From mboxrd@z Thu Jan 1 00:00:00 1970 From: Wu Fengguang Subject: Re: [PATCH 30/45] vmscan: lumpy pageout Date: Wed, 7 Oct 2009 22:20:54 +0800 Message-ID: <20091007142054.GA6798@localhost> References: <8acda98c0910070338n7220fdabo8c7f8f9e7d21ef6c@mail.gmail.com> <20091007111454.GB15936@localhost> <8acda98c0910070437g1498f99eua9a5ad71e6c2bae4@mail.gmail.com> <20091007132924.GB20855@localhost> <20091007134254.GA26244@localhost> Mime-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: Andrew Morton , Theodore Tso , Christoph Hellwig , Dave Chinner , Chris Mason , Peter Zijlstra , "Li, Shaohua" , Myklebust Trond , "jens.axboe@oracle.com" , Jan Kara , Nick Piggin , "linux-fsdevel@vger.kernel.org" To: Nikita Danilov Return-path: Received: from mga14.intel.com ([143.182.124.37]:56040 "EHLO mga14.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752692AbZJGOVt (ORCPT ); Wed, 7 Oct 2009 10:21:49 -0400 Content-Disposition: inline In-Reply-To: <20091007134254.GA26244@localhost> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Wed, Oct 07, 2009 at 09:42:54PM +0800, Wu Fengguang wrote: > On Wed, Oct 07, 2009 at 09:29:24PM +0800, Wu Fengguang wrote: > > On Wed, Oct 07, 2009 at 07:37:35PM +0800, Nikita Danilov wrote: > > > 2009/10/7 Wu Fengguang : > > > > On Wed, Oct 07, 2009 at 06:38:37PM +0800, Nikita Danilov wrote: > > > >> Hello, > > > >> > > >=20 > > > [...] > > >=20 > > > > > > > > Glad to know about your experiences :) Interestingly I started = with > > > > ->writepage() and then switch to ->writepages() because filesys= tems > > > > behave better with the latter (i.e. less file fragmentation). > > >=20 > > > By the way, why is your patch doing > > >=20 > > > ->writepage(page->index); > > > generic_writepages(page->index + 1, LUMPY_PAGEOUT_PAGES -= 1); > > >=20 > > > instead of > > >=20 > > > generic_writepages(page->index, LUMPY_PAGEOUT_PAGES); > > >=20 > > > ? Is this because of the difficulties with passing back page spec= ific > > > errors from generic_writepages()? > >=20 > > Yes. It's possible to tell write_cache_pages() to return > > AOP_WRITEPAGE_ACTIVATE. Other ->writepages() don't have to deal wit= h > > this because their ->writepage() won't return AOP_WRITEPAGE_ACTIVAT= E > > at all. > >=20 > > But it is going to be ugly to specialize the first locked page in > > every ->writepages() functions.. > >=20 > > > > > > > > I'd like to just ignore the shmem case, by adding a > > > > bdi_cap_writeback_dirty() check. Because clustered writing to s= wap > > > > device may be a less gain. > > >=20 > > > Or you can just call try_to_unmap() from shmem_writepage() when > > > wbc->for_reclaim is true. > >=20 > > Hmm, it's more comfortable to stay away from shmem for the initial = version. > > But feel free to submit a patch for it in future :) > >=20 > > > > Page filtering should also be possible in write_cache_pages(). = =C2=A0But > > > > what do you mean by "hard-to-fix races against inode reclamatio= n"? > > >=20 > > > vmscan.c pageout path doesn't take a reference on inode, so the > > > instant ->writepage() releases lock on the page, the inode can be > > > freed. > >=20 > > Ah, then we could just do igrab() if inode_lock is not locked? > > A bit ugly though. >=20 > Since writepage() can sleep, we don't need to worry about inode_lock. >=20 > Here is the updated patch. Sorry forget to check return value of igrab().. --- vmscan: lumpy pageout When pageout a dirty page, try to piggy back more consecutive dirty pages (up to 512KB) to improve IO efficiency. Only ext3/reiserfs which don't have its own aops->writepages are supported in this initial version. CC: Dave Chinner CC: Nikita Danilov Signed-off-by: Wu Fengguang --- mm/page-writeback.c | 12 ++++++++++++ mm/vmscan.c | 16 ++++++++++++++++ 2 files changed, 28 insertions(+) --- linux.orig/mm/vmscan.c 2009-10-07 21:39:13.000000000 +0800 +++ linux/mm/vmscan.c 2009-10-07 22:18:55.000000000 +0800 @@ -344,6 +344,8 @@ typedef enum { PAGE_CLEAN, } pageout_t; =20 +#define LUMPY_PAGEOUT_PAGES (512 * 1024 / PAGE_CACHE_SIZE) + /* * pageout is called by shrink_page_list() for each dirty page. * Calls ->writepage(). @@ -398,6 +400,7 @@ static pageout_t pageout(struct page *pa .nonblocking =3D 1, .for_reclaim =3D 1, }; + struct inode *inode =3D igrab(mapping->host); =20 SetPageReclaim(page); res =3D mapping->a_ops->writepage(page, &wbc); @@ -405,10 +408,23 @@ static pageout_t pageout(struct page *pa handle_write_error(mapping, page, res); if (res =3D=3D AOP_WRITEPAGE_ACTIVATE) { ClearPageReclaim(page); + iput(inode); return PAGE_ACTIVATE; } =20 /* + * only write_cache_pages() supports for_reclaim for now + * ignore shmem for now, thanks to Nikita. + */ + if (bdi_cap_writeback_dirty(mapping->backing_dev_info) && + !mapping->a_ops->writepages) { + wbc.range_start =3D (page->index + 1) << PAGE_CACHE_SHIFT; + wbc.nr_to_write =3D LUMPY_PAGEOUT_PAGES - 1; + generic_writepages(mapping, &wbc); + iput(inode); + } + + /* * Wait on writeback if requested to. This happens when * direct reclaiming a large contiguous area and the * first attempt to free a range of pages fails. --- linux.orig/mm/page-writeback.c 2009-10-07 21:39:13.000000000 +0800 +++ linux/mm/page-writeback.c 2009-10-07 21:39:14.000000000 +0800 @@ -805,6 +805,11 @@ int write_cache_pages(struct address_spa break; } =20 + if (wbc->for_reclaim && done_index !=3D page->index) { + done =3D 1; + break; + } + if (nr_to_write !=3D wbc->nr_to_write && done_index + WB_SEGMENT_DIST < page->index && --wbc->nr_segments <=3D 0) { @@ -846,6 +851,13 @@ continue_unlock: if (!clear_page_dirty_for_io(page)) goto continue_unlock; =20 + /* + * active and unevictable pages will be checked at + * rotate time + */ + if (wbc->for_reclaim) + SetPageReclaim(page); + ret =3D (*writepage)(page, wbc, data); if (unlikely(ret)) { if (ret =3D=3D AOP_WRITEPAGE_ACTIVATE) { -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel= " in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html