1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef BLK_MQ_H
3 #define BLK_MQ_H
4
5 #include <linux/blkdev.h>
6 #include <linux/sbitmap.h>
7 #include <linux/srcu.h>
8
9 struct blk_mq_tags;
10 struct blk_flush_queue;
11
12 /**
13 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware block device
14 */
15 struct blk_mq_hw_ctx {
16 struct {
17 spinlock_t lock;
18 struct list_head dispatch;
19 unsigned long state; /* BLK_MQ_S_* flags */
20 } ____cacheline_aligned_in_smp;
21
22 struct delayed_work run_work;
23 cpumask_var_t cpumask;
24 int next_cpu;
25 int next_cpu_batch;
26
27 unsigned long flags; /* BLK_MQ_F_* flags */
28
29 void *sched_data;
30 struct request_queue *queue;
31 struct blk_flush_queue *fq;
32
33 void *driver_data;
34
35 struct sbitmap ctx_map;
36
37 struct blk_mq_ctx *dispatch_from;
38 unsigned int dispatch_busy;
39
40 unsigned int nr_ctx;
41 struct blk_mq_ctx **ctxs;
42
43 spinlock_t dispatch_wait_lock;
44 wait_queue_entry_t dispatch_wait;
45 atomic_t wait_index;
46
47 struct blk_mq_tags *tags;
48 struct blk_mq_tags *sched_tags;
49
50 unsigned long queued;
51 unsigned long run;
52 #define BLK_MQ_MAX_DISPATCH_ORDER 7
53 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
54
55 unsigned int numa_node;
56 unsigned int queue_num;
57
58 atomic_t nr_active;
59 unsigned int nr_expired;
60
61 struct hlist_node cpuhp_dead;
62 struct kobject kobj;
63
64 unsigned long poll_considered;
65 unsigned long poll_invoked;
66 unsigned long poll_success;
67
68 #ifdef CONFIG_BLK_DEBUG_FS
69 struct dentry *debugfs_dir;
70 struct dentry *sched_debugfs_dir;
71 #endif
72
73 /* Must be the last member - see also blk_mq_hw_ctx_size(). */
74 struct srcu_struct srcu[0];
75 };
76
77 struct blk_mq_tag_set {
78 unsigned int *mq_map;
79 const struct blk_mq_ops *ops;
80 unsigned int nr_hw_queues;
81 unsigned int queue_depth; /* max hw supported */
82 unsigned int reserved_tags;
83 unsigned int cmd_size; /* per-request extra data */
84 int numa_node;
85 unsigned int timeout;
86 unsigned int flags; /* BLK_MQ_F_* */
87 void *driver_data;
88
89 struct blk_mq_tags **tags;
90
91 struct mutex tag_list_lock;
92 struct list_head tag_list;
93 };
94
95 struct blk_mq_queue_data {
96 struct request *rq;
97 bool last;
98 };
99
100 typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *,
101 const struct blk_mq_queue_data *);
102 typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *);
103 typedef void (put_budget_fn)(struct blk_mq_hw_ctx *);
104 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
105 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
106 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
107 typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
108 unsigned int, unsigned int);
109 typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
110 unsigned int);
111
112 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
113 bool);
114 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
115 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
116 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
117 typedef void (cleanup_rq_fn)(struct request *);
118
119
120 struct blk_mq_ops {
121 /*
122 * Queue request
123 */
124 queue_rq_fn *queue_rq;
125
126 /*
127 * Reserve budget before queue request, once .queue_rq is
128 * run, it is driver's responsibility to release the
129 * reserved budget. Also we have to handle failure case
130 * of .get_budget for avoiding I/O deadlock.
131 */
132 get_budget_fn *get_budget;
133 put_budget_fn *put_budget;
134
135 /*
136 * Called on request timeout
137 */
138 timeout_fn *timeout;
139
140 /*
141 * Called to poll for completion of a specific tag.
142 */
143 poll_fn *poll;
144
145 softirq_done_fn *complete;
146
147 /*
148 * Called when the block layer side of a hardware queue has been
149 * set up, allowing the driver to allocate/init matching structures.
150 * Ditto for exit/teardown.
151 */
152 init_hctx_fn *init_hctx;
153 exit_hctx_fn *exit_hctx;
154
155 /*
156 * Called for every command allocated by the block layer to allow
157 * the driver to set up driver specific data.
158 *
159 * Tag greater than or equal to queue_depth is for setting up
160 * flush request.
161 *
162 * Ditto for exit/teardown.
163 */
164 init_request_fn *init_request;
165 exit_request_fn *exit_request;
166 /* Called from inside blk_get_request() */
167 void (*initialize_rq_fn)(struct request *rq);
168
169 /*
170 * Called before freeing one request which isn't completed yet,
171 * and usually for freeing the driver private data
172 */
173 cleanup_rq_fn *cleanup_rq;
174
175 map_queues_fn *map_queues;
176
177 #ifdef CONFIG_BLK_DEBUG_FS
178 /*
179 * Used by the debugfs implementation to show driver-specific
180 * information about a request.
