blk-mq.h 8.9 KB
Newer Older
1 2 3 4
#ifndef BLK_MQ_H
#define BLK_MQ_H

#include <linux/blkdev.h>
5
#include <linux/sbitmap.h>
6
#include <linux/srcu.h>
7 8

struct blk_mq_tags;
9
struct blk_flush_queue;
10 11 12 13 14

struct blk_mq_hw_ctx {
	struct {
		spinlock_t		lock;
		struct list_head	dispatch;
15
		unsigned long		state;		/* BLK_MQ_S_* flags */
16 17
	} ____cacheline_aligned_in_smp;

18
	struct delayed_work	run_work;
19
	cpumask_var_t		cpumask;
20 21
	int			next_cpu;
	int			next_cpu_batch;
22 23 24

	unsigned long		flags;		/* BLK_MQ_F_* flags */

25
	void			*sched_data;
26
	struct request_queue	*queue;
27
	struct blk_flush_queue	*fq;
28 29 30

	void			*driver_data;

31
	struct sbitmap		ctx_map;
32

33 34
	struct blk_mq_ctx	*dispatch_from;

35
	struct blk_mq_ctx	**ctxs;
36
	unsigned int		nr_ctx;
37

38
	wait_queue_entry_t	dispatch_wait;
39
	atomic_t		wait_index;
40 41

	struct blk_mq_tags	*tags;
42
	struct blk_mq_tags	*sched_tags;
43 44 45

	unsigned long		queued;
	unsigned long		run;
46
#define BLK_MQ_MAX_DISPATCH_ORDER	7
47 48 49
	unsigned long		dispatched[BLK_MQ_MAX_DISPATCH_ORDER];

	unsigned int		numa_node;
50
	unsigned int		queue_num;
51

52 53
	atomic_t		nr_active;

54
	struct hlist_node	cpuhp_dead;
55
	struct kobject		kobj;
J
Jens Axboe 已提交
56

57
	unsigned long		poll_considered;
J
Jens Axboe 已提交
58 59
	unsigned long		poll_invoked;
	unsigned long		poll_success;
60 61 62

#ifdef CONFIG_BLK_DEBUG_FS
	struct dentry		*debugfs_dir;
63
	struct dentry		*sched_debugfs_dir;
64
#endif
65 66 67

	/* Must be the last member - see also blk_mq_hw_ctx_size(). */
	struct srcu_struct	queue_rq_srcu[0];
68 69
};

70
struct blk_mq_tag_set {
71
	unsigned int		*mq_map;
J
Jens Axboe 已提交
72
	const struct blk_mq_ops	*ops;
73
	unsigned int		nr_hw_queues;
74
	unsigned int		queue_depth;	/* max hw supported */
75 76 77 78 79
	unsigned int		reserved_tags;
	unsigned int		cmd_size;	/* per-request extra data */
	int			numa_node;
	unsigned int		timeout;
	unsigned int		flags;		/* BLK_MQ_F_* */
80 81 82
	void			*driver_data;

	struct blk_mq_tags	**tags;
83 84 85

	struct mutex		tag_list_lock;
	struct list_head	tag_list;
86 87
};

88 89 90 91 92
struct blk_mq_queue_data {
	struct request *rq;
	bool last;
};

93 94
typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *,
		const struct blk_mq_queue_data *);
95
typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *);
96
typedef void (put_budget_fn)(struct blk_mq_hw_ctx *);
97
typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
98 99
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);
100
typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
101
		unsigned int, unsigned int);
102
typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
103
		unsigned int);
104

105 106
typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
		bool);
K
Keith Busch 已提交
107
typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
J
Jens Axboe 已提交
108
typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
109
typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
J
Jens Axboe 已提交
110

111

112 113 114 115 116 117
struct blk_mq_ops {
	/*
	 * Queue request
	 */
	queue_rq_fn		*queue_rq;

118 119 120 121 122 123 124 125 126
	/*
	 * Reserve budget before queue request, once .queue_rq is
	 * run, it is driver's responsibility to release the
	 * reserved budget. Also we have to handle failure case
	 * of .get_budget for avoiding I/O deadlock.
	 */
	get_budget_fn		*get_budget;
	put_budget_fn		*put_budget;

127 128 129
	/*
	 * Called on request timeout
	 */
130
	timeout_fn		*timeout;
131

J
Jens Axboe 已提交
132 133 134 135 136
	/*
	 * Called to poll for completion of a specific tag.
	 */
	poll_fn			*poll;

137 138
	softirq_done_fn		*complete;

139 140 141 142 143 144 145
	/*
	 * 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;
146 147 148 149

