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         atomic_t                wait_index;
  37
  38         struct blk_mq_tags      *tags;
  39         struct blk_mq_tags      *sched_tags;
  40
  41         struct srcu_struct      queue_rq_srcu;
  42
  43         unsigned long           queued;
  44         unsigned long           run;
  45 #define BLK_MQ_MAX_DISPATCH_ORDER       7
  46         unsigned long           dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
  47
  48         unsigned int            numa_node;
  49         unsigned int            queue_num;
  50
  51         atomic_t                nr_active;
  52
  53         struct delayed_work     delay_work;
  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
  63 struct blk_mq_tag_set {
  64         unsigned int            *mq_map;
  65         const struct blk_mq_ops *ops;
  66         unsigned int            nr_hw_queues;
  67         unsigned int            queue_depth;    /* max hw supported */
  68         unsigned int            reserved_tags;
  69         unsigned int            cmd_size;       /* per-request extra data */
  70         int                     numa_node;
  71         unsigned int            timeout;
  72         unsigned int            flags;          /* BLK_MQ_F_* */
  73         void                    *driver_data;
  74
  75         struct blk_mq_tags      **tags;
  76
  77         struct mutex            tag_list_lock;
  78         struct list_head        tag_list;
  79 };
  80
  81 struct blk_mq_queue_data {
  82         struct request *rq;
  83         struct list_head *list;
  84         bool last;
  85 };
  86
  87 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
  88 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
  89 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
  90 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
  91 typedef int (init_request_fn)(void *, struct request *, unsigned int,
  92                 unsigned int, unsigned int);
  93 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
  94                 unsigned int);
  95 typedef int (reinit_request_fn)(void *, struct request *);
  96
  97 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
  98                 bool);
  99 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
 100 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
 101 typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
 102
 103
 104 struct blk_mq_ops {
 105         /*
 106          * Queue request
 107          */
 108         queue_rq_fn             *queue_rq;
 109
 110         /*
 111          * Called on request timeout
 112          */
 113         timeout_fn              *timeout;
 114
 115         /*
 116          * Called to poll for completion of a specific tag.
 117          */
 118         poll_fn                 *poll;
 119
 120         softirq_done_fn         *complete;
 121
 122         /*
 123          * Called when the block layer side of a hardware queue has been
 124          * set up, allowing the driver to allocate/init matching structures.
 125          * Ditto for exit/teardown.
 126          */
 127         init_hctx_fn            *init_hctx;
 128         exit_hctx_fn            *exit_hctx;
 129
 130         /*
 131          * Called for every command allocated by the block layer to allow
 132          * the driver to set up driver specific data.
 133          *
 134          * Tag greater than or equal to queue_depth is for setting up
 135          * flush request.
 136          *
 137          * Ditto for exit/teardown.
 138          */
 139         init_request_fn         *init_request;
 140         exit_request_fn         *exit_request;
 141         reinit_request_fn       *reinit_request;
 142
 143         map_queues_fn           *map_queues;
 144 };
 145
 146 enum {
 147         BLK_MQ_RQ_QUEUE_OK      = 0,    /* queued fine */
 148         BLK_MQ_RQ_QUEUE_BUSY    = 1,    /* requeue IO for later */
 149         BLK_MQ_RQ_QUEUE_ERROR   = 2,    /* end IO with error */
 150
 151         BLK_MQ_F_SHOULD_MERGE   = 1 << 0,
 152         BLK_MQ_F_TAG_SHARED     = 1 << 1,
 153         BLK_MQ_F_SG_MERGE       = 1 << 2,
 154         BLK_MQ_F_DEFER_ISSUE    = 1 << 4,
 155         BLK_MQ_F_BLOCKING       = 1 << 5,
 156         BLK_MQ_F_NO_SCHED       = 1 << 6,
 157         BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
 158         BLK_MQ_F_ALLOC_POLICY_BITS = 1,
 159
 160         BLK_MQ_S_STOPPED        = 0,
 161         BLK_MQ_S_TAG_ACTIVE     = 1,
 162         BLK_MQ_S_SCHED_RESTART  = 2,
 163
 164         BLK_MQ_MAX_DEPTH        = 10240,
 165
 166         BLK_MQ_CPU_WORK_BATCH   = 8,
 167 };
 168 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
 169         ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
 170                 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
 171 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
 172         ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
 173                 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
 174
 175 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
 176 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
 177                                                   struct request_queue *q);
 178 int blk_mq_register_dev(struct device *, struct request_queue *);
 179 void blk_mq_unregister_dev(struct device *, struct request_queue *);
 180
 181 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
 182 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
 183
 184 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
 185
 186 void blk_mq_free_request(struct request *rq);
 187 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
 188
 189 enum {
 190         BLK_MQ_REQ_NOWAIT       = (1 << 0), /* return when out of requests */
 191         BLK_MQ_REQ_RESERVED     = (1 << 1), /* allocate from reserved pool */
 192         BLK_MQ_REQ_INTERNAL     = (1 << 2), /* allocate internal/sched tag */
 193 };
 194
 195 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 196                 unsigned int flags);
 197 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
 198                 unsigned int flags, unsigned int hctx_idx);
 199 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 200
 201 enum {
 202         BLK_MQ_UNIQUE_TAG_BITS = 16,
 203         BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
 204 };
 205
 206 u32 blk_mq_unique_tag(struct request *rq);
 207
 208 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
 209 {
 210         return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
 211 }
 212
 213 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
 214 {
 215         return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
 216 }
 217
 218
 219 int blk_mq_request_started(struct request *rq);
 220 void blk_mq_start_request(struct request *rq);
 221 void blk_mq_end_request(struct request *rq, int error);
 222 void __blk_mq_end_request(struct request *rq, int error);
 223
 224 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
 225 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
 226                                 bool kick_requeue_list);
 227 void blk_mq_kick_requeue_list(struct request_queue *q);
 228 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
 229 void blk_mq_abort_requeue_list(struct request_queue *q);
 230 void blk_mq_complete_request(struct request *rq, int error);
 231
 232 bool blk_mq_queue_stopped(struct request_queue *q);
 233 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
 234 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
 235 void blk_mq_stop_hw_queues(struct request_queue *q);
 236 void blk_mq_start_hw_queues(struct request_queue *q);
 237 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
 238 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 239 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
 240 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
 241 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
 242                 busy_tag_iter_fn *fn, void *priv);
 243 void blk_mq_freeze_queue(struct request_queue *q);
 244 void blk_mq_unfreeze_queue(struct request_queue *q);
 245 void blk_mq_freeze_queue_start(struct request_queue *q);
 246 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
 247
 248 int blk_mq_map_queues(struct blk_mq_tag_set *set);
 249 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
 250
 251 /*
 252  * Driver command data is immediately after the request. So subtract request
 253  * size to get back to the original request, add request size to get the PDU.
 254  */
 255 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
 256 {
 257         return pdu - sizeof(struct request);
 258 }
 259 static inline void *blk_mq_rq_to_pdu(struct request *rq)
 260 {
 261         return rq + 1;
 262 }
 263
 264 #define queue_for_each_hw_ctx(q, hctx, i)                               \
 265         for ((i) = 0; (i) < (q)->nr_hw_queues &&                        \
 266              ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
 267
 268 #define hctx_for_each_ctx(hctx, ctx, i)                                 \
 269         for ((i) = 0; (i) < (hctx)->nr_ctx &&                           \
 270              ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
 271
 272 #endif