drivers/nvme/host/nvme.h

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * Copyright (c) 2011-2014, Intel Corporation.
   4  */
   5
   6 #ifndef _NVME_H
   7 #define _NVME_H
   8
   9 #include <linux/nvme.h>
  10 #include <linux/cdev.h>
  11 #include <linux/pci.h>
  12 #include <linux/kref.h>
  13 #include <linux/blk-mq.h>
  14 #include <linux/sed-opal.h>
  15 #include <linux/fault-inject.h>
  16 #include <linux/rcupdate.h>
  17 #include <linux/wait.h>
  18 #include <linux/t10-pi.h>
  19
  20 #include <trace/events/block.h>
  21
  22 extern unsigned int nvme_io_timeout;
  23 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
  24
  25 extern unsigned int admin_timeout;
  26 #define NVME_ADMIN_TIMEOUT      (admin_timeout * HZ)
  27
  28 #define NVME_DEFAULT_KATO       5
  29
  30 #ifdef CONFIG_ARCH_NO_SG_CHAIN
  31 #define  NVME_INLINE_SG_CNT  0
  32 #define  NVME_INLINE_METADATA_SG_CNT  0
  33 #else
  34 #define  NVME_INLINE_SG_CNT  2
  35 #define  NVME_INLINE_METADATA_SG_CNT  1
  36 #endif
  37
  38 /*
  39  * Default to a 4K page size, with the intention to update this
  40  * path in the future to accommodate architectures with differing
  41  * kernel and IO page sizes.
  42  */
  43 #define NVME_CTRL_PAGE_SHIFT    12
  44 #define NVME_CTRL_PAGE_SIZE     (1 << NVME_CTRL_PAGE_SHIFT)
  45
  46 extern struct workqueue_struct *nvme_wq;
  47 extern struct workqueue_struct *nvme_reset_wq;
  48 extern struct workqueue_struct *nvme_delete_wq;
  49
  50 /*
  51  * List of workarounds for devices that required behavior not specified in
  52  * the standard.
  53  */
  54 enum nvme_quirks {
  55         /*
  56          * Prefers I/O aligned to a stripe size specified in a vendor
  57          * specific Identify field.
  58          */
  59         NVME_QUIRK_STRIPE_SIZE                  = (1 << 0),
  60
  61         /*
  62          * The controller doesn't handle Identify value others than 0 or 1
  63          * correctly.
  64          */
  65         NVME_QUIRK_IDENTIFY_CNS                 = (1 << 1),
  66
  67         /*
  68          * The controller deterministically returns O's on reads to
  69          * logical blocks that deallocate was called on.
  70          */
  71         NVME_QUIRK_DEALLOCATE_ZEROES            = (1 << 2),
  72
  73         /*
  74          * The controller needs a delay before starts checking the device
  75          * readiness, which is done by reading the NVME_CSTS_RDY bit.
  76          */
  77         NVME_QUIRK_DELAY_BEFORE_CHK_RDY         = (1 << 3),
  78
  79         /*
  80          * APST should not be used.
  81          */
  82         NVME_QUIRK_NO_APST                      = (1 << 4),
  83
  84         /*
  85          * The deepest sleep state should not be used.
  86          */
  87         NVME_QUIRK_NO_DEEPEST_PS                = (1 << 5),
  88
  89         /*
  90          * Set MEDIUM priority on SQ creation
  91          */
  92         NVME_QUIRK_MEDIUM_PRIO_SQ               = (1 << 7),
  93
  94         /*
  95          * Ignore device provided subnqn.
  96          */
  97         NVME_QUIRK_IGNORE_DEV_SUBNQN            = (1 << 8),
  98
  99         /*
 100          * Broken Write Zeroes.
 101          */
 102         NVME_QUIRK_DISABLE_WRITE_ZEROES         = (1 << 9),
 103
 104         /*
 105          * Force simple suspend/resume path.
 106          */
 107         NVME_QUIRK_SIMPLE_SUSPEND               = (1 << 10),
 108
 109         /*
 110          * Use only one interrupt vector for all queues
 111          */
 112         NVME_QUIRK_SINGLE_VECTOR                = (1 << 11),
 113
 114         /*
 115          * Use non-standard 128 bytes SQEs.
 116          */
 117         NVME_QUIRK_128_BYTES_SQES               = (1 << 12),
 118
 119         /*
 120          * Prevent tag overlap between queues
 121          */
 122         NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
 123
 124         /*
 125          * Don't change the value of the temperature threshold feature
 126          */
 127         NVME_QUIRK_NO_TEMP_THRESH_CHANGE        = (1 << 14),
 128
 129         /*
 130          * The controller doesn't handle the Identify Namespace
 131          * Identification Descriptor list subcommand despite claiming
 132          * NVMe 1.3 compliance.
 133          */
 134         NVME_QUIRK_NO_NS_DESC_LIST              = (1 << 15),
 135
 136         /*
 137          * The controller does not properly handle DMA addresses over
 138          * 48 bits.
 139          */
 140         NVME_QUIRK_DMA_ADDRESS_BITS_48          = (1 << 16),
 141
 142         /*
 143          * The controller requires the command_id value be be limited, so skip
 144          * encoding the generation sequence number.
 145          */
 146         NVME_QUIRK_SKIP_CID_GEN                 = (1 << 17),
 147 };
 148
 149 /*
 150  * Common request structure for NVMe passthrough.  All drivers must have
 151  * this structure as the first member of their request-private data.
 152  */
 153 struct nvme_request {
 154         struct nvme_command     *cmd;
 155         union nvme_result       result;
 156         u8                      genctr;
 157         u8                      retries;
 158         u8                      flags;
 159         u16                     status;
 160         struct nvme_ctrl        *ctrl;
 161 };
 162
 163 /*
 164  * Mark a bio as coming in through the mpath node.
 165  */
 166 #define REQ_NVME_MPATH          REQ_DRV
 167
 168 enum {
 169         NVME_REQ_CANCELLED              = (1 << 0),
 170         NVME_REQ_USERCMD                = (1 << 1),
 171 };
 172
 173 static inline struct nvme_request *nvme_req(struct request *req)
 174 {
 175         return blk_mq_rq_to_pdu(req);
 176 }
 177
 178 static inline u16 nvme_req_qid(struct request *req)
 179 {
 180         if (!req->q->queuedata)
 181                 return 0;
 182
 183         return req->mq_hctx->queue_num + 1;
 184 }
 185
 186 /* The below value is the specific amount of delay needed before checking
 187  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
 188  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
 189  * found empirically.
 190  */
 191 #define NVME_QUIRK_DELAY_AMOUNT         2300
 192
 193 /*
 194  * enum nvme_ctrl_state: Controller state
 195  *
 196  * @NVME_CTRL_NEW:              New controller just allocated, initial state
 197  * @NVME_CTRL_LIVE:             Controller is connected and I/O capable
 198  * @NVME_CTRL_RESETTING:        Controller is resetting (or scheduled reset)
 199  * @NVME_CTRL_CONNECTING:       Controller is disconnected, now connecting the
 200  *                              transport
 201  * @NVME_CTRL_DELETING:         Controller is deleting (or scheduled deletion)
 202  * @NVME_CTRL_DELETING_NOIO:    Controller is deleting and I/O is not
 203  *                              disabled/failed immediately. This state comes
 204  *                              after all async event processing took place and
 205  *                              before ns removal and the controller deletion
 206  *                              progress
 207  * @NVME_CTRL_DEAD:             Controller is non-present/unresponsive during
 208  *                              shutdown or removal. In this case we forcibly
 209  *                              kill all inflight I/O as they have no chance to
 210  *                              complete
 211  */
 212 enum nvme_ctrl_state {
 213         NVME_CTRL_NEW,
 214         NVME_CTRL_LIVE,
 215         NVME_CTRL_RESETTING,
 216         NVME_CTRL_CONNECTING,
 217         NVME_CTRL_DELETING,
 218         NVME_CTRL_DELETING_NOIO,
 219         NVME_CTRL_DEAD,
 220 };
 221
 222 struct nvme_fault_inject {
 223 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
 224         struct fault_attr attr;
 225         struct dentry *parent;
