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 53 /* 54 * List of workarounds for devices that required behavior not specified in 55 * the standard. 56 */ 57 enum nvme_quirks { 58 /* 59 * Prefers I/O aligned to a stripe size specified in a vendor 60 * specific Identify field. 61 */ 62 NVME_QUIRK_STRIPE_SIZE = (1 << 0), 63 64 /* 65 * The controller doesn't handle Identify value others than 0 or 1 66 * correctly. 67 */ 68 NVME_QUIRK_IDENTIFY_CNS = (1 << 1), 69 70 /* 71 * The controller deterministically returns O's on reads to 72 * logical blocks that deallocate was called on. 73 */ 74 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2), 75 76 /* 77 * The controller needs a delay before starts checking the device 78 * readiness, which is done by reading the NVME_CSTS_RDY bit. 79 */ 80 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3), 81 82 /* 83 * APST should not be used. 84 */ 85 NVME_QUIRK_NO_APST = (1 << 4), 86 87 /* 88 * The deepest sleep state should not be used. 89 */ 90 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5), 91 92 /* 93 * Set MEDIUM priority on SQ creation 94 */ 95 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7), 96 97 /* 98 * Ignore device provided subnqn. 99 */ 100 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8), 101 102 /* 103 * Broken Write Zeroes. 104 */ 105 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9), 106 107 /* 108 * Force simple suspend/resume path. 109 */ 110 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10), 111 112 /* 113 * Use only one interrupt vector for all queues 114 */ 115 NVME_QUIRK_SINGLE_VECTOR = (1 << 11), 116 117 /* 118 * Use non-standard 128 bytes SQEs. 119 */ 120 NVME_QUIRK_128_BYTES_SQES = (1 << 12), 121 122 /* 123 * Prevent tag overlap between queues 124 */ 125 NVME_QUIRK_SHARED_TAGS = (1 << 13), 126 127 /* 128 * Don't change the value of the temperature threshold feature 129 */ 130 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14), 131 132 /* 133 * The controller doesn't handle the Identify Namespace 134 * Identification Descriptor list subcommand despite claiming 135 * NVMe 1.3 compliance. 136 */ 137 NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15), 138 139 /* 140 * The controller does not properly handle DMA addresses over 141 * 48 bits. 142 */ 143 NVME_QUIRK_DMA_ADDRESS_BITS_48 = (1 << 16), 144 145 /* 146 * The controller requires the command_id value be limited, so skip 147 * encoding the generation sequence number. 148 */ 149 NVME_QUIRK_SKIP_CID_GEN = (1 << 17), 150 151 /* 152 * Reports garbage in the namespace identifiers (eui64, nguid, uuid). 153 */ 154 NVME_QUIRK_BOGUS_NID = (1 << 18), 155 156 /* 157 * No temperature thresholds for channels other than 0 (Composite). 158 */ 159 NVME_QUIRK_NO_SECONDARY_TEMP_THRESH = (1 << 19), 160 161 /* 162 * Disables simple suspend/resume path. 163 */ 164 NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND = (1 << 20), 165 166 /* 167 * MSI (but not MSI-X) interrupts are broken and never fire. 168 */ 169 NVME_QUIRK_BROKEN_MSI = (1 << 21), 170 }; 171 172 /* 173 * Common request structure for NVMe passthrough. All drivers must have 174 * this structure as the first member of their request-private data. 175 */ 176 struct nvme_request { 177 struct nvme_command *cmd; 178 union nvme_result result; 179 u8 genctr; 180 u8 retries; 181 u8 flags; 182 u16 status; 183 #ifdef CONFIG_NVME_MULTIPATH 184 unsigned long start_time; 185 #endif 186 struct nvme_ctrl *ctrl; 187 }; 188 189 /* 190 * Mark a bio as coming in through the mpath node. 191 */ 192 #define REQ_NVME_MPATH REQ_DRV 193 194 enum { 195 NVME_REQ_CANCELLED = (1 << 0), 196 NVME_REQ_USERCMD = (1 << 1), 197 NVME_MPATH_IO_STATS = (1 << 2), 198 }; 199 200 static inline struct nvme_request *nvme_req(struct request *req) 201 { 202 return blk_mq_rq_to_pdu(req); 203 } 204 205 static inline u16 nvme_req_qid(struct request *req) 206 { 207 if (!req->q->queuedata) 208 return 0; 209 210 return req->mq_hctx->queue_num + 1; 211 } 212 213 /* The below value is the specific amount of delay needed before checking 214 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the 215 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was 216 * found empirically. 