From: Minchan Kim <minchan@kernel.org>
To: Rik van Riel <riel@redhat.com>
Cc: Marek Szyprowski <m.szyprowski@samsung.com>,
Mel Gorman <mgorman@suse.de>,
Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>,
Andrew Morton <akpm@linux-foundation.org>,
linux-kernel@vger.kernel.org, linux-mm@kvack.org
Subject: Re: [RFC 2/2] cma: support MIGRATE_DISCARD
Date: Thu, 16 Aug 2012 09:15:17 +0900 [thread overview]
Message-ID: <20120816001517.GC15225@bbox> (raw)
In-Reply-To: <20120815233323.GB15225@bbox>
On Thu, Aug 16, 2012 at 08:33:23AM +0900, Minchan Kim wrote:
> Hi Rik,
>
> On Wed, Aug 15, 2012 at 02:58:01PM -0400, Rik van Riel wrote:
> > On 08/14/2012 04:57 AM, Minchan Kim wrote:
> > >This patch introudes MIGRATE_DISCARD mode in migration.
> > >It drop clean cache pages instead of migration so that
> > >migration latency could be reduced. Of course, it could
> > >evict code pages but latency of big contiguous memory
> > >is more important than some background application's slow down
> > >in mobile embedded enviroment.
> >
> > Would it be an idea to only drop clean UNMAPPED
> > page cache pages?
>
> Firstly I thougt about that but I chose more agressive thing.
> Namely, even drop mapped page cache.
> Because it can reduce latency more(ex, memcpy + remapping cost
> during migration) and it could not trivial if migration range is big.
>
> >
> > >Signed-off-by: Minchan Kim <minchan@kernel.org>
> >
> > >@@ -799,12 +802,39 @@ static int __unmap_and_move(struct page *page, struct page *newpage,
> > > goto skip_unmap;
> > > }
> > >
> > >+ file = page_is_file_cache(page);
> > >+ ttu_flags = TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS;
> > >+
> > >+ if (!(mode & MIGRATE_DISCARD) || !file || PageDirty(page))
> > >+ ttu_flags |= TTU_MIGRATION;
> > >+ else
> > >+ discard_mode = true;
> > >+
> > > /* Establish migration ptes or remove ptes */
> > >- try_to_unmap(page, TTU_MIGRATION|TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS);
> > >+ try_to_unmap(page, ttu_flags);
> >
> > This bit looks wrong, because you end up ignoring
> > mlock and then discarding the page.
>
> Argh, Thanks!
> I will fix it in next spin.
>
> >
> > Only dropping clean page cache pages that are not
> > mapped would avoid that problem, without introducing
> > much complexity in the code.
>
> Hmm, I don't think it makes code much complex.
> How about this?
>
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 0f3b7cd..0909d79 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1223,7 +1223,8 @@ out:
> * repeatedly from try_to_unmap_ksm, try_to_unmap_anon or try_to_unmap_file.
> */
> int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> - unsigned long address, enum ttu_flags flags)
> + unsigned long address, enum ttu_flags flags,
> + unsigned long *vm_flags)
> {
> struct mm_struct *mm = vma->vm_mm;
> pte_t *pte;
> @@ -1235,6 +1236,7 @@ int try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
> if (!pte)
> goto out;
>
> + vm_flags |= vma->vm_flags;
> /*
> * If the page is mlock()d, we cannot swap it out.
> * If it's recently referenced (perhaps page_referenced
> @@ -1652,7 +1654,7 @@ out:
> * SWAP_FAIL - the page is unswappable
> * SWAP_MLOCK - page is mlocked.
> */
> -int try_to_unmap(struct page *page, enum ttu_flags flags)
> +int try_to_unmap(struct page *page, enum ttu_flags flags, unsigned long *vm_flags)
> {
> int ret;
>
> <snip>
>
> + file = page_is_file_cache(page);
> + ttu_flags = TTU_IGNORE_MLOCK|TTU_IGNORE_ACCESS;
> +
> + if (!(mode & MIGRATE_DISCARD) || !file || PageDirty(page) ||
> + vm_flags & VM_LOCKED)
We do try_to_unmap after this piece so we can't get the information in advance. :(
I don't have better idea which doesn't have a drawback so I will accept your idea.
Thanks, Rik.
> + ttu_flags |= TTU_MIGRATION;
> + else
> + discard_mode = true;
> +
>
>
--
Kind regards,
Minchan Kim
prev parent reply other threads:[~2012-08-16 0:13 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-08-14 8:57 [RFC 0/2] Reduce alloc_contig_range latency Minchan Kim
2012-08-14 8:57 ` [RFC 1/2] cma: remove __reclaim_pages Minchan Kim
2012-08-15 18:49 ` Rik van Riel
2012-08-16 13:58 ` Mel Gorman
2012-08-17 1:05 ` Minchan Kim
2012-08-17 14:48 ` Marek Szyprowski
2012-08-14 8:57 ` [RFC 2/2] cma: support MIGRATE_DISCARD Minchan Kim
2012-08-14 14:19 ` Michal Nazarewicz
2012-08-15 23:20 ` Minchan Kim
2012-08-16 13:17 ` Michal Nazarewicz
2012-08-15 18:58 ` Rik van Riel
2012-08-15 23:33 ` Minchan Kim
2012-08-16 0:15 ` Minchan Kim [this message]
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=20120816001517.GC15225@bbox \
--to=minchan@kernel.org \
--cc=akpm@linux-foundation.org \
--cc=kamezawa.hiroyu@jp.fujitsu.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-mm@kvack.org \
--cc=m.szyprowski@samsung.com \
--cc=mgorman@suse.de \
--cc=riel@redhat.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).