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 work_struct      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 delayed_work     delay_work;
  55
  56         struct hlist_node       cpuhp_dead;
  57         struct kobject          kobj;
  58
  59         unsigned long           poll_considered;
  60         unsigned long           poll_invoked;
  61         unsigned long           poll_success;
  62 };
  63
  64 struct blk_mq_tag_set {
  65         unsigned int            *mq_map;
  66         const 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 struct blk_mq_queue_data {
  83         struct request *rq;
  84         struct list_head *list;
  85         bool last;
  86 };
  87
  88 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
  89 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
  90 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
  91 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
  92 typedef int (init_request_fn)(void *, struct request *, unsigned int,
  93                 unsigned int, unsigned int);
  94 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
  95                 unsigned int);
  96 typedef int (reinit_request_fn)(void *, struct request *);
  97
  98 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
  99                 bool);
 100 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
 101 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
 102 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
 103
 104
 105 struct blk_mq_ops {
 106         /*
 107          * Queue request
 108          */
 109         queue_rq_fn             *queue_rq;
 110
 111         /*
 112          * Called on request timeout
 113          */
 114         timeout_fn              *timeout;
 115
 116         /*
 117          * Called to poll for completion of a specific tag.
 118          */
 119         poll_fn                 *poll;
 120
 121         softirq_done_fn         *complete;
 122
 123         /*
 124          * Called when the block layer side of a hardware queue has been
 125          * set up, allowing the driver to allocate/init matching structures.
 126          * Ditto for exit/teardown.
 127          */
 128         init_hctx_fn            *init_hctx;
 129         exit_hctx_fn            *exit_hctx;
 130
 131         /*
 132          * Called for every command allocated by the block layer to allow
 133          * the driver to set up driver specific data.
 134          *
 135          * Tag greater than or equal to queue_depth is for setting up
 136          * flush request.
 137          *
 138          * Ditto for exit/teardown.
 139          */
 140         init_request_fn         *init_request;
 141         exit_request_fn         *exit_request;
 142         reinit_request_fn       *reinit_request;
 143
 144         map_queues_fn           *map_queues;
 145 };
 146
 147 enum {
 148         BLK_MQ_RQ_QUEUE_OK      = 0,    /* queued fine */
 149         BLK_MQ_RQ_QUEUE_BUSY    = 1,    /* requeue IO for later */
 150         BLK_MQ_RQ_QUEUE_ERROR   = 2,    /* end IO with error */
 151
 152         BLK_MQ_F_SHOULD_MERGE   = 1 << 0,
 153         BLK_MQ_F_TAG_SHARED     = 1 << 1,
 154         BLK_MQ_F_SG_MERGE       = 1 << 2,
 155         BLK_MQ_F_DEFER_ISSUE    = 1 << 4,
 156         BLK_MQ_F_BLOCKING       = 1 << 5,
 157         BLK_MQ_F_NO_SCHED       = 1 << 6,
 158         BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
 159         BLK_MQ_F_ALLOC_POLICY_BITS = 1,
 160
 161         BLK_MQ_S_STOPPED        = 0,
 162         BLK_MQ_S_TAG_ACTIVE     = 1,
 163         BLK_MQ_S_SCHED_RESTART  = 2,
 164         BLK_MQ_S_TAG_WAITING    = 3,
 165
 166         BLK_MQ_MAX_DEPTH        = 10240,
 167
 168         BLK_MQ_CPU_WORK_BATCH   = 8,
 169 };
 170 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
 171         ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
 172                 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
 173 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
 174         ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
 175                 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
 176
 177 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
 178 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 179                                                   struct request_queue *q);
 180 int blk_mq_register_dev(struct device *, struct request_queue *);
 181 void blk_mq_unregister_dev(struct device *, struct request_queue *);
 182
 183 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
 184 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
 185
 186 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
 187
 188 void blk_mq_free_request(struct request *rq);
 189 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 190
 191 enum {
 192         BLK_MQ_REQ_NOWAIT       = (1 << 0), /* return when out of requests */
 193         BLK_MQ_REQ_RESERVED     = (1 << 1), /* allocate from reserved pool */
 194         BLK_MQ_REQ_INTERNAL     = (1 << 2), /* allocate internal/sched tag */
 195 };
 196
 197 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 198                 unsigned int flags);
 199 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
 200                 unsigned int flags, unsigned int hctx_idx);
 201 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 202
 203 enum {
 204         BLK_MQ_UNIQUE_TAG_BITS = 16,
 205         BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
 206 };
 207
 208 u32 blk_mq_unique_tag(struct request *rq);
 209
 210 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
 211 {
 212         return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
 213 }
 214
 215 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
 216 {
 217         return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
 218 }
 219
 220
 221 int blk_mq_request_started(struct request *rq);
 222 void blk_mq_start_request(struct request *rq);
 223 void blk_mq_end_request(struct request *rq, int error);
 224 void __blk_mq_end_request(struct request *rq, int error);
 225
 226 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
 227 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 228                                 bool kick_requeue_list);
 229 void blk_mq_kick_requeue_list(struct request_queue *q);
 230 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
 231 void blk_mq_abort_requeue_list(struct request_queue *q);
 232 void blk_mq_complete_request(struct request *rq, int error);
 233
 234 bool blk_mq_queue_stopped(struct request_queue *q);
 235 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
 236 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
 237 void blk_mq_stop_hw_queues(struct request_queue *q);
 238 void blk_mq_start_hw_queues(struct request_queue *q);
 239 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 240 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 241 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
 242 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 243 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 244                 busy_tag_iter_fn *fn, void *priv);
 245 void blk_mq_freeze_queue(struct request_queue *q);
 246 void blk_mq_unfreeze_queue(struct request_queue *q);
 247 void blk_mq_freeze_queue_start(struct request_queue *q);
 248 void blk_mq_freeze_queue_wait(struct request_queue *q);
 249 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
 250                                      unsigned long timeout);
 251 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
 252
 253 int blk_mq_map_queues(struct blk_mq_tag_set *set);
 254 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 255
 256 /*
 257  * Driver command data is immediately after the request. So subtract request
 258  * size to get back to the original request, add request size to get the PDU.
 259  */
 260 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
 261 {
 262         return pdu - sizeof(struct request);
 263 }
 264 static inline void *blk_mq_rq_to_pdu(struct request *rq)
 265 {
 266         return rq + 1;
 267 }
 268
 269 #define queue_for_each_hw_ctx(q, hctx, i)                               \
 270         for ((i) = 0; (i) < (q)->nr_hw_queues &&                        \
 271              ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
 272
 273 #define hctx_for_each_ctx(hctx, ctx, i)                                 \
 274         for ((i) = 0; (i) < (hctx)->nr_ctx &&                           \
 275              ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
 276
 277 #endif