217 */ 218 #define NVME_QUIRK_DELAY_AMOUNT 2300 219 220 /* 221 * enum nvme_ctrl_state: Controller state 222 * 223 * @NVME_CTRL_NEW: New controller just allocated, initial state 224 * @NVME_CTRL_LIVE: Controller is connected and I/O capable 225 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset) 226 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the 227 * transport 228 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion) 229 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not 230 * disabled/failed immediately. This state comes 231 * after all async event processing took place and 232 * before ns removal and the controller deletion 233 * progress 234 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during 235 * shutdown or removal. In this case we forcibly 236 * kill all inflight I/O as they have no chance to 237 * complete 238 */ 239 enum nvme_ctrl_state { 240 NVME_CTRL_NEW, 241 NVME_CTRL_LIVE, 242 NVME_CTRL_RESETTING, 243 NVME_CTRL_CONNECTING, 244 NVME_CTRL_DELETING, 245 NVME_CTRL_DELETING_NOIO, 246 NVME_CTRL_DEAD, 247 }; 248 249 struct nvme_fault_inject { 250 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 251 struct fault_attr attr; 252 struct dentry *parent; 253 bool dont_retry; /* DNR, do not retry */ 254 u16 status; /* status code */ 255 #endif 256 }; 257 258 enum nvme_ctrl_flags { 259 NVME_CTRL_FAILFAST_EXPIRED = 0, 260 NVME_CTRL_ADMIN_Q_STOPPED = 1, 261 NVME_CTRL_STARTED_ONCE = 2, 262 NVME_CTRL_STOPPED = 3, 263 NVME_CTRL_SKIP_ID_CNS_CS = 4, 264 NVME_CTRL_DIRTY_CAPABILITY = 5, 265 NVME_CTRL_FROZEN = 6, 266 }; 267 268 struct nvme_ctrl { 269 bool comp_seen; 270 bool identified; 271 bool passthru_err_log_enabled; 272 enum nvme_ctrl_state state; 273 spinlock_t lock; 274 struct mutex scan_lock; 275 const struct nvme_ctrl_ops *ops; 276 struct request_queue *admin_q; 277 struct request_queue *connect_q; 278 struct request_queue *fabrics_q; 279 struct device *dev; 280 int instance; 281 int numa_node; 282 struct blk_mq_tag_set *tagset; 283 struct blk_mq_tag_set *admin_tagset; 284 struct list_head namespaces; 285 struct mutex namespaces_lock; 286 struct srcu_struct srcu; 287 struct device ctrl_device; 288 struct device *device; /* char device */ 289 #ifdef CONFIG_NVME_HWMON 290 struct device *hwmon_device; 291 #endif 292 struct cdev cdev; 293 struct work_struct reset_work; 294 struct work_struct delete_work; 295 wait_queue_head_t state_wq; 296 297 struct nvme_subsystem *subsys; 298 struct list_head subsys_entry; 299 300 struct opal_dev *opal_dev; 301 302 char name[12]; 303 u16 cntlid; 304 305 u16 mtfa; 306 u32 ctrl_config; 307 u32 queue_count; 308 309 u64 cap; 310 u32 max_hw_sectors; 311 u32 max_segments; 312 u32 max_integrity_segments; 313 u32 max_zeroes_sectors; 314 #ifdef CONFIG_BLK_DEV_ZONED 315 u32 max_zone_append; 316 #endif 317 u16 crdt[3]; 318 u16 oncs; 319 u8 dmrl; 320 u32 dmrsl; 321 u16 oacs; 322 u16 sqsize; 323 u32 max_namespaces; 324 atomic_t abort_limit; 325 u8 vwc; 326 u32 vs; 327 u32 sgls; 328 u16 kas; 329 u8 npss; 330 u8 apsta; 331 u16 wctemp; 332 u16 cctemp; 333 u32 oaes; 334 u32 aen_result; 335 u32 ctratt; 336 unsigned int shutdown_timeout; 337 unsigned int kato; 338 bool subsystem; 339 unsigned long quirks; 340 struct nvme_id_power_state psd[32]; 341 struct nvme_effects_log *effects; 342 struct xarray cels; 343 struct work_struct scan_work; 344 struct work_struct async_event_work; 345 struct delayed_work ka_work; 346 struct delayed_work failfast_work; 347 struct nvme_command ka_cmd; 348 unsigned long ka_last_check_time; 349 struct work_struct fw_act_work; 350 unsigned long events; 351 352 #ifdef CONFIG_NVME_MULTIPATH 353 /* asymmetric namespace access: */ 354 u8 anacap; 355 u8 anatt; 356 u32 anagrpmax; 357 u32 nanagrpid; 358 struct mutex ana_lock; 359 struct nvme_ana_rsp_hdr *ana_log_buf; 360 size_t ana_log_size; 361 struct timer_list anatt_timer; 362 struct work_struct ana_work; 363 #endif 364 365 #ifdef CONFIG_NVME_HOST_AUTH 366 struct work_struct dhchap_auth_work; 367 struct mutex dhchap_auth_mutex; 368 struct nvme_dhchap_queue_context *dhchap_ctxs; 369 struct nvme_dhchap_key *host_key; 370 struct nvme_dhchap_key *ctrl_key; 371 u16 transaction; 372 #endif 373 struct key *tls_key; 374 375 /* Power saving configuration */ 376 u64 ps_max_latency_us; 377 