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 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 struct blk_flush_queue *fq;
38
39 void *driver_data;
40
41 struct blk_mq_ctxmap ctx_map;
42
43 unsigned int nr_ctx;
44 struct blk_mq_ctx **ctxs;
45
46 atomic_t wait_index;
47
48 struct blk_mq_tags *tags;
49
50 unsigned long queued;
51 unsigned long run;
52 #define BLK_MQ_MAX_DISPATCH_ORDER 10
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
60 struct blk_mq_cpu_notifier cpu_notifier;
61 struct kobject kobj;
62
63 unsigned long poll_invoked;
64 unsigned long poll_success;
65 };
66
67 struct blk_mq_tag_set {
68 struct blk_mq_ops *ops;
69 unsigned int nr_hw_queues;
70 unsigned int queue_depth; /* max hw supported */
71 unsigned int reserved_tags;
72 unsigned int cmd_size; /* per-request extra data */
73 int numa_node;
74 unsigned int timeout;
75 unsigned int flags; /* BLK_MQ_F_* */
76 void *driver_data;
77
78 struct blk_mq_tags **tags;
79
80 struct mutex tag_list_lock;
81 struct list_head tag_list;
82 };
83
84 struct blk_mq_queue_data {
85 struct request *rq;
86 struct list_head *list;
87 bool last;
88 };
89
90 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
91 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
92 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
93 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
94 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
95 typedef int (init_request_fn)(void *, struct request *, unsigned int,
96 unsigned int, unsigned int);
97 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
98 unsigned int);
99
100 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
101 bool);
102 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
103 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
104
105
106 struct blk_mq_ops {
107 /*
108 * Queue request
109 */
110 queue_rq_fn *queue_rq;
111
112 /*
113 * Map to specific hardware queue
114 */
115 map_queue_fn *map_queue;
116
117 /*
118 * Called on request timeout
119 */
120 timeout_fn *timeout;
121
122 /*
123 * Called to poll for completion of a specific tag.
124 */
125 poll_fn *poll;
126
127 softirq_done_fn *complete;
128
129 /*
130 * Called when the block layer side of a hardware queue has been
131 * set up, allowing the driver to allocate/init matching structures.
132 * Ditto for exit/teardown.
133 */
134 init_hctx_fn *init_hctx;
135 exit_hctx_fn *exit_hctx;
136
137 /*
138 * Called for every command allocated by the block layer to allow
139 * the driver to set up driver specific data.
140 *
141 * Tag greater than or equal to queue_depth is for setting up
142 * flush request.
143 *
144 * Ditto for exit/teardown.
145 */
146 init_request_fn *init_request;
147 exit_request_fn *exit_request;
148 };
149
150 enum {
151 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
152 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
153 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
154
155 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
156 BLK_MQ_F_TAG_SHARED = 1 << 1,
157 BLK_MQ_F_SG_MERGE = 1 << 2,
158 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
159 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
160 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
161
162 BLK_MQ_S_STOPPED = 0,
163 BLK_MQ_S_TAG_ACTIVE = 1,
164
165 BLK_MQ_MAX_DEPTH = 10240,
166
167 BLK_MQ_CPU_WORK_BATCH = 8,
168 };
169 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
170 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
171 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
172 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
173 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
174 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
175
176 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
177 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
178 struct request_queue *q);
179 int blk_mq_register_disk(struct gendisk *);
180 void blk_mq_unregister_disk(struct gendisk *);
181
182 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
183 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
184
185 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
186
187 void blk_mq_insert_request(struct request *, bool, bool, bool);
188 void blk_mq_free_request(struct request *rq);
189 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
190 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
191 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
192 gfp_t gfp, bool reserved);
193 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
194 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
195
196 enum {
197 BLK_MQ_UNIQUE_TAG_BITS = 16,
198 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
199 };
200
201 u32 blk_mq_unique_tag(struct request *rq);
202
blk_mq_unique_tag_to_hwq(u32 unique_tag)203 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
204 {
205 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
206 }
207
blk_mq_unique_tag_to_tag(u32 unique_tag)208 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
209 {
210 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
211 }
212
213 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
214 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
215
216 int blk_mq_request_started(struct request *rq);
217 void blk_mq_start_request(struct request *rq);
218 void blk_mq_end_request(struct request *rq, int error);
219 void __blk_mq_end_request(struct request *rq, int error);
220
221 void blk_mq_requeue_request(struct request *rq);
222 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
223 void blk_mq_cancel_requeue_work(struct request_queue *q);
224 void blk_mq_kick_requeue_list(struct request_queue *q);
225 void blk_mq_abort_requeue_list(struct request_queue *q);
226 void blk_mq_complete_request(struct request *rq, int error);
227
228 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
229 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
230 void blk_mq_stop_hw_queues(struct request_queue *q);
231 void blk_mq_start_hw_queues(struct request_queue *q);
232 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
233 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
234 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
235 void blk_mq_all_tag_busy_iter(struct blk_mq_tags *tags, busy_tag_iter_fn *fn,
236 void *priv);
237 void blk_mq_freeze_queue(struct request_queue *q);
238 void blk_mq_unfreeze_queue(struct request_queue *q);
239 void blk_mq_freeze_queue_start(struct request_queue *q);
240
241 /*
242 * Driver command data is immediately after the request. So subtract request
243 * size to get back to the original request, add request size to get the PDU.
244 */
blk_mq_rq_from_pdu(void * pdu)245 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
246 {
247 return pdu - sizeof(struct request);
248 }
blk_mq_rq_to_pdu(struct request * rq)249 static inline void *blk_mq_rq_to_pdu(struct request *rq)
250 {
251 return rq + 1;
252 }
253
254 #define queue_for_each_hw_ctx(q, hctx, i) \
255 for ((i) = 0; (i) < (q)->nr_hw_queues && \
256 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
257
258 #define queue_for_each_ctx(q, ctx, i) \
259 for ((i) = 0; (i) < (q)->nr_queues && \
260 ({ ctx = per_cpu_ptr((q)->queue_ctx, (i)); 1; }); (i)++)
261
262 #define hctx_for_each_ctx(hctx, ctx, i) \
263 for ((i) = 0; (i) < (hctx)->nr_ctx && \
264 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
265
266 #define blk_ctx_sum(q, sum) \
267 ({ \
268 struct blk_mq_ctx *__x; \
269 unsigned int __ret = 0, __i; \
270 \
271 queue_for_each_ctx((q), __x, __i) \
272 __ret += sum; \
273 __ret; \
274 })
275
276 #endif
277