blk-mq.h revision 6fca6a611c27f1f0d90fbe1cc3c229dbf8c09e48 
1#ifndef BLK_MQ_H
2#define BLK_MQ_H
3
4#include <linux/blkdev.h>
5
6struct blk_mq_tags;
7
8struct blk_mq_cpu_notifier {
9	struct list_head list;
10	void *data;
11	int (*notify)(void *data, unsigned long action, unsigned int cpu);
12};
13
14struct blk_mq_ctxmap {
15	unsigned int map_size;
16	unsigned int bits_per_word;
17	struct blk_align_bitmap *map;
18};
19
20struct blk_mq_hw_ctx {
21	struct {
22		spinlock_t		lock;
23		struct list_head	dispatch;
24	} ____cacheline_aligned_in_smp;
25
26	unsigned long		state;		/* BLK_MQ_S_* flags */
27	struct delayed_work	run_work;
28	struct delayed_work	delay_work;
29	cpumask_var_t		cpumask;
30	int			next_cpu;
31	int			next_cpu_batch;
32
33	unsigned long		flags;		/* BLK_MQ_F_* flags */
34
35	struct request_queue	*queue;
36	unsigned int		queue_num;
37
38	void			*driver_data;
39
40	struct blk_mq_ctxmap	ctx_map;
41
42	unsigned int		nr_ctx;
43	struct blk_mq_ctx	**ctxs;
44
45	unsigned int		wait_index;
46
47	struct blk_mq_tags	*tags;
48
49	unsigned long		queued;
50	unsigned long		run;
51#define BLK_MQ_MAX_DISPATCH_ORDER	10
52	unsigned long		dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
53
54	unsigned int		numa_node;
55	unsigned int		cmd_size;	/* per-request extra data */
56
57	atomic_t		nr_active;
58
59	struct blk_mq_cpu_notifier	cpu_notifier;
60	struct kobject		kobj;
61};
62
63struct blk_mq_tag_set {
64	struct blk_mq_ops	*ops;
65	unsigned int		nr_hw_queues;
66	unsigned int		queue_depth;	/* max hw supported */
67	unsigned int		reserved_tags;
68	unsigned int		cmd_size;	/* per-request extra data */
69	int			numa_node;
70	unsigned int		timeout;
71	unsigned int		flags;		/* BLK_MQ_F_* */
72	void			*driver_data;
73
74	struct blk_mq_tags	**tags;
75
76	struct mutex		tag_list_lock;
77	struct list_head	tag_list;
78};
79
80typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, struct request *);
81typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
82typedef struct blk_mq_hw_ctx *(alloc_hctx_fn)(struct blk_mq_tag_set *,
83		unsigned int, int);
84typedef void (free_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
85typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
86typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
87typedef int (init_request_fn)(void *, struct request *, unsigned int,
88		unsigned int, unsigned int);
89typedef void (exit_request_fn)(void *, struct request *, unsigned int,
90		unsigned int);
91
92struct blk_mq_ops {
93	/*
94	 * Queue request
95	 */
96	queue_rq_fn		*queue_rq;
97
98	/*
99	 * Map to specific hardware queue
100	 */
101	map_queue_fn		*map_queue;
102
103	/*
104	 * Called on request timeout
105	 */
106	rq_timed_out_fn		*timeout;
107
108	softirq_done_fn		*complete;
109
110	/*
111	 * Override for hctx allocations (should probably go)
112	 */
113	alloc_hctx_fn		*alloc_hctx;
114	free_hctx_fn		*free_hctx;
115
116	/*
117	 * Called when the block layer side of a hardware queue has been
118	 * set up, allowing the driver to allocate/init matching structures.
119	 * Ditto for exit/teardown.
120	 */
121	init_hctx_fn		*init_hctx;
122	exit_hctx_fn		*exit_hctx;
123
124	/*
125	 * Called for every command allocated by the block layer to allow
126	 * the driver to set up driver specific data.
127	 * Ditto for exit/teardown.
128	 */
129	init_request_fn		*init_request;
130	exit_request_fn		*exit_request;
131};
132
133enum {
134	BLK_MQ_RQ_QUEUE_OK	= 0,	/* queued fine */
135	BLK_MQ_RQ_QUEUE_BUSY	= 1,	/* requeue IO for later */
136	BLK_MQ_RQ_QUEUE_ERROR	= 2,	/* end IO with error */
137
138	BLK_MQ_F_SHOULD_MERGE	= 1 << 0,
139	BLK_MQ_F_SHOULD_SORT	= 1 << 1,
140	BLK_MQ_F_TAG_SHARED	= 1 << 2,
141
142	BLK_MQ_S_STOPPED	= 0,
143	BLK_MQ_S_TAG_ACTIVE	= 1,
144
145	BLK_MQ_MAX_DEPTH	= 2048,
146
147	BLK_MQ_CPU_WORK_BATCH	= 8,
148};
149
150struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
151int blk_mq_register_disk(struct gendisk *);
152void blk_mq_unregister_disk(struct gendisk *);
153
154int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
155void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
156
157void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
158
159void blk_mq_insert_request(struct request *, bool, bool, bool);
160void blk_mq_run_queues(struct request_queue *q, bool async);
161void blk_mq_free_request(struct request *rq);
162bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
163struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp);
164struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, gfp_t gfp);
165struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
166
167struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
168struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
169void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *, unsigned int);
170
171void blk_mq_end_io(struct request *rq, int error);
172void __blk_mq_end_io(struct request *rq, int error);
173
174void blk_mq_requeue_request(struct request *rq);
175void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
176void blk_mq_kick_requeue_list(struct request_queue *q);
177void blk_mq_complete_request(struct request *rq);
178
179void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
180void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
181void blk_mq_stop_hw_queues(struct request_queue *q);
182void blk_mq_start_hw_queues(struct request_queue *q);
183void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
184void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
185void blk_mq_tag_busy_iter(struct blk_mq_tags *tags, void (*fn)(void *data, unsigned long *), void *data);
186
187/*
188 * Driver command data is immediately after the request. So subtract request
189 * size to get back to the original request.
190 */
191static inline struct request *blk_mq_rq_from_pdu(void *pdu)
192{
193	return pdu - sizeof(struct request);
194}
195static inline void *blk_mq_rq_to_pdu(struct request *rq)
196{
197	return (void *) rq + sizeof(*rq);
198}
199
200#define queue_for_each_hw_ctx(q, hctx, i)				\
201	for ((i) = 0; (i) < (q)->nr_hw_queues &&			\
202	     ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
203
204#define queue_for_each_ctx(q, ctx, i)					\
205	for ((i) = 0; (i) < (q)->nr_queues &&				\
206	     ({ ctx = per_cpu_ptr((q)->queue_ctx, (i)); 1; }); (i)++)
207
208#define hctx_for_each_ctx(hctx, ctx, i)					\
209	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
210	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
211
212#define blk_ctx_sum(q, sum)						\
213({									\
214	struct blk_mq_ctx *__x;						\
215	unsigned int __ret = 0, __i;					\
216									\
217	queue_for_each_ctx((q), __x, __i)				\
218		__ret += sum;						\
219	__ret;								\
220})
221
222#endif
223