Re: [patch]vmscan: make kswapd use a correct order

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

On Thu, Dec 2, 2010 at 10:05 AM, Shaohua Li <shaohua.li@intel.com> wrote:
> On Thu, 2010-12-02 at 08:54 +0800, Minchan Kim wrote:
>> On Thu, Dec 2, 2010 at 9:29 AM, Shaohua Li <shaohua.li@intel.com> wrote:
>> > On Wed, 2010-12-01 at 23:58 +0800, Minchan Kim wrote:
>> >> On Wed, Dec 01, 2010 at 06:44:27PM +0900, KOSAKI Motohiro wrote:
>> >> > > T0: Task1 wakeup_kswapd(order=3)
>> >> > > T1: kswapd enters balance_pgdat
>> >> > > T2: Task2 wakeup_kswapd(order=2), because pages reclaimed by kswapd are used
>> >> > > quickly
>> >> > > T3: kswapd exits balance_pgdat. kswapd will do check. Now new order=2,
>> >> > > pgdat->kswapd_max_order will become 0, but order=3, if sleeping_prematurely,
>> >> > > then order will become pgdat->kswapd_max_order(0), while at this time the
>> >> > > order should 2
>> >> > > This isn't a big deal, but we do have a small window the order is wrong.
>> >> > >
>> >> > > Signed-off-by: Shaohua Li <shaohua.li@intel.com>
>> >> > >
>> >> > > diff --git a/mm/vmscan.c b/mm/vmscan.c
>> >> > > index d31d7ce..15cd0d2 100644
>> >> > > --- a/mm/vmscan.c
>> >> > > +++ b/mm/vmscan.c
>> >> > > @@ -2450,7 +2450,7 @@ static int kswapd(void *p)
>> >> > >                           }
>> >> > >                   }
>> >> > >
>> >> > > -                 order = pgdat->kswapd_max_order;
>> >> > > +                 order = max_t(unsigned long, new_order, pgdat->kswapd_max_order);
>> >> > >           }
>> >> > >           finish_wait(&pgdat->kswapd_wait, &wait);
>> >> >
>> >> > Good catch!
>> >> >
>> >> > But unfortunatelly, the code is not correct. At least, don't fit corrent
>> >> > design.
>> >> >
>> >> > 1) if "order < new_order" condition is false, we already decided to don't
>> >> >    use new_order. So, we shouldn't use new_order after kswapd_try_to_sleep()
>> >> > 2) if sleeping_prematurely() return false, it probably mean
>> >> >    zone_watermark_ok_safe(zone, order, high_wmark) return false