[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170511024123.GB3967@jaegeuk.local>
Date: Wed, 10 May 2017 19:41:23 -0700
From: Jaegeuk Kim <jaegeuk@...nel.org>
To: Chao Yu <yuchao0@...wei.com>
Cc: linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
linux-f2fs-devel@...ts.sourceforge.net
Subject: Re: [f2fs-dev] [PATCH 1/3] f2fs: use f2fs_submit_page_bio for
ra_meta_pages
On 05/11, Chao Yu wrote:
> Hi Jaegeuk,
>
> On 2017/5/11 7:48, Jaegeuk Kim wrote:
> > This patch avoids to use f2fs_submit_merged_bio for read, which was the only
> > read case.
>
> This makes f2fs losing the chance to merge multiple pages into one bio during
> reading continuous physical blocks, it may cause potential performance
> regression, how about using a local bio in ra_meta_pages to cache more pages?
This is a readahead flow, which is asynchronous and not a performance critical
flow. And, I expect blk_plug is still able to merge them. We're using this in
readahead of node blocks as well.
Thanks,
>
> Thanks,
>
> >
> > Signed-off-by: Jaegeuk Kim <jaegeuk@...nel.org>
> > ---
> > fs/f2fs/checkpoint.c | 4 +---
> > 1 file changed, 1 insertion(+), 3 deletions(-)
> >
> > diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c
> > index ea9c317b5916..8d92f8249000 100644
> > --- a/fs/f2fs/checkpoint.c
> > +++ b/fs/f2fs/checkpoint.c
> > @@ -207,12 +207,10 @@ int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
> > }
> >
> > fio.page = page;
> > - fio.old_blkaddr = fio.new_blkaddr;
> > - f2fs_submit_page_mbio(&fio);
> > + f2fs_submit_page_bio(&fio);
> > f2fs_put_page(page, 0);
> > }
> > out:
> > - f2fs_submit_merged_bio(sbi, META, READ);
> > blk_finish_plug(&plug);
> > return blkno - start;
> > }
> >
Powered by blists - more mailing lists