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 <shli@fusionio.com>
Alexander Gordeev <agordeev@redhat.com>
Christoph Hellwig <hch@infradead.org>
Mike Christie <michaelc@cs.wisc.edu>
Matias Bjorling <m@bjorling.me>
Jeff Moyer <jmoyer@redhat.com>

Acked-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

4 files changed, 233 insertions, 7 deletions

diff --git a/include/linux/bio.h b/include/linux/bio.h
index ec48bac5b039..4c2775443dcf 100644
--- a/include/linux/bio.h
+++ b/include/linux/bio.h
@@ -419,6 +419,8 @@ static inline void bio_list_init(struct bio_list *bl)
         bl->head = bl->tail = NULL;
 }
 
+#define BIO_EMPTY_LIST  { NULL, NULL }
+
 #define bio_list_for_each(bio, bl) \
         for (bio = (bl)->head; bio; bio = bio->bi_next)
 
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 <linux/blkdev.h>
+
+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
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index c26801e14788..238ef0ed62f8 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -178,6 +178,7 @@ enum rq_flag_bits {
         __REQ_MIXED_MERGE,      /* merge of different types, fail separately */
         __REQ_KERNEL,           /* direct IO to kernel pages */
         __REQ_PM,               /* runtime pm request */
+        __REQ_END,              /* last of chain of requests */
         __REQ_NR_BITS,          /* stops here */
 };
 
@@ -229,5 +230,6 @@ enum rq_flag_bits {
 #define REQ_SECURE              (1ULL << __REQ_SECURE)
 #define REQ_KERNEL              (1ULL << __REQ_KERNEL)
 #define REQ_PM                  (1ULL << __REQ_PM)
+#define REQ_END                 (1ULL << __REQ_END)
 
 #endif /* __LINUX_BLK_TYPES_H */
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 0a8da96274c3..f26ec20f6354 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -8,6 +8,7 @@
 #include <linux/major.h>
 #include <linux/genhd.h>
 #include <linux/list.h>
+#include <linux/llist.h>
 #include <linux/timer.h>
 #include <linux/workqueue.h>
 #include <linux/pagemap.h>
@@ -94,10 +95,17 @@ enum rq_cmd_type_bits {
  * as well!
  */
 struct request {
-        struct list_head queuelist;
-        struct call_single_data csd;
+        union {
+                struct list_head queuelist;
+                struct llist_node ll_list;
+        };
+        union {
+                struct call_single_data csd;
+                struct work_struct mq_flush_data;
+        };
 
         struct request_queue *q;
+        struct blk_mq_ctx *mq_ctx;
 
         u64 cmd_flags;
         enum rq_cmd_type_bits cmd_type;
@@ -213,6 +221,8 @@ struct request_pm_state
 
 #include <linux/elevator.h>
 
+struct blk_queue_ctx;
+
 typedef void (request_fn_proc) (struct request_queue *q);
 typedef void (make_request_fn) (struct request_queue *q, struct bio *bio);
 typedef int (prep_rq_fn) (struct request_queue *, struct request *);
@@ -311,6 +321,18 @@ struct request_queue {
         dma_drain_needed_fn     *dma_drain_needed;
         lld_busy_fn             *lld_busy_fn;
 
+        struct blk_mq_ops       *mq_ops;
+
+        unsigned int            *mq_map;
+
+        /* sw queues */
+        struct blk_mq_ctx       *queue_ctx;
+        unsigned int            nr_queues;
+
+        /* hw dispatch queues */
+        struct blk_mq_hw_ctx    **queue_hw_ctx;
+        unsigned int            nr_hw_queues;
+
         /*
          * Dispatch queue sorting
          */
@@ -359,6 +381,11 @@ struct request_queue {
          */
         struct kobject kobj;
 
