blob: 2498fdf3a503a261cf663469cabae48d46d5c24d [file] [log] [blame]
Jens Axboe320ae512013-10-24 09:20:05 +01001#ifndef BLK_MQ_H
2#define BLK_MQ_H
3
4#include <linux/blkdev.h>
5
6struct blk_mq_tags;
Ming Leif70ced02014-09-25 23:23:47 +08007struct blk_flush_queue;
Jens Axboe320ae512013-10-24 09:20:05 +01008
9struct blk_mq_cpu_notifier {
10 struct list_head list;
11 void *data;
Jens Axboee814e712014-05-21 13:59:08 -060012 int (*notify)(void *data, unsigned long action, unsigned int cpu);
Jens Axboe320ae512013-10-24 09:20:05 +010013};
14
Jens Axboe1429d7c2014-05-19 09:23:55 -060015struct blk_mq_ctxmap {
Jens Axboe569fd0c2015-04-17 08:28:50 -060016 unsigned int size;
Jens Axboe1429d7c2014-05-19 09:23:55 -060017 unsigned int bits_per_word;
18 struct blk_align_bitmap *map;
19};
20
Jens Axboe320ae512013-10-24 09:20:05 +010021struct blk_mq_hw_ctx {
22 struct {
23 spinlock_t lock;
24 struct list_head dispatch;
25 } ____cacheline_aligned_in_smp;
26
27 unsigned long state; /* BLK_MQ_S_* flags */
Christoph Hellwig70f4db62014-04-16 10:48:08 -060028 struct delayed_work run_work;
29 struct delayed_work delay_work;
Jens Axboee4043dc2014-04-09 10:18:23 -060030 cpumask_var_t cpumask;
Jens Axboe506e9312014-05-07 10:26:44 -060031 int next_cpu;
32 int next_cpu_batch;
Jens Axboe320ae512013-10-24 09:20:05 +010033
34 unsigned long flags; /* BLK_MQ_F_* flags */
35
36 struct request_queue *queue;
Ming Leif70ced02014-09-25 23:23:47 +080037 struct blk_flush_queue *fq;
Jens Axboe320ae512013-10-24 09:20:05 +010038
39 void *driver_data;
40
Jens Axboe1429d7c2014-05-19 09:23:55 -060041 struct blk_mq_ctxmap ctx_map;
42
Jens Axboe4bb659b2014-05-09 09:36:49 -060043 unsigned int nr_ctx;
44 struct blk_mq_ctx **ctxs;
45
Alexander Gordeev8537b122014-06-17 22:12:35 -070046 atomic_t wait_index;
Jens Axboe320ae512013-10-24 09:20:05 +010047
Jens Axboe320ae512013-10-24 09:20:05 +010048 struct blk_mq_tags *tags;
49
50 unsigned long queued;
51 unsigned long run;
52#define BLK_MQ_MAX_DISPATCH_ORDER 10
53 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
54
Jens Axboe320ae512013-10-24 09:20:05 +010055 unsigned int numa_node;
Jens Axboe17ded322015-01-07 10:44:04 -070056 unsigned int queue_num;
Jens Axboe320ae512013-10-24 09:20:05 +010057
Jens Axboe0d2602c2014-05-13 15:10:52 -060058 atomic_t nr_active;
59
Jens Axboe320ae512013-10-24 09:20:05 +010060 struct blk_mq_cpu_notifier cpu_notifier;
61 struct kobject kobj;
Jens Axboe05229be2015-11-05 10:44:55 -070062
63 unsigned long poll_invoked;
64 unsigned long poll_success;
Jens Axboe320ae512013-10-24 09:20:05 +010065};
66
Christoph Hellwig24d2f902014-04-15 14:14:00 -060067struct blk_mq_tag_set {
Jens Axboe320ae512013-10-24 09:20:05 +010068 struct blk_mq_ops *ops;
69 unsigned int nr_hw_queues;
Jens Axboee3a2b3f2014-05-20 11:49:02 -060070 unsigned int queue_depth; /* max hw supported */
Jens Axboe320ae512013-10-24 09:20:05 +010071 unsigned int reserved_tags;
72 unsigned int cmd_size; /* per-request extra data */
73 int numa_node;
74 unsigned int timeout;
75 unsigned int flags; /* BLK_MQ_F_* */
Christoph Hellwig24d2f902014-04-15 14:14:00 -060076 void *driver_data;
77
78 struct blk_mq_tags **tags;
Jens Axboe0d2602c2014-05-13 15:10:52 -060079
80 struct mutex tag_list_lock;
81 struct list_head tag_list;
Jens Axboe320ae512013-10-24 09:20:05 +010082};
83
Jens Axboe74c45052014-10-29 11:14:52 -060084struct blk_mq_queue_data {
85 struct request *rq;
86 struct list_head *list;
87 bool last;
88};
89
90typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
Jens Axboe320ae512013-10-24 09:20:05 +010091typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
Christoph Hellwig0152fb62014-09-13 16:40:13 -070092typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
Jens Axboe320ae512013-10-24 09:20:05 +010093typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
94typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
Christoph Hellwig24d2f902014-04-15 14:14:00 -060095typedef int (init_request_fn)(void *, struct request *, unsigned int,
96 unsigned int, unsigned int);
97typedef void (exit_request_fn)(void *, struct request *, unsigned int,
98 unsigned int);
Jens Axboe320ae512013-10-24 09:20:05 +010099
Christoph Hellwig81481eb2014-09-13 16:40:11 -0700100typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
101 bool);
Keith Buschf26cdc82015-06-01 09:29:53 -0600102typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
Jens Axboe05229be2015-11-05 10:44:55 -0700103typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
104
Christoph Hellwig81481eb2014-09-13 16:40:11 -0700105
Jens Axboe320ae512013-10-24 09:20:05 +0100106struct blk_mq_ops {
107 /*
108 * Queue request
109 */
110 queue_rq_fn *queue_rq;
111
112 /*
113 * Map to specific hardware queue
114 */
115 map_queue_fn *map_queue;
116
117 /*
118 * Called on request timeout
119 */
Christoph Hellwig0152fb62014-09-13 16:40:13 -0700120 timeout_fn *timeout;
Jens Axboe320ae512013-10-24 09:20:05 +0100121
Jens Axboe05229be2015-11-05 10:44:55 -0700122 /*
123 * Called to poll for completion of a specific tag.
124 */
125 poll_fn *poll;
126
Christoph Hellwig30a91cb2014-02-10 03:24:38 -0800127 softirq_done_fn *complete;
128
Jens Axboe320ae512013-10-24 09:20:05 +0100129 /*
Jens Axboe320ae512013-10-24 09:20:05 +0100130 * Called when the block layer side of a hardware queue has been
131 * set up, allowing the driver to allocate/init matching structures.
132 * Ditto for exit/teardown.
133 */
134 init_hctx_fn *init_hctx;
135 exit_hctx_fn *exit_hctx;
Christoph Hellwige9b267d2014-04-15 13:59:10 -0600136
137 /*
138 * Called for every command allocated by the block layer to allow
139 * the driver to set up driver specific data.
Ming Leif70ced02014-09-25 23:23:47 +0800140 *
141 * Tag greater than or equal to queue_depth is for setting up
142 * flush request.
143 *
Christoph Hellwige9b267d2014-04-15 13:59:10 -0600144 * Ditto for exit/teardown.
145 */
146 init_request_fn *init_request;
147 exit_request_fn *exit_request;
Jens Axboe320ae512013-10-24 09:20:05 +0100148};
149
150enum {
151 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
152 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
153 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
154
155 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
Jens Axboe8a58d1f2014-08-15 12:38:41 -0600156 BLK_MQ_F_TAG_SHARED = 1 << 1,
157 BLK_MQ_F_SG_MERGE = 1 << 2,
Jens Axboee167dfb2014-10-29 11:18:26 -0600158 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
Shaohua Li24391c02015-01-23 14:18:00 -0700159 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
160 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
Jens Axboe320ae512013-10-24 09:20:05 +0100161
Jens Axboe5d12f902014-03-19 15:25:02 -0600162 BLK_MQ_S_STOPPED = 0,
Jens Axboe0d2602c2014-05-13 15:10:52 -0600163 BLK_MQ_S_TAG_ACTIVE = 1,
Jens Axboe320ae512013-10-24 09:20:05 +0100164
Jens Axboea4391c62014-06-05 15:21:56 -0600165 BLK_MQ_MAX_DEPTH = 10240,
Jens Axboe506e9312014-05-07 10:26:44 -0600166
167 BLK_MQ_CPU_WORK_BATCH = 8,
Jens Axboe320ae512013-10-24 09:20:05 +0100168};
Shaohua Li24391c02015-01-23 14:18:00 -0700169#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
170 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
171 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
172#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
173 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
174 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
Jens Axboe320ae512013-10-24 09:20:05 +0100175
Christoph Hellwig24d2f902014-04-15 14:14:00 -0600176struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
Mike Snitzerb62c21b2015-03-12 23:56:02 -0400177struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
178 struct request_queue *q);
Jens Axboe320ae512013-10-24 09:20:05 +0100179int blk_mq_register_disk(struct gendisk *);
180void blk_mq_unregister_disk(struct gendisk *);
Jens Axboe320ae512013-10-24 09:20:05 +0100181
Christoph Hellwig24d2f902014-04-15 14:14:00 -0600182int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
183void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
184
Jens Axboe320ae512013-10-24 09:20:05 +0100185void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
186
Christoph Hellwigfeb71da2014-02-20 15:32:37 -0800187void blk_mq_insert_request(struct request *, bool, bool, bool);
Jens Axboe320ae512013-10-24 09:20:05 +0100188void blk_mq_free_request(struct request *rq);
Jens Axboe7c7f2f22014-11-17 10:41:57 -0700189void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
Jens Axboe320ae512013-10-24 09:20:05 +0100190bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100191
192enum {
193 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
194 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
195};
196
Christoph Hellwig4ce01dd2014-05-27 20:59:46 +0200197struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
Christoph Hellwig6f3b0e82015-11-26 09:13:05 +0100198 unsigned int flags);
Jens Axboe0e62f512014-06-04 10:23:49 -0600199struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
Keith Buschf26cdc82015-06-01 09:29:53 -0600200struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
Jens Axboe320ae512013-10-24 09:20:05 +0100201
Bart Van Assche205fb5f2014-10-30 14:45:11 +0100202enum {
203 BLK_MQ_UNIQUE_TAG_BITS = 16,