	/*
	 * Called for every command allocated by the block layer to allow
	 * the driver to set up driver specific data.
150 151 152 153
	 *
	 * Tag greater than or equal to queue_depth is for setting up
	 * flush request.
	 *
154 155 156 157
	 * Ditto for exit/teardown.
	 */
	init_request_fn		*init_request;
	exit_request_fn		*exit_request;
158 159
	/* Called from inside blk_get_request() */
	void (*initialize_rq_fn)(struct request *rq);
160 161

	map_queues_fn		*map_queues;
162 163 164 165 166 167 168 169

#ifdef CONFIG_BLK_DEBUG_FS
	/*
	 * Used by the debugfs implementation to show driver-specific
	 * information about a request.
	 */
	void (*show_rq)(struct seq_file *m, struct request *rq);
#endif
170 171 172 173
};

enum {
	BLK_MQ_F_SHOULD_MERGE	= 1 << 0,
174 175
	BLK_MQ_F_TAG_SHARED	= 1 << 1,
	BLK_MQ_F_SG_MERGE	= 1 << 2,
176
	BLK_MQ_F_BLOCKING	= 1 << 5,
177
	BLK_MQ_F_NO_SCHED	= 1 << 6,
S
Shaohua Li 已提交
178 179
	BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
	BLK_MQ_F_ALLOC_POLICY_BITS = 1,
180

181
	BLK_MQ_S_STOPPED	= 0,
182
	BLK_MQ_S_TAG_ACTIVE	= 1,
183
	BLK_MQ_S_SCHED_RESTART	= 2,
184
	BLK_MQ_S_START_ON_RUN	= 3,
185

186
	BLK_MQ_MAX_DEPTH	= 10240,
187 188

	BLK_MQ_CPU_WORK_BATCH	= 8,
189
};
S
Shaohua Li 已提交
190 191 192 193 194 195
#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
	((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
		((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
	((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
		<< BLK_MQ_F_ALLOC_POLICY_START_BIT)
196

197
struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
198 199
struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
						  struct request_queue *q);
200 201
int blk_mq_register_dev(struct device *, struct request_queue *);
void blk_mq_unregister_dev(struct device *, struct request_queue *);
202

203 204 205
int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
void blk_mq_free_tag_set(struct blk_mq_tag_set *set);

206 207 208 209
void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);

void blk_mq_free_request(struct request *rq);
bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
210 211 212 213

enum {
	BLK_MQ_REQ_NOWAIT	= (1 << 0), /* return when out of requests */
	BLK_MQ_REQ_RESERVED	= (1 << 1), /* allocate from reserved pool */
214
	BLK_MQ_REQ_INTERNAL	= (1 << 2), /* allocate internal/sched tag */
215
	BLK_MQ_REQ_PREEMPT	= (1 << 3), /* set RQF_PREEMPT */
216 217
};

218
struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
219
		unsigned int flags);
220 221
struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
		unsigned int op, unsigned int flags, unsigned int hctx_idx);
222
struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
223

B
Bart Van Assche 已提交
224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
enum {
	BLK_MQ_UNIQUE_TAG_BITS = 16,
	BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
};

u32 blk_mq_unique_tag(struct request *rq);

static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
{
	return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
}

static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
{
	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
}

241

242
int blk_mq_request_started(struct request *rq);
243
void blk_mq_start_request(struct request *rq);
244 245
void blk_mq_end_request(struct request *rq, blk_status_t error);
void __blk_mq_end_request(struct request *rq, blk_status_t error);
246

247 248 249
void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
				bool kick_requeue_list);
250
void blk_mq_kick_requeue_list(struct request_queue *q);
251
void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
252
void blk_mq_complete_request(struct request *rq);
253

254
bool blk_mq_queue_stopped(struct request_queue *q);
255 256
void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
257
void blk_mq_stop_hw_queues(struct request_queue *q);
258
void blk_mq_start_hw_queues(struct request_queue *q);
259
void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
260
void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
261
void blk_mq_quiesce_queue(struct request_queue *q);
262
void blk_mq_unquiesce_queue(struct request_queue *q);
263
void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
O
Omar Sandoval 已提交
264
void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
265
void blk_mq_run_hw_queues(struct request_queue *q, bool async);
266
void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
267 268
void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
		busy_tag_iter_fn *fn, void *priv);
269
void blk_mq_freeze_queue(struct request_queue *q);
270
void blk_mq_unfreeze_queue(struct request_queue *q);
271
void blk_freeze_queue_start(struct request_queue *q);
272
void blk_mq_freeze_queue_wait(struct request_queue *q);
273 274
int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
				     unsigned long timeout);
275 276
int blk_mq_tagset_iter(struct blk_mq_tag_set *set, void *data,
		int (reinit_request)(void *, struct request *));
277

278
int blk_mq_map_queues(struct blk_mq_tag_set *set);
K
Keith Busch 已提交
279 280
void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);

281
void blk_mq_quiesce_queue_nowait(struct request_queue *q);
282

283 284
/*
 * Driver command data is immediately after the request. So subtract request
J
Jens Axboe 已提交
285
 * size to get back to the original request, add request size to get the PDU.
286 287 288 289 290 291 292
 */
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)
{
J
Jens Axboe 已提交
293
	return rq + 1;
294 295 296
}

#define queue_for_each_hw_ctx(q, hctx, i)				\
297 298
	for ((i) = 0; (i) < (q)->nr_hw_queues &&			\
	     ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
299 300

#define hctx_for_each_ctx(hctx, ctx, i)					\
301 302
	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
303 304

#endif