 226         bool dont_retry;        /* DNR, do not retry */
 227         u16 status;             /* status code */
 228 #endif
 229 };
 230
 231 struct nvme_ctrl {
 232         bool comp_seen;
 233         enum nvme_ctrl_state state;
 234         bool identified;
 235         spinlock_t lock;
 236         struct mutex scan_lock;
 237         const struct nvme_ctrl_ops *ops;
 238         struct request_queue *admin_q;
 239         struct request_queue *connect_q;
 240         struct request_queue *fabrics_q;
 241         struct device *dev;
 242         int instance;
 243         int numa_node;
 244         struct blk_mq_tag_set *tagset;
 245         struct blk_mq_tag_set *admin_tagset;
 246         struct list_head namespaces;
 247         struct rw_semaphore namespaces_rwsem;
 248         struct device ctrl_device;
 249         struct device *device;  /* char device */
 250 #ifdef CONFIG_NVME_HWMON
 251         struct device *hwmon_device;
 252 #endif
 253         struct cdev cdev;
 254         struct work_struct reset_work;
 255         struct work_struct delete_work;
 256         wait_queue_head_t state_wq;
 257
 258         struct nvme_subsystem *subsys;
 259         struct list_head subsys_entry;
 260
 261         struct opal_dev *opal_dev;
 262
 263         char name[12];
 264         u16 cntlid;
 265
 266         u32 ctrl_config;
 267         u16 mtfa;
 268         u32 queue_count;
 269
 270         u64 cap;
 271         u32 max_hw_sectors;
 272         u32 max_segments;
 273         u32 max_integrity_segments;
 274         u32 max_discard_sectors;
 275         u32 max_discard_segments;
 276         u32 max_zeroes_sectors;
 277 #ifdef CONFIG_BLK_DEV_ZONED
 278         u32 max_zone_append;
 279 #endif
 280         u16 crdt[3];
 281         u16 oncs;
 282         u16 oacs;
 283         u16 nssa;
 284         u16 nr_streams;
 285         u16 sqsize;
 286         u32 max_namespaces;
 287         atomic_t abort_limit;
 288         u8 vwc;
 289         u32 vs;
 290         u32 sgls;
 291         u16 kas;
 292         u8 npss;
 293         u8 apsta;
 294         u16 wctemp;
 295         u16 cctemp;
 296         u32 oaes;
 297         u32 aen_result;
 298         u32 ctratt;
 299         unsigned int shutdown_timeout;
 300         unsigned int kato;
 301         bool subsystem;
 302         unsigned long quirks;
 303         struct nvme_id_power_state psd[32];
 304         struct nvme_effects_log *effects;
 305         struct xarray cels;
 306         struct work_struct scan_work;
 307         struct work_struct async_event_work;
 308         struct delayed_work ka_work;
 309         struct delayed_work failfast_work;
 310         struct nvme_command ka_cmd;
 311         struct work_struct fw_act_work;
 312         unsigned long events;
 313
 314 #ifdef CONFIG_NVME_MULTIPATH
 315         /* asymmetric namespace access: */
 316         u8 anacap;
 317         u8 anatt;
 318         u32 anagrpmax;
 319         u32 nanagrpid;
 320         struct mutex ana_lock;
 321         struct nvme_ana_rsp_hdr *ana_log_buf;
 322         size_t ana_log_size;
 323         struct timer_list anatt_timer;
 324         struct work_struct ana_work;
 325 #endif
 326
 327         /* Power saving configuration */
 328         u64 ps_max_latency_us;
 329         bool apst_enabled;
 330
 331         /* PCIe only: */
 332         u32 hmpre;
 333         u32 hmmin;
 334         u32 hmminds;
 335         u16 hmmaxd;
 336
 337         /* Fabrics only */
 338         u32 ioccsz;
 339         u32 iorcsz;
 340         u16 icdoff;
 341         u16 maxcmd;
 342         int nr_reconnects;
 343         unsigned long flags;
 344 #define NVME_CTRL_FAILFAST_EXPIRED      0
 345         struct nvmf_ctrl_options *opts;
 346
 347         struct page *discard_page;
 348         unsigned long discard_page_busy;
 349
 350         struct nvme_fault_inject fault_inject;
 351 };
 352
 353 enum nvme_iopolicy {
 354         NVME_IOPOLICY_NUMA,
 355         NVME_IOPOLICY_RR,
 356 };
 357
 358 struct nvme_subsystem {
 359         int                     instance;
 360         struct device           dev;
 361         /*
 362          * Because we unregister the device on the last put we need
 363          * a separate refcount.
