Set chunk_sectors to the full stripe width (io_opt) so that the block
layer splits I/O at full stripe boundaries. This ensures that large
writes are aligned to full stripes, avoiding the read-modify-write
overhead that occurs with partial stripe writes in RAID-5/6.

When chunk_sectors is set, the block layer's bio splitting logic in
get_max_io_size() uses blk_boundary_sectors_left() to limit I/O size
to the boundary. This naturally aligns split bios to full stripe
boundaries, enabling more efficient full stripe writes.

Test results with 24-disk RAID5 (chunk_size=64k):
  dd if=/dev/zero of=/dev/md0 bs=10M oflag=direct
  Before: 461 MB/s
  After:  520 MB/s  (+12.8%)

Suggested-by: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Yu Kuai <yukuai@fnnas.com>
---
 drivers/md/raid5.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 8854e024f311..810d936560d1 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -7783,6 +7783,7 @@ static int raid5_set_limits(struct mddev *mddev)
 	lim.logical_block_size = mddev->logical_block_size;
 	lim.io_min = mddev->chunk_sectors << 9;
 	lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
+	lim.chunk_sectors = lim.io_opt >> 9;
 	lim.features |= BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE;
 	lim.discard_granularity = stripe;
 	lim.max_write_zeroes_sectors = 0;
-- 
2.51.0

在 2026/2/23 11:58, Yu Kuai 写道:

> Set chunk_sectors to the full stripe width (io_opt) so that the block
> layer splits I/O at full stripe boundaries. This ensures that large
> writes are aligned to full stripes, avoiding the read-modify-write
> overhead that occurs with partial stripe writes in RAID-5/6.
>
> When chunk_sectors is set, the block layer's bio splitting logic in
> get_max_io_size() uses blk_boundary_sectors_left() to limit I/O size
> to the boundary. This naturally aligns split bios to full stripe
> boundaries, enabling more efficient full stripe writes.
>
> Test results with 24-disk RAID5 (chunk_size=64k):
>    dd if=/dev/zero of=/dev/md0 bs=10M oflag=direct
>    Before: 461 MB/s
>    After:  520 MB/s  (+12.8%)
>
> Suggested-by: Christoph Hellwig<hch@infradead.org>
> Signed-off-by: Yu Kuai<yukuai@fnnas.com>
> ---
>   drivers/md/raid5.c | 1 +
>   1 file changed, 1 insertion(+)

Applied to md-7.0

-- 
Thansk,
Kuai

Looks good:

Reviewed-by: Christoph Hellwig <hch@lst.de>

Dear Kuai,


Thank you for your patch.

Am 23.02.26 um 04:58 schrieb Yu Kuai:
> Set chunk_sectors to the full stripe width (io_opt) so that the block
> layer splits I/O at full stripe boundaries. This ensures that large
> writes are aligned to full stripes, avoiding the read-modify-write
> overhead that occurs with partial stripe writes in RAID-5/6.
> 
> When chunk_sectors is set, the block layer's bio splitting logic in
> get_max_io_size() uses blk_boundary_sectors_left() to limit I/O size
> to the boundary. This naturally aligns split bios to full stripe
> boundaries, enabling more efficient full stripe writes.
> 
> Test results with 24-disk RAID5 (chunk_size=64k):
>    dd if=/dev/zero of=/dev/md0 bs=10M oflag=direct
>    Before: 461 MB/s
>    After:  520 MB/s  (+12.8%)

Sweet.

> Suggested-by: Christoph Hellwig <hch@infradead.org>
> Signed-off-by: Yu Kuai <yukuai@fnnas.com>

Once committed, should this be backported to the longterm and stable series?

> ---
>   drivers/md/raid5.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index 8854e024f311..810d936560d1 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -7783,6 +7783,7 @@ static int raid5_set_limits(struct mddev *mddev)
>   	lim.logical_block_size = mddev->logical_block_size;
>   	lim.io_min = mddev->chunk_sectors << 9;
>   	lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
> +	lim.chunk_sectors = lim.io_opt >> 9;
>   	lim.features |= BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE;
>   	lim.discard_granularity = stripe;
>   	lim.max_write_zeroes_sectors = 0;

Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>


Kind regards,

Paul

Hi,

在 2026/2/23 15:26, Paul Menzel 写道:
> Dear Kuai,
>
>
> Thank you for your patch.
>
> Am 23.02.26 um 04:58 schrieb Yu Kuai:
>> Set chunk_sectors to the full stripe width (io_opt) so that the block
>> layer splits I/O at full stripe boundaries. This ensures that large
>> writes are aligned to full stripes, avoiding the read-modify-write
>> overhead that occurs with partial stripe writes in RAID-5/6.
>>
>> When chunk_sectors is set, the block layer's bio splitting logic in
>> get_max_io_size() uses blk_boundary_sectors_left() to limit I/O size
>> to the boundary. This naturally aligns split bios to full stripe
>> boundaries, enabling more efficient full stripe writes.
>>
>> Test results with 24-disk RAID5 (chunk_size=64k):
>>    dd if=/dev/zero of=/dev/md0 bs=10M oflag=direct
>>    Before: 461 MB/s
>>    After:  520 MB/s  (+12.8%)
>
> Sweet.
>
>> Suggested-by: Christoph Hellwig <hch@infradead.org>
>> Signed-off-by: Yu Kuai <yukuai@fnnas.com>
>
> Once committed, should this be backported to the longterm and stable 
> series?

Sorry for the delay, this is just performance improvement, I think this will
not be backported automatically.This patch need to be send to LTS with some
explanation if someone is interested.

>
>> ---
>>   drivers/md/raid5.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
>> index 8854e024f311..810d936560d1 100644
>> --- a/drivers/md/raid5.c
>> +++ b/drivers/md/raid5.c
>> @@ -7783,6 +7783,7 @@ static int raid5_set_limits(struct mddev *mddev)
>>       lim.logical_block_size = mddev->logical_block_size;
>>       lim.io_min = mddev->chunk_sectors << 9;
>>       lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
>> +    lim.chunk_sectors = lim.io_opt >> 9;
>>       lim.features |= BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE;
>>       lim.discard_granularity = stripe;
>>       lim.max_write_zeroes_sectors = 0;
>
> Reviewed-by: Paul Menzel <pmenzel@molgen.mpg.de>
>
>
> Kind regards,
>
> Paul
>
-- 
Thansk,
Kuai