+        /*
+         * mq queue kobject
+         */
+        struct kobject mq_kobj;
+
 #ifdef CONFIG_PM_RUNTIME
         struct device           *dev;
         int                     rpm_status;
@@ -423,7 +450,13 @@ struct request_queue {
         unsigned long           flush_pending_since;
         struct list_head        flush_queue[2];
         struct list_head        flush_data_in_flight;
-        struct request          flush_rq;
+        union {
+                struct request  flush_rq;
+                struct {
+                        spinlock_t mq_flush_lock;
+                        struct work_struct mq_flush_work;
+                };
+        };
 
         struct mutex            sysfs_lock;
 
@@ -435,14 +468,14 @@ struct request_queue {
         struct bsg_class_device bsg_dev;
 #endif
 
-#ifdef CONFIG_BLK_CGROUP
-        struct list_head        all_q_node;
-#endif
 #ifdef CONFIG_BLK_DEV_THROTTLING
         /* Throttle data */
         struct throtl_data *td;
 #endif
         struct rcu_head         rcu_head;
+        wait_queue_head_t       mq_freeze_wq;
+        struct percpu_counter   mq_usage_counter;
+        struct list_head        all_q_node;
 };
 
 #define QUEUE_FLAG_QUEUED       1       /* uses generic tag queueing */
@@ -465,6 +498,7 @@ struct request_queue {
 #define QUEUE_FLAG_SECDISCARD  17       /* supports SECDISCARD */
 #define QUEUE_FLAG_SAME_FORCE  18       /* force complete on same CPU */
 #define QUEUE_FLAG_DEAD        19       /* queue tear-down finished */
+#define QUEUE_FLAG_INIT_DONE   20       /* queue is initialized */
 
 #define QUEUE_FLAG_DEFAULT      ((1 << QUEUE_FLAG_IO_STAT) |            \
                                  (1 << QUEUE_FLAG_STACKABLE)    |       \
@@ -537,6 +571,7 @@ static inline void queue_flag_clear(unsigned int flag, struct request_queue *q)
 #define blk_queue_dying(q)      test_bit(QUEUE_FLAG_DYING, &(q)->queue_flags)
 #define blk_queue_dead(q)       test_bit(QUEUE_FLAG_DEAD, &(q)->queue_flags)
 #define blk_queue_bypass(q)     test_bit(QUEUE_FLAG_BYPASS, &(q)->queue_flags)
+#define blk_queue_init_done(q)  test_bit(QUEUE_FLAG_INIT_DONE, &(q)->queue_flags)
 #define blk_queue_nomerges(q)   test_bit(QUEUE_FLAG_NOMERGES, &(q)->queue_flags)
 #define blk_queue_noxmerges(q)  \
         test_bit(QUEUE_FLAG_NOXMERGES, &(q)->queue_flags)
@@ -1011,6 +1046,7 @@ static inline void blk_post_runtime_resume(struct request_queue *q, int err) {}
 struct blk_plug {
         unsigned long magic; /* detect uninitialized use-cases */
         struct list_head list; /* requests */
+        struct list_head mq_list; /* blk-mq requests */
         struct list_head cb_list; /* md requires an unplug callback */
 };
 #define BLK_MAX_REQUEST_COUNT 16
@@ -1048,7 +1084,10 @@ static inline bool blk_needs_flush_plug(struct task_struct *tsk)
 {
         struct blk_plug *plug = tsk->plug;
 
-        return plug && (!list_empty(&plug->list) || !list_empty(&plug->cb_list));
+        return plug &&
+                (!list_empty(&plug->list) ||
+                 !list_empty(&plug->mq_list) ||
+                 !list_empty(&plug->cb_list));
 }
 
 /*
@@ -1323,6 +1362,7 @@ static inline void put_dev_sector(Sector p)
 
 struct work_struct;
 int kblockd_schedule_work(struct request_queue *q, struct work_struct *work);
+int kblockd_schedule_delayed_work(struct request_queue *q, struct delayed_work *dwork, unsigned long delay);
 
 #ifdef CONFIG_BLK_CGROUP
 /*