 364          */
 365         struct kref             ref;
 366         struct list_head        entry;
 367         struct mutex            lock;
 368         struct list_head        ctrls;
 369         struct list_head        nsheads;
 370         char                    subnqn[NVMF_NQN_SIZE];
 371         char                    serial[20];
 372         char                    model[40];
 373         char                    firmware_rev[8];
 374         u8                      cmic;
 375         u16                     vendor_id;
 376         u16                     awupf;  /* 0's based awupf value. */
 377         struct ida              ns_ida;
 378 #ifdef CONFIG_NVME_MULTIPATH
 379         enum nvme_iopolicy      iopolicy;
 380 #endif
 381 };
 382
 383 /*
 384  * Container structure for uniqueue namespace identifiers.
 385  */
 386 struct nvme_ns_ids {
 387         u8      eui64[8];
 388         u8      nguid[16];
 389         uuid_t  uuid;
 390         u8      csi;
 391 };
 392
 393 /*
 394  * Anchor structure for namespaces.  There is one for each namespace in a
 395  * NVMe subsystem that any of our controllers can see, and the namespace
 396  * structure for each controller is chained of it.  For private namespaces
 397  * there is a 1:1 relation to our namespace structures, that is ->list
 398  * only ever has a single entry for private namespaces.
 399  */
 400 struct nvme_ns_head {
 401         struct list_head        list;
 402         struct srcu_struct      srcu;
 403         struct nvme_subsystem   *subsys;
 404         unsigned                ns_id;
 405         struct nvme_ns_ids      ids;
 406         struct list_head        entry;
 407         struct kref             ref;
 408         bool                    shared;
 409         int                     instance;
 410         struct nvme_effects_log *effects;
 411
 412         struct cdev             cdev;
 413         struct device           cdev_device;
 414
 415         struct gendisk          *disk;
 416 #ifdef CONFIG_NVME_MULTIPATH
 417         struct bio_list         requeue_list;
 418         spinlock_t              requeue_lock;
 419         struct work_struct      requeue_work;
 420         struct mutex            lock;
 421         unsigned long           flags;
 422 #define NVME_NSHEAD_DISK_LIVE   0
 423         struct nvme_ns __rcu    *current_path[];
 424 #endif
 425 };
 426
 427 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
 428 {
 429         return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
 430 }
 431
 432 enum nvme_ns_features {
 433         NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
 434         NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
 435 };
 436
 437 struct nvme_ns {
 438         struct list_head list;
 439
 440         struct nvme_ctrl *ctrl;
 441         struct request_queue *queue;
 442         struct gendisk *disk;
 443 #ifdef CONFIG_NVME_MULTIPATH
 444         enum nvme_ana_state ana_state;
 445         u32 ana_grpid;
 446 #endif
 447         struct list_head siblings;
 448         struct kref kref;
 449         struct nvme_ns_head *head;
 450
 451         int lba_shift;
 452         u16 ms;
 453         u16 sgs;
 454         u32 sws;
 455         u8 pi_type;
 456 #ifdef CONFIG_BLK_DEV_ZONED
 457         u64 zsze;
 458 #endif
 459         unsigned long features;
 460         unsigned long flags;
 461 #define NVME_NS_REMOVING        0
 462 #define NVME_NS_DEAD            1
 463 #define NVME_NS_ANA_PENDING     2
 464 #define NVME_NS_FORCE_RO        3
 465 #define NVME_NS_READY           4
 466
 467         struct cdev             cdev;
 468         struct device           cdev_device;
 469
 470         struct nvme_fault_inject fault_inject;
 471
 472 };
 473
 474 /* NVMe ns supports metadata actions by the controller (generate/strip) */
 475 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
 476 {
 477         return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
 478 }
 479
 480 struct nvme_ctrl_ops {
 481         const char *name;
 482         struct module *module;
 483         unsigned int flags;
 484 #define NVME_F_FABRICS                  (1 << 0)
 485 #define NVME_F_METADATA_SUPPORTED       (1 << 1)
 486 #define NVME_F_PCI_P2PDMA               (1 << 2)
 487         int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
