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 #include <linux/ratelimit_types.h> 20 21 #include <trace/events/block.h> 22 23 extern const struct pr_ops nvme_pr_ops; 24 25 extern unsigned int nvme_io_timeout; 26 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ) 27 28 extern unsigned int admin_timeout; 29 #define NVME_ADMIN_TIMEOUT (admin_timeout * HZ) 30 31 #define NVME_DEFAULT_KATO 5 32 33 #ifdef CONFIG_ARCH_NO_SG_CHAIN 34 #define NVME_INLINE_SG_CNT 0 35 #define NVME_INLINE_METADATA_SG_CNT 0 36 #else 37 #define NVME_INLINE_SG_CNT 2 38 #define NVME_INLINE_METADATA_SG_CNT 1 39 #endif 40 41 /* 42 * Default to a 4K page size, with the intention to update this 43 * path in the future to accommodate architectures with differing 44 * kernel and IO page sizes. 45 */ 46 #define NVME_CTRL_PAGE_SHIFT 12 47 #define NVME_CTRL_PAGE_SIZE (1 << NVME_CTRL_PAGE_SHIFT) 48 49 extern struct workqueue_struct *nvme_wq; 50 extern struct workqueue_struct *nvme_reset_wq; 51 extern struct workqueue_struct *nvme_delete_wq; 52 extern struct mutex nvme_subsystems_lock; 53 54 /* 55 * List of workarounds for devices that required behavior not specified in 56 * the standard. 57 */ 58 enum nvme_quirks { 59 /* 60 * Prefers I/O aligned to a stripe size specified in a vendor 61 * specific Identify field. 62 */ 63 NVME_QUIRK_STRIPE_SIZE = (1 << 0), 64 65 /* 66 * The controller doesn't handle Identify value others than 0 or 1 67 * correctly. 68 */ 69 NVME_QUIRK_IDENTIFY_CNS = (1 << 1), 70 71 /* 72 * The controller deterministically returns O's on reads to 73 * logical blocks that deallocate was called on. 74 */ 75 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2), 76 77 /* 78 * The controller needs a delay before starts checking the device 79 * readiness, which is done by reading the NVME_CSTS_RDY bit. 80 */ 81 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3), 82 83 /* 84 * APST should not be used. 85 */ 86 NVME_QUIRK_NO_APST = (1 << 4), 87 88 /* 89 * The deepest sleep state should not be used. 90 */ 91 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5), 92 93 /* 94 * Set MEDIUM priority on SQ creation 95 */ 96 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7), 97 98 /* 99 * Ignore device provided subnqn. 100 */ 101 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8), 102 103 /* 104 * Broken Write Zeroes. 105 */ 106 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9), 107 108 /* 109 * Force simple suspend/resume path. 110 */ 111 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10), 112 113 /* 114 * Use only one interrupt vector for all queues 115 */ 116 NVME_QUIRK_SINGLE_VECTOR = (1 << 11), 117 118 /* 119 * Use non-standard 128 bytes SQEs. 120 */ 121 NVME_QUIRK_128_BYTES_SQES = (1 << 12), 122 123 /* 124 * Prevent tag overlap between queues 125 */ 126 NVME_QUIRK_SHARED_TAGS = (1 << 13), 127 128 /* 129 * Don't change the value of the temperature threshold feature 130 */ 131 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14), 132 133 /* 134 * The controller doesn't handle the Identify Namespace 135 * Identification Descriptor list subcommand despite claiming 136 * NVMe 1.3 compliance. 137 */ 138 NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15), 139 140 /* 141 * The controller does not properly handle DMA addresses over 142 * 48 bits. 143 */ 144 NVME_QUIRK_DMA_ADDRESS_BITS_48 = (1 << 16), 145 146 /* 147 * The controller requires the command_id value be limited, so skip 148 * encoding the generation sequence number. 149 */ 150 NVME_QUIRK_SKIP_CID_GEN = (1 << 17), 151 152 /* 153 * Reports garbage in the namespace identifiers (eui64, nguid, uuid). 154 */ 155 NVME_QUIRK_BOGUS_NID = (1 << 18), 156 157 /* 158 * No temperature thresholds for channels other than 0 (Composite). 159 */ 160 NVME_QUIRK_NO_SECONDARY_TEMP_THRESH = (1 << 19), 161 162 /* 163 * Disables simple suspend/resume path. 164 */ 165 NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND = (1 << 20), 166 167 /* 168 * MSI (but not MSI-X) interrupts are broken and never fire. 169 */ 170 NVME_QUIRK_BROKEN_MSI = (1 << 21), 171 }; 172 173 /* 174 * Common request structure for NVMe passthrough. All drivers must have 175 * this structure as the first member of their request-private data. 176 */ 177 struct nvme_request { 178 struct nvme_command *cmd; 179 union nvme_result result; 180 u8 genctr; 181 u8 retries; 182 u8 flags; 183 u16 status; 184 #ifdef CONFIG_NVME_MULTIPATH 185 unsigned long start_time; 186 #endif 187 struct nvme_ctrl *ctrl; 188 }; 189 190 /* 191 * Mark