Re: [PATCH v3 16/21] mm: handle lruvec relocks in compaction

[Konstantin Khlebnikov]

KAMEZAWA Hiroyuki wrote:
> On Thu, 23 Feb 2012 17:52:56 +0400
> Konstantin Khlebnikov<khlebnikov@xxxxxxxxxx>  wrote:
>
> > Prepare for lru_lock splitting in memory compaction code.
> >
> > * disable irqs in acct_isolated() for __mod_zone_page_state(),
> >    lru_lock isn't required there.
> >
> > Signed-off-by: Konstantin Khlebnikov<khlebnikov@xxxxxxxxxx>
> > ---
> >   mm/compaction.c |   30 ++++++++++++++++--------------
> >   1 files changed, 16 insertions(+), 14 deletions(-)
> >
> > diff --git a/mm/compaction.c b/mm/compaction.c
> > index a976b28..54340e4 100644
> > --- a/mm/compaction.c
> > +++ b/mm/compaction.c
> > @@ -224,8 +224,10 @@ static void acct_isolated(struct zone *zone, struct compact_control *cc)
> >   	list_for_each_entry(page,&cc->migratepages, lru)
> >   		count[!!page_is_file_cache(page)]++;
> >
> > +	local_irq_disable();
> >   	__mod_zone_page_state(zone, NR_ISOLATED_ANON, count[0]);
> >   	__mod_zone_page_state(zone, NR_ISOLATED_FILE, count[1]);
> > +	local_irq_enable();
>
> Why we need to disable Irq here ??

__mod_zone_page_state() want this to protect per-cpu counters, maybe preempt_disable() is enough.

> >   }
> >
> >   /* Similar to reclaim, but different enough that they don't share logic */
> > @@ -262,7 +264,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
> >   	unsigned long nr_scanned = 0, nr_isolated = 0;
> >   	struct list_head *migratelist =&cc->migratepages;
> >   	isolate_mode_t mode = ISOLATE_ACTIVE|ISOLATE_INACTIVE;
> > -	struct lruvec *lruvec;
> > +	struct lruvec *lruvec = NULL;
> >
> >   	/* Do not scan outside zone boundaries */
> >   	low_pfn = max(cc->migrate_pfn, zone->zone_start_pfn);
> > @@ -294,25 +296,24 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
> >
> >   	/* Time to isolate some pages for migration */
> >   	cond_resched();
> > -	spin_lock_irq(&zone->lru_lock);
> >   	for (; low_pfn<  end_pfn; low_pfn++) {
> >   		struct page *page;
> > -		bool locked = true;
> >
> >   		/* give a chance to irqs before checking need_resched() */
> >   		if (!((low_pfn+1) % SWAP_CLUSTER_MAX)) {
> > -			spin_unlock_irq(&zone->lru_lock);
> > -			locked = false;
> > +			if (lruvec)
> > +				unlock_lruvec_irq(lruvec);
> > +			lruvec = NULL;
> >   		}
> > -		if (need_resched() || spin_is_contended(&zone->lru_lock)) {
> > -			if (locked)
> > -				spin_unlock_irq(&zone->lru_lock);
> > +		if (need_resched() ||
> > +		    (lruvec&&  spin_is_contended(&zone->lru_lock))) {
> > +			if (lruvec)
> > +				unlock_lruvec_irq(lruvec);
> > +			lruvec = NULL;
> >   			cond_resched();
> > -			spin_lock_irq(&zone->lru_lock);
> >   			if (fatal_signal_pending(current))
> >   				break;
> > -		} else if (!locked)
> > -			spin_lock_irq(&zone->lru_lock);
> > +		}
> >
> >   		/*
> >   		 * migrate_pfn does not necessarily start aligned to a
> > @@ -359,7 +360,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
> >   			continue;
> >   		}
> >
> > -		if (!PageLRU(page))
> > +		if (!__lock_page_lruvec_irq(&lruvec, page))
> >   			continue;
>
> Could you add more comments onto __lock_page_lruvec_irq() ?

Actually there is a very unlikely race with page free-realloc,
(which is fixed in Hugh's patchset, and surprisingly fixed in my old memory controller)
thus this part will be redesigned.

> Thanks,
> -Kame

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