bool apst_enabled; 378 379 /* PCIe only: */ 380 u16 hmmaxd; 381 u32 hmpre; 382 u32 hmmin; 383 u32 hmminds; 384 385 /* Fabrics only */ 386 u32 ioccsz; 387 u32 iorcsz; 388 u16 icdoff; 389 u16 maxcmd; 390 int nr_reconnects; 391 unsigned long flags; 392 struct nvmf_ctrl_options *opts; 393 394 struct page *discard_page; 395 unsigned long discard_page_busy; 396 397 struct nvme_fault_inject fault_inject; 398 399 enum nvme_ctrl_type cntrltype; 400 enum nvme_dctype dctype; 401 }; 402 403 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl) 404 { 405 return READ_ONCE(ctrl->state); 406 } 407 408 enum nvme_iopolicy { 409 NVME_IOPOLICY_NUMA, 410 NVME_IOPOLICY_RR, 411 }; 412 413 struct nvme_subsystem { 414 int instance; 415 struct device dev; 416 /* 417 * Because we unregister the device on the last put we need 418 * a separate refcount. 419 */ 420 struct kref ref; 421 struct list_head entry; 422 struct mutex lock; 423 struct list_head ctrls; 424 struct list_head nsheads; 425 char subnqn[NVMF_NQN_SIZE]; 426 char serial[20]; 427 char model[40]; 428 char firmware_rev[8]; 429 u8 cmic; 430 enum nvme_subsys_type subtype; 431 u16 vendor_id; 432 u16 awupf; /* 0's based awupf value. */ 433 struct ida ns_ida; 434 #ifdef CONFIG_NVME_MULTIPATH 435 enum nvme_iopolicy iopolicy; 436 #endif 437 }; 438 439 /* 440 * Container structure for uniqueue namespace identifiers. 441 */ 442 struct nvme_ns_ids { 443 u8 eui64[8]; 444 u8 nguid[16]; 445 uuid_t uuid; 446 u8 csi; 447 }; 448 449 /* 450 * Anchor structure for namespaces. There is one for each namespace in a 451 * NVMe subsystem that any of our controllers can see, and the namespace 452 * structure for each controller is chained of it. For private namespaces 453 * there is a 1:1 relation to our namespace structures, that is ->list 454 * only ever has a single entry for private namespaces. 455 */ 456 struct nvme_ns_head { 457 struct list_head list; 458 struct srcu_struct srcu; 459 struct nvme_subsystem *subsys; 460 struct nvme_ns_ids ids; 461 struct list_head entry; 462 struct kref ref; 463 bool shared; 464 bool passthru_err_log_enabled; 465 int instance; 466 struct nvme_effects_log *effects; 467 u64 nuse; 468 unsigned ns_id; 469 int lba_shift; 470 u16 ms; 471 u16 pi_size; 472 u8 pi_type; 473 u8 pi_offset; 474 u8 guard_type; 475 #ifdef CONFIG_BLK_DEV_ZONED 476 u64 zsze; 477 #endif 478 unsigned long features; 479 480 struct ratelimit_state rs_nuse; 481 482 struct cdev cdev; 483 struct device cdev_device; 484 485 struct gendisk *disk; 486 #ifdef CONFIG_NVME_MULTIPATH 487 struct bio_list requeue_list; 488 spinlock_t requeue_lock; 489 struct work_struct requeue_work; 490 struct mutex lock; 491 unsigned long flags; 492 #define NVME_NSHEAD_DISK_LIVE 0 493 struct nvme_ns __rcu *current_path[]; 494 #endif 495 }; 496 497 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head) 498 { 499 return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk; 500 } 501 502 enum nvme_ns_features { 503 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */ 504 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */ 505 NVME_NS_DEAC, /* DEAC bit in Write Zeores supported */ 506 }; 507 508 struct nvme_ns { 509 struct list_head list; 510 511 struct nvme_ctrl *ctrl; 512 struct request_queue *queue; 513 struct gendisk *disk; 514 #ifdef CONFIG_NVME_MULTIPATH 515 enum nvme_ana_state ana_state; 516 u32 ana_grpid; 517 #endif 518 struct list_head siblings; 519 struct kref kref; 520 struct nvme_ns_head *head; 521 522 unsigned long flags; 523 #define NVME_NS_REMOVING 0 524 #define NVME_NS_ANA_PENDING 2 525 #define NVME_NS_FORCE_RO 3 526 #define NVME_NS_READY 4 527 528 struct cdev cdev; 529 struct device cdev_device; 530 531 struct nvme_fault_inject fault_inject; 532 }; 533 534 /* NVMe ns supports metadata actions by the controller (generate/strip) */ 535 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head) 536 { 537 return head->pi_type && head->ms == head->pi_size; 538 } 539 540 struct nvme_ctrl_ops { 541 const char *name; 542 struct module *module; 543 unsigned int flags; 544 #define NVME_F_FABRICS (1 << 0) 545 #define NVME_F_METADATA_SUPPORTED (1 << 1) 546 #define NVME_F_BLOCKING (1 << 2) 547 548 const struct attribute_group **dev_attr_groups; 549 