From: Minchan Kim <minchan@kernel.org>
To: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrew Morton <akpm@linux-foundation.org>,
linux-kernel@vger.kernel.org, linux-mm@kvack.org,
jlayton@poochiereds.net, bfields@fieldses.org,
Joonsoo Kim <iamjoonsoo.kim@lge.com>,
koct9i@gmail.com, aquini@redhat.com,
virtualization@lists.linux-foundation.org,
Mel Gorman <mgorman@suse.de>, Hugh Dickins <hughd@google.com>,
Sergey Senozhatsky <sergey.senozhatsky@gmail.com>,
Rik van Riel <riel@redhat.com>,
rknize@motorola.com, Gioh Kim <gi-oh.kim@profitbricks.com>,
Sangseok Lee <sangseok.lee@lge.com>,
Chan Gyun Jeong <chan.jeong@lge.com>,
Al Viro <viro@ZenIV.linux.org.uk>,
YiPing Xu <xuyiping@hisilicon.com>,
dri-devel@lists.freedesktop.org, Gioh Kim <gurugio@hanmail.net>
Subject: Re: [PATCH v3 02/16] mm/compaction: support non-lru movable page migration
Date: Thu, 7 Apr 2016 11:35:32 +0900 [thread overview]
Message-ID: <20160407023532.GD15178@bbox> (raw)
In-Reply-To: <57026B12.6060000@suse.cz>
On Mon, Apr 04, 2016 at 03:24:34PM +0200, Vlastimil Babka wrote:
> On 04/04/2016 07:12 AM, Minchan Kim wrote:
> >On Fri, Apr 01, 2016 at 11:29:14PM +0200, Vlastimil Babka wrote:
> >>Might have been better as a separate migration patch and then a
> >>compaction patch. It's prefixed mm/compaction, but most changed are
> >>in mm/migrate.c
> >
> >Indeed. The title is rather misleading but not sure it's a good idea
> >to separate compaction and migration part.
>
> Guess it's better to see the new functions together with its user
> after all, OK.
>
> >I will just resend to change the tile from "mm/compaction" to
> >"mm/migration".
>
> OK!
>
> >>Also I'm a bit uncomfortable how isolate_movable_page() blindly expects that
> >>page->mapping->a_ops->isolate_page exists for PageMovable() pages.
> >>What if it's a false positive on a PG_reclaim page? Can we rely on
> >>PG_reclaim always (and without races) implying PageLRU() so that we
> >>don't even attempt isolate_movable_page()?
> >
> >For now, we shouldn't have such a false positive because PageMovable
> >checks page->_mapcount == PAGE_MOVABLE_MAPCOUNT_VALUE as well as PG_movable
> >under PG_lock.
> >
> >But I read your question about user-mapped drvier pages so we cannot
> >use _mapcount anymore so I will find another thing. A option is this.
> >
> >static inline int PageMovable(struct page *page)
> >{
> > int ret = 0;
> > struct address_space *mapping;
> > struct address_space_operations *a_op;
> >
> > if (!test_bit(PG_movable, &(page->flags))
> > goto out;
> >
> > mapping = page->mapping;
> > if (!mapping)
> > goto out;
> >
> > a_op = mapping->a_op;
> > if (!aop)
> > goto out;
> > if (a_op->isolate_page)
> > ret = 1;
> >out:
> > return ret;
> >
> >}
> >
> >It works under PG_lock but with this, we need trylock_page to peek
> >whether it's movable non-lru or not for scanning pfn.
>
> Hm I hoped that with READ_ONCE() we could do the peek safely without
> trylock_page, if we use it only as a heuristic. But I guess it would
> require at least RCU-level protection of the
> page->mapping->a_op->isolate_page chain.
>
> >For avoiding that, we need another function to peek which just checks
> >PG_movable bit instead of all things.
> >
> >
> >/*
> > * If @page_locked is false, we cannot guarantee page->mapping's stability
> > * so just the function checks with PG_movable which could be false positive
> > * so caller should check it again under PG_lock to check a_ops->isolate_page.
> > */
> >static inline int PageMovable(struct page *page, bool page_locked)
> >{
> > int ret = 0;
> > struct address_space *mapping;
> > struct address_space_operations *a_op;
> >
> > if (!test_bit(PG_movable, &(page->flags))
> > goto out;
> >
> > if (!page_locked) {
> > ret = 1;
> > goto out;
> > }
> >
> > mapping = page->mapping;
> > if (!mapping)
> > goto out;
> >
> > a_op = mapping->a_op;
> > if (!aop)
> > goto out;
> > if (a_op->isolate_page)
> > ret = 1;
> >out:
> > return ret;
> >}
>
> I wouldn't put everything into single function, but create something
> like __PageMovable() just for the unlocked peek. Unlike the
> zone->lru_lock, we don't keep page_lock() across iterations in
> isolate_migratepages_block(), as obviously each page has different
> lock.
> So the page_locked parameter would be always passed as constant, and
> at that point it's better to have separate functions.
Agree.
>
> So I guess the question is how many false positives from overlap
> with PG_reclaim the scanner will hit if we give up on
> PAGE_MOVABLE_MAPCOUNT_VALUE, as that will increase number of page
> locks just to realize that it's not actual PageMovable() page...
I don't think it's too many because PG_reclaim bit is set to only
LRU pages at the moment and we can check PageMovable after !PageLRU
check.