204 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
205};
206
207u32 blk_mq_unique_tag(struct request *rq);
208
209static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
210{
211 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
212}
213
214static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
215{
216 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
217}
218
Jens Axboe320ae512013-10-24 09:20:05 +0100219struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
Jens Axboef14bbe72014-05-27 12:06:53 -0600220struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
Jens Axboe320ae512013-10-24 09:20:05 +0100221
Keith Busch973c0192015-01-07 18:55:43 -0700222int blk_mq_request_started(struct request *rq);
Christoph Hellwige2490072014-09-13 16:40:09 -0700223void blk_mq_start_request(struct request *rq);
Christoph Hellwigc8a446a2014-09-13 16:40:10 -0700224void blk_mq_end_request(struct request *rq, int error);
225void __blk_mq_end_request(struct request *rq, int error);
Jens Axboe320ae512013-10-24 09:20:05 +0100226
Christoph Hellwiged0791b2014-04-16 09:44:57 +0200227void blk_mq_requeue_request(struct request *rq);
Christoph Hellwig6fca6a62014-05-28 08:08:02 -0600228void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
Keith Buschc68ed592015-01-07 18:55:44 -0700229void blk_mq_cancel_requeue_work(struct request_queue *q);
Christoph Hellwig6fca6a62014-05-28 08:08:02 -0600230void blk_mq_kick_requeue_list(struct request_queue *q);
Jens Axboe1885b242015-01-07 18:55:45 -0700231void blk_mq_abort_requeue_list(struct request_queue *q);
Christoph Hellwigf4829a92015-09-27 21:01:50 +0200232void blk_mq_complete_request(struct request *rq, int error);
Christoph Hellwig30a91cb2014-02-10 03:24:38 -0800233
Jens Axboe320ae512013-10-24 09:20:05 +0100234void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
235void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
Christoph Hellwig280d45f2013-10-25 14:45:58 +0100236void blk_mq_stop_hw_queues(struct request_queue *q);
Christoph Hellwig2f268552014-04-16 09:44:56 +0200237void blk_mq_start_hw_queues(struct request_queue *q);
Christoph Hellwig1b4a3252014-04-16 09:44:54 +0200238void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
Mike Snitzerb94ec292015-03-11 23:56:38 -0400239void blk_mq_run_hw_queues(struct request_queue *q, bool async);
Christoph Hellwig70f4db62014-04-16 10:48:08 -0600240void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
Sagi Grimberge0489482016-03-10 13:58:46 +0200241void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
242 busy_tag_iter_fn *fn, void *priv);
Jens Axboec761d962015-01-02 15:05:12 -0700243void blk_mq_freeze_queue(struct request_queue *q);
Keith Buschb4c6a022014-12-19 17:54:14 -0700244void blk_mq_unfreeze_queue(struct request_queue *q);
245void blk_mq_freeze_queue_start(struct request_queue *q);
Jens Axboe320ae512013-10-24 09:20:05 +0100246
Keith Busch868f2f02015-12-17 17:08:14 -0700247void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
248
Jens Axboe320ae512013-10-24 09:20:05 +0100249/*
250 * Driver command data is immediately after the request. So subtract request
Jens Axboe2963e3f2015-04-09 15:54:05 -0600251 * size to get back to the original request, add request size to get the PDU.
Jens Axboe320ae512013-10-24 09:20:05 +0100252 */
253static inline struct request *blk_mq_rq_from_pdu(void *pdu)
254{
255 return pdu - sizeof(struct request);
256}
257static inline void *blk_mq_rq_to_pdu(struct request *rq)
258{
Jens Axboe2963e3f2015-04-09 15:54:05 -0600259 return rq + 1;
Jens Axboe320ae512013-10-24 09:20:05 +0100260}
261
Jens Axboe320ae512013-10-24 09:20:05 +0100262#define queue_for_each_hw_ctx(q, hctx, i) \
Jose Alonso0d0b7d42014-01-28 08:09:46 -0700263 for ((i) = 0; (i) < (q)->nr_hw_queues && \
264 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
Jens Axboe320ae512013-10-24 09:20:05 +0100265
Jens Axboe320ae512013-10-24 09:20:05 +0100266#define hctx_for_each_ctx(hctx, ctx, i) \
Jose Alonso0d0b7d42014-01-28 08:09:46 -0700267 for ((i) = 0; (i) < (hctx)->nr_ctx && \
268 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
Jens Axboe320ae512013-10-24 09:20:05 +0100269
Jens Axboe320ae512013-10-24 09:20:05 +0100270#endif