1 #ifndef BLK_MQ_H
2 #define BLK_MQ_H
3
4 #include <linux/blkdev.h>
5
6 struct blk_mq_tags;
7 struct blk_flush_queue;
8
9 struct blk_mq_cpu_notifier {
10 struct list_head list;
11 void *data;
12 int (*notify)(void *data, unsigned long action, unsigned int cpu);
13 };
14
15 struct blk_mq_ctxmap {
16 unsigned int map_size;
17 unsigned int bits_per_word;
18 struct blk_align_bitmap *map;
19 };
20
21 struct blk_mq_hw_ctx {
22 struct {
23 spinlock_t lock;
24 struct list_head dispatch;
25 } ____cacheline_aligned_in_smp;
26
27 unsigned long state; /* BLK_MQ_S_* flags */
28 struct delayed_work run_work;
29 struct delayed_work delay_work;
30 cpumask_var_t cpumask;
31 int next_cpu;
32 int next_cpu_batch;
33
34 unsigned long flags; /* BLK_MQ_F_* flags */
35
36 struct request_queue *queue;
37 unsigned int queue_num;
38 struct blk_flush_queue *fq;
39
40 void *driver_data;
41
42 struct blk_mq_ctxmap ctx_map;
43
44 unsigned int nr_ctx;
45 struct blk_mq_ctx **ctxs;
46
47 atomic_t wait_index;
48
49 struct blk_mq_tags *tags;
50
51 unsigned long queued;
52 unsigned long run;
53 #define BLK_MQ_MAX_DISPATCH_ORDER 10
54 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
55
56 unsigned int numa_node;
57 unsigned int cmd_size; /* per-request extra data */
58
59 atomic_t nr_active;
60
61 struct blk_mq_cpu_notifier cpu_notifier;
62 struct kobject kobj;
63 };
64
65 struct blk_mq_tag_set {
66 struct blk_mq_ops *ops;
67 unsigned int nr_hw_queues;
68 unsigned int queue_depth; /* max hw supported */
69 unsigned int reserved_tags;
70 unsigned int cmd_size; /* per-request extra data */
71 int numa_node;
72 unsigned int timeout;
73 unsigned int flags; /* BLK_MQ_F_* */
74 void *driver_data;
75
76 struct blk_mq_tags **tags;
77
78 struct mutex tag_list_lock;
79 struct list_head tag_list;
80 };
81
82 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, struct request *, bool);
83 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
84 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
85 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
86 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
87 typedef int (init_request_fn)(void *, struct request *, unsigned int,
88 unsigned int, unsigned int);
89 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
90 unsigned int);
91
92 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
93 bool);
94
95 struct blk_mq_ops {
96 /*
97 * Queue request
98 */
99 queue_rq_fn *queue_rq;
100
101 /*
102 * Map to specific hardware queue
103 */
104 map_queue_fn *map_queue;
105
106 /*
107 * Called on request timeout
108 */
109 timeout_fn *timeout;
110
111 softirq_done_fn *complete;
112
113 /*
114 * Called when the block layer side of a hardware queue has been
115 * set up, allowing the driver to allocate/init matching structures.
116 * Ditto for exit/teardown.
117 */
118 init_hctx_fn *init_hctx;
119 exit_hctx_fn *exit_hctx;
120
121 /*
122 * Called for every command allocated by the block layer to allow
123 * the driver to set up driver specific data.
124 *
125 * Tag greater than or equal to queue_depth is for setting up
126 * flush request.
127 *
128 * Ditto for exit/teardown.
129 */
130 init_request_fn *init_request;
131 exit_request_fn *exit_request;
132 };
133
134 enum {
135 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
136 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
137 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
138
139 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
140 BLK_MQ_F_TAG_SHARED = 1 << 1,
141 BLK_MQ_F_SG_MERGE = 1 << 2,
142 BLK_MQ_F_SYSFS_UP = 1 << 3,
143
144 BLK_MQ_S_STOPPED = 0,
145 BLK_MQ_S_TAG_ACTIVE = 1,
146
147 BLK_MQ_MAX_DEPTH = 10240,
148
149 BLK_MQ_CPU_WORK_BATCH = 8,
150 };
151
152 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
153 void blk_mq_finish_init(struct request_queue *q);
154 int blk_mq_register_disk(struct gendisk *);
155 void blk_mq_unregister_disk(struct gendisk *);
156
157 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
158 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
159
160 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
161
162 void blk_mq_insert_request(struct request *, bool, bool, bool);
163 void blk_mq_run_queues(struct request_queue *q, bool async);
164 void blk_mq_free_request(struct request *rq);
165 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
166 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
167 gfp_t gfp, bool reserved);
168 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
169
170 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
171 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
172
173 void blk_mq_start_request(struct request *rq);
174 void blk_mq_end_request(struct request *rq, int error);
175 void __blk_mq_end_request(struct request *rq, int error);
176
177 void blk_mq_requeue_request(struct request *rq);
178 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
179 void blk_mq_kick_requeue_list(struct request_queue *q);
180 void blk_mq_complete_request(struct request *rq);
181
182 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
183 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
184 void blk_mq_stop_hw_queues(struct request_queue *q);
185 void blk_mq_start_hw_queues(struct request_queue *q);
186 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
187 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
188 void blk_mq_tag_busy_iter(struct blk_mq_hw_ctx *hctx, busy_iter_fn *fn,
189 void *priv);
190
191 /*
192 * Driver command data is immediately after the request. So subtract request
193 * size to get back to the original request.
194 */
blk_mq_rq_from_pdu(void * pdu)195 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
196 {
197 return pdu - sizeof(struct request);
198 }
blk_mq_rq_to_pdu(struct request * rq)199 static inline void *blk_mq_rq_to_pdu(struct request *rq)
200 {
201 return (void *) rq + sizeof(*rq);
202 }
203
204 #define queue_for_each_hw_ctx(q, hctx, i) \
205 for ((i) = 0; (i) < (q)->nr_hw_queues && \
206 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
207
208 #define queue_for_each_ctx(q, ctx, i) \
209 for ((i) = 0; (i) < (q)->nr_queues && \
210 ({ ctx = per_cpu_ptr((q)->queue_ctx, (i)); 1; }); (i)++)
211
212 #define hctx_for_each_ctx(hctx, ctx, i) \
213 for ((i) = 0; (i) < (hctx)->nr_ctx && \
214 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
215
216 #define blk_ctx_sum(q, sum) \
217 ({ \
218 struct blk_mq_ctx *__x; \
219 unsigned int __ret = 0, __i; \
220 \
221 queue_for_each_ctx((q), __x, __i) \
222 __ret += sum; \
223 __ret; \
224 })
225
226 #endif
227