include/linux/blk-mq.h

   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                 unsigned long           state;          /* BLK_MQ_S_* flags */
  26         } ____cacheline_aligned_in_smp;
  27
  28         struct work_struct      run_work;
  29         cpumask_var_t           cpumask;
  30         int                     next_cpu;
  31         int                     next_cpu_batch;
  32
  33         unsigned long           flags;          /* BLK_MQ_F_* flags */
  34
  35         struct request_queue    *queue;
  36         struct blk_flush_queue  *fq;
  37
  38         void                    *driver_data;
  39
  40         struct blk_mq_ctxmap    ctx_map;
  41
  42         struct blk_mq_ctx       **ctxs;
  43         unsigned int            nr_ctx;
  44
  45         atomic_t                wait_index;
  46
  47         struct blk_mq_tags      *tags;
  48
  49         unsigned long           queued;
  50         unsigned long           run;
  51 #define BLK_MQ_MAX_DISPATCH_ORDER       7
  52         unsigned long           dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
  53
  54         unsigned int            numa_node;
  55         unsigned int            queue_num;
  56
  57         atomic_t                nr_active;
  58
  59         struct delayed_work     delay_work;
  60
  61         struct blk_mq_cpu_notifier      cpu_notifier;
  62         struct kobject          kobj;
  63
  64         unsigned long           poll_invoked;
  65         unsigned long           poll_success;
  66 };
  67
  68 struct blk_mq_tag_set {
  69         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         struct list_head *list;
  88         bool last;
  89 };
  90
  91 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
  92 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
  93 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
  94 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
  95 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
  96 typedef int (init_request_fn)(void *, struct request *, unsigned int,
  97                 unsigned int, unsigned int);
  98 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
  99                 unsigned int);
 100 typedef int (reinit_request_fn)(void *, struct request *);
 101
 102 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
 103                 bool);
 104 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
 105 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
 106
 107
 108 struct blk_mq_ops {
 109         /*
 110          * Queue request
 111          */
 112         queue_rq_fn             *queue_rq;
 113
 114         /*
 115          * Map to specific hardware queue
 116          */
 117         map_queue_fn            *map_queue;
 118
 119         /*
 120          * Called on request timeout
 121          */
 122         timeout_fn              *timeout;
 123
 124         /*
 125          * Called to poll for completion of a specific tag.
 126          */
 127         poll_fn                 *poll;
 128
 129         softirq_done_fn         *complete;
 130
 131         /*
 132          * Called when the block layer side of a hardware queue has been
 133          * set up, allowing the driver to allocate/init matching structures.
 134          * Ditto for exit/teardown.
 135          */
 136         init_hctx_fn            *init_hctx;
 137         exit_hctx_fn            *exit_hctx;
 138
 139         /*
 140          * Called for every command allocated by the block layer to allow
 141          * the driver to set up driver specific data.
 142          *
 143          * Tag greater than or equal to queue_depth is for setting up
 144          * flush request.
 145          *
 146          * Ditto for exit/teardown.
 147          */
 148         init_request_fn         *init_request;
 149         exit_request_fn         *exit_request;
 150         reinit_request_fn       *reinit_request;
 151 };
 152
 153 enum {
 154         BLK_MQ_RQ_QUEUE_OK      = 0,    /* queued fine */
 155         BLK_MQ_RQ_QUEUE_BUSY    = 1,    /* requeue IO for later */
 156         BLK_MQ_RQ_QUEUE_ERROR   = 2,    /* end IO with error */
 157
 158         BLK_MQ_F_SHOULD_MERGE   = 1 << 0,
 159         BLK_MQ_F_TAG_SHARED     = 1 << 1,
 160         BLK_MQ_F_SG_MERGE       = 1 << 2,
 161         BLK_MQ_F_DEFER_ISSUE    = 1 << 4,
 162         BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
 163         BLK_MQ_F_ALLOC_POLICY_BITS = 1,
 164
 165         BLK_MQ_S_STOPPED        = 0,
 166         BLK_MQ_S_TAG_ACTIVE     = 1,
 167
 168         BLK_MQ_MAX_DEPTH        = 10240,
 169
 170         BLK_MQ_CPU_WORK_BATCH   = 8,
 171 };
 172 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
 173         ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
 174                 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
 175 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
 176         ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
 177                 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
 178
 179 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
 180 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 181                                                   struct request_queue *q);
 182 int blk_mq_register_disk(struct gendisk *);
 183 void blk_mq_unregister_disk(struct gendisk *);
 184
 185 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
 186 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
 187
 188 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
 189
 190 void blk_mq_insert_request(struct request *, bool, bool, bool);
 191 void blk_mq_free_request(struct request *rq);
 192 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
 193 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 194
 195 enum {
 196         BLK_MQ_REQ_NOWAIT       = (1 << 0), /* return when out of requests */
 197         BLK_MQ_REQ_RESERVED     = (1 << 1), /* allocate from reserved pool */
 198 };
 199
 200 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 201                 unsigned int flags);
 202 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
 203                 unsigned int flags, unsigned int hctx_idx);
 204 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 205 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
 206
 207 enum {
 208         BLK_MQ_UNIQUE_TAG_BITS = 16,
 209         BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
 210 };
 211
 212 u32 blk_mq_unique_tag(struct request *rq);
 213
 214 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
 215 {
 216         return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
 217 }
 218
 219 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
 220 {
 221         return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
 222 }
 223
 224 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
 225 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
 226
 227 int blk_mq_request_started(struct request *rq);
 228 void blk_mq_start_request(struct request *rq);
 229 void blk_mq_end_request(struct request *rq, int error);
 230 void __blk_mq_end_request(struct request *rq, int error);
 231
 232 void blk_mq_requeue_request(struct request *rq);
 233 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
 234 void blk_mq_cancel_requeue_work(struct request_queue *q);
 235 void blk_mq_kick_requeue_list(struct request_queue *q);
 236 void blk_mq_abort_requeue_list(struct request_queue *q);
 237 void blk_mq_complete_request(struct request *rq, int error);
 238
 239 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
 240 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
 241 void blk_mq_stop_hw_queues(struct request_queue *q);
 242 void blk_mq_start_hw_queues(struct request_queue *q);
 243 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 244 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
 245 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 246 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 247                 busy_tag_iter_fn *fn, void *priv);
 248 void blk_mq_freeze_queue(struct request_queue *q);
 249 void blk_mq_unfreeze_queue(struct request_queue *q);
 250 void blk_mq_freeze_queue_start(struct request_queue *q);
 251 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
 252
 253 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 254
 255 /*
 256  * Driver command data is immediately after the request. So subtract request
 257  * size to get back to the original request, add request size to get the PDU.
 258  */
 259 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
 260 {
 261         return pdu - sizeof(struct request);
 262 }
 263 static inline void *blk_mq_rq_to_pdu(struct request *rq)
 264 {
 265         return rq + 1;
 266 }
 267
 268 #define queue_for_each_hw_ctx(q, hctx, i)                               \
 269         for ((i) = 0; (i) < (q)->nr_hw_queues &&                        \
 270              ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
 271
 272 #define hctx_for_each_ctx(hctx, ctx, i)                                 \
 273         for ((i) = 0; (i) < (hctx)->nr_ctx &&                           \
 274              ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
 275
 276 #endif