During sequential workload, if no full-stripe write occurs, each parity
will be written data_disks (raid_disks-max_degraded) times. The lifetime
of the parity written before the data_disks update will be short, so
RWH_WRITE_LIFE_SHORT write hint can be assigned to those writes.
New counter has been added to the stripe_head struct to count number of
parity updates in a stripe. If this counter is equal to data_disks or this
is the full stripe write, a write hint from the original bio is assigned.
Otherwise RWH_WRITE_LIFE_SHORT write hint will be set.
Signed-off-by: Mariusz Dabrowski <mariusz.dabrowski@xxxxxxxxx>
Reviewed-by: Artur Paszkiewicz <artur.paszkiewicz@xxxxxxxxx>
Reviewed-by: Pawel Baldysiak <pawel.baldysiak@xxxxxxxxx>
---
drivers/md/raid5.c | 67 ++++++++++++++++++++++++++++++++++++++++++++++++++----
drivers/md/raid5.h | 2 ++
2 files changed, 65 insertions(+), 4 deletions(-)
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 088b97bbcbe2..1c581b0bbc44 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -978,6 +978,36 @@ static void defer_issue_bios(struct r5conf *conf, sector_t sector,
dispatch_bio_list(&tmp);
}
+static void raid5_set_parity_write_hint(struct bio *bio,
+ struct stripe_head *sh,
+ int *updates_cnt)
+{
+ enum rw_hint hint = WRITE_LIFE_NOT_SET;
+ struct r5conf *conf = sh->raid_conf;
+ int data_disks = conf->raid_disks - conf->max_degraded;
+
+ if (!test_bit(STRIPE_FULL_WRITE, &sh->state) &&
+ (++*updates_cnt < data_disks)) {
+ hint = WRITE_LIFE_SHORT;
+ } else {
+ int i;
+
+ for (i = 0; i < sh->disks; i++) {
+ if ((i != sh->pd_idx) && (i != sh->qd_idx)) {
+ if (sh->dev[i].req.bi_write_hint > 0)
+ hint = sh->dev[i].req.bi_write_hint;
+ else if (sh->dev[i].write_hint > 0)
+ hint = sh->dev[i].write_hint;
+ if (hint)
+ break;
+ }
+ }
+
+ *updates_cnt = 0;
+ }
+ bio->bi_write_hint = hint;
+}
+
static void
raid5_end_read_request(struct bio *bi);
static void
@@ -1003,6 +1037,9 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
int replace_only = 0;
struct bio *bi, *rbi;
struct md_rdev *rdev, *rrdev = NULL;
+ int sh_syncing = s->syncing || s->expanding ||
+ s->expanded || s->replacing;
+
sh = head_sh;
if (test_and_clear_bit(R5_Wantwrite, &sh->dev[i].flags)) {
@@ -1094,8 +1131,7 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
}
if (rdev) {
- if (s->syncing || s->expanding || s->expanded
- || s->replacing)
+ if (sh_syncing)
md_sync_acct(rdev->bdev, STRIPE_SECTORS);
set_bit(STRIPE_IO_STARTED, &sh->state);
@@ -1155,14 +1191,25 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
trace_block_bio_remap(bi->bi_disk->queue,
bi, disk_devt(conf->mddev->gendisk),
sh->dev[i].sector);
+
+ if (op_is_write(op) && !sh_syncing &&
+ test_bit(WH_POLICY_PARITY,
+ &conf->write_hint_flags)) {
+ if (i == sh->pd_idx)
+ raid5_set_parity_write_hint(bi, sh,
+ &sh->p_updates);
+ else if (i == sh->qd_idx)
+ raid5_set_parity_write_hint(bi, sh,
+ &sh->q_updates);
+ }
+
if (should_defer && op_is_write(op))
bio_list_add(&pending_bios, bi);
else
generic_make_request(bi);
}
if (rrdev) {
- if (s->syncing || s->expanding || s->expanded
- || s->replacing)
+ if (sh_syncing)
md_sync_acct(rrdev->bdev, STRIPE_SECTORS);
set_bit(STRIPE_IO_STARTED, &sh->state);
@@ -1205,6 +1252,18 @@ static void ops_run_io(struct stripe_head *sh, struct stripe_head_state *s)
trace_block_bio_remap(rbi->bi_disk->queue,
rbi, disk_devt(conf->mddev->gendisk),
sh->dev[i].sector);
+
+ if (op_is_write(op) && !sh_syncing &&
+ test_bit(WH_POLICY_PARITY,
+ &conf->write_hint_flags)) {
+ if (i == sh->pd_idx)
+ raid5_set_parity_write_hint(rbi, sh,
+ &sh->p_updates);
+ else if (i == sh->qd_idx)
+ raid5_set_parity_write_hint(rbi, sh,
+ &sh->q_updates);
+ }
+
if (should_defer && op_is_write(op))
bio_list_add(&pending_bios, rbi);
else
diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h
index e08071dc4202..4a5fa44ffecf 100644
--- a/drivers/md/raid5.h
+++ b/drivers/md/raid5.h
@@ -219,6 +219,8 @@ struct stripe_head {
* this is only checked when stripe
* has STRIPE_BATCH_READY
*/
+ int p_updates;
+ int q_updates;
enum check_states check_state;
enum reconstruct_states reconstruct_state;
spinlock_t stripe_lock;
--
2.16.1
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
