1 /* SPDX-License-Identifier: GPL-2.0 */
5 #include <linux/blkdev.h>
6 #include <linux/sbitmap.h>
7 #include <linux/srcu.h>
10 struct blk_flush_queue;
13 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware block device
15 struct blk_mq_hw_ctx {
18 struct list_head dispatch;
19 unsigned long state; /* BLK_MQ_S_* flags */
20 } ____cacheline_aligned_in_smp;
22 struct delayed_work run_work;
23 cpumask_var_t cpumask;
27 unsigned long flags; /* BLK_MQ_F_* flags */
30 struct request_queue *queue;
31 struct blk_flush_queue *fq;
35 struct sbitmap ctx_map;
37 struct blk_mq_ctx *dispatch_from;
38 unsigned int dispatch_busy;
41 unsigned short nr_ctx;
42 struct blk_mq_ctx **ctxs;
44 spinlock_t dispatch_wait_lock;
45 wait_queue_entry_t dispatch_wait;
48 struct blk_mq_tags *tags;
49 struct blk_mq_tags *sched_tags;
53 #define BLK_MQ_MAX_DISPATCH_ORDER 7
54 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
56 unsigned int numa_node;
57 unsigned int queue_num;
60 unsigned int nr_expired;
62 struct hlist_node cpuhp_dead;
65 unsigned long poll_considered;
66 unsigned long poll_invoked;
67 unsigned long poll_success;
69 #ifdef CONFIG_BLK_DEBUG_FS
70 struct dentry *debugfs_dir;
71 struct dentry *sched_debugfs_dir;
74 /* Must be the last member - see also blk_mq_hw_ctx_size(). */
75 struct srcu_struct srcu[0];
78 struct blk_mq_queue_map {
80 unsigned int nr_queues;
81 unsigned int queue_offset;
88 struct blk_mq_tag_set {
90 * map[] holds ctx -> hctx mappings, one map exists for each type
91 * that the driver wishes to support. There are no restrictions
92 * on maps being of the same size, and it's perfectly legal to
93 * share maps between types.
95 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
96 unsigned int nr_maps; /* nr entries in map[] */
97 const struct blk_mq_ops *ops;
98 unsigned int nr_hw_queues; /* nr hw queues across maps */
99 unsigned int queue_depth; /* max hw supported */
100 unsigned int reserved_tags;
101 unsigned int cmd_size; /* per-request extra data */
103 unsigned int timeout;
104 unsigned int flags; /* BLK_MQ_F_* */
107 struct blk_mq_tags **tags;
109 struct mutex tag_list_lock;
110 struct list_head tag_list;
113 struct blk_mq_queue_data {
118 typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *,
119 const struct blk_mq_queue_data *);
120 /* takes rq->cmd_flags as input, returns a hardware type index */
121 typedef int (rq_flags_to_type_fn)(struct request_queue *, unsigned int);
122 typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *);
123 typedef void (put_budget_fn)(struct blk_mq_hw_ctx *);
124 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
125 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
126 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
127 typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
128 unsigned int, unsigned int);
129 typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
132 typedef bool (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
134 typedef bool (busy_tag_iter_fn)(struct request *, void *, bool);
135 typedef int (poll_fn)(struct blk_mq_hw_ctx *);
136 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
137 typedef bool (busy_fn)(struct request_queue *);
138 typedef void (complete_fn)(struct request *);
145 queue_rq_fn *queue_rq;
148 * Return a queue map type for the given request/bio flags
150 rq_flags_to_type_fn *rq_flags_to_type;
153 * Reserve budget before queue request, once .queue_rq is
154 * run, it is driver's responsibility to release the
155 * reserved budget. Also we have to handle failure case
156 * of .get_budget for avoiding I/O deadlock.
158 get_budget_fn *get_budget;
159 put_budget_fn *put_budget;
162 * Called on request timeout
167 * Called to poll for completion of a specific tag.
171 complete_fn *complete;
174 * Called when the block layer side of a hardware queue has been
175 * set up, allowing the driver to allocate/init matching structures.
176 * Ditto for exit/teardown.
178 init_hctx_fn *init_hctx;
179 exit_hctx_fn *exit_hctx;
182 * Called for every command allocated by the block layer to allow
183 * the driver to set up driver specific data.
185 * Tag greater than or equal to queue_depth is for setting up
188 * Ditto for exit/teardown.
