From: Tao Cui child_lat.lock is acquired with irqsave semantics in both iolatency_check_latencies() (called from the blkcg_iolatency_done_bio() softirq path) and blkiolatency_timer_fn() (timer softirq). iolatency_clear_scaling() instead uses a plain spin_lock(), which only happens to be safe because both of its callers enter with interrupts already disabled: * iolatency_set_limit() runs under queue_lock via blkg_conf_prep(), which returns with the lock held and interrupts disabled; * iolatency_pd_offline() runs under queue_lock from blkg_destroy() and blkcg_deactivate_policy(), both of which take it with spin_lock_irq(). Take the lock with guard(spinlock_irqsave)() so the locking is self-contained and consistent with the other two sites, instead of relying on an undocumented caller precondition. No functional change. Signed-off-by: Tao Cui --- Changes in v2: - use guard(spinlock_irqsave)() (suggested by Leon Hwang) --- block/blk-iolatency.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/block/blk-iolatency.c b/block/blk-iolatency.c index cef02b6c5fa9..1cc33aa0b669 100644 --- a/block/blk-iolatency.c +++ b/block/blk-iolatency.c @@ -815,12 +815,11 @@ static void iolatency_clear_scaling(struct blkcg_gq *blkg) return; lat_info = &iolat->child_lat; - spin_lock(&lat_info->lock); + guard(spinlock_irqsave)(&lat_info->lock); atomic_set(&lat_info->scale_cookie, DEFAULT_SCALE_COOKIE); lat_info->last_scale_event = 0; lat_info->scale_grp = NULL; lat_info->scale_lat = 0; - spin_unlock(&lat_info->lock); } } -- 2.43.0