a bio as coming in through the mpath node. 192 */ 193 #define REQ_NVME_MPATH REQ_DRV 194 195 enum { 196 NVME_REQ_CANCELLED = (1 << 0), 197 NVME_REQ_USERCMD = (1 << 1), 198 NVME_MPATH_IO_STATS = (1 << 2), 199 NVME_MPATH_CNT_ACTIVE = (1 << 3), 200 }; 201 202 static inline struct nvme_request *nvme_req(struct request *req) 203 { 204 return blk_mq_rq_to_pdu(req); 205 } 206 207 static inline u16 nvme_req_qid(struct request *req) 208 { 209 if (!req->q->queuedata) 210 return 0; 211 212 return req->mq_hctx->queue_num + 1; 213 } 214 215 /* The below value is the specific amount of delay needed before checking 216 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the 217 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was 218 * found empirically. 219 */ 220 #define NVME_QUIRK_DELAY_AMOUNT 2300 221 222 /* 223 * enum nvme_ctrl_state: Controller state 224 * 225 * @NVME_CTRL_NEW: New controller just allocated, initial state 226 * @NVME_CTRL_LIVE: Controller is connected and I/O capable 227 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset) 228 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the 229 * transport 230 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion) 231 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not 232 * disabled/failed immediately. This state comes 233 * after all async event processing took place and 234 * before ns removal and the controller deletion 235 * progress 236 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during 237 * shutdown or removal. In this case we forcibly 238 * kill all inflight I/O as they have no chance to 239 * complete 240 */ 241 enum nvme_ctrl_state { 242 NVME_CTRL_NEW, 243 NVME_CTRL_LIVE, 244 NVME_CTRL_RESETTING, 245 NVME_CTRL_CONNECTING, 246 NVME_CTRL_DELETING, 247 NVME_CTRL_DELETING_NOIO, 248 NVME_CTRL_DEAD, 249 }; 250 251 struct nvme_fault_inject { 252 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 253 struct fault_attr attr; 254 struct dentry *parent; 255 bool dont_retry; /* DNR, do not retry */ 256 u16 status; /* status code */ 257 #endif 258 }; 259 260 enum nvme_ctrl_flags { 261 NVME_CTRL_FAILFAST_EXPIRED = 0, 262 NVME_CTRL_ADMIN_Q_STOPPED = 1, 263 NVME_CTRL_STARTED_ONCE = 2, 264 NVME_CTRL_STOPPED = 3, 265 NVME_CTRL_SKIP_ID_CNS_CS = 4, 266 NVME_CTRL_DIRTY_CAPABILITY = 5, 267 NVME_CTRL_FROZEN = 6, 268 }; 269 270 struct nvme_ctrl { 271 bool comp_seen; 272 bool identified; 273 bool passthru_err_log_enabled; 274 enum nvme_ctrl_state state; 275 spinlock_t lock; 276 struct mutex scan_lock; 277 const struct nvme_ctrl_ops *ops; 278 struct request_queue *admin_q; 279 struct request_queue *connect_q; 280 struct request_queue *fabrics_q; 281 struct device *dev; 282 int instance; 283 int numa_node; 284 struct blk_mq_tag_set *tagset; 285 struct blk_mq_tag_set *admin_tagset; 286 struct list_head namespaces; 287 struct mutex namespaces_lock; 288 struct srcu_struct srcu; 289 struct device ctrl_device; 290 struct device *device; /* char device */ 291 #ifdef CONFIG_NVME_HWMON 292 struct device *hwmon_device; 293 #endif 294 struct cdev cdev; 295 struct work_struct reset_work; 296 struct work_struct delete_work; 297 wait_queue_head_t state_wq; 298 299 struct nvme_subsystem *subsys; 300 struct list_head subsys_entry; 301 302 struct opal_dev *opal_dev; 303 304 u16 cntlid; 305 306 u16 mtfa; 307 u32 ctrl_config; 308 u32 queue_count; 309 310 u64 cap; 311 u32 max_hw_sectors; 312 u32 max_segments; 313 u32 max_integrity_segments; 314 u32 max_zeroes_sectors; 315 #ifdef CONFIG_BLK_DEV_ZONED 316 u32 max_zone_append; 317 #endif 318 u16 crdt[3]; 319 u16 oncs; 320 u8 dmrl; 321 u32 dmrsl; 322 u16 oacs; 323 u16 sqsize; 324 u32 max_namespaces; 325 atomic_t abort_limit; 326 u8 vwc; 327 u32 vs; 328 u32 sgls; 329 u16 kas; 330 u8 npss; 331 u8 apsta; 332 u16 wctemp; 333 u16 cctemp; 334 u32 oaes; 335 u32 aen_result; 336 u32 ctratt; 337 unsigned int shutdown_timeout; 338 unsigned int kato; 339 bool subsystem; 340 unsigned long quirks; 341 struct nvme_id_power_state psd[32]; 342 struct nvme_effects_log *effects; 343 struct xarray cels; 344 struct work_struct scan_work; 345 struct work_struct async_event_work; 346 struct delayed_work ka_work; 347 struct delayed_work failfast_work; 348 struct nvme_command ka_cmd; 349 unsigned long ka_last_check_time; 350 struct work_struct fw_act_work; 351 unsigned long events; 352 353 #ifdef CONFIG_NVME_MULTIPATH 354 /* asymmetric namespace access: */ 355 u8 anacap; 356 u8 anatt; 357 u32 anagrpmax; 358 u32 nanagrpid; 359 struct mutex ana_lock; 360 struct nvme_ana_rsp_hdr *ana_log_buf; 361 size_t ana_log_size; 362 struct timer_list anatt_timer; 363 struct work_struct ana_work; 364 atomic_t nr_active; 365 #endif 366 367 #ifdef CONFIG_NVME_HOST_AUTH 368 struct work_struct dhchap_auth_work; 369 struct mutex dhchap_auth_mutex; 370 struct nvme_dhchap_queue_context *dhchap_ctxs; 371 struct nvme_dhchap_key *host_key; 372 struct nvme_dhchap_key *ctrl_key; 373 u16 transaction; 374 #endif 375 struct key *tls_key; 376 377 /* Power saving configuration */ 378 u64 ps_max_latency_us; 379 bool apst_enabled; 380 381 /* PCIe only: */ 382 u16 hmmaxd; 383 u32 hmpre; 384 u32 hmmin; 385 u32 hmminds; 386 387 /* Fabrics only */ 388 u32 ioccsz; 389 u32 iorcsz; 390 u16 icdoff; 391 u16 maxcmd; 392 int nr_reconnects; 393 unsigned long flags; 394 struct nvmf_ctrl_options *opts; 395 396 struct page *discard_page; 397 unsigned long discard_page_busy; 398 399 struct nvme_fault_inject fault_inject; 400 401 enum nvme_ctrl_type cntrltype; 402 enum nvme_dctype dctype; 403 }; 404 405 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl) 406 { 407 return READ_ONCE(ctrl->state); 408 } 409 410 enum nvme_iopolicy { 411 NVME_IOPOLICY_NUMA, 412 NVME_IOPOLICY_RR, 413 NVME_IOPOLICY_QD, 414 }; 415 416 struct nvme_subsystem { 417 int instance; 418 struct device dev; 419 /* 420 * Because we unregister the device on the last put we need 421 * a separate refcount. 422 */ 423 struct kref ref; 424 struct list_head entry; 425 struct mutex lock; 426 struct list_head ctrls; 427 struct list_head nsheads; 428 char subnqn[NVMF_NQN_SIZE]; 429 char serial[20]; 430 char model[40]; 431 char firmware_rev[8]; 432 u8 cmic; 433 enum nvme_subsys_type subtype; 434 u16 vendor_id; 435 u16 awupf; /* 0's based awupf value. */ 436 struct ida ns_ida; 437 #ifdef CONFIG_NVME_MULTIPATH 438 enum nvme_iopolicy iopolicy; 439 #endif 440 }; 441 442 /* 443 * Container structure for uniqueue namespace identifiers. 444 */ 445 struct nvme_ns_ids { 446 u8 eui64[8]; 447 u8 nguid[16]; 448 uuid_t uuid; 449 u8 csi; 450 }; 451 452 /* 453 * Anchor structure for namespaces. There is one for each namespace in a 454 * NVMe subsystem that any of our controllers can see, and the namespace 455 * structure for each controller is chained of it. For private namespaces 456 * there is a 1:1 relation to our namespace structures, that is ->list 457 * only ever has a single entry for private namespaces. 458 */ 459 struct nvme_ns_head { 460 struct list_head list; 461 struct srcu_struct srcu; 462 struct nvme_subsystem *subsys; 463 struct nvme_ns_ids ids; 464 u8 lba_shift; 465 u16 ms; 466 u16 pi_size; 467 u8 pi_type; 468 u8 guard_type; 469 struct list_head entry; 470 struct kref ref; 471 bool shared; 472 bool passthru_err_log_enabled; 473 struct nvme_effects_log *effects; 474 u64 nuse; 475 unsigned ns_id; 476 int instance; 477 #ifdef CONFIG_BLK_DEV_ZONED 478 u64 zsze; 479 #endif 480 unsigned long features; 481 482 struct ratelimit_state rs_nuse; 483 484 struct cdev cdev; 485 struct device cdev_device; 486 487 struct gendisk *disk; 488 #ifdef CONFIG_NVME_MULTIPATH 489 struct bio_list requeue_list; 490 spinlock_t requeue_lock; 491 struct work_struct requeue_work; 492 struct mutex lock; 493 unsigned long flags; 494 #define NVME_NSHEAD_DISK_LIVE 0 495 struct nvme_ns __rcu *current_path[]; 496 #endif 497 }; 498 499 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head) 500 { 501 return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk; 502 } 503 504 enum nvme_ns_features { 505 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */ 506 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */ 507 NVME_NS_DEAC = 1 << 2, /* DEAC bit in Write Zeores supported */ 508 }; 509 510 struct nvme_ns { 511 struct list_head list; 512 513 struct nvme_ctrl *ctrl; 514 struct request_queue *queue; 515 struct gendisk *disk; 516 #ifdef CONFIG_NVME_MULTIPATH 517 enum nvme_ana_state ana_state; 518 u32 ana_grpid; 519 #endif 520 struct list_head siblings; 521 struct kref kref; 522 struct nvme_ns_head *head; 523 524 unsigned long flags; 525 #define NVME_NS_REMOVING 0 526 #define NVME_NS_ANA_PENDING 2 527 #define NVME_NS_FORCE_RO 3 528 #define NVME_NS_READY 4 529 530 struct cdev cdev; 531 struct device cdev_device; 532 533 struct nvme_fault_inject fault_inject; 534 }; 535 536 /* NVMe ns supports metadata actions by the controller (generate/strip) */ 537 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head) 538 { 539 return head->pi_type && head->ms == head->pi_size; 540 } 541 542 struct nvme_ctrl_ops { 543 const char *name; 544 