[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20240508014545.mf7pexpctfl44pq3@hippo>
Date: Wed, 8 May 2024 09:45:45 +0800
From: Xu Yang <xu.yang_2@....com>
To: Christoph Hellwig <hch@....de>
Cc: Jens Axboe <axboe@...nel.dk>, "Darrick J. Wong" <djwong@...nel.org>,
Andrew Morton <akpm@...ux-foundation.org>,
Matthew Wilcox <willy@...radead.org>,
Christian Brauner <christian@...uner.io>,
linux-block@...r.kernel.org, linux-fsdevel@...r.kernel.org,
linux-xfs@...r.kernel.org, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, Luis Chamberlain <mcgrof@...nel.org>,
Pankaj Raghav <p.raghav@...sung.com>,
Hannes Reinecke <hare@...e.de>, jun.li@....com, haibo.chen@....com
Subject: Re: [PATCH 5/6] block: use iomap for writes to block devices
On Fri, Apr 26, 2024 at 06:37:27PM +0800, Xu Yang wrote:
> Hi Christoph,
>
> On Tue, Aug 01, 2023 at 07:22:00PM +0200, Christoph Hellwig wrote:
> > Use iomap in buffer_head compat mode to write to block devices.
> >
> > Signed-off-by: Christoph Hellwig <hch@....de>
> > Reviewed-by: Luis Chamberlain <mcgrof@...nel.org>
> > Reviewed-by: Pankaj Raghav <p.raghav@...sung.com>
> > Reviewed-by: Hannes Reinecke <hare@...e.de>
> > ---
> > block/Kconfig | 1 +
> > block/fops.c | 31 +++++++++++++++++++++++++++++--
> > 2 files changed, 30 insertions(+), 2 deletions(-)
> >
> > diff --git a/block/Kconfig b/block/Kconfig
> > index 86122e459fe046..1a13ef0b1ca10c 100644
> > --- a/block/Kconfig
> > +++ b/block/Kconfig
> > @@ -5,6 +5,7 @@
> > menuconfig BLOCK
> > bool "Enable the block layer" if EXPERT
> > default y
> > + select FS_IOMAP
> > select SBITMAP
> > help
> > Provide block layer support for the kernel.
> > diff --git a/block/fops.c b/block/fops.c
> > index f0b822c28ddfe2..063ece37d44e44 100644
> > --- a/block/fops.c
> > +++ b/block/fops.c
> > @@ -15,6 +15,7 @@
> > #include <linux/falloc.h>
> > #include <linux/suspend.h>
> > #include <linux/fs.h>
> > +#include <linux/iomap.h>
> > #include <linux/module.h>
> > #include "blk.h"
> >
> > @@ -386,6 +387,27 @@ static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
> > return __blkdev_direct_IO(iocb, iter, bio_max_segs(nr_pages));
> > }
> >
> > +static int blkdev_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
> > + unsigned int flags, struct iomap *iomap, struct iomap *srcmap)
> > +{
> > + struct block_device *bdev = I_BDEV(inode);
> > + loff_t isize = i_size_read(inode);
> > +
> > + iomap->bdev = bdev;
> > + iomap->offset = ALIGN_DOWN(offset, bdev_logical_block_size(bdev));
> > + if (iomap->offset >= isize)
> > + return -EIO;
> > + iomap->type = IOMAP_MAPPED;
> > + iomap->addr = iomap->offset;
> > + iomap->length = isize - iomap->offset;
> > + iomap->flags |= IOMAP_F_BUFFER_HEAD;
> > + return 0;
> > +}
> > +
> > +static const struct iomap_ops blkdev_iomap_ops = {
> > + .iomap_begin = blkdev_iomap_begin,
> > +};
> > +
> > static int blkdev_writepage(struct page *page, struct writeback_control *wbc)
> > {
> > return block_write_full_page(page, blkdev_get_block, wbc);
> > @@ -556,6 +578,11 @@ blkdev_direct_write(struct kiocb *iocb, struct iov_iter *from)
> > return written;
> > }
> >
> > +static ssize_t blkdev_buffered_write(struct kiocb *iocb, struct iov_iter *from)
> > +{
> > + return iomap_file_buffered_write(iocb, from, &blkdev_iomap_ops);
> > +}
> > +
> > /*
> > * Write data to the block device. Only intended for the block device itself
> > * and the raw driver which basically is a fake block device.
> > @@ -605,9 +632,9 @@ static ssize_t blkdev_write_iter(struct kiocb *iocb, struct iov_iter *from)
> > ret = blkdev_direct_write(iocb, from);
> > if (ret >= 0 && iov_iter_count(from))
> > ret = direct_write_fallback(iocb, from, ret,
> > - generic_perform_write(iocb, from));
> > + blkdev_buffered_write(iocb, from));
> > } else {
> > - ret = generic_perform_write(iocb, from);
> > + ret = blkdev_buffered_write(iocb, from);
> > }
> >
> > if (ret > 0)
>
> I'm testing SSD block device write performance recently. I found the write
> speed descrased greatly on my board (330MB/s -> 130MB/s). Then I spent some
> time to find cause, finally find that it's caused by this patch and if I
> revert this patch, write speed can recover to 330MB/s.
>
> I'm using below command to test write performance:
> dd if=/dev/zero of=/dev/sda bs=4M count=1024
>
> And I also do more tests to get more findings. In short, I found write
> speed changes with the "bs=" parameter.
>
> I totally write 4GB data to sda for each test, the results as below:
>
> - dd if=/dev/zero of=/dev/sda bs=400K count=10485 (334 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=800K count=5242 (278 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=1600K count=2621 (204 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=2200K count=1906 (170 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=3000K count=1398 (150 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=4500K count=932 (139 MB/s)
>
> When this patch reverted, I got below results:
>
> - dd if=/dev/zero of=/dev/sda bs=400K count=10485 (339 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=800K count=5242 (330 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=1600K count=2621 (332 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=2200K count=1906 (333 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=3000K count=1398 (333 MB/s)
> - dd if=/dev/zero of=/dev/sda bs=4500K count=932 (333 MB/s)
>
> I just want to know if this results is expected when uses iomap, or it's
> a real issue?
>
> Many thanks in advance!
A gentle ping.
>
> Best Regards,
> Xu Yang
>
> > --
> > 2.39.2
> >
Powered by blists - more mailing lists