include/linux/blk-mq.h

   1 #ifndef BLK_MQ_H
   2 #define BLK_MQ_H
   3
   4 #include <linux/blkdev.h>
   5 #include <linux/sbitmap.h>
   6 #include <linux/srcu.h>
   7
   8 struct blk_mq_tags;
   9 struct blk_flush_queue;
  10
  11 struct blk_mq_hw_ctx {
  12         struct {
  13                 spinlock_t              lock;
  14                 struct list_head        dispatch;
  15                 unsigned long           state;          /* BLK_MQ_S_* flags */
  16         } ____cacheline_aligned_in_smp;
  17
  18         struct delayed_work     run_work;
  19         cpumask_var_t           cpumask;
  20         int                     next_cpu;
  21         int                     next_cpu_batch;
  22
  23         unsigned long           flags;          /* BLK_MQ_F_* flags */
  24
  25         void                    *sched_data;
  26         struct request_queue    *queue;
  27         struct blk_flush_queue  *fq;
  28
  29         void                    *driver_data;
  30
  31         struct sbitmap          ctx_map;
  32
  33         struct blk_mq_ctx       **ctxs;
  34         unsigned int            nr_ctx;
  35
  36         wait_queue_t            dispatch_wait;
  37         atomic_t                wait_index;
  38
  39         struct blk_mq_tags      *tags;
  40         struct blk_mq_tags      *sched_tags;
  41
  42         struct srcu_struct      queue_rq_srcu;
  43
  44         unsigned long           queued;
  45         unsigned long           run;
  46 #define BLK_MQ_MAX_DISPATCH_ORDER       7
  47         unsigned long           dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
  48
  49         unsigned int            numa_node;
  50         unsigned int            queue_num;
  51
  52         atomic_t                nr_active;
  53
  54         struct hlist_node       cpuhp_dead;
  55         struct kobject          kobj;
  56
  57         unsigned long           poll_considered;
  58         unsigned long           poll_invoked;
  59         unsigned long           poll_success;
  60
  61 #ifdef CONFIG_BLK_DEBUG_FS
  62         struct dentry           *debugfs_dir;
  63         struct dentry           *sched_debugfs_dir;
  64 #endif
  65 };
  66
  67 struct blk_mq_tag_set {
  68         unsigned int            *mq_map;
  69         const struct blk_mq_ops *ops;
  70         unsigned int            nr_hw_queues;
  71         unsigned int            queue_depth;    /* max hw supported */
  72         unsigned int            reserved_tags;
  73         unsigned int            cmd_size;       /* per-request extra data */
  74         int                     numa_node;
  75         unsigned int            timeout;
  76         unsigned int            flags;          /* BLK_MQ_F_* */
  77         void                    *driver_data;
  78
  79         struct blk_mq_tags      **tags;
  80
  81         struct mutex            tag_list_lock;
  82         struct list_head        tag_list;
  83 };
  84
  85 struct blk_mq_queue_data {
  86         struct request *rq;
  87         bool last;
  88 };
  89
  90 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
  91 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
  92 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
  93 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
  94 typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
  95                 unsigned int, unsigned int);
  96 typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
  97                 unsigned int);
  98 typedef int (reinit_request_fn)(void *, struct request *);
  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 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
 105
 106
 107 struct blk_mq_ops {
 108         /*
 109          * Queue request
 110          */
 111         queue_rq_fn             *queue_rq;
 112
 113         /*
 114          * Called on request timeout
 115          */
 116         timeout_fn              *timeout;
 117
 118         /*
 119          * Called to poll for completion of a specific tag.
 120          */
 121         poll_fn                 *poll;
 122
 123         softirq_done_fn         *complete;
 124
 125         /*
 126          * Called when the block layer side of a hardware queue has been
 127          * set up, allowing the driver to allocate/init matching structures.
 128          * Ditto for exit/teardown.
 129          */
 130         init_hctx_fn            *init_hctx;
 131         exit_hctx_fn            *exit_hctx;
 132
 133         /*
 134          * Called for every command allocated by the block layer to allow
 135          * the driver to set up driver specific data.
 136          *
 137          * Tag greater than or equal to queue_depth is for setting up
 138          * flush request.
 139          *
 140          * Ditto for exit/teardown.
 141          */
 142         init_request_fn         *init_request;
 143         exit_request_fn         *exit_request;
 144         reinit_request_fn       *reinit_request;
 145
 146         map_queues_fn           *map_queues;
 147
 148 #ifdef CONFIG_BLK_DEBUG_FS
 149         /*
 150          * Used by the debugfs implementation to show driver-specific
 151          * information about a request.