struct module *module; 545 unsigned int flags; 546 #define NVME_F_FABRICS (1 << 0) 547 #define NVME_F_METADATA_SUPPORTED (1 << 1) 548 #define NVME_F_BLOCKING (1 << 2) 549 550 const struct attribute_group **dev_attr_groups; 551 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val); 552 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val); 553 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val); 554 void (*free_ctrl)(struct nvme_ctrl *ctrl); 555 void (*submit_async_event)(struct nvme_ctrl *ctrl); 556 int (*subsystem_reset)(struct nvme_ctrl *ctrl); 557 void (*delete_ctrl)(struct nvme_ctrl *ctrl); 558 void (*stop_ctrl)(struct nvme_ctrl *ctrl); 559 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); 560 void (*print_device_info)(struct nvme_ctrl *ctrl); 561 bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl); 562 }; 563 564 /* 565 * nvme command_id is constructed as such: 566 * | xxxx | xxxxxxxxxxxx | 567 * gen request tag 568 */ 569 #define nvme_genctr_mask(gen) (gen & 0xf) 570 #define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12) 571 #define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12) 572 #define nvme_tag_from_cid(cid) (cid & 0xfff) 573 574 static inline u16 nvme_cid(struct request *rq) 575 { 576 return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag; 577 } 578 579 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags, 580 u16 command_id) 581 { 582 u8 genctr = nvme_genctr_from_cid(command_id); 583 u16 tag = nvme_tag_from_cid(command_id); 584 struct request *rq; 585 586 rq = blk_mq_tag_to_rq(tags, tag); 587 if (unlikely(!rq)) { 588 pr_err("could not locate request for tag %#x\n", 589 tag); 590 return NULL; 591 } 592 if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) { 593 dev_err(nvme_req(rq)->ctrl->device, 594 "request %#x genctr mismatch (got %#x expected %#x)\n", 595 tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr)); 596 return NULL; 597 } 598 return rq; 599 } 600 601 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags, 602 u16 command_id) 603 { 604 return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id)); 605 } 606 607 /* 608 * Return the length of the string without the space padding 609 */ 610 static inline int nvme_strlen(char *s, int len) 611 { 612 while (s[len - 1] == ' ') 613 len--; 614 return len; 615 } 616 617 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl) 618 { 619 struct nvme_subsystem *subsys = ctrl->subsys; 620 621 if (ctrl->ops->print_device_info) { 622 ctrl->ops->print_device_info(ctrl); 623 return; 624 } 625 626 dev_err(ctrl->device, 627 "VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id, 628 nvme_strlen(subsys->model, sizeof(subsys->model)), 629 subsys->model, nvme_strlen(subsys->firmware_rev, 630 sizeof(subsys->firmware_rev)), 631 subsys->firmware_rev); 632 } 633 634 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 635 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 636 const char *dev_name); 637 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject); 638 void nvme_should_fail(struct request *req); 639 #else 640 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 641 const char *dev_name) 642 { 643 } 644 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj) 645 { 646 } 647 static inline void nvme_should_fail(struct request *req) {} 648 #endif 649 650 bool nvme_wait_reset(struct nvme_ctrl *ctrl); 651 int nvme_try_sched_reset(struct nvme_ctrl *ctrl); 652 653 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl) 654 { 655 if (!ctrl->subsystem || !ctrl->ops->subsystem_reset) 656 return -ENOTTY; 657 return ctrl->ops->subsystem_reset(ctrl); 658 } 659 660 /* 661 * Convert a 512B sector number to a device logical block number. 662 */ 663 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector) 664 { 665 return sector >> (head->lba_shift - SECTOR_SHIFT); 666 } 667 668 /* 669 * Convert a device logical block number to a 512B sector number. 670 */ 671 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba) 672 { 673 return lba << (head->lba_shift - SECTOR_SHIFT); 674 } 675 676 /* 677 * Convert byte length to nvme's 0-based num dwords 678 */ 679 static inline u32 nvme_bytes_to_numd(size_t len) 680 { 681 return (len >> 2) - 1; 682 } 683 684 static inline bool nvme_is_ana_error(u16 status) 685 { 686 switch (status & NVME_SCT_SC_MASK) { 687 case NVME_SC_ANA_TRANSITION: 688 case NVME_SC_ANA_INACCESSIBLE: 689 case NVME_SC_ANA_PERSISTENT_LOSS: 690 return true; 691 default: 692 return false; 693 } 694 } 695 696 static inline bool nvme_is_path_error(u16 status) 697 { 698 /* check for a status code type of 'path related status' */ 699 return (status & NVME_SCT_MASK) == NVME_SCT_PATH; 700 } 701 702 /* 703 * Fill in the status and result information from the CQE, and then figure out 704 * if blk-mq will need to use IPI magic to complete the request, and if yes do 705 * so. If not let the caller complete the request without an indirect function 706 * call. 707 */ 708 static inline bool nvme_try_complete_req(struct request *req, __le16 status, 709 union nvme_result result) 710 { 711 struct nvme_request *rq = nvme_req(req); 712 struct nvme_ctrl *ctrl = rq->ctrl; 713 714 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN)) 715 rq->genctr++; 716 717 rq->status = le16_to_cpu(status) >> 1; 718 rq->result = result; 719 /* inject error when permitted by fault injection framework */ 720 nvme_should_fail(req); 721 if (unlikely(blk_should_fake_timeout(req->q))) 722 return true; 723 return blk_mq_complete_request_remote(req); 724 } 725 726 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl) 727 { 728 get_device(ctrl->device); 729 } 730 731 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl) 732 { 733 put_device(ctrl->device); 734 } 735 736 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id) 737 { 738 return !qid && 739 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH; 740 } 741 742 /* 743 * Returns true for sink states that can't ever transition back to live. 744 */ 745 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl) 746 { 747 switch (nvme_ctrl_state(ctrl)) { 748 case NVME_CTRL_NEW: 749 case NVME_CTRL_LIVE: 750 case NVME_CTRL_RESETTING: 751 case NVME_CTRL_CONNECTING: 752 return false; 753 case NVME_CTRL_DELETING: 754 case NVME_CTRL_DELETING_NOIO: 755 case NVME_CTRL_DEAD: 756 return true; 757 default: 758 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state); 759 return true; 760 } 761 } 762 763 void nvme_end_req(struct request *req); 764 void nvme_complete_rq(struct request *req); 765 void nvme_complete_batch_req(struct request *req); 766 767 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob, 768 void (*fn)(struct request *rq)) 769 { 770 struct request *req; 771 772 rq_list_for_each(&iob->req_list, req) { 773 fn(req); 774 nvme_complete_batch_req(req); 775 } 776 blk_mq_end_request_batch(iob); 777 } 778 779 blk_status_t nvme_host_path_error(struct request *req); 780 bool nvme_cancel_request(struct request *req, void *data); 781 void nvme_cancel_tagset(struct nvme_ctrl *ctrl); 782 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl); 783 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, 784 enum nvme_ctrl_state new_state); 785 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown); 786 int nvme_enable_ctrl(struct nvme_ctrl *ctrl); 787 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, 788 const struct nvme_ctrl_ops *ops, unsigned long quirks); 789 int nvme_add_ctrl(struct nvme_ctrl *ctrl); 790 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl); 791 void nvme_start_ctrl(struct nvme_ctrl *ctrl); 792 void nvme_stop_ctrl(struct nvme_ctrl *ctrl); 793 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended); 794 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 795 const struct blk_mq_ops *ops, unsigned int cmd_size); 796 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl); 797 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 798 const struct blk_mq_ops *ops, unsigned int nr_maps, 799 unsigned int cmd_size); 800 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl); 801 802 void nvme_remove_namespaces(struct nvme_ctrl *ctrl); 803 804 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, 805 volatile union nvme_result *res); 806 807 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl); 808 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl); 809 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl); 810 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl); 811 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl); 812 void nvme_sync_queues(struct nvme_ctrl *ctrl); 813 void nvme_sync_io_queues(struct nvme_ctrl *ctrl); 814 void nvme_unfreeze(struct nvme_ctrl *ctrl); 815 void nvme_wait_freeze(struct nvme_ctrl *ctrl); 816 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); 817 void nvme_start_freeze(struct nvme_ctrl *ctrl); 818 819 static inline enum req_op nvme_req_op(struct nvme_command *cmd) 820 { 821 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; 822 } 823 824 #define NVME_QID_ANY -1 825 void nvme_init_request(struct request *req, struct nvme_command *cmd); 826 void nvme_cleanup_cmd(struct request *req); 827 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req); 828 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl, 829 struct request *req); 830 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 831 bool queue_live, enum nvme_ctrl_state state); 832 833 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 834 bool queue_live) 835 { 836 enum nvme_ctrl_state state = nvme_ctrl_state(ctrl); 837 838 if (likely(state == NVME_CTRL_LIVE)) 839 return true; 840 if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING) 841 return queue_live; 842 return __nvme_check_ready(ctrl, rq, queue_live, state); 843 } 844 845 /* 846 * NSID shall be unique for all shared namespaces, or if at least one of the 847 * following conditions is met: 848 * 1. Namespace Management is supported by the controller 849 * 2. ANA is supported by the controller 850 * 3. NVM Set are supported by the controller 851 * 852 * In other case, private namespace are not required to report a unique NSID. 853 */ 854 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl, 855 struct nvme_ns_head *head) 856 { 857 return head->shared || 858 (ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) || 859 (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) || 860 (ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS); 861 } 862 863 /* 864 * Flags for __nvme_submit_sync_cmd() 865 */ 866 typedef __u32 __bitwise nvme_submit_flags_t; 867 868 enum { 869 /* Insert request at the head of the queue */ 870 NVME_SUBMIT_AT_HEAD = (__force nvme_submit_flags_t)(1 << 0), 871 /* Set BLK_MQ_REQ_NOWAIT when allocating request */ 872 NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1), 873 /* Set BLK_MQ_REQ_RESERVED when allocating request */ 874 NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2), 875 /* Retry command when NVME_STATUS_DNR is not set in the result */ 876 NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3), 877 }; 878 879 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 880 void *buf, unsigned bufflen); 881 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 882 union nvme_result *result, void *buffer, unsigned bufflen, 883 int qid, nvme_submit_flags_t flags); 884 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, 885 unsigned int dword11, void *buffer, size_t buflen, 886 u32 *result); 887 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, 888 unsigned int dword11, void *buffer, size_t buflen, 889 u32 *result); 890 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count); 891 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl); 892 int nvme_reset_ctrl(struct nvme_ctrl *ctrl); 893 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl); 894 int nvme_delete_ctrl(struct nvme_ctrl *ctrl); 895 void nvme_queue_scan(struct nvme_ctrl *ctrl); 896 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi, 897 void *log, size_t size, u64 offset); 898 bool nvme_tryget_ns_head(struct nvme_ns_head *head); 899 void nvme_put_ns_head(struct nvme_ns_head *head); 900 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device, 901 const struct file_operations *fops, struct module *owner); 902 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device); 903 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode, 904 unsigned int cmd, unsigned long arg); 905 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 906 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode, 907 unsigned int cmd, unsigned long arg); 908 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd, 909 unsigned long arg); 910 long nvme_dev_ioctl(struct file *file, unsigned int cmd, 911 unsigned long arg); 912 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd, 913 struct io_comp_batch *iob, unsigned int poll_flags); 914 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd, 915 unsigned int issue_flags); 916 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd, 917 unsigned int issue_flags); 918 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid, 919 struct nvme_id_ns **id); 920 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo); 921 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags); 922 923 extern const struct attribute_group *nvme_ns_attr_groups[]; 924 extern const struct pr_ops nvme_pr_ops; 925 extern const struct block_device_operations nvme_ns_head_ops; 926 extern const struct attribute_group nvme_dev_attrs_group; 927 extern const