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val); 550 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val); 551 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val); 552 void (*free_ctrl)(struct nvme_ctrl *ctrl); 553 void (*submit_async_event)(struct nvme_ctrl *ctrl); 554 void (*delete_ctrl)(struct nvme_ctrl *ctrl); 555 void (*stop_ctrl)(struct nvme_ctrl *ctrl); 556 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); 557 void (*print_device_info)(struct nvme_ctrl *ctrl); 558 bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl); 559 }; 560 561 /* 562 * nvme command_id is constructed as such: 563 * | xxxx | xxxxxxxxxxxx | 564 * gen request tag 565 */ 566 #define nvme_genctr_mask(gen) (gen & 0xf) 567 #define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12) 568 #define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12) 569 #define nvme_tag_from_cid(cid) (cid & 0xfff) 570 571 static inline u16 nvme_cid(struct request *rq) 572 { 573 return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag; 574 } 575 576 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags, 577 u16 command_id) 578 { 579 u8 genctr = nvme_genctr_from_cid(command_id); 580 u16 tag = nvme_tag_from_cid(command_id); 581 struct request *rq; 582 583 rq = blk_mq_tag_to_rq(tags, tag); 584 if (unlikely(!rq)) { 585 pr_err("could not locate request for tag %#x\n", 586 tag); 587 return NULL; 588 } 589 if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) { 590 dev_err(nvme_req(rq)->ctrl->device, 591 "request %#x genctr mismatch (got %#x expected %#x)\n", 592 tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr)); 593 return NULL; 594 } 595 return rq; 596 } 597 598 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags, 599 u16 command_id) 600 { 601 return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id)); 602 } 603 604 /* 605 * Return the length of the string without the space padding 606 */ 607 static inline int nvme_strlen(char *s, int len) 608 { 609 while (s[len - 1] == ' ') 610 len--; 611 return len; 612 } 613 614 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl) 615 { 616 struct nvme_subsystem *subsys = ctrl->subsys; 617 618 if (ctrl->ops->print_device_info) { 619 ctrl->ops->print_device_info(ctrl); 620 return; 621 } 622 623 dev_err(ctrl->device, 624 "VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id, 625 nvme_strlen(subsys->model, sizeof(subsys->model)), 626 subsys->model, nvme_strlen(subsys->firmware_rev, 627 sizeof(subsys->firmware_rev)), 628 subsys->firmware_rev); 629 } 630 631 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 632 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 633 const char *dev_name); 634 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject); 635 void nvme_should_fail(struct request *req); 636 #else 637 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, 638 const char *dev_name) 639 { 640 } 641 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj) 642 { 643 } 644 static inline void nvme_should_fail(struct request *req) {} 645 #endif 646 647 bool nvme_wait_reset(struct nvme_ctrl *ctrl); 648 int nvme_try_sched_reset(struct nvme_ctrl *ctrl); 649 650 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl) 651 { 652 int ret; 653 654 if (!ctrl->subsystem) 655 return -ENOTTY; 656 if (!nvme_wait_reset(ctrl)) 657 return -EBUSY; 658 659 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65); 660 if (ret) 661 return ret; 662 663 return nvme_try_sched_reset(ctrl); 664 } 665 666 /* 667 * Convert a 512B sector number to a device logical block number. 668 */ 669 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector) 670 { 671 return sector >> (head->lba_shift - SECTOR_SHIFT); 672 } 673 674 /* 675 * Convert a device logical block number to a 512B sector number. 676 */ 677 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba) 678 { 679 return lba << (head->lba_shift - SECTOR_SHIFT); 680 } 681 682 /* 683 * Convert byte length to nvme's 0-based num dwords 684 */ 685 static inline u32 nvme_bytes_to_numd(size_t len) 686 { 687 return (len >> 2) - 1; 688 } 689 690 static inline bool nvme_is_ana_error(u16 status) 691 { 692 switch (status & 0x7ff) { 693 case NVME_SC_ANA_TRANSITION: 694 case NVME_SC_ANA_INACCESSIBLE: 695 case NVME_SC_ANA_PERSISