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 limited, so skip 144 * encoding the generation sequence number. 145 */ 146 NVME_QUIRK_SKIP_CID_GEN = (1 << 17), 147 148 /* 149 * Reports garbage in the namespace identifiers (eui64, nguid, uuid). 150 */ 151 NVME_QUIRK_BOGUS_NID = (1 << 18), 152 153 /* 154 * No temperature thresholds for channels other than 0 (Composite). 155 */ 156 NVME_QUIRK_NO_SECONDARY_TEMP_THRESH = (1 << 19), 157 }; 158 159 /* 160 * Common request structure for NVMe passthrough. All drivers must have 161 * this structure as the first member of their request-private data. 162 */ 163 struct nvme_request { 164 struct nvme_command *cmd; 165 union nvme_result result; 166 u8 genctr; 167 u8 retries; 168 u8 flags; 169 u16 status; 170 #ifdef CONFIG_NVME_MULTIPATH 171 unsigned long start_time; 172 #endif 173 struct nvme_ctrl *ctrl; 174 }; 175 176 /* 177 * Mark a bio as coming in through the mpath node. 178 */ 179 #define REQ_NVME_MPATH REQ_DRV 180 181 enum { 182 NVME_REQ_CANCELLED = (1 << 0), 183 NVME_REQ_USERCMD = (1 << 1), 184 NVME_MPATH_IO_STATS = (1 << 2), 185 }; 186 187 static inline struct nvme_request *nvme_req(struct request *req) 188 { 189 return blk_mq_rq_to_pdu(req); 190 } 191 192 static inline u16 nvme_req_qid(struct request *req) 193 { 194 if (!req->q->queuedata) 195 return 0; 196 197 return req->mq_hctx->queue_num + 1; 198 } 199 200 /* The below value is the specific amount of delay needed before checking 201 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the 202 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was 203 * found empirically. 204 */ 205 #define NVME_QUIRK_DELAY_AMOUNT 2300 206 207 /* 208 * enum nvme_ctrl_state: Controller state 209 * 210 * @NVME_CTRL_NEW: New controller just allocated, initial state 211 * @NVME_CTRL_LIVE: Controller is connected and I/O capable 212 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset) 213 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the 214 * transport 215 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion) 216 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not 217 * disabled/failed immediately. This state comes 218 * after all async event processing took place and 219 * before ns removal and the controller deletion 220 * progress 221 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during 222 * shutdown or removal. In this case we forcibly 223 * kill all inflight I/O as they have no chance to 224 * complete 225 */ 226 enum nvme_ctrl_state { 227 NVME_CTRL_NEW, 228 NVME_CTRL_LIVE, 229 NVME_CTRL_RESETTING, 230 NVME_CTRL_CONNECTING, 231 NVME_CTRL_DELETING, 232 NVME_CTRL_DELETING_NOIO, 233 NVME_CTRL_DEAD, 234 }; 235 236 struct nvme_fault_inject { 237 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 238 struct fault_attr attr; 239 struct dentry *parent; 240 bool dont_retry; /* DNR, do not retry */ 241 u16 status; /* status code */ 242 #endif 243 }; 244 245 enum nvme_ctrl_flags { 246 NVME_CTRL_FAILFAST_EXPIRED = 0, 247 NVME_CTRL_ADMIN_Q_STOPPED = 1, 248 NVME_CTRL_STARTED_ONCE = 2, 249 NVME_CTRL_STOPPED = 3, 250 }; 251 252 struct nvme_ctrl { 253 bool comp_seen; 254 enum nvme_ctrl_state state; 255 bool identified; 256 spinlock_t lock; 257 struct mutex scan_lock; 258 const struct nvme_ctrl_ops *ops; 259 struct request_queue *admin_q; 260 struct request_queue *connect_q; 261 struct request_queue *fabrics_q; 262 struct device *dev; 263 int instance; 264 int numa_node; 265 struct blk_mq_tag_set *tagset; 266 struct blk_mq_tag_set *admin_tagset; 267 struct list_head namespaces; 268 struct rw_semaphore namespaces_rwsem; 269 struct device ctrl_device; 270 struct device *device; /* char device */ 271 #ifdef CONFIG_NVME_HWMON 272 struct device *hwmon_device; 273 #endif 274 struct cdev cdev; 275 struct work_struct reset_work; 276 struct work_struct delete_work; 277 wait_queue_head_t state_wq; 278 279 struct nvme_subsystem *subsys; 280 struct list_head subsys_entry; 281 282 struct opal_dev *opal_dev; 283 284 char name[12]; 285 u16 cntlid; 286 287 u32 ctrl_config; 288 u16 mtfa; 289 u32 queue_count; 290 291 u64 cap; 292 u32 max_hw_sectors; 293 u32 max_segments; 294 u32 max_integrity_segments; 295 u32 max_discard_sectors; 296 u32 max_discard_segments; 297 u32 max_zeroes_sectors; 298 #ifdef CONFIG_BLK_DEV_ZONED 299 u32 max_zone_append; 300 #endif 301 u16 crdt[3]; 302 u16 oncs; 303 u32 dmrsl; 304 u16 oacs; 305 u16 sqsize; 306 u32 max_namespaces; 307 atomic_t abort_limit; 308 u8 vwc; 309 u32 vs; 310 u32 sgls; 311 u16 kas; 312 u8 npss; 313 u8 apsta; 314 u16 wctemp; 315 u16 cctemp; 316 u32 oaes; 317 u32 aen_result; 318 u32 ctratt; 319 unsigned int shutdown_timeout; 320 unsigned int kato; 321 bool subsystem; 322 unsigned long quirks; 323 struct nvme_id_power_state psd[32]; 324 struct nvme_effects_log *effects; 325 struct xarray cels; 326 struct work_struct scan_work; 327 struct work_struct async_event_work; 328 struct delayed_work ka_work; 329 struct delayed_work failfast_work; 330 struct nvme_command ka_cmd; 331 unsigned long ka_last_check_time; 332 struct work_struct fw_act_work; 333 unsigned long events; 334 335 #ifdef CONFIG_NVME_MULTIPATH 336 /* asymmetric namespace access: */ 337 u8 anacap; 338 u8 anatt; 339 u32 anagrpmax; 340 u32 nanagrpid; 341 struct mutex ana_lock; 342 struct nvme_ana_rsp_hdr *ana_log_buf; 343 size_t ana_log_size; 344 struct timer_list anatt_timer; 345 struct work_struct ana_work; 346 #endif 347 348 #ifdef CONFIG_NVME_AUTH 349 struct work_struct dhchap_auth_work; 350 struct mutex dhchap_auth_mutex; 351 struct nvme_dhchap_queue_context *dhchap_ctxs; 352 struct nvme_dhchap_key *host_key; 353 struct nvme_dhchap_key *ctrl_key; 354 u16 transaction; 355 #endif 356 357 /* Power saving configuration */ 358 u64 ps_max_latency_us; 359 bool apst_enabled; 360 361 /* PCIe only: */ 362 u32 hmpre; 363 u32 hmmin; 364 u32 hmminds; 365 u16 hmmaxd; 366 367 /* Fabrics only */ 368 u32 ioccsz; 369 u32 iorcsz; 370 u16 icdoff; 371 u16 maxcmd; 372 int nr_reconnects; 373 unsigned long flags; 374 struct nvmf_ctrl_options *opts; 375 376 struct page *discard_page; 377 unsigned long discard_page_busy; 378 379 struct nvme_fault_inject fault_inject; 380 381 enum nvme_ctrl_type cntrltype; 382 enum nvme_dctype dctype; 383 }; 384 385 enum nvme_iopolicy { 386 NVME_IOPOLICY_NUMA, 387 NVME_IOPOLICY_RR, 388 }; 389 390 struct nvme_subsystem { 391 int instance; 392 struct device dev; 393 /* 394 * Because we unregister the device on the last put we need 395 * a separate refcount. 396 */ 397 struct kref ref; 398 struct list_head entry; 399 struct mutex lock; 400 struct list_head ctrls; 401 struct list_head nsheads; 402 char subnqn[NVMF_NQN_SIZE]; 403 char serial[20]; 404 char model[40]; 405 char firmware_rev[8]; 406 u8 cmic; 407 enum nvme_subsys_type subtype; 408 u16 vendor_id; 409 u16 awupf; /* 0's based awupf value. */ 410 struct ida ns_ida; 411 #ifdef CONFIG_NVME_MULTIPATH 412 enum nvme_iopolicy iopolicy; 413 #endif 414 }; 415 416 /* 417 * Container structure for uniqueue namespace identifiers. 418 */ 419 struct nvme_ns_ids { 420 u8 eui64[8]; 421 u8 nguid[16]; 422 uuid_t uuid; 423 u8 csi; 424 }; 425 426 /* 427 * Anchor structure for namespaces. There is one for each namespace in a 428 * NVMe subsystem that any of our controllers can see, and the namespace 429 * structure for each controller is chained of it. For private namespaces 430 * there is a 1:1 relation to our namespace structures, that is ->list 431 * only ever has a single entry for private namespaces. 432 */ 433 struct nvme_ns_head { 434 struct list_head list; 435 struct srcu_struct srcu; 436 struct nvme_subsystem *subsys; 437 unsigned ns_id; 438 struct nvme_ns_ids ids; 439 struct list_head entry; 440 struct kref ref; 441 bool shared; 442 int instance; 443 struct nvme_effects_log *effects; 444 445 struct cdev cdev; 446 struct device cdev_device; 447 448 struct gendisk *disk; 449 #ifdef CONFIG_NVME_MULTIPATH 450 struct bio_list requeue_list; 451 spinlock_t requeue_lock; 452 struct work_struct requeue_work; 453 struct mutex lock; 454 unsigned long flags; 455 #define NVME_NSHEAD_DISK_LIVE 0 456 struct nvme_ns __rcu *current_path[]; 457 #endif 458 }; 459 460 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head) 461 { 462 return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk; 463 } 464 465 enum nvme_ns_features { 466 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */ 467 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */ 468 NVME_NS_DEAC, /* DEAC bit in Write Zeores supported */ 469 }; 470 471 struct nvme_ns { 472 struct list_head list; 473 474 struct nvme_ctrl *ctrl; 475 struct request_queue *queue; 476 struct gendisk *disk; 477 #ifdef CONFIG_NVME_MULTIPATH 478 enum nvme_ana_state ana_state; 479 u32 ana_grpid; 480 #endif 481 struct list_head siblings; 482 struct kref kref; 483 struct nvme_ns_head *head; 484 485 int lba_shift; 486 u16 ms; 487 u16 pi_size; 488 u16 sgs; 489 u32 sws; 490 u8 pi_type; 491 u8 guard_type; 492 #ifdef CONFIG_BLK_DEV_ZONED 493 u64 zsze; 494 #endif 495 unsigned long features; 496 unsigned long flags; 497 #define NVME_NS_REMOVING 0 498 #define NVME_NS_ANA_PENDING 2 499 #define NVME_NS_FORCE_RO 3 500 #define NVME_NS_READY 4 501 502 struct cdev cdev; 503 struct device cdev_device; 504 505 struct nvme_fault_inject fault_inject; 506 507 }; 508 509 /* NVMe ns supports metadata actions by the controller (generate/strip) */ 510 static inline bool nvme_ns_has_pi(struct nvme_ns *ns) 511 { 512 return ns->pi_type && ns->ms == ns->pi_size; 513 } 514 515 struct nvme_ctrl_ops { 516 const char *name; 517 struct module *module; 518 unsigned int flags; 519 #define NVME_F_FABRICS (1 << 0) 520 #define NVME_F_METADATA_SUPPORTED (1 << 1) 521 #define NVME_F_BLOCKING (1 << 2) 522 523 const struct attribute_group **dev_attr_groups; 524 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val); 525 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val); 526 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val); 527 void (*free_ctrl)(struct nvme_ctrl *ctrl); 528 void (*submit_async_event)(struct nvme_ctrl *ctrl); 529 void (*delete_ctrl)(struct nvme_ctrl *ctrl); 530 void (*stop_ctrl)(struct nvme_ctrl *ctrl); 531 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); 532 void (*print_device_info)(struct nvme_ctrl *ctrl); 533 bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl); 534 }; 535 536 /* 537 * nvme command_id is constructed as such: 538 * | xxxx | xxxxxxxxxxxx | 539 * gen request tag 540 */ 541 #define nvme_genctr_mask(gen) (gen & 0xf) 542 #define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12) 543 #define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12) 544 #define nvme_tag_from_cid(cid) (cid & 0xfff) 545 546 static inline u16 nvme_cid(struct request *rq) 547 { 548 return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag; 549 } 550 551 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags, 552 u16 command_id) 553 { 554 u8 genctr = nvme_genctr_from_cid(command_id); 555 u16 tag = nvme_tag_from_cid(command_id); 556 struct request *rq; 557 558 rq = blk_mq_tag_to_rq(tags, tag); 559 if (unlikely(!rq)) { 560 pr_err("could not locate request for tag %#x\n", 561 tag); 562 return NULL; 563 } 564 if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) { 565 dev_err(nvme_req(rq)->ctrl->device, 566 "request %#x genctr mismatch (got %#x expected %#x)\n", 567 tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr)); 568 return NULL; 569 } 570 return rq; 571 } 572 573 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags, 574 u16 command_id) 575 { 576 return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id)); 577 } 578 579 /* 580 * Return the length of the string without the space padding 581 */ 582 static inline int nvme_strlen(char *s, int len) 583 { 584 while (s[len - 1] == ' ') 585 len--; 586 return len; 587 } 588 589 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl) 590 { 591 struct nvme_subsystem *subsys = ctrl->subsys; 592 593 if (ctrl->ops->print_device_info) { 594 ctrl->ops->print_device_info(ctrl); 595 return; 596 } 597 598 dev_err(ctrl->device, 599 "VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id, 600 nvme_strlen(subsys->model, sizeof(subsys->model)), 601 subsys->model, nvme_strlen(subsys->firmware_rev, 602 sizeof(subsys->firmware_rev)), 603 subsys->firmware_rev); 604 } 605 606 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 607 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 608 const char *dev_name); 609 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject); 610 void nvme_should_fail(struct request *req); 611 #else 612 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 613 const char *dev_name) 614 { 615 } 616 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj) 617 { 618 } 619 static inline void nvme_should_fail(struct request *req) {} 620 #endif 621 622 bool nvme_wait_reset(struct nvme_ctrl *ctrl); 623 int nvme_try_sched_reset(struct nvme_ctrl *ctrl); 624 625 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl) 626 { 627 int ret; 628 629 if (!ctrl->subsystem) 630 return -ENOTTY; 631 if (!nvme_wait_reset(ctrl)) 632 return -EBUSY; 633 634 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65); 635 if (ret) 636 return ret; 637 638 return nvme_try_sched_reset(ctrl); 639 } 640 641 /* 642 * Convert a 512B sector number to a device logical block number. 643 */ 644 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector) 645 { 646 return sector >> (ns->lba_shift - SECTOR_SHIFT); 647 } 648 649 /* 650 * Convert a device logical block number to a 512B sector number. 651 */ 652 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba) 653 { 654 return lba << (ns->lba_shift - SECTOR_SHIFT); 655 } 656 657 /* 658 * Convert byte length to nvme's 0-based num dwords 659 */ 660 static inline u32 nvme_bytes_to_numd(size_t len) 661 { 662 return (len >> 2) - 1; 663 } 664 665 static inline bool nvme_is_ana_error(u16 status) 666 { 667 switch (status & 0x7ff) { 668 case NVME_SC_ANA_TRANSITION: 669 case NVME_SC_ANA_INACCESSIBLE: 670 case NVME_SC_ANA_PERSISTENT_LOSS: 671 return true; 672 default: 673 return false; 674 } 675 } 676 677 static inline bool nvme_is_path_error(u16 status) 678 { 679 /* check for a status code type of 'path related status' */ 680 return (status & 0x700) == 0x300; 681 } 682 683 /* 684 * Fill in the status and result information from the CQE, and then figure out 685 * if blk-mq will need to use IPI magic to complete the request, and if yes do 686 * so. If not let the caller complete the request without an indirect function 687 * call. 688 */ 689 static inline bool nvme_try_complete_req(struct request *req, __le16 status, 690 union nvme_result result) 691 { 692 struct nvme_request *rq = nvme_req(req); 693 struct nvme_ctrl *ctrl = rq->ctrl; 694 695 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN)) 696 rq->genctr++; 697 698 rq->status = le16_to_cpu(status) >> 1; 699 rq->result = result; 700 /* inject error when permitted by fault injection framework */ 701 nvme_should_fail(req); 702 if (unlikely(blk_should_fake_timeout(req->q))) 703 return true; 704 return blk_mq_complete_request_remote(req); 705 } 706 707 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl) 708 { 709 get_device(ctrl->device); 710 } 711 712 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl) 713 { 714 put_device(ctrl->device); 715 } 716 717 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id) 718 { 719 return !qid && 720 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH; 721 } 722 723 void nvme_complete_rq(struct request *req); 724 void nvme_complete_batch_req(struct request *req); 725 726 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob, 727 void (*fn)(struct request *rq)) 728 { 729 struct request *req; 730 731 rq_list_for_each(&iob->req_list, req) { 732 fn(req); 733 nvme_complete_batch_req(req); 734 } 735 blk_mq_end_request_batch(iob); 736 } 737 738 blk_status_t nvme_host_path_error(struct request *req); 739 bool nvme_cancel_request(struct request *req, void *data); 740 void nvme_cancel_tagset(struct nvme_ctrl *ctrl); 741 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl); 742 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, 743 enum nvme_ctrl_state new_state); 744 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown); 745 int nvme_enable_ctrl(struct nvme_ctrl *ctrl); 746 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, 747 const struct nvme_ctrl_ops *ops, unsigned long quirks); 748 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl); 749 void nvme_start_ctrl(struct nvme_ctrl *ctrl); 750 void nvme_stop_ctrl(struct nvme_ctrl *ctrl); 751 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended); 752 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 753 const struct blk_mq_ops *ops, unsigned int cmd_size); 754 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl); 755 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 756 const struct blk_mq_ops *ops, unsigned int nr_maps, 757 unsigned int cmd_size); 758 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl); 759 760 void nvme_remove_namespaces(struct nvme_ctrl *ctrl); 761 762 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, 763 volatile union nvme_result *res); 764 765 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl); 766 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl); 767 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl); 768 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl); 769 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl); 770 void nvme_sync_queues(struct nvme_ctrl *ctrl); 771 void nvme_sync_io_queues(struct nvme_ctrl *ctrl); 772 void nvme_unfreeze(struct nvme_ctrl *ctrl); 773 void nvme_wait_freeze(struct nvme_ctrl *ctrl); 774 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); 775 void nvme_start_freeze(struct nvme_ctrl *ctrl); 776 777 static inline enum req_op nvme_req_op(struct nvme_command *cmd) 778 { 779 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; 780 } 781 782 #define NVME_QID_ANY -1 783 void nvme_init_request(struct request *req, struct nvme_command *cmd); 784 void nvme_cleanup_cmd(struct request *req); 785 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req); 786 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl, 787 struct request *req); 788 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 789 bool queue_live); 790 791 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 792 bool queue_live) 793 { 794 if (likely(ctrl->state == NVME_CTRL_LIVE)) 795 return true; 796 if (ctrl->ops->flags & NVME_F_FABRICS && 797 ctrl->state == NVME_CTRL_DELETING) 798 return queue_live; 799 return __nvme_check_ready(ctrl, rq, queue_live); 800 } 801 802 /* 803 * NSID shall be unique for all shared namespaces, or if at least one of the 804 * following conditions is met: 805 * 1. Namespace Management is supported by the controller 806 * 2. ANA is supported by the controller 807 * 3. NVM Set are supported by the controller 808 * 809 * In other case, private namespace are not required to report a unique NSID. 810 */ 811 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl, 812 struct nvme_ns_head *head) 813 { 814 return head->shared || 815 (ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) || 816 (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) || 817 (ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS); 818 } 819 820 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 821 void *buf, unsigned bufflen); 822 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 823 union nvme_result *result, void *buffer, unsigned bufflen, 824 int qid, int at_head, 825 blk_mq_req_flags_t flags); 826 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, 827 unsigned int dword11, void *buffer, size_t buflen, 828 u32 *result); 829 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, 830 unsigned int dword11, void *buffer, size_t buflen, 831 u32 *result); 832 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count); 833 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl); 834 int nvme_reset_ctrl(struct nvme_ctrl *ctrl); 835 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl); 836 int nvme_delete_ctrl(struct nvme_ctrl *ctrl); 837 void nvme_queue_scan(struct nvme_ctrl *ctrl); 838 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi, 839 void *log, size_t size, u64 offset); 840 bool nvme_tryget_ns_head(struct nvme_ns_head *head); 841 void nvme_put_ns_head(struct nvme_ns_head *head); 842 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device, 843 const struct file_operations *fops, struct module *owner); 844 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device); 845 int nvme_ioctl(struct block_device *bdev, fmode_t mode, 846 unsigned int cmd, unsigned long arg); 847 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 848 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode, 849 unsigned int cmd, unsigned long arg); 850 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd, 851 unsigned long arg); 852 long nvme_dev_ioctl(struct file *file, unsigned int cmd, 853 unsigned long arg); 854 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd, 855 struct io_comp_batch *iob, unsigned int poll_flags); 856 int nvme_ns_head_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd, 857 struct io_comp_batch *iob, unsigned int poll_flags); 858 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd, 859 unsigned int issue_flags); 860 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd, 861 unsigned int issue_flags); 862 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo); 863 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags); 864 865 extern const struct attribute_group *nvme_ns_id_attr_groups[]; 866 extern const struct pr_ops nvme_pr_ops; 867 extern const struct block_device_operations nvme_ns_head_ops; 868 extern const struct attribute_group nvme_dev_attrs_group; 869 870 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head); 871 #ifdef CONFIG_NVME_MULTIPATH 872 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 873 { 874 return ctrl->ana_log_buf != NULL; 875 } 876 877 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys); 878 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys); 879 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys); 880 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys); 881 void nvme_failover_req(struct request *req); 882 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl); 883 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head); 884 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid); 885 void nvme_mpath_remove_disk(struct nvme_ns_head *head); 886 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); 887 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl); 888 void nvme_mpath_update(struct nvme_ctrl *ctrl); 889 void nvme_mpath_uninit(struct nvme_ctrl *ctrl); 890 void nvme_mpath_stop(struct nvme_ctrl *ctrl); 891 bool nvme_mpath_clear_current_path(struct nvme_ns *ns); 892 void nvme_mpath_revalidate_paths(struct nvme_ns *ns); 893 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl); 894 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head); 895 void nvme_mpath_start_request(struct request *rq); 896 void nvme_mpath_end_request(struct request *rq); 897 898 static inline void nvme_trace_bio_complete(struct request *req) 899 { 900 struct nvme_ns *ns = req->q->queuedata; 901 902 if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio) 903 trace_block_bio_complete(ns->head->disk->queue, req->bio); 904 } 905 906 extern bool multipath; 907 extern struct device_attribute dev_attr_ana_grpid; 908 extern struct device_attribute dev_attr_ana_state; 909 extern struct device_attribute subsys_attr_iopolicy; 910 911 #else 912 #define multipath false 913 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 914 { 915 return false; 916 } 917 static inline void nvme_failover_req(struct request *req) 918 { 919 } 920 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl) 921 { 922 } 923 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, 924 struct nvme_ns_head *head) 925 { 926 return 0; 927 } 928 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid) 929 { 930 } 931 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head) 932 { 933 } 934 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns) 935 { 936 return false; 937 } 938 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns) 939 { 940 } 941 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl) 942 { 943 } 944 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head) 945 { 946 } 947 static inline void nvme_trace_bio_complete(struct request *req) 948 { 949 } 950 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl) 951 { 952 } 953 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, 954 struct nvme_id_ctrl *id) 955 { 956 if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) 957 dev_warn(ctrl->device, 958 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n"); 959 return 0; 960 } 961 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl) 962 { 963 } 964 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl) 965 { 966 } 967 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl) 968 { 969 } 970 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys) 971 { 972 } 973 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys) 974 { 975 } 976 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys) 977 { 978 } 979 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys) 980 { 981 } 982 static inline void nvme_mpath_start_request(struct request *rq) 983 { 984 } 985 static inline void nvme_mpath_end_request(struct request *rq) 986 { 987 } 988 #endif /* CONFIG_NVME_MULTIPATH */ 989 990 int nvme_revalidate_zones(struct