomic_read(&bc->bc_refcount) < 0))
> > +			printk(KERN_ERR "BC: Bad refcount: bc=%p, "
> > +					"luid=%d, ref=%d\n",
> > +					bc, bc->bc_id,
> > +					atomic_read(&bc->bc_refcount));
> > +		return;
> > +	}
> > +
> > +	BUG_ON(bc == &init_bc);
> > +	verify_held(bc);
> > +	hlist_del(&bc->hash);
> > +	spin_unlock_irqrestore(&bc_hash_lock, flags);
> > +	kmem_cache_free(bc_cachep, bc);
> > +}
> 
> I wonder if it's safe and worthwhile to optimise away the local_irq_save():

Suppose ->bc_refcount == 1

> 	if (atomic_dec_and_test(&bc->bc_refcount)) {

Yes, preempted or blocks on spin_lock() below.

another cpu locks bc_hash_lock, does get_beancounter() (beancounter_findcreate),
then does put_beancounter(), and frees it.

> 		spin_lock_irqsave(&bc_hash_lock, flags);
> 		if (atomic_read(&bc->bc_refcount) == 0) {

Yes,

> 			free it
>

Already freed.

Oleg.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/