Thanks.
next prev parent reply other threads:[~2016-04-07 2:35 UTC|newest]
Thread overview: 65+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-03-30 7:11 [PATCH v3 00/16] Support non-lru page migration Minchan Kim
2016-03-30 7:12 ` [PATCH v3 01/16] mm: use put_page to free page instead of putback_lru_page Minchan Kim
2016-04-01 12:58 ` Vlastimil Babka
2016-04-04 1:39 ` Minchan Kim
2016-04-04 4:45 ` Naoya Horiguchi
2016-04-04 14:46 ` Vlastimil Babka
2016-04-05 1:54 ` Naoya Horiguchi
2016-04-05 8:20 ` Vlastimil Babka
2016-04-06 0:54 ` Naoya Horiguchi
2016-04-06 7:57 ` Vlastimil Babka
2016-04-04 5:53 ` Balbir Singh
2016-04-04 6:01 ` Minchan Kim
2016-04-05 3:10 ` Balbir Singh
2016-03-30 7:12 ` [PATCH v3 02/16] mm/compaction: support non-lru movable page migration Minchan Kim
2016-04-01 21:29 ` Vlastimil Babka
2016-04-04 5:12 ` Minchan Kim
2016-04-04 13:24 ` Vlastimil Babka
2016-04-07 2:35 ` Minchan Kim [this message]
2016-04-12 8:00 ` Chulmin Kim
2016-04-12 14:25 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 03/16] mm: add non-lru movable page support document Minchan Kim
2016-04-01 14:38 ` Vlastimil Babka
2016-04-04 2:25 ` Minchan Kim
2016-04-04 13:09 ` Vlastimil Babka
2016-04-07 2:27 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 04/16] mm/balloon: use general movable page feature into balloon Minchan Kim
2016-04-05 12:03 ` Vlastimil Babka
2016-04-11 4:29 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 05/16] zsmalloc: keep max_object in size_class Minchan Kim
2016-04-17 15:08 ` Sergey Senozhatsky
2016-03-30 7:12 ` [PATCH v3 06/16] zsmalloc: squeeze inuse into page->mapping Minchan Kim
2016-04-17 15:08 ` Sergey Senozhatsky
2016-04-19 7:40 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 07/16] zsmalloc: remove page_mapcount_reset Minchan Kim
2016-04-17 15:11 ` Sergey Senozhatsky
2016-03-30 7:12 ` [PATCH v3 08/16] zsmalloc: squeeze freelist into page->mapping Minchan Kim
2016-04-17 15:56 ` Sergey Senozhatsky
2016-04-19 7:42 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 09/16] zsmalloc: move struct zs_meta from mapping to freelist Minchan Kim
2016-04-17 15:22 ` Sergey Senozhatsky
2016-03-30 7:12 ` [PATCH v3 10/16] zsmalloc: factor page chain functionality out Minchan Kim
2016-04-18 0:33 ` Sergey Senozhatsky
2016-04-19 7:46 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 11/16] zsmalloc: separate free_zspage from putback_zspage Minchan Kim
2016-04-18 1:04 ` Sergey Senozhatsky
2016-04-19 7:51 ` Minchan Kim
2016-04-19 7:53 ` Sergey Senozhatsky
2016-03-30 7:12 ` [PATCH v3 12/16] zsmalloc: zs_compact refactoring Minchan Kim
2016-04-04 8:04 ` Chulmin Kim
2016-04-04 9:01 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 13/16] zsmalloc: migrate head page of zspage Minchan Kim
2016-04-06 13:01 ` Chulmin Kim
2016-04-07 0:34 ` Chulmin Kim
2016-04-07 0:43 ` Minchan Kim
2016-04-19 6:08 ` Chulmin Kim
2016-04-19 6:15 ` Minchan Kim
2016-03-30 7:12 ` [PATCH v3 14/16] zsmalloc: use single linked list for page chain Minchan Kim
2016-03-30 7:12 ` [PATCH v3 15/16] zsmalloc: migrate tail pages in zspage Minchan Kim
2016-03-30 7:12 ` [PATCH v3 16/16] zram: use __GFP_MOVABLE for memory allocation Minchan Kim
2016-03-30 23:11 ` [PATCH v3 00/16] Support non-lru page migration Andrew Morton
2016-03-31 0:29 ` Sergey Senozhatsky
2016-03-31 0:57 ` Minchan Kim
2016-03-31 0:57 ` Minchan Kim
2016-04-04 13:17 ` John Einar Reitan
2016-04-11 4:35 ` Minchan Kim
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20160407023532.GD15178@bbox \
--to=minchan@kernel.org \
--cc=akpm@linux-foundation.org \
--cc=aquini@redhat.com \
--cc=bfields@fieldses.org \
--cc=chan.jeong@lge.com \
--cc=dri-devel@lists.freedesktop.org \
--cc=gi-oh.kim@profitbricks.com \
--cc=gurugio@hanmail.net \
--cc=hughd@google.com \
--cc=iamjoonsoo.kim@lge.com \
--cc=jlayton@poochiereds.net \
--cc=koct9i@gmail.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=mgorman@suse.de \
--cc=riel@redhat.com \
--cc=rknize@motorola.com \
--cc=sangseok.lee@lge.com \
--cc=sergey.senozhatsky@gmail.com \
--cc=vbabka@suse.cz \
--cc=viro@ZenIV.linux.org.uk \
--cc=virtualization@lists.linux-foundation.org \
--cc=xuyiping@hisilicon.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).