nvme_ns *ns); 991 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector, 992 unsigned int nr_zones, report_zones_cb cb, void *data); 993 #ifdef CONFIG_BLK_DEV_ZONED 994 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf); 995 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req, 996 struct nvme_command *cmnd, 997 enum nvme_zone_mgmt_action action); 998 #else 999 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, 1000 struct request *req, struct nvme_command *cmnd, 1001 enum nvme_zone_mgmt_action action) 1002 { 1003 return BLK_STS_NOTSUPP; 1004 } 1005 1006 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf) 1007 { 1008 dev_warn(ns->ctrl->device, 1009 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n"); 1010 return -EPROTONOSUPPORT; 1011 } 1012 #endif 1013 1014 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev) 1015 { 1016 return dev_to_disk(dev)->private_data; 1017 } 1018 1019 #ifdef CONFIG_NVME_HWMON 1020 int nvme_hwmon_init(struct nvme_ctrl *ctrl); 1021 void nvme_hwmon_exit(struct nvme_ctrl *ctrl); 1022 #else 1023 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl) 1024 { 1025 return 0; 1026 } 1027 1028 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl) 1029 { 1030 } 1031 #endif 1032 1033 static inline void nvme_start_request(struct request *rq) 1034 { 1035 if (rq->cmd_flags & REQ_NVME_MPATH) 1036 nvme_mpath_start_request(rq); 1037 blk_mq_start_request(rq); 1038 } 1039 1040 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl) 1041 { 1042 return ctrl->sgls & ((1 << 0) | (1 << 1)); 1043 } 1044 1045 #ifdef CONFIG_NVME_AUTH 1046 int __init nvme_init_auth(void); 1047 void __exit nvme_exit_auth(void); 1048 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl); 1049 void nvme_auth_stop(struct nvme_ctrl *ctrl); 1050 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid); 1051 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid); 1052 void nvme_auth_free(struct nvme_ctrl *ctrl); 1053 #else 1054 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl) 1055 { 1056 return 0; 1057 } 1058 static inline int __init nvme_init_auth(void) 1059 { 1060 return 0; 1061 } 1062 static inline void __exit nvme_exit_auth(void) 1063 { 1064 } 1065 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {}; 1066 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid) 1067 { 1068 return -EPROTONOSUPPORT; 1069 } 1070 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid) 1071 { 1072 return NVME_SC_AUTH_REQUIRED; 1073 } 1074 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {}; 1075 #endif 1076 1077 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, 1078 u8 opcode); 1079 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode); 1080 int nvme_execute_rq(struct request *rq, bool at_head); 1081 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects, 1082 struct nvme_command *cmd, int status); 1083 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file); 1084 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid); 1085 void nvme_put_ns(struct nvme_ns *ns); 1086 1087 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl) 1088 { 1089 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI; 1090 } 1091 1092 #ifdef CONFIG_NVME_VERBOSE_ERRORS 1093 const unsigned char *nvme_get_error_status_str(u16 status); 1094 const unsigned char *nvme_get_opcode_str(u8 opcode); 1095 const unsigned char *nvme_get_admin_opcode_str(u8 opcode); 1096 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode); 1097 #else /* CONFIG_NVME_VERBOSE_ERRORS */ 1098 static inline const unsigned char *nvme_get_error_status_str(u16 status) 1099 { 1100 return "I/O Error"; 1101 } 1102 static inline const unsigned char *nvme_get_opcode_str(u8 opcode) 1103 { 1104 return "I/O Cmd"; 1105 } 1106 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode) 1107 { 1108 return "Admin Cmd"; 1109 } 1110 1111 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode) 1112 { 1113 return "Fabrics Cmd"; 1114 } 1115 #endif /* CONFIG_NVME_VERBOSE_ERRORS */ 1116 1117 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype) 1118 { 1119 if (opcode == nvme_fabrics_command) 1120 return nvme_get_fabrics_opcode_str(fctype); 1121 return qid ? nvme_get_opcode_str(opcode) : 1122 nvme_get_admin_opcode_str(opcode); 1123 } 1124 #endif /* _NVME_H */ 1125