190 init_request_fn *init_request;
191 exit_request_fn *exit_request;
192 /* Called from inside blk_get_request() */
193 void (*initialize_rq_fn)(struct request *rq);
196 * If set, returns whether or not this queue currently is busy
200 map_queues_fn *map_queues;
202 #ifdef CONFIG_BLK_DEBUG_FS
204 * Used by the debugfs implementation to show driver-specific
205 * information about a request.
207 void (*show_rq)(struct seq_file *m, struct request *rq);
212 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
213 BLK_MQ_F_TAG_SHARED = 1 << 1,
214 BLK_MQ_F_SG_MERGE = 1 << 2,
215 BLK_MQ_F_BLOCKING = 1 << 5,
216 BLK_MQ_F_NO_SCHED = 1 << 6,
217 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
218 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
220 BLK_MQ_S_STOPPED = 0,
221 BLK_MQ_S_TAG_ACTIVE = 1,
222 BLK_MQ_S_SCHED_RESTART = 2,
224 BLK_MQ_MAX_DEPTH = 10240,
226 BLK_MQ_CPU_WORK_BATCH = 8,
228 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
229 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
230 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
231 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
232 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
233 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
235 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
236 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
237 struct request_queue *q);
238 struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set,
239 const struct blk_mq_ops *ops,
240 unsigned int queue_depth,
241 unsigned int set_flags);
242 int blk_mq_register_dev(struct device *, struct request_queue *);
243 void blk_mq_unregister_dev(struct device *, struct request_queue *);
245 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
246 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
248 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
250 void blk_mq_free_request(struct request *rq);
251 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
253 bool blk_mq_queue_busy(struct request_queue *q);
256 /* return when out of requests */
257 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
258 /* allocate from reserved pool */
259 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
260 /* allocate internal/sched tag */
261 BLK_MQ_REQ_INTERNAL = (__force blk_mq_req_flags_t)(1 << 2),
262 /* set RQF_PREEMPT */
263 BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3),
266 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
267 blk_mq_req_flags_t flags);
268 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
269 unsigned int op, blk_mq_req_flags_t flags,
270 unsigned int hctx_idx);
271 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
274 BLK_MQ_UNIQUE_TAG_BITS = 16,
275 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
278 u32 blk_mq_unique_tag(struct request *rq);
280 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
282 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
285 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
287 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
291 int blk_mq_request_started(struct request *rq);
292 void blk_mq_start_request(struct request *rq);
293 void blk_mq_end_request(struct request *rq, blk_status_t error);
294 void __blk_mq_end_request(struct request *rq, blk_status_t error);
296 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
297 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
298 bool kick_requeue_list);
299 void blk_mq_kick_requeue_list(struct request_queue *q);
300 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
301 bool blk_mq_complete_request(struct request *rq);
302 bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list,
304 bool blk_mq_queue_stopped(struct request_queue *q);
305 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
306 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
307 void blk_mq_stop_hw_queues(struct request_queue *q);
308 void blk_mq_start_hw_queues(struct request_queue *q);
309 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
310 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
311 void blk_mq_quiesce_queue(struct request_queue *q);
312 void blk_mq_unquiesce_queue(struct request_queue *q);
313 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
314 bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
315 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
316 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
317 busy_tag_iter_fn *fn, void *priv);
318 void blk_mq_freeze_queue(struct request_queue *q);
319 void blk_mq_unfreeze_queue(struct request_queue *q);
320 void blk_freeze_queue_start(struct request_queue *q);
321 void blk_mq_freeze_queue_wait(struct request_queue *q);
322 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
323 unsigned long timeout);
325 int blk_mq_map_queues(struct blk_mq_queue_map *qmap);
326 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
328 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
330 unsigned int blk_mq_rq_cpu(struct request *rq);
333 * Driver command data is immediately after the request. So subtract request
334 * size to get back to the original request, add request size to get the PDU.
336 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
338 return pdu - sizeof(struct request);
340 static inline void *blk_mq_rq_to_pdu(struct request *rq)
345 #define queue_for_each_hw_ctx(q, hctx, i) \
346 for ((i) = 0; (i) < (q)->nr_hw_queues && \
347 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
349 #define hctx_for_each_ctx(hctx, ctx, i) \
350 for ((i) = 0; (i) < (hctx)->nr_ctx && \
351 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)