 488         int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
 489         int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
 490         void (*free_ctrl)(struct nvme_ctrl *ctrl);
 491         void (*submit_async_event)(struct nvme_ctrl *ctrl);
 492         void (*delete_ctrl)(struct nvme_ctrl *ctrl);
 493         int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
 494 };
 495
 496 /*
 497  * nvme command_id is constructed as such:
 498  * | xxxx | xxxxxxxxxxxx |
 499  *   gen    request tag
 500  */
 501 #define nvme_genctr_mask(gen)                   (gen & 0xf)
 502 #define nvme_cid_install_genctr(gen)            (nvme_genctr_mask(gen) << 12)
 503 #define nvme_genctr_from_cid(cid)               ((cid & 0xf000) >> 12)
 504 #define nvme_tag_from_cid(cid)                  (cid & 0xfff)
 505
 506 static inline u16 nvme_cid(struct request *rq)
 507 {
 508         return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
 509 }
 510
 511 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
 512                 u16 command_id)
 513 {
 514         u8 genctr = nvme_genctr_from_cid(command_id);
 515         u16 tag = nvme_tag_from_cid(command_id);
 516         struct request *rq;
 517
 518         rq = blk_mq_tag_to_rq(tags, tag);
 519         if (unlikely(!rq)) {
 520                 pr_err("could not locate request for tag %#x\n",
 521                         tag);
 522                 return NULL;
 523         }
 524         if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
 525                 dev_err(nvme_req(rq)->ctrl->device,
 526                         "request %#x genctr mismatch (got %#x expected %#x)\n",
 527                         tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
 528                 return NULL;
 529         }
 530         return rq;
 531 }
 532
 533 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
 534                 u16 command_id)
 535 {
 536         return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
 537 }
 538
 539 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
 540 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
 541                             const char *dev_name);
 542 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
 543 void nvme_should_fail(struct request *req);
 544 #else
 545 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
 546                                           const char *dev_name)
 547 {
 548 }
 549 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
 550 {
 551 }
 552 static inline void nvme_should_fail(struct request *req) {}
 553 #endif
 554
 555 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
 556 {
 557         if (!ctrl->subsystem)
 558                 return -ENOTTY;
 559         return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
 560 }
 561
 562 /*
 563  * Convert a 512B sector number to a device logical block number.
 564  */
 565 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
 566 {
 567         return sector >> (ns->lba_shift - SECTOR_SHIFT);
 568 }
 569
 570 /*
 571  * Convert a device logical block number to a 512B sector number.
 572  */
 573 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
 574 {
 575         return lba << (ns->lba_shift - SECTOR_SHIFT);
 576 }
 577
 578 /*
 579  * Convert byte length to nvme's 0-based num dwords
 580  */
 581 static inline u32 nvme_bytes_to_numd(size_t len)
 582 {
 583         return (len >> 2) - 1;
 584 }
 585
 586 static inline bool nvme_is_ana_error(u16 status)
 587 {
 588         switch (status & 0x7ff) {
 589         case NVME_SC_ANA_TRANSITION:
 590         case NVME_SC_ANA_INACCESSIBLE:
 591         case NVME_SC_ANA_PERSISTENT_LOSS:
 592                 return true;
 593         default:
 594                 return false;
 595         }
 596 }
 597
 598 static inline bool nvme_is_path_error(u16 status)
 599 {
 600         /* check for a status code type of 'path related status' */
 601         return (status & 0x700) == 0x300;
 602 }
 603
 604 /*
 605  * Fill in the status and result information from the CQE, and then figure out
 606  * if blk-mq will need to use IPI magic to complete the request, and if yes do
 607  * so.  If not let the caller complete the request without an indirect function
 608  * call.
 609  */
 610 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
 611                 union nvme_result result)
 612 {
 613         struct nvme_request *rq = nvme_req(req);
 614
 615         rq->status = le16_to_cpu(status) >> 1;
 616         rq->result = result;
 617         /* inject error when permitted by fault injection framework */
 618         nvme_should_fail(req);
 619         if (unlikely(blk_should_fake_timeout(req->q)))
 620                 return true;
 621         return blk_mq_complete_request_remote(req);
 622 }
 623
 624 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
 625 {
 626         get_device(ctrl->device);
 627 }
 628
 629 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
 630 {
 631         put_device(ctrl->device);
 632 }
 633
 634 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
 635 {
 636         return !qid &&
 637                 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
 638 }
 639
 640 void nvme_complete_rq(struct request *req);
 641 blk_status_t nvme_host_path_error(struct request *req);
 642 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
 643 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
 644 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
 645 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
 646                 enum nvme_ctrl_state new_state);
 647 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
 648 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
 649 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
 650 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
 651 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
 652                 const struct nvme_ctrl_ops *ops, unsigned long quirks);
 653 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
 654 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
 655 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
 656 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl);
 657
 658 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
 659
 660 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
 661                 bool send);
 662
 663 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
 664                 volatile union nvme_result *res);
 665
 666 void nvme_stop_queues(struct nvme_ctrl *ctrl);
 667 void nvme_start_queues(struct nvme_ctrl *ctrl);
 668 void nvme_kill_queues(struct nvme_ctrl *ctrl);
 669 void nvme_sync_queues(struct nvme_ctrl *ctrl);
 670 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
 671 void nvme_unfreeze(struct nvme_ctrl *ctrl);
 672 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
 673 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
 674 void nvme_start_freeze(struct nvme_ctrl *ctrl);
 675
 676 #define NVME_QID_ANY -1
 677 struct request *nvme_alloc_request(struct request_queue *q,
 678                 struct nvme_command *cmd, blk_mq_req_flags_t flags);
 679 void nvme_cleanup_cmd(struct request *req);
 680 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
 681 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
 682                 struct request *req);
 683 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 684                 bool queue_live);
 685
 686 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 687                 bool queue_live)
 688 {
 689         if (likely(ctrl->state == NVME_CTRL_LIVE))
 690                 return true;
 691         if (ctrl->ops->flags & NVME_F_FABRICS &&
 692             ctrl->state == NVME_CTRL_DELETING)
 693                 return true;
 694         return __nvme_check_ready(ctrl, rq, queue_live);
 695 }
 696 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 697                 void *buf, unsigned bufflen);
 698 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 699                 union nvme_result *result, void *buffer, unsigned bufflen,
 700                 unsigned timeout, int qid, int at_head,
 701                 blk_mq_req_flags_t flags);
 702 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
 703                       unsigned int dword11, void *buffer, size_t buflen,
 704                       u32 *result);
 705 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
 706                       unsigned int dword11, void *buffer, size_t buflen,
 707                       u32 *result);
 708 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
 709 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
 710 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
 711 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
 712 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
 713 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
 714 void nvme_queue_scan(struct nvme_ctrl *ctrl);
 715 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
 716                 void *log, size_t size, u64 offset);
 717 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
 718 void nvme_put_ns_head(struct nvme_ns_head *head);
 719 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
 720                 const struct file_operations *fops, struct module *owner);
 721 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
 722 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
 723                 unsigned int cmd, unsigned long arg);
 724 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
 725 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
 726                 unsigned int cmd, unsigned long arg);
 727 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
 728                 unsigned long arg);
 729 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
 730                 unsigned long arg);
 731 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
 732
 733 extern const struct attribute_group *nvme_ns_id_attr_groups[];
 734 extern const struct pr_ops nvme_pr_ops;