struct attribute_group *nvme_subsys_attrs_groups[]; 928 extern const struct attribute_group *nvme_dev_attr_groups[]; 929 extern const struct block_device_operations nvme_bdev_ops; 930 931 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl); 932 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head); 933 #ifdef CONFIG_NVME_MULTIPATH 934 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 935 { 936 return ctrl->ana_log_buf != NULL; 937 } 938 939 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys); 940 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys); 941 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys); 942 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys); 943 void nvme_failover_req(struct request *req); 944 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl); 945 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head); 946 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid); 947 void nvme_mpath_remove_disk(struct nvme_ns_head *head); 948 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); 949 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl); 950 void nvme_mpath_update(struct nvme_ctrl *ctrl); 951 void nvme_mpath_uninit(struct nvme_ctrl *ctrl); 952 void nvme_mpath_stop(struct nvme_ctrl *ctrl); 953 bool nvme_mpath_clear_current_path(struct nvme_ns *ns); 954 void nvme_mpath_revalidate_paths(struct nvme_ns *ns); 955 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl); 956 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head); 957 void nvme_mpath_start_request(struct request *rq); 958 void nvme_mpath_end_request(struct request *rq); 959 960 static inline void nvme_trace_bio_complete(struct request *req) 961 { 962 struct nvme_ns *ns = req->q->queuedata; 963 964 if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio) 965 trace_block_bio_complete(ns->head->disk->queue, req->bio); 966 } 967 968 extern bool multipath; 969 extern struct device_attribute dev_attr_ana_grpid; 970 extern struct device_attribute dev_attr_ana_state; 971 extern struct device_attribute subsys_attr_iopolicy; 972 973 static inline bool nvme_disk_is_ns_head(struct gendisk *disk) 974 { 975 return disk->fops == &nvme_ns_head_ops; 976 } 977 #else 978 #define multipath false 979 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 980 { 981 return false; 982 } 983 static inline void nvme_failover_req(struct request *req) 984 { 985 } 986 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl) 987 { 988 } 989 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, 990 struct nvme_ns_head *head) 991 { 992 return 0; 993 } 994 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid) 995 { 996 } 997 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head) 998 { 999 } 1000 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns) 1001 { 1002 return false; 1003 } 1004 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns) 1005 { 1006 } 1007 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl) 1008 { 1009 } 1010 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head) 1011 { 1012 } 1013 static inline void nvme_trace_bio_complete(struct request *req) 1014 { 1015 } 1016 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl) 1017 { 1018 } 1019 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, 1020 struct nvme_id_ctrl *id) 1021 { 1022 if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) 1023 dev_warn(ctrl->device, 1024 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n"); 1025 return 0; 1026 } 1027 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl) 1028 { 1029 } 1030 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl) 1031 { 1032 } 1033 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl) 1034 { 1035 } 1036 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys) 1037 { 1038 } 1039 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys) 1040 { 1041 } 1042 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys) 1043 { 1044 } 1045 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys) 1046 { 1047 } 1048 static inline void nvme_mpath_start_request(struct request *rq) 1049 { 1050 } 1051 static inline void nvme_mpath_end_request(struct request *rq) 1052 { 1053 } 1054 static inline bool nvme_disk_is_ns_head(struct gendisk *disk) 1055 { 1056 return false; 1057 } 1058 #endif /* CONFIG_NVME_MULTIPATH */ 1059 1060 int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16], 1061 enum blk_unique_id type); 1062 1063 struct nvme_zone_info { 1064 u64 zone_size; 1065 unsigned int max_open_zones; 1066 