include/linux/blk-mq.h

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