Re: [PATCH 2/2] mm: compaction: Minimise the time IRQs are disabled while isolating pages for migration

[Minchan Kim <minchan.kim@gmail.com>]

On Tue, Mar 01, 2011 at 12:07:12AM +0100, Andrea Arcangeli wrote:
> On Tue, Mar 01, 2011 at 08:01:31AM +0900, Minchan Kim wrote:
> > I am not sure it's good if we release the lock whenever lru->lock was contended
> > unconditionally? There are many kinds of lru_lock operations(add to lru, 
> > del from lru, isolation, reclaim, activation, deactivation and so on).
> 
> This is mostly to mirror cond_resched_lock (which actually uses
> spin_needbreak but it's ok to have it also when preempt is off). I
> doubt it makes a big difference but I tried to mirror
> cond_resched_lock.

But what's the benefit of releasing lock in here when lock contentionn happen where
activate_page for example? 
> 
> > Do we really need to release the lock whenever all such operations were contened?
> > I think what we need is just spin_is_contended_irqcontext.
> > Otherwise, please write down the comment for justifying for it.
> 
> What is spin_is_contended_irqcontext?

I thought what we need function is to check lock contention happened in only irq context
for short irq latency.

> 
> > This patch is for reducing for irq latency but do we have to check signal 
> > in irq hold time?
> 
> I think it's good idea to check the signal in case the loop is very
> long and this is run in direct compaction context.

I don't oppose the signal check.
I am not sure why we should check by just sign of lru_lock contention.

How about this by coarse-grained?

               /* give a chance to irqs before checking need_resched() */
               if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
                       if (fatal_signal_pending(current))
                               break;
                       spin_unlock_irq(&zone->lru_lock);
                       unlocked = true;
               }
               if (need_resched() || spin_is_contended(&zone->lru_lock)) {
                       if (!unlocked)
                               spin_unlock_irq(&zone->lru_lock);
                       cond_resched();
                       spin_lock_irq(&zone->lru_lock);
               } else if (unlocked)
                       spin_lock_irq(&zone->lru_lock);


> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

-- 
Kind regards,
Minchan Kim

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>