 152          */
 153         void (*show_rq)(struct seq_file *m, struct request *rq);
 154 #endif
 155 };
 156
 157 enum {
 158         BLK_MQ_RQ_QUEUE_OK      = 0,    /* queued fine */
 159         BLK_MQ_RQ_QUEUE_BUSY    = 1,    /* requeue IO for later */
 160         BLK_MQ_RQ_QUEUE_ERROR   = 2,    /* end IO with error */
 161
 162         BLK_MQ_F_SHOULD_MERGE   = 1 << 0,
 163         BLK_MQ_F_TAG_SHARED     = 1 << 1,
 164         BLK_MQ_F_SG_MERGE       = 1 << 2,
 165         BLK_MQ_F_BLOCKING       = 1 << 5,
 166         BLK_MQ_F_NO_SCHED       = 1 << 6,
 167         BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
 168         BLK_MQ_F_ALLOC_POLICY_BITS = 1,
 169
 170         BLK_MQ_S_STOPPED        = 0,
 171         BLK_MQ_S_TAG_ACTIVE     = 1,
 172         BLK_MQ_S_SCHED_RESTART  = 2,
 173         BLK_MQ_S_TAG_WAITING    = 3,
 174         BLK_MQ_S_START_ON_RUN   = 4,
 175
 176         BLK_MQ_MAX_DEPTH        = 10240,
 177
 178         BLK_MQ_CPU_WORK_BATCH   = 8,
 179 };
 180 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
 181         ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
 182                 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
 183 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
 184         ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
 185                 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
 186
 187 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
 188 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 189                                                   struct request_queue *q);
 190 int blk_mq_register_dev(struct device *, struct request_queue *);
 191 void blk_mq_unregister_dev(struct device *, struct request_queue *);
 192
 193 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
 194 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
 195
 196 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
 197
 198 void blk_mq_free_request(struct request *rq);
 199 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 200
 201 enum {
 202         BLK_MQ_REQ_NOWAIT       = (1 << 0), /* return when out of requests */
 203         BLK_MQ_REQ_RESERVED     = (1 << 1), /* allocate from reserved pool */
 204         BLK_MQ_REQ_INTERNAL     = (1 << 2), /* allocate internal/sched tag */
 205 };
 206
 207 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 208                 unsigned int flags);
 209 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
 210                 unsigned int flags, unsigned int hctx_idx);
 211 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 212
 213 enum {
 214         BLK_MQ_UNIQUE_TAG_BITS = 16,
 215         BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
 216 };
 217
 218 u32 blk_mq_unique_tag(struct request *rq);
 219
 220 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
 221 {
 222         return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
 223 }
 224
 225 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
 226 {
 227         return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
 228 }
 229
 230
 231 int blk_mq_request_started(struct request *rq);
 232 void blk_mq_start_request(struct request *rq);
 233 void blk_mq_end_request(struct request *rq, int error);
 234 void __blk_mq_end_request(struct request *rq, int error);
 235
 236 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
 237 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 238                                 bool kick_requeue_list);
 239 void blk_mq_kick_requeue_list(struct request_queue *q);
 240 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
 241 void blk_mq_abort_requeue_list(struct request_queue *q);
 242 void blk_mq_complete_request(struct request *rq);
 243
 244 bool blk_mq_queue_stopped(struct request_queue *q);
 245 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
 246 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
 247 void blk_mq_stop_hw_queues(struct request_queue *q);
 248 void blk_mq_start_hw_queues(struct request_queue *q);
 249 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 250 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 251 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 252 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 253 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
 254 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 255 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 256                 busy_tag_iter_fn *fn, void *priv);
 257 void blk_mq_freeze_queue(struct request_queue *q);
 258 void blk_mq_unfreeze_queue(struct request_queue *q);
 259 void blk_freeze_queue_start(struct request_queue *q);
 260 void blk_mq_freeze_queue_wait(struct request_queue *q);
 261 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
 262                                      unsigned long timeout);
 263 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
 264
 265 int blk_mq_map_queues(struct blk_mq_tag_set *set);
 266 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 267
 268 /*
 269  * Driver command data is immediately after the request. So subtract request
 270  * size to get back to the original request, add request size to get the PDU.
 271  */
 272 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
 273 {
 274         return pdu - sizeof(struct request);
 275 }
 276 static inline void *blk_mq_rq_to_pdu(struct request *rq)
 277 {
 278         return rq + 1;
 279 }
 280
 281 #define queue_for_each_hw_ctx(q, hctx, i)                               \
 282         for ((i) = 0; (i) < (q)->nr_hw_queues &&                        \
 283              ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
 284
 285 #define hctx_for_each_ctx(hctx, ctx, i)                                 \
 286         for ((i) = 0; (i) < (hctx)->nr_ctx &&                           \
 287              ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
 288
 289 #endif