From 320ae51feed5c2f13664aa05a76bec198967e04d Mon Sep 17 00:00:00 2001 From: Jens Axboe Date: Thu, 24 Oct 2013 09:20:05 +0100 Subject: blk-mq: new multi-queue block IO queueing mechanism Linux currently has two models for block devices: - The classic request_fn based approach, where drivers use struct request units for IO. The block layer provides various helper functionalities to let drivers share code, things like tag management, timeout handling, queueing, etc. - The "stacked" approach, where a driver squeezes in between the block layer and IO submitter. Since this bypasses the IO stack, driver generally have to manage everything themselves. With drivers being written for new high IOPS devices, the classic request_fn based driver doesn't work well enough. The design dates back to when both SMP and high IOPS was rare. It has problems with scaling to bigger machines, and runs into scaling issues even on smaller machines when you have IOPS in the hundreds of thousands per device. The stacked approach is then most often selected as the model for the driver. But this means that everybody has to re-invent everything, and along with that we get all the problems again that the shared approach solved. This commit introduces blk-mq, block multi queue support. The design is centered around per-cpu queues for queueing IO, which then funnel down into x number of hardware submission queues. We might have a 1:1 mapping between the two, or it might be an N:M mapping. That all depends on what the hardware supports. blk-mq provides various helper functions, which include: - Scalable support for request tagging. Most devices need to be able to uniquely identify a request both in the driver and to the hardware. The tagging uses per-cpu caches for freed tags, to enable cache hot reuse. - Timeout handling without tracking request on a per-device basis. Basically the driver should be able to get a notification, if a request happens to fail. - Optional support for non 1:1 mappings between issue and submission queues. blk-mq can redirect IO completions to the desired location. - Support for per-request payloads. Drivers almost always need to associate a request structure with some driver private command structure. Drivers can tell blk-mq this at init time, and then any request handed to the driver will have the required size of memory associated with it. - Support for merging of IO, and plugging. The stacked model gets neither of these. Even for high IOPS devices, merging sequential IO reduces per-command overhead and thus increases bandwidth. For now, this is provided as a potential 3rd queueing model, with the hope being that, as it matures, it can replace both the classic and stacked model. That would get us back to having just 1 real model for block devices, leaving the stacked approach to dm/md devices (as it was originally intended). Contributions in this patch from the following people: Shaohua Li Alexander Gordeev Christoph Hellwig Mike Christie Matias Bjorling Jeff Moyer Acked-by: Christoph Hellwig Signed-off-by: Jens Axboe --- include/linux/blk-mq.h | 182 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 182 insertions(+) create mode 100644 include/linux/blk-mq.h (limited to 'include/linux/blk-mq.h') diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h new file mode 100644 index 000000000000..746042ff321a --- /dev/null +++ b/include/linux/blk-mq.h @@ -0,0 +1,182 @@ +#ifndef BLK_MQ_H +#define BLK_MQ_H + +#include + +struct blk_mq_tags; + +struct blk_mq_cpu_notifier { + struct list_head list; + void *data; + void (*notify)(void *data, unsigned long action, unsigned int cpu); +}; + +struct blk_mq_hw_ctx { + struct { + spinlock_t lock; + struct list_head dispatch; + } ____cacheline_aligned_in_smp; + + unsigned long state; /* BLK_MQ_S_* flags */ + struct delayed_work delayed_work; + + unsigned long flags; /* BLK_MQ_F_* flags */ + + struct request_queue *queue; + unsigned int queue_num; + + void *driver_data; + + unsigned int nr_ctx; + struct blk_mq_ctx **ctxs; + unsigned int nr_ctx_map; + unsigned long *ctx_map; + + struct request **rqs; + struct list_head page_list; + struct blk_mq_tags *tags; + + unsigned long queued; + unsigned long run; +#define BLK_MQ_MAX_DISPATCH_ORDER 10 + unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; + + unsigned int queue_depth; + unsigned int numa_node; + unsigned int cmd_size; /* per-request extra data */ + + struct blk_mq_cpu_notifier cpu_notifier; + struct kobject kobj; +}; + +struct blk_mq_reg { + struct blk_mq_ops *ops; + unsigned int nr_hw_queues; + unsigned int queue_depth; + unsigned int reserved_tags; + unsigned int cmd_size; /* per-request extra data */ + int numa_node; + unsigned int timeout; + unsigned int flags; /* BLK_MQ_F_* */ +}; + +typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, struct request *); +typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int); +typedef struct blk_mq_hw_ctx *(alloc_hctx_fn)(struct blk_mq_reg *,unsigned int); +typedef void (free_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); +typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); +typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); + +struct blk_mq_ops { + /* + * Queue request + */ + queue_rq_fn *queue_rq; + + /* + * Map to specific hardware queue + */ + map_queue_fn *map_queue; + + /* + * Called on request timeout + */ + rq_timed_out_fn *timeout; + + /* + * Override for hctx allocations (should probably go) + */ + alloc_hctx_fn *alloc_hctx; + free_hctx_fn *free_hctx; + + /* + * Called when the block layer side of a hardware queue has been + * set up, allowing the driver to allocate/init matching structures. + * Ditto for exit/teardown. + */ + init_hctx_fn *init_hctx; + exit_hctx_fn *exit_hctx; +}; + +enum { + BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */ + BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */ + BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */ + + BLK_MQ_F_SHOULD_MERGE = 1 << 0, + BLK_MQ_F_SHOULD_SORT = 1 << 1, + BLK_MQ_F_SHOULD_IPI = 1 << 2, + + BLK_MQ_S_STOPPED = 1 << 0, + + BLK_MQ_MAX_DEPTH = 2048, +}; + +struct request_queue *blk_mq_init_queue(struct blk_mq_reg *, void *); +void blk_mq_free_queue(struct request_queue *); +int blk_mq_register_disk(struct gendisk *); +void blk_mq_unregister_disk(struct gendisk *); +void blk_mq_init_commands(struct request_queue *, void (*init)(void *data, struct blk_mq_hw_ctx *, struct request *, unsigned int), void *data); + +void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); + +void blk_mq_insert_request(struct request_queue *, struct request *, bool); +void blk_mq_run_queues(struct request_queue *q, bool async); +void blk_mq_free_request(struct request *rq); +bool blk_mq_can_queue(struct blk_mq_hw_ctx *); +struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp); +struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, gfp_t gfp); +struct request *blk_mq_rq_from_tag(struct request_queue *q, unsigned int tag); + +struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index); +struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_reg *, unsigned int); +void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *, unsigned int); + +void blk_mq_end_io(struct request *rq, int error); + +void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); +void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); +void blk_mq_start_stopped_hw_queues(struct request_queue *q); + +/* + * Driver command data is immediately after the request. So subtract request + * size to get back to the original request. + */ +static inline struct request *blk_mq_rq_from_pdu(void *pdu) +{ + return pdu - sizeof(struct request); +} +static inline void *blk_mq_rq_to_pdu(struct request *rq) +{ + return (void *) rq + sizeof(*rq); +} + +static inline struct request *blk_mq_tag_to_rq(struct blk_mq_hw_ctx *hctx, + unsigned int tag) +{ + return hctx->rqs[tag]; +} + +#define queue_for_each_hw_ctx(q, hctx, i) \ + for ((i) = 0, hctx = (q)->queue_hw_ctx[0]; \ + (i) < (q)->nr_hw_queues; (i)++, hctx = (q)->queue_hw_ctx[i]) + +#define queue_for_each_ctx(q, ctx, i) \ + for ((i) = 0, ctx = per_cpu_ptr((q)->queue_ctx, 0); \ + (i) < (q)->nr_queues; (i)++, ctx = per_cpu_ptr(q->queue_ctx, (i))) + +#define hctx_for_each_ctx(hctx, ctx, i) \ + for ((i) = 0, ctx = (hctx)->ctxs[0]; \ + (i) < (hctx)->nr_ctx; (i)++, ctx = (hctx)->ctxs[(i)]) + +#define blk_ctx_sum(q, sum) \ +({ \ + struct blk_mq_ctx *__x; \ + unsigned int __ret = 0, __i; \ + \ + queue_for_each_ctx((q), __x, __i) \ + __ret += sum; \ + __ret; \ +}) + +#endif -- cgit v1.2.3 From 280d45f6c35d8d7a0fe20c36caf426e3ac139cf9 Mon Sep 17 00:00:00 2001 From: Christoph Hellwig Date: Fri, 25 Oct 2013 14:45:58 +0100 Subject: blk-mq: add blk_mq_stop_hw_queues Add a helper to iterate over all hw queues and stop them. This is useful for driver that implement PM suspend functionality. Signed-off-by: Christoph Hellwig Modified to just call blk_mq_stop_hw_queue() by Jens. Signed-off-by: Jens Axboe --- block/blk-mq.c | 10 ++++++++++ include/linux/blk-mq.h | 1 + 2 files changed, 11 insertions(+) (limited to 'include/linux/blk-mq.h') diff --git a/block/blk-mq.c b/block/blk-mq.c index f21ec964e411..ac804c635040 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -672,6 +672,16 @@ void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx) } EXPORT_SYMBOL(blk_mq_stop_hw_queue); +void blk_mq_stop_hw_queues(struct request_queue *q) +{ + struct blk_mq_hw_ctx *hctx; + int i; + + queue_for_each_hw_ctx(q, hctx, i) + blk_mq_stop_hw_queue(hctx); +} +EXPORT_SYMBOL(blk_mq_stop_hw_queues); + void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx) { clear_bit(BLK_MQ_S_STOPPED, &hctx->state); diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h index 746042ff321a..3368b97bee73 100644 --- a/include/linux/blk-mq.h +++ b/include/linux/blk-mq.h @@ -136,6 +136,7 @@ void blk_mq_end_io(struct request *rq, int error); void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); +void blk_mq_stop_hw_queues(struct request_queue *q); void blk_mq_start_stopped_hw_queues(struct request_queue *q); /* -- cgit v1.2.3 From 3228f48be2d19b2dd90db96ec16a40187a2946f3 Mon Sep 17 00:00:00 2001 From: Christoph Hellwig Date: Mon, 28 Oct 2013 13:33:58 -0600 Subject: blk-mq: fix for flush deadlock The flush state machine takes in a struct request, which then is submitted multiple times to the underling driver. The old block code requeses the same request for each of those, so it does not have an issue with tapping into the request pool. The new one on the other hand allocates a new request for each of the actualy steps of the flush sequence. If have already allocated all of the tags for IO, we will fail allocating the flush request. Set aside a reserved request just for flushes. Signed-off-by: Jens Axboe --- block/blk-core.c | 2 +- block/blk-flush.c | 2 +- block/blk-mq.c | 14 ++++++++++++-- include/linux/blk-mq.h | 2 +- 4 files changed, 15 insertions(+), 5 deletions(-) (limited to 'include/linux/blk-mq.h') diff --git a/block/blk-core.c b/block/blk-core.c index 3bb9e9f7f87e..9677c6525ed8 100644 --- a/block/blk-core.c +++ b/block/blk-core.c @@ -1102,7 +1102,7 @@ static struct request *blk_old_get_request(struct request_queue *q, int rw, struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask) { if (q->mq_ops) - return blk_mq_alloc_request(q, rw, gfp_mask); + return blk_mq_alloc_request(q, rw, gfp_mask, false); else return blk_old_get_request(q, rw, gfp_mask); } diff --git a/block/blk-flush.c b/block/blk-flush.c index 3e4cc9c7890a..331e627301ea 100644 --- a/block/blk-flush.c +++ b/block/blk-flush.c @@ -286,7 +286,7 @@ static void mq_flush_work(struct work_struct *work) /* We don't need set REQ_FLUSH_SEQ, it's for consistency */ rq = blk_mq_alloc_request(q, WRITE_FLUSH|REQ_FLUSH_SEQ, - __GFP_WAIT|GFP_ATOMIC); + __GFP_WAIT|GFP_ATOMIC, true); rq->cmd_type = REQ_TYPE_FS; rq->end_io = flush_end_io; diff --git a/block/blk-mq.c b/block/blk-mq.c index ac804c635040..2dc8de86d0d2 100644 --- a/block/blk-mq.c +++ b/block/blk-mq.c @@ -210,14 +210,15 @@ static struct request *blk_mq_alloc_request_pinned(struct request_queue *q, return rq; } -struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp) +struct request *blk_mq_alloc_request(struct request_queue *q, int rw, + gfp_t gfp, bool reserved) { struct request *rq; if (blk_mq_queue_enter(q)) return NULL; - rq = blk_mq_alloc_request_pinned(q, rw, gfp, false); + rq = blk_mq_alloc_request_pinned(q, rw, gfp, reserved); blk_mq_put_ctx(rq->mq_ctx); return rq; } @@ -1327,6 +1328,15 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_reg *reg, reg->queue_depth = BLK_MQ_MAX_DEPTH; } + /* + * Set aside a tag for flush requests. It will only be used while + * another flush request is in progress but outside the driver. + * + * TODO: only allocate if flushes are supported + */ + reg->queue_depth++; + reg->reserved_tags++; + if (reg->queue_depth < (reg->reserved_tags + BLK_MQ_TAG_MIN)) return ERR_PTR(-EINVAL); diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h index 3368b97bee73..ab0e9b2025b3 100644 --- a/include/linux/blk-mq.h +++ b/include/linux/blk-mq.h @@ -124,7 +124,7 @@ void blk_mq_insert_request(struct request_queue *, struct request *, bool); void blk_mq_run_queues(struct request_queue *q, bool async); void blk_mq_free_request(struct request *rq); bool blk_mq_can_queue(struct blk_mq_hw_ctx *); -struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp); +struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp, bool reserved); struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, gfp_t gfp); struct request *blk_mq_rq_from_tag(struct request_queue *q, unsigned int tag); -- cgit v1.2.3