unsigned int max_active_zones; 1067 }; 1068 1069 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector, 1070 unsigned int nr_zones, report_zones_cb cb, void *data); 1071 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf, 1072 struct nvme_zone_info *zi); 1073 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim, 1074 struct nvme_zone_info *zi); 1075 #ifdef CONFIG_BLK_DEV_ZONED 1076 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req, 1077 struct nvme_command *cmnd, 1078 enum nvme_zone_mgmt_action action); 1079 #else 1080 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, 1081 struct request *req, struct nvme_command *cmnd, 1082 enum nvme_zone_mgmt_action action) 1083 { 1084 return BLK_STS_NOTSUPP; 1085 } 1086 #endif 1087 1088 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev) 1089 { 1090 struct gendisk *disk = dev_to_disk(dev); 1091 1092 WARN_ON(nvme_disk_is_ns_head(disk)); 1093 return disk->private_data; 1094 } 1095 1096 #ifdef CONFIG_NVME_HWMON 1097 int nvme_hwmon_init(struct nvme_ctrl *ctrl); 1098 void nvme_hwmon_exit(struct nvme_ctrl *ctrl); 1099 #else 1100 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl) 1101 { 1102 return 0; 1103 } 1104 1105 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl) 1106 { 1107 } 1108 #endif 1109 1110 static inline void nvme_start_request(struct request *rq) 1111 { 1112 if (rq->cmd_flags & REQ_NVME_MPATH) 1113 nvme_mpath_start_request(rq); 1114 blk_mq_start_request(rq); 1115 } 1116 1117 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl) 1118 { 1119 return ctrl->sgls & ((1 << 0) | (1 << 1)); 1120 } 1121 1122 #ifdef CONFIG_NVME_HOST_AUTH 1123 int __init nvme_init_auth(void); 1124 void __exit nvme_exit_auth(void); 1125 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl); 1126 void nvme_auth_stop(struct nvme_ctrl *ctrl); 1127 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid); 1128 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid); 1129 void nvme_auth_free(struct nvme_ctrl *ctrl); 1130 #else 1131 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl) 1132 { 1133 return 0; 1134 } 1135 static inline int __init nvme_init_auth(void) 1136 { 1137 return 0; 1138 } 1139 static inline void __exit nvme_exit_auth(void) 1140 { 1141 } 1142 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {}; 1143 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid) 1144 { 1145 return -EPROTONOSUPPORT; 1146 } 1147 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid) 1148 { 1149 return -EPROTONOSUPPORT; 1150 } 1151 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {}; 1152 #endif 1153 1154 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, 1155 u8 opcode); 1156 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode); 1157 int nvme_execute_rq(struct request *rq, bool at_head); 1158 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects, 1159 struct nvme_command *cmd, int status); 1160 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file); 1161 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid); 1162 bool nvme_get_ns(struct nvme_ns *ns); 1163 void nvme_put_ns(struct nvme_ns *ns); 1164 1165 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl) 1166 { 1167 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI; 1168 } 1169 1170 #ifdef CONFIG_NVME_VERBOSE_ERRORS 1171 const char *nvme_get_error_status_str(u16 status); 1172 const char *nvme_get_opcode_str(u8 opcode); 1173 const char *nvme_get_admin_opcode_str(u8 opcode); 1174 const char *nvme_get_fabrics_opcode_str(u8 opcode); 1175 #else /* CONFIG_NVME_VERBOSE_ERRORS */ 1176 static inline const char *nvme_get_error_status_str(u16 status) 1177 { 1178 return "I/O Error"; 1179 } 1180 static inline const char *nvme_get_opcode_str(u8 opcode) 1181 { 1182 return "I/O Cmd"; 1183 } 1184 static inline const char *nvme_get_admin_opcode_str(u8 opcode) 1185 { 1186 return "Admin Cmd"; 1187 } 1188 1189 static inline const char *nvme_get_fabrics_opcode_str(u8 opcode) 1190 { 1191 return "Fabrics Cmd"; 1192 } 1193 #endif /* CONFIG_NVME_VERBOSE_ERRORS */ 1194 1195 static inline const char *nvme_opcode_str(int qid, u8 opcode) 1196 { 1197 return qid ? nvme_get_opcode_str(opcode) : 1198 nvme_get_admin_opcode_str(opcode); 1199 } 1200 1201 static inline const char *nvme_fabrics_opcode_str( 1202 int qid, const struct nvme_command *cmd) 1203 { 1204 if (nvme_is_fabrics(cmd)) 1205 return nvme_get_fabrics_opcode_str(cmd->fabrics.fctype); 1206 1207 return nvme_opcode_str(qid, cmd->common.opcode); 1208 } 1209 #endif /* _NVME_H */ 1210