181 */
182 void (*show_rq)(struct seq_file *m, struct request *rq);
183 #endif
184 };
185
186 enum {
187 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
188 BLK_MQ_F_TAG_SHARED = 1 << 1,
189 BLK_MQ_F_SG_MERGE = 1 << 2,
190 BLK_MQ_F_BLOCKING = 1 << 5,
191 BLK_MQ_F_NO_SCHED = 1 << 6,
192 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
193 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
194
195 BLK_MQ_S_STOPPED = 0,
196 BLK_MQ_S_TAG_ACTIVE = 1,
197 BLK_MQ_S_SCHED_RESTART = 2,
198
199 BLK_MQ_MAX_DEPTH = 10240,
200
201 BLK_MQ_CPU_WORK_BATCH = 8,
202 };
203 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
204 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
205 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
206 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
207 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
208 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
209
210 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
211 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
212 struct request_queue *q);
213 int blk_mq_register_dev(struct device *, struct request_queue *);
214 void blk_mq_unregister_dev(struct device *, struct request_queue *);
215
216 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
217 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
218
219 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
220
221 void blk_mq_free_request(struct request *rq);
222 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
223
224 enum {
225 /* return when out of requests */
226 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
227 /* allocate from reserved pool */
228 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
229 /* allocate internal/sched tag */
230 BLK_MQ_REQ_INTERNAL = (__force blk_mq_req_flags_t)(1 << 2),
231 /* set RQF_PREEMPT */
232 BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3),
233 };
234
235 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
236 blk_mq_req_flags_t flags);
237 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
238 unsigned int op, blk_mq_req_flags_t flags,
239 unsigned int hctx_idx);
240 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
241
242 enum {
243 BLK_MQ_UNIQUE_TAG_BITS = 16,
244 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
245 };
246
247 u32 blk_mq_unique_tag(struct request *rq);
248
blk_mq_unique_tag_to_hwq(u32 unique_tag)249 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
250 {
251 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
252 }
253
blk_mq_unique_tag_to_tag(u32 unique_tag)254 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
255 {
256 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
257 }
258
259
260 int blk_mq_request_started(struct request *rq);
261 void blk_mq_start_request(struct request *rq);
262 void blk_mq_end_request(struct request *rq, blk_status_t error);
263 void __blk_mq_end_request(struct request *rq, blk_status_t error);
264
265 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
266 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
267 bool kick_requeue_list);
268 void blk_mq_kick_requeue_list(struct request_queue *q);
269 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
270 void blk_mq_complete_request(struct request *rq);
271 bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list,
272 struct bio *bio);
273 bool blk_mq_queue_stopped(struct request_queue *q);
274 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
275 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
276 void blk_mq_stop_hw_queues(struct request_queue *q);
277 void blk_mq_start_hw_queues(struct request_queue *q);
278 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
279 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
280 void blk_mq_quiesce_queue(struct request_queue *q);
281 void blk_mq_unquiesce_queue(struct request_queue *q);
282 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
283 bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
284 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
285 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
286 busy_tag_iter_fn *fn, void *priv);
287 void blk_mq_freeze_queue(struct request_queue *q);
288 void blk_mq_unfreeze_queue(struct request_queue *q);
289 void blk_freeze_queue_start(struct request_queue *q);
290 void blk_mq_freeze_queue_wait(struct request_queue *q);
291 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
292 unsigned long timeout);
293
294 int blk_mq_map_queues(struct blk_mq_tag_set *set);
295 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
296
297 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
298
299 /**
300 * blk_mq_mark_complete() - Set request state to complete
301 * @rq: request to set to complete state
302 *
303 * Returns true if request state was successfully set to complete. If
304 * successful, the caller is responsibile for seeing this request is ended, as
305 * blk_mq_complete_request will not work again.
306 */
blk_mq_mark_complete(struct request * rq)307 static inline bool blk_mq_mark_complete(struct request *rq)
308 {
309 return cmpxchg(&rq->state, MQ_RQ_IN_FLIGHT, MQ_RQ_COMPLETE) ==
310 MQ_RQ_IN_FLIGHT;
311 }
312
313 /*
314 * Driver command data is immediately after the request. So subtract request
315 * size to get back to the original request, add request size to get the PDU.
316 */
blk_mq_rq_from_pdu(void * pdu)317 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
318 {
319 return pdu - sizeof(struct request);
320 }
blk_mq_rq_to_pdu(struct request * rq)321 static inline void *blk_mq_rq_to_pdu(struct request *rq)
322 {
323 return rq + 1;
324 }
325
326 #define queue_for_each_hw_ctx(q, hctx, i) \
327 for ((i) = 0; (i) < (q)->nr_hw_queues && \
328 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
329
330 #define hctx_for_each_ctx(hctx, ctx, i) \
331 for ((i) = 0; (i) < (hctx)->nr_ctx && \
332 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
333
blk_mq_cleanup_rq(struct request * rq)334 static inline void blk_mq_cleanup_rq(struct request *rq)
335 {
336 if (rq->q->mq_ops->cleanup_rq)
337 rq->q->mq_ops->cleanup_rq(rq);
338 }
339
340 #endif
341