 735 extern const struct block_device_operations nvme_ns_head_ops;
 736
 737 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
 738 #ifdef CONFIG_NVME_MULTIPATH
 739 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
 740 {
 741         return ctrl->ana_log_buf != NULL;
 742 }
 743
 744 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
 745 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
 746 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
 747 bool nvme_mpath_set_disk_name(struct nvme_ns *ns, char *disk_name, int *flags);
 748 void nvme_failover_req(struct request *req);
 749 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
 750 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
 751 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
 752 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
 753 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
 754 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
 755 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
 756 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
 757 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
 758 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
 759 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
 760 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
 761
 762 static inline void nvme_trace_bio_complete(struct request *req)
 763 {
 764         struct nvme_ns *ns = req->q->queuedata;
 765
 766         if (req->cmd_flags & REQ_NVME_MPATH)
 767                 trace_block_bio_complete(ns->head->disk->queue, req->bio);
 768 }
 769
 770 extern struct device_attribute dev_attr_ana_grpid;
 771 extern struct device_attribute dev_attr_ana_state;
 772 extern struct device_attribute subsys_attr_iopolicy;
 773
 774 #else
 775 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
 776 {
 777         return false;
 778 }
 779 static inline bool nvme_mpath_set_disk_name(struct nvme_ns *ns, char *disk_name,
 780                 int *flags)
 781 {
 782         return false;
 783 }
 784 static inline void nvme_failover_req(struct request *req)
 785 {
 786 }
 787 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
 788 {
 789 }
 790 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
 791                 struct nvme_ns_head *head)
 792 {
 793         return 0;
 794 }
 795 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
 796                 struct nvme_id_ns *id)
 797 {
 798 }
 799 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
 800 {
 801 }
 802 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
 803 {
 804         return false;
 805 }
 806 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
 807 {
 808 }
 809 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
 810 {
 811 }
 812 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
 813 {
 814 }
 815 static inline void nvme_trace_bio_complete(struct request *req)
 816 {
 817 }
 818 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
 819 {
 820 }
 821 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
 822                 struct nvme_id_ctrl *id)
 823 {
 824         if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
 825                 dev_warn(ctrl->device,
 826 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
 827         return 0;
 828 }
 829 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
 830 {
 831 }
 832 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
 833 {
 834 }
 835 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
 836 {
 837 }
 838 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
 839 {
 840 }
 841 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
 842 {
 843 }
 844 #endif /* CONFIG_NVME_MULTIPATH */
 845
 846 int nvme_revalidate_zones(struct nvme_ns *ns);
 847 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
 848                 unsigned int nr_zones, report_zones_cb cb, void *data);
 849 #ifdef CONFIG_BLK_DEV_ZONED
 850 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
 851 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
 852                                        struct nvme_command *cmnd,
 853                                        enum nvme_zone_mgmt_action action);
 854 #else
 855 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
 856                 struct request *req, struct nvme_command *cmnd,
 857                 enum nvme_zone_mgmt_action action)
 858 {
 859         return BLK_STS_NOTSUPP;
 860 }
 861
 862 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
 863 {
 864         dev_warn(ns->ctrl->device,
 865                  "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
 866         return -EPROTONOSUPPORT;
 867 }
 868 #endif
 869
 870 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
 871 {
 872         return dev_to_disk(dev)->private_data;
 873 }
 874
 875 #ifdef CONFIG_NVME_HWMON
 876 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
 877 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
 878 #else
 879 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
 880 {
 881         return 0;
 882 }
 883
 884 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
 885 {
 886 }
 887 #endif
 888
 889 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
 890 {
 891         return ctrl->sgls & ((1 << 0) | (1 << 1));
 892 }
 893
 894 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
 895                          u8 opcode);
 896 int nvme_execute_passthru_rq(struct request *rq);
 897 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
 898 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
 899 void nvme_put_ns(struct nvme_ns *ns);
 900
 901 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
 902 {
 903         return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
 904 }
 905
 906 #endif /* _NVME_H */