1 #ifndef BLK_MQ_H
2 #define BLK_MQ_H
3
4 #include <linux/blkdev.h>
5 #include <linux/sbitmap.h>
6
7 struct blk_mq_tags;
8 struct blk_flush_queue;
9
10 struct blk_mq_hw_ctx {
11 struct {
12 spinlock_t lock;
13 struct list_head dispatch;
14 unsigned long state; /* BLK_MQ_S_* flags */
15 } ____cacheline_aligned_in_smp;
16
17 struct work_struct run_work;
18 cpumask_var_t cpumask;
19 int next_cpu;
20 int next_cpu_batch;
21
22 unsigned long flags; /* BLK_MQ_F_* flags */
23
24 struct request_queue *queue;
25 struct blk_flush_queue *fq;
26
27 void *driver_data;
28
29 struct sbitmap ctx_map;
30
31 struct blk_mq_ctx **ctxs;
32 unsigned int nr_ctx;
33
34 atomic_t wait_index;
35
36 struct blk_mq_tags *tags;
37
38 unsigned long queued;
39 unsigned long run;
40 #define BLK_MQ_MAX_DISPATCH_ORDER 7
41 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
42
43 unsigned int numa_node;
44 unsigned int queue_num;
45
46 atomic_t nr_active;
47
48 struct delayed_work delay_work;
49
50 struct hlist_node cpuhp_dead;
51 struct kobject kobj;
52
53 unsigned long poll_considered;
54 unsigned long poll_invoked;
55 unsigned long poll_success;
56 };
57
58 struct blk_mq_tag_set {
59 unsigned int *mq_map;
60 struct blk_mq_ops *ops;
61 unsigned int nr_hw_queues;
62 unsigned int queue_depth; /* max hw supported */
63 unsigned int reserved_tags;
64 unsigned int cmd_size; /* per-request extra data */
65 int numa_node;
66 unsigned int timeout;
67 unsigned int flags; /* BLK_MQ_F_* */
68 void *driver_data;
69
70 struct blk_mq_tags **tags;
71
72 struct mutex tag_list_lock;
73 struct list_head tag_list;
74 };
75
76 struct blk_mq_queue_data {
77 struct request *rq;
78 struct list_head *list;
79 bool last;
80 };
81
82 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
83 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
84 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
85 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
86 typedef int (init_request_fn)(void *, struct request *, unsigned int,
87 unsigned int, unsigned int);
88 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
89 unsigned int);
90 typedef int (reinit_request_fn)(void *, struct request *);
91
92 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
93 bool);
94 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
95 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
96 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
97
98
99 struct blk_mq_ops {
100 /*
101 * Queue request
102 */
103 queue_rq_fn *queue_rq;
104
105 /*
106 * Called on request timeout
107 */
108 timeout_fn *timeout;
109
110 /*
111 * Called to poll for completion of a specific tag.
112 */
113 poll_fn *poll;
114
115 softirq_done_fn *complete;
116
117 /*
118 * Called when the block layer side of a hardware queue has been
119 * set up, allowing the driver to allocate/init matching structures.
120 * Ditto for exit/teardown.
121 */
122 init_hctx_fn *init_hctx;
123 exit_hctx_fn *exit_hctx;
124
125 /*
126 * Called for every command allocated by the block layer to allow
127 * the driver to set up driver specific data.
128 *
129 * Tag greater than or equal to queue_depth is for setting up
130 * flush request.
131 *
132 * Ditto for exit/teardown.
133 */
134 init_request_fn *init_request;
135 exit_request_fn *exit_request;
136 reinit_request_fn *reinit_request;
137
138 map_queues_fn *map_queues;
139 };
140
141 enum {
142 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
143 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
144 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
145
146 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
147 BLK_MQ_F_TAG_SHARED = 1 << 1,
148 BLK_MQ_F_SG_MERGE = 1 << 2,
149 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
150 BLK_MQ_F_BLOCKING = 1 << 5,
151 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
152 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
153
154 BLK_MQ_S_STOPPED = 0,
155 BLK_MQ_S_TAG_ACTIVE = 1,
156
157 BLK_MQ_MAX_DEPTH = 10240,
158
159 BLK_MQ_CPU_WORK_BATCH = 8,
160 };
161 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
162 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
163 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
164 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
165 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
166 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
167
168 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
169 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
170 struct request_queue *q);
171 int blk_mq_register_dev(struct device *, struct request_queue *);
172 void blk_mq_unregister_dev(struct device *, struct request_queue *);
173
174 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
175 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
176
177 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
178
179 void blk_mq_insert_request(struct request *, bool, bool, bool);
180 void blk_mq_free_request(struct request *rq);
181 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
182 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
183
184 enum {
185 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
186 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
187 };
188
189 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
190 unsigned int flags);
191 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
192 unsigned int flags, unsigned int hctx_idx);
193 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
194
195 enum {
196 BLK_MQ_UNIQUE_TAG_BITS = 16,
197 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
198 };
199
200 u32 blk_mq_unique_tag(struct request *rq);
201
blk_mq_unique_tag_to_hwq(u32 unique_tag)202 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
203 {
204 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
205 }
206
blk_mq_unique_tag_to_tag(u32 unique_tag)207 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
208 {
209 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
210 }
211
212
213 int blk_mq_request_started(struct request *rq);
214 void blk_mq_start_request(struct request *rq);
215 void blk_mq_end_request(struct request *rq, int error);
216 void __blk_mq_end_request(struct request *rq, int error);
217
218 void blk_mq_requeue_request(struct request *rq);
219 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
220 void blk_mq_cancel_requeue_work(struct request_queue *q);
221 void blk_mq_kick_requeue_list(struct request_queue *q);
222 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
223 void blk_mq_abort_requeue_list(struct request_queue *q);
224 void blk_mq_complete_request(struct request *rq, int error);
225
226 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
227 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
228 void blk_mq_stop_hw_queues(struct request_queue *q);
229 void blk_mq_start_hw_queues(struct request_queue *q);
230 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
231 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
232 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
233 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
234 busy_tag_iter_fn *fn, void *priv);
235 void blk_mq_freeze_queue(struct request_queue *q);
236 void blk_mq_unfreeze_queue(struct request_queue *q);
237 void blk_mq_freeze_queue_start(struct request_queue *q);
238 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
239
240 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
241
242 /*
243 * Driver command data is immediately after the request. So subtract request
244 * size to get back to the original request, add request size to get the PDU.
245 */
blk_mq_rq_from_pdu(void * pdu)246 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
247 {
248 return pdu - sizeof(struct request);
249 }
blk_mq_rq_to_pdu(struct request * rq)250 static inline void *blk_mq_rq_to_pdu(struct request *rq)
251 {
252 return rq + 1;
253 }
254
255 #define queue_for_each_hw_ctx(q, hctx, i) \
256 for ((i) = 0; (i) < (q)->nr_hw_queues && \
257 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
258
259 #define hctx_for_each_ctx(hctx, ctx, i) \
260 for ((i) = 0; (i) < (hctx)->nr_ctx && \
261 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
262
263 #endif
264