[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 03 Mar 2012 12:27:39 +0400
From: Konstantin Khlebnikov <khlebnikov@...nvz.org>
To: Hugh Dickins <hughd@...gle.com>
CC: Andrew Morton <akpm@...ux-foundation.org>,
Johannes Weiner <jweiner@...hat.com>,
KAMEZAWA Hiroyuki <kamezawa.hiroyu@...fujitsu.com>,
"linux-mm@...ck.org" <linux-mm@...ck.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 3/7] mm: rework __isolate_lru_page() file/anon filter
Hugh Dickins wrote:
> On Wed, 29 Feb 2012, Konstantin Khlebnikov wrote:
>
>> This patch adds file/anon filter bits into isolate_mode_t,
>> this allows to simplify checks in __isolate_lru_page().
>>
>> Signed-off-by: Konstantin Khlebnikov<khlebnikov@...nvz.org>
>
> Almost-Acked-by: Hugh Dickins<hughd@...gle.com>
>
> with one whitespace nit, and one functional addition requested.
>
> I'm perfectly happy with your :?s myself, but some people do dislike
> them. I'm happy with the switch alternative if it's as efficient:
> something that surprised me very much when trying to get convincing
> performance numbers for per-memcg per-zone lru_lock at home...
>
> ... __isolate_lru_page() featured astonishly high on the perf report
> of streaming from files on ext4 on /dev/ram0 to /dev/null, coming
> immediately below the obvious zeroing and copying: okay, the zeroing
> and copying were around 30% each, and __isolate_lru_page() down around
> 2% or below, but even so it seemed very odd that it should feature so
> high, and any optimizations to it very welcome - unless it was purely
> some bogus result.
Actually ANON/FILE ACTIVE/INACTIVE checks does not required at non-lumpy reclaim
(all pages are picked from right lru list) and compaction (it does not care).
But seems like removing these two bit-checks cannot give noticeable performance gain.
This patch can be postponed. It does not so important and
it does not share context with other patches in this set.
>
>> ---
>> include/linux/mmzone.h | 4 ++++
>> include/linux/swap.h | 2 +-
>> mm/compaction.c | 5 +++--
>> mm/vmscan.c | 27 +++++++++++++--------------
>> 4 files changed, 21 insertions(+), 17 deletions(-)
>>
>> diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
>> index eff4918..2fed935 100644
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -193,6 +193,10 @@ struct lruvec {
>> #define ISOLATE_UNMAPPED ((__force isolate_mode_t)0x8)
>> /* Isolate for asynchronous migration */
>> #define ISOLATE_ASYNC_MIGRATE ((__force isolate_mode_t)0x10)
>> +/* Isolate swap-backed pages */
>> +#define ISOLATE_ANON ((__force isolate_mode_t)0x20)
>> +/* Isolate file-backed pages */
>> +#define ISOLATE_FILE ((__force isolate_mode_t)0x40)
>
> From the patch you can see that the #defines above yours used a
> space where you have used a tab: better to use a space as above.
>
>> @@ -375,7 +376,7 @@ static isolate_migrate_t isolate_migratepages(struct zone *zone,
>> mode |= ISOLATE_ASYNC_MIGRATE;
>>
>> /* Try isolate the page */
>> - if (__isolate_lru_page(page, mode, 0) != 0)
>> + if (__isolate_lru_page(page, mode) != 0)
>> continue;
>
> I thought you were missing something there, but no, that's rather
> the case you are simplifying. However...
>
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index af6cfe7..1b70338 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -1520,6 +1511,10 @@ shrink_inactive_list(unsigned long nr_to_scan, struct mem_cgroup_zone *mz,
>> isolate_mode |= ISOLATE_UNMAPPED;
>> if (!sc->may_writepage)
>> isolate_mode |= ISOLATE_CLEAN;
>> + if (file)
>> + isolate_mode |= ISOLATE_FILE;
>> + else
>> + isolate_mode |= ISOLATE_ANON;
>
> Above here, under "if (sc->reclaim_mode& RECLAIM_MODE_LUMPYRECLAIM)",
> don't you need
>
> isolate_mode |= ISOLATE_ACTIVE | ISOLATE_FILE | ISOLATE_ANON;
>
> now to reproduce the same "all_lru_mode" behaviour as before?
Yes, I missed this. Thanks.
>
> Hugh
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists