From 52c6ede6919c47bdf04ebacd1d1ef94ed00859be Mon Sep 17 00:00:00 2001 From: Shaohua Li Date: Wed, 24 Apr 2013 11:42:43 +1000 Subject: [PATCH] raid5: make release_stripe lockless release_stripe still has big lock contention. We just add the stripe to a llist without taking device_lock. We let the raid5d thread to do the real stripe release, which must hold device_lock anyway. In this way, release_stripe doesn't hold any locks. The side effect is the released stripes order is changed. But sounds not a big deal, stripes are never handled in order. And I thought block layer can already do nice request merge, which means order isn't that important. I kept the unplug release batch, which is unnecessary with this patch from lock contention avoid point of view, and actually if we delete it, the stripe_head release_list and lru can share storage. But the unplug release batch is also helpful for request merge. We probably can delay wakeup raid5d till unplug, but I'm still afraid of the case which raid5d is running. Signed-off-by: Shaohua Li Signed-off-by: NeilBrown --- drivers/md/raid5.c | 54 +++++++++++++++++++++++++++++++++++++++++----- drivers/md/raid5.h | 3 +++ 2 files changed, 52 insertions(+), 5 deletions(-) diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c index 2e6538a7174f..76742268ccec 100644 --- a/drivers/md/raid5.c +++ b/drivers/md/raid5.c @@ -264,12 +264,46 @@ static void __release_stripe(struct r5conf *conf, struct stripe_head *sh) do_release_stripe(conf, sh); } +/* should hold conf->device_lock already */ +static int release_stripe_list(struct r5conf *conf) +{ + struct stripe_head *sh; + int count = 0; + + llist_for_each_entry(sh, llist_del_all(&conf->released_stripes), + release_list) { + /* + * llist_del_all() uses cmpxchg, so implies a memory fence. + * It's guaranteed the stripe isn't in released_stripes list + * now, clearing STRIPE_ON_RELEASE_LIST is safe. + */ + clear_bit(STRIPE_ON_RELEASE_LIST, &sh->state); + /* + * Don't worry the bit is set here, because if the bit is set + * again, the count is always > 1. This is true for + * STRIPE_ON_UNPLUG_LIST bit too. + */ + __release_stripe(conf, sh); + count++; + } + return count; +} + static void release_stripe(struct stripe_head *sh) { struct r5conf *conf = sh->raid_conf; unsigned long flags; + bool wakeup; + if (test_and_set_bit(STRIPE_ON_RELEASE_LIST, &sh->state)) + goto slow_path; + wakeup = llist_add(&sh->release_list, &conf->released_stripes); + if (wakeup) + md_wakeup_thread(conf->mddev->thread); + return; +slow_path: local_irq_save(flags); + /* we are ok here if STRIPE_ON_RELEASE_LIST is set or not */ if (atomic_dec_and_lock(&sh->count, &conf->device_lock)) { do_release_stripe(conf, sh); spin_unlock(&conf->device_lock); @@ -517,7 +551,8 @@ get_active_stripe(struct r5conf *conf, sector_t sector, if (atomic_read(&sh->count)) { BUG_ON(!list_empty(&sh->lru) && !test_bit(STRIPE_EXPANDING, &sh->state) - && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state)); + && !test_bit(STRIPE_ON_UNPLUG_LIST, &sh->state) + && !test_bit(STRIPE_ON_RELEASE_LIST, &sh->state)); } else { if (!test_bit(STRIPE_HANDLE, &sh->state)) atomic_inc(&conf->active_stripes); @@ -4177,6 +4212,10 @@ static void raid5_unplug(struct blk_plug_cb *blk_cb, bool from_schedule) */ smp_mb__before_clear_bit(); clear_bit(STRIPE_ON_UNPLUG_LIST, &sh->state); + /* + * STRIPE_ON_RELEASE_LIST could be set here. In that + * case, the count is always > 1 here + */ __release_stripe(conf, sh); cnt++; } @@ -4877,10 +4916,12 @@ static void raid5auxd(struct md_thread *thread) handled = 0; spin_lock_irq(&conf->device_lock); while (1) { - int batch_size; + int batch_size, released; + + released = release_stripe_list(conf); batch_size = handle_active_stripes(conf, &auxth->work_mask); - if (!batch_size) + if (!batch_size && !released) break; handled += batch_size; } @@ -4915,7 +4956,9 @@ static void raid5d(struct md_thread *thread) spin_lock_irq(&conf->device_lock); while (1) { struct bio *bio; - int batch_size; + int batch_size, released; + + released = release_stripe_list(conf); if ( !list_empty(&conf->bitmap_list)) { @@ -4940,7 +4983,7 @@ static void raid5d(struct md_thread *thread) } batch_size = handle_active_stripes(conf, &conf->work_mask); - if (!batch_size) + if (!batch_size && !released) break; handled += batch_size; @@ -5531,6 +5574,7 @@ static struct r5conf *setup_conf(struct mddev *mddev) INIT_LIST_HEAD(&conf->delayed_list); INIT_LIST_HEAD(&conf->bitmap_list); INIT_LIST_HEAD(&conf->inactive_list); + init_llist_head(&conf->released_stripes); atomic_set(&conf->active_stripes, 0); atomic_set(&conf->preread_active_stripes, 0); atomic_set(&conf->active_aligned_reads, 0); diff --git a/drivers/md/raid5.h b/drivers/md/raid5.h index b83967f83079..217cb19a31f4 100644 --- a/drivers/md/raid5.h +++ b/drivers/md/raid5.h @@ -197,6 +197,7 @@ enum reconstruct_states { struct stripe_head { struct hlist_node hash; struct list_head lru; /* inactive_list or handle_list */ + struct llist_node release_list; struct r5conf *raid_conf; short generation; /* increments with every * reshape */ @@ -321,6 +322,7 @@ enum { STRIPE_OPS_REQ_PENDING, STRIPE_ON_UNPLUG_LIST, STRIPE_DISCARD, + STRIPE_ON_RELEASE_LIST, }; /* @@ -459,6 +461,7 @@ struct r5conf { */ atomic_t active_stripes; struct list_head inactive_list; + struct llist_head released_stripes; wait_queue_head_t wait_for_stripe; wait_queue_head_t wait_for_overlap; int inactive_blocked; /* release of inactive stripes blocked, -- 2.39.5