Re: [PATCH v2] mm: remove MIGRATE_ISOLATE check in hotpath
From: Andrew Morton
Date: Mon Feb 25 2013 - 17:50:27 EST
On Mon, 25 Feb 2013 11:13:08 +0900
Minchan Kim <minchan@xxxxxxxxxx> wrote:
> > >
> > > ...
> > >
> > > @@ -683,7 +683,7 @@ static void free_one_page(struct zone *zone, struct page *page, int order,
> > > zone->pages_scanned = 0;
> > >
> > > __free_one_page(page, zone, order, migratetype);
> > > - if (unlikely(migratetype != MIGRATE_ISOLATE))
> > > + if (unlikely(!is_migrate_isolate(migratetype)))
> > > __mod_zone_freepage_state(zone, 1 << order, migratetype);
> > > spin_unlock(&zone->lock);
> > > }
> > The code both before and after this patch is assuming that the
> > migratetype in free_one_page is likely to be MIGRATE_ISOLATE. Seems
> > wrong. If CONFIG_MEMORY_ISOLATION=n this ends up doing
> > if(unlikely(true)) which is harmless-but-amusing.
> >From the beginning of [2139cbe627, cma: fix counting of isolated pages],
> it was wrong. We can't make sure it's very likely.
> If it is called by order-0 page free path, it is but if it is called by
> high order page free path, we can't.
> So I think it would be better to remove unlikley.
Order-0 pages surely preponderate, so I'd say that "likely" is the way
I don't recall anyone ever demonstrating that likely/unlikely actually
does anything useful. It would be interesting to have a play around,
see if it actually does good things to the code generation.
I think someone (perhaps in or near Dave Jones?) once had a patch which
added counters to likely/unlikely, so the kernel can accumulate and
then report upon the hit/miss ratio at each site. iirc, an alarmingly
large number of the sites were deoptimisations!
> They are trivial patch so send it now or send it after you release
> first mmotm after finishing merge window?
It's in mainline now.
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/