On Mon, Aug 12, 2019 at 04:24:01PM +0200, Hannes Reinecke wrote: > On 8/12/19 3:43 PM, Ming Lei wrote: > > Most of blk-mq drivers depend on managed IRQ's auto-affinity to setup > > up queue mapping. Thomas mentioned the following point[1]: > > > > " > > That was the constraint of managed interrupts from the very beginning: > > > > The driver/subsystem has to quiesce the interrupt line and the associated > > queue _before_ it gets shutdown in CPU unplug and not fiddle with it > > until it's restarted by the core when the CPU is plugged in again. > > " > > > > However, current blk-mq implementation doesn't quiesce hw queue before > > the last CPU in the hctx is shutdown. Even worse, CPUHP_BLK_MQ_DEAD is > > one cpuhp state handled after the CPU is down, so there isn't any chance > > to quiesce hctx for blk-mq wrt. CPU hotplug. > > > > Add new cpuhp state of CPUHP_AP_BLK_MQ_ONLINE for blk-mq to stop queues > > and wait for completion of in-flight requests. > > > > [1] https://lore.kernel.org/linux-block/alpine.DEB.2.21.1904051331270.1802@xxxxxxxxxxxxxxxxxxxxxxx/ > > > > Cc: Bart Van Assche <bvanassche@xxxxxxx> > > Cc: Hannes Reinecke <hare@xxxxxxxx> > > Cc: Christoph Hellwig <hch@xxxxxx> > > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx> > > Cc: Keith Busch <keith.busch@xxxxxxxxx> > > Signed-off-by: Ming Lei <ming.lei@xxxxxxxxxx> > > --- > > block/blk-mq-tag.c | 2 +- > > block/blk-mq-tag.h | 2 ++ > > block/blk-mq.c | 65 ++++++++++++++++++++++++++++++++++++++ > > include/linux/blk-mq.h | 1 + > > include/linux/cpuhotplug.h | 1 + > > 5 files changed, 70 insertions(+), 1 deletion(-) > > > > diff --git a/block/blk-mq-tag.c b/block/blk-mq-tag.c > > index 008388e82b5c..31828b82552b 100644 > > --- a/block/blk-mq-tag.c > > +++ b/block/blk-mq-tag.c > > @@ -325,7 +325,7 @@ static void bt_tags_for_each(struct blk_mq_tags *tags, struct sbitmap_queue *bt, > > * true to continue iterating tags, false to stop. > > * @priv: Will be passed as second argument to @fn. > > */ > > -static void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags, > > +void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags, > > busy_tag_iter_fn *fn, void *priv) > > { > > if (tags->nr_reserved_tags) > > diff --git a/block/blk-mq-tag.h b/block/blk-mq-tag.h > > index 61deab0b5a5a..321fd6f440e6 100644 > > --- a/block/blk-mq-tag.h > > +++ b/block/blk-mq-tag.h > > @@ -35,6 +35,8 @@ extern int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx, > > extern void blk_mq_tag_wakeup_all(struct blk_mq_tags *tags, bool); > > void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn, > > void *priv); > > +void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags, > > + busy_tag_iter_fn *fn, void *priv); > > > > static inline struct sbq_wait_state *bt_wait_ptr(struct sbitmap_queue *bt, > > struct blk_mq_hw_ctx *hctx) > > diff --git a/block/blk-mq.c b/block/blk-mq.c > > index 6968de9d7402..6931b2ba2776 100644 > > --- a/block/blk-mq.c > > +++ b/block/blk-mq.c > > @@ -2206,6 +2206,61 @@ int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, > > return -ENOMEM; > > } > > > > +static bool blk_mq_count_inflight_rq(struct request *rq, void *data, > > + bool reserved) > > +{ > > + unsigned *count = data; > > + > > + if ((blk_mq_rq_state(rq) == MQ_RQ_IN_FLIGHT)) > > + (*count)++; > > + > > + return true; > > +} > > + > > +static unsigned blk_mq_tags_inflight_rqs(struct blk_mq_tags *tags) > > +{ > > + unsigned count = 0; > > + > > + blk_mq_all_tag_busy_iter(tags, blk_mq_count_inflight_rq, &count); > > + > > + return count; > > +} > > + > > +static void blk_mq_drain_inflight_rqs(struct blk_mq_hw_ctx *hctx) > > +{ > > + while (1) { > > + if (!blk_mq_tags_inflight_rqs(hctx->tags)) > > + break; > > + msleep(5); > > + } > > +} > > + > > +static int blk_mq_hctx_notify_online(unsigned int cpu, struct hlist_node *node) > > +{ > > + struct blk_mq_hw_ctx *hctx = hlist_entry_safe(node, > > + struct blk_mq_hw_ctx, cpuhp_online); > > + unsigned prev_cpu = -1; > > + > > + while (true) { > > + unsigned next_cpu = cpumask_next_and(prev_cpu, hctx->cpumask, > > + cpu_online_mask); > > + > > + if (next_cpu >= nr_cpu_ids) > > + break; > > + > > + /* return if there is other online CPU on this hctx */ > > + if (next_cpu != cpu) > > + return 0; > > + > > + prev_cpu = next_cpu; > > + } > > + > > + set_bit(BLK_MQ_S_INTERNAL_STOPPED, &hctx->state); > > + blk_mq_drain_inflight_rqs(hctx); > > + > > + return 0; > > +} > > + > > /* > > * 'cpu' is going away. splice any existing rq_list entries from this > > * software queue to the hw queue dispatch list, and ensure that it > > Isn't that inverted? The above comment is wrong, and you will see it is fixed in patch 4. blk_mq_hctx_notify_dead() is called when the specified CPU is dead, and blk_mq_hctx_notify_online() is called before the CPU goes away during cpuhp path. > From the function I would assume it'll be called once the CPU is being > set toe 'online', yet from the description I would have assumed the No, blk_mq_init() only registers teardown callback for CPUHP_AP_BLK_MQ_ONLINE, that means blk_mq_hctx_notify_online() is only called when CPU is going away, still online, so we can stop queue for quiescing IO on this queue. > INTERNAL_STOPPED bit is set when the cpu goes offline. > Care to elaborate? The idea is to quiesce queue in two stages: 1) in blk_mq_hctx_notify_online(), this CPU isn't dead yet, but is going to become dead, so we can wait for completion of in-flight IOs. Meantime stop the hw queue. 2) in blk_mq_hctx_notify_dead(), all CPUs of this hw queue have been dead, what we can do is to end the request and re-submit the IO, and new request will be mapped to another active hw queue because blk_mq_hctx_notify_dead() is always called from one online CPU. Thanks, Ming
