| Message ID | 1534478043-7170-3-git-send-email-jianchao.w.wang@oracle.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | fixes for the updating nr_hw_queues | expand |
On Fri, Aug 17, 2018 at 11:54 AM, Jianchao Wang <jianchao.w.wang@oracle.com> wrote: > For blk-mq, part_in_flight/rw will invoke blk_mq_in_flight/rw to > account the inflight requests. It will access the queue_hw_ctx and > nr_hw_queues w/o any protection. When updating nr_hw_queues and > blk_mq_in_flight/rw occur concurrently, panic comes up. > > Before update nr_hw_queues, the q will be frozen. So we could use > q_usage_counter to avoid the race. percpu_ref_is_zero is used here > so that we will not miss any in-flight request. And also both the > check and blk_mq_queue_tag_busy_iter are under rcu critical section, > then __blk_mq_update_nr_hw_queues could ensure the zeroed q_usage_counter > to be globally visible. > > Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com> > --- > block/blk-mq.c | 31 ++++++++++++++++++++++++++++++- > 1 file changed, 30 insertions(+), 1 deletion(-) > > diff --git a/block/blk-mq.c b/block/blk-mq.c > index de7027f..9ec98bd 100644 > --- a/block/blk-mq.c > +++ b/block/blk-mq.c > @@ -112,7 +112,22 @@ void blk_mq_in_flight(struct request_queue *q, struct hd_struct *part, > struct mq_inflight mi = { .part = part, .inflight = inflight, }; > > inflight[0] = inflight[1] = 0; > + > + /* > + * __blk_mq_update_nr_hw_queues will update the nr_hw_queues and > + * queue_hw_ctx after freeze the queue. So we could use q_usage_counter > + * to avoid race with it. __blk_mq_update_nr_hw_queues will use > + * synchronize_rcu to ensure all of the users of blk_mq_in_flight > + * go out of the critical section and see zeroed q_usage_counter. > + */ > + rcu_read_lock(); > + if (percpu_ref_is_zero(&q->q_usage_counter)) { > + rcu_read_unlock(); > + return; > + } > + > blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight, &mi); > + rcu_read_unlock(); > } > > static void blk_mq_check_inflight_rw(struct blk_mq_hw_ctx *hctx, > @@ -131,7 +146,18 @@ void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part, > struct mq_inflight mi = { .part = part, .inflight = inflight, }; > > inflight[0] = inflight[1] = 0; > + > + /* > + * See comment of blk_mq_in_flight. > + */ > + rcu_read_lock(); > + if (percpu_ref_is_zero(&q->q_usage_counter)) { > + rcu_read_unlock(); > + return; > + } > + > blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight_rw, &mi); > + rcu_read_unlock(); I'd suggest to put the rcu_* and percpu_ref_is_zero() into blk_mq_queue_tag_busy_iter(). Thanks, Ming Lei
Hi Ming On 08/17/2018 05:36 PM, Ming Lei wrote: > I'd suggest to put the rcu_* and percpu_ref_is_zero() into > blk_mq_queue_tag_busy_iter(). Yes, it's fine for me. :) I will change it in next version. Thanks Jianchao
diff --git a/block/blk-mq.c b/block/blk-mq.c index de7027f..9ec98bd 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -112,7 +112,22 @@ void blk_mq_in_flight(struct request_queue *q, struct hd_struct *part, struct mq_inflight mi = { .part = part, .inflight = inflight, }; inflight[0] = inflight[1] = 0; + + /* + * __blk_mq_update_nr_hw_queues will update the nr_hw_queues and + * queue_hw_ctx after freeze the queue. So we could use q_usage_counter + * to avoid race with it. __blk_mq_update_nr_hw_queues will use + * synchronize_rcu to ensure all of the users of blk_mq_in_flight + * go out of the critical section and see zeroed q_usage_counter. + */ + rcu_read_lock(); + if (percpu_ref_is_zero(&q->q_usage_counter)) { + rcu_read_unlock(); + return; + } + blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight, &mi); + rcu_read_unlock(); } static void blk_mq_check_inflight_rw(struct blk_mq_hw_ctx *hctx, @@ -131,7 +146,18 @@ void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part, struct mq_inflight mi = { .part = part, .inflight = inflight, }; inflight[0] = inflight[1] = 0; + + /* + * See comment of blk_mq_in_flight. + */ + rcu_read_lock(); + if (percpu_ref_is_zero(&q->q_usage_counter)) { + rcu_read_unlock(); + return; + } + blk_mq_queue_tag_busy_iter(q, blk_mq_check_inflight_rw, &mi); + rcu_read_unlock(); } void blk_freeze_queue_start(struct request_queue *q) @@ -2905,7 +2931,10 @@ static void __blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, list_for_each_entry(q, &set->tag_list, tag_set_list) blk_mq_freeze_queue(q); - + /* + * Sync with blk_mq_in_flight and blk_mq_in_flight_rw + */ + synchronize_rcu(); /* * switch io scheduler to NULL to clean up the data in it. * will get it back after update mapping between cpu and hw queues.
For blk-mq, part_in_flight/rw will invoke blk_mq_in_flight/rw to account the inflight requests. It will access the queue_hw_ctx and nr_hw_queues w/o any protection. When updating nr_hw_queues and blk_mq_in_flight/rw occur concurrently, panic comes up. Before update nr_hw_queues, the q will be frozen. So we could use q_usage_counter to avoid the race. percpu_ref_is_zero is used here so that we will not miss any in-flight request. And also both the check and blk_mq_queue_tag_busy_iter are under rcu critical section, then __blk_mq_update_nr_hw_queues could ensure the zeroed q_usage_counter to be globally visible. Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com> --- block/blk-mq.c | 31 ++++++++++++++++++++++++++++++- 1 file changed, 30 insertions(+), 1 deletion(-)