Hi,
在 2025/7/30 0:12, Tony Battersby 写道:
md-raid currently sets io_min and io_opt to the RAID chunk and stripe
sizes and then calls queue_limits_stack_bdev() to combine the io_min and
io_opt values with those of the component devices. The io_opt size is
notably combined using the least common multiple (lcm), which does not
work well in practice for some drives (1), resulting in overflow or
unreasonable values.
dm-raid, on the other hand, sets io_min and io_opt through the
raid_io_hints() function, which is called after stacking all the queue
limits of the component drives, so the RAID chunk and stripe sizes
override the values of the stacking.
Change md-raid to be more like dm-raid by setting io_min and io_opt to
the RAID chunk and stripe sizes after stacking the queue limits of the
component devies. This fixes /sys/block/md0/queue/optimal_io_size from
being a bogus value like 3221127168 to being the correct RAID stripe
size.
This is already discussed, and mtp3sas should fix this strange value.
(1) SATA disks attached to mpt3sas report io_opt = 16776704, or
2^24 - 512. See also commit 9c0ba14828d6 ("blk-settings: round down
io_opt to physical_block_size").
Signed-off-by: Tony Battersby <tonyb@xxxxxxxxxxxxxxx>
---
drivers/md/md.c | 15 +++++++++++++++
drivers/md/raid0.c | 4 ++--
drivers/md/raid10.c | 4 ++--
drivers/md/raid5.c | 4 ++--
4 files changed, 21 insertions(+), 6 deletions(-)
diff --git a/drivers/md/md.c b/drivers/md/md.c
index 0f03b21e66e4..decf593d3bd7 100644
--- a/drivers/md/md.c
+++ b/drivers/md/md.c
@@ -5837,11 +5837,15 @@ EXPORT_SYMBOL_GPL(mddev_stack_rdev_limits);
int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev)
{
struct queue_limits lim;
+ unsigned int io_min;
+ unsigned int io_opt;
if (mddev_is_dm(mddev))
return 0;
lim = queue_limits_start_update(mddev->gendisk->queue);
+ io_min = lim.io_min;
+ io_opt = lim.io_opt;
queue_limits_stack_bdev(&lim, rdev->bdev, rdev->data_offset,
mddev->gendisk->disk_name);
@@ -5851,6 +5855,17 @@ int mddev_stack_new_rdev(struct mddev *mddev, struct md_rdev *rdev)
queue_limits_cancel_update(mddev->gendisk->queue);
return -ENXIO;
}
+ switch (mddev->level) {
+ case 0:
+ case 4:
+ case 5:
+ case 6:
+ case 10:
+ /* Preserve original chunk size and stripe size. */
+ lim.io_min = io_min;
+ lim.io_opt = io_opt;
+ break;
+ }
return queue_limits_commit_update(mddev->gendisk->queue, &lim);
}
diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c
index d8f639f4ae12..657e66e92e14 100644
--- a/drivers/md/raid0.c
+++ b/drivers/md/raid0.c
@@ -382,12 +382,12 @@ static int raid0_set_limits(struct mddev *mddev)
md_init_stacking_limits(&lim);
lim.max_hw_sectors = mddev->chunk_sectors;
lim.max_write_zeroes_sectors = mddev->chunk_sectors;
- lim.io_min = mddev->chunk_sectors << 9;
- lim.io_opt = lim.io_min * mddev->raid_disks;
lim.features |= BLK_FEAT_ATOMIC_WRITES;
err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
if (err)
return err;
+ lim.io_min = mddev->chunk_sectors << 9;
+ lim.io_opt = lim.io_min * mddev->raid_disks;
This is too hacky, and I'm sure will break existing users, at least this
will be a
mess to build raid array on the top of another array.
Thanks,
Kuai
return queue_limits_set(mddev->gendisk->queue, &lim);
}
diff --git a/drivers/md/raid10.c b/drivers/md/raid10.c
index c9bd2005bfd0..ea5147531ceb 100644
--- a/drivers/md/raid10.c
+++ b/drivers/md/raid10.c
@@ -4011,12 +4011,12 @@ static int raid10_set_queue_limits(struct mddev *mddev)
md_init_stacking_limits(&lim);
lim.max_write_zeroes_sectors = 0;
- lim.io_min = mddev->chunk_sectors << 9;
- lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
lim.features |= BLK_FEAT_ATOMIC_WRITES;
err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
if (err)
return err;
+ lim.io_min = mddev->chunk_sectors << 9;
+ lim.io_opt = lim.io_min * raid10_nr_stripes(conf);
return queue_limits_set(mddev->gendisk->queue, &lim);
}
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index ca5b0e8ba707..bba647c38cff 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -7727,8 +7727,6 @@ static int raid5_set_limits(struct mddev *mddev)
stripe = roundup_pow_of_two(data_disks * (mddev->chunk_sectors << 9));
md_init_stacking_limits(&lim);
- lim.io_min = mddev->chunk_sectors << 9;
- lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
lim.features |= BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE;
lim.discard_granularity = stripe;
lim.max_write_zeroes_sectors = 0;
@@ -7736,6 +7734,8 @@ static int raid5_set_limits(struct mddev *mddev)
rdev_for_each(rdev, mddev)
queue_limits_stack_bdev(&lim, rdev->bdev, rdev->new_data_offset,
mddev->gendisk->disk_name);
+ lim.io_min = mddev->chunk_sectors << 9;
+ lim.io_opt = lim.io_min * (conf->raid_disks - conf->max_degraded);
/*
* Zeroing is required for discard, otherwise data could be lost.
base-commit: 038d61fd642278bab63ee8ef722c50d10ab01e8f
