Commit be04c14a authored by Mike Snitzer's avatar Mike Snitzer

dm: use op specific max_sectors when splitting abnormal io

Split abnormal IO in terms of the corresponding operation specific
max_sectors (max_discard_sectors, max_secure_erase_sectors or
max_write_zeroes_sectors).

This fixes a significant dm-thinp discard performance regression that
was introduced with commit e2dd8aca ("dm bio prison v1: improve
concurrent IO performance"). Relative to discard: max_discard_sectors
is used instead of max_sectors; which fixes excessive discard splitting
(e.g. max_sectors=128K vs max_discard_sectors=64M).

Tested by discarding an 1 Petabyte dm-thin device:
lvcreate -V 1125899906842624B -T test/pool -n thin
time blkdiscard /dev/test/thin

Before this fix (splitting discards every 128K): ~116m
 After this fix (splitting discards every 64M) : 0m33.460s
Reported-by: default avatarZorro Lang <zlang@redhat.com>
Fixes: 06961c48 ("dm: split discards further if target sets max_discard_granularity")
Requires: 13f6facf ("dm: allow targets to require splitting WRITE_ZEROES and SECURE_ERASE")
Fixes: e2dd8aca ("dm bio prison v1: improve concurrent IO performance")
Signed-off-by: default avatarMike Snitzer <snitzer@kernel.org>
parent 722d9082
......@@ -1172,7 +1172,8 @@ static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
}
static sector_t __max_io_len(struct dm_target *ti, sector_t sector,
unsigned int max_granularity)
unsigned int max_granularity,
unsigned int max_sectors)
{
sector_t target_offset = dm_target_offset(ti, sector);
sector_t len = max_io_len_target_boundary(ti, target_offset);
......@@ -1186,13 +1187,13 @@ static sector_t __max_io_len(struct dm_target *ti, sector_t sector,
if (!max_granularity)
return len;
return min_t(sector_t, len,
min(queue_max_sectors(ti->table->md->queue),
min(max_sectors ? : queue_max_sectors(ti->table->md->queue),
blk_chunk_sectors_left(target_offset, max_granularity)));
}
static inline sector_t max_io_len(struct dm_target *ti, sector_t sector)
{
return __max_io_len(ti, sector, ti->max_io_len);
return __max_io_len(ti, sector, ti->max_io_len, 0);
}
int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
......@@ -1581,12 +1582,13 @@ static void __send_empty_flush(struct clone_info *ci)
static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
unsigned int num_bios,
unsigned int max_granularity)
unsigned int max_granularity,
unsigned int max_sectors)
{
unsigned int len, bios;
len = min_t(sector_t, ci->sector_count,
__max_io_len(ti, ci->sector, max_granularity));
__max_io_len(ti, ci->sector, max_granularity, max_sectors));
atomic_add(num_bios, &ci->io->io_count);
bios = __send_duplicate_bios(ci, ti, num_bios, &len);
......@@ -1623,23 +1625,27 @@ static blk_status_t __process_abnormal_io(struct clone_info *ci,
{
unsigned int num_bios = 0;
unsigned int max_granularity = 0;
unsigned int max_sectors = 0;
struct queue_limits *limits = dm_get_queue_limits(ti->table->md);
switch (bio_op(ci->bio)) {
case REQ_OP_DISCARD:
num_bios = ti->num_discard_bios;
max_sectors = limits->max_discard_sectors;
if (ti->max_discard_granularity)
max_granularity = limits->max_discard_sectors;
max_granularity = max_sectors;
break;
case REQ_OP_SECURE_ERASE:
num_bios = ti->num_secure_erase_bios;
max_sectors = limits->max_secure_erase_sectors;
if (ti->max_secure_erase_granularity)
max_granularity = limits->max_secure_erase_sectors;
max_granularity = max_sectors;
break;
case REQ_OP_WRITE_ZEROES:
num_bios = ti->num_write_zeroes_bios;
max_sectors = limits->max_write_zeroes_sectors;
if (ti->max_write_zeroes_granularity)
max_granularity = limits->max_write_zeroes_sectors;
max_granularity = max_sectors;
break;
default:
break;
......@@ -1654,7 +1660,8 @@ static blk_status_t __process_abnormal_io(struct clone_info *ci,
if (unlikely(!num_bios))
return BLK_STS_NOTSUPP;
__send_changing_extent_only(ci, ti, num_bios, max_granularity);
__send_changing_extent_only(ci, ti, num_bios,
max_granularity, max_sectors);
return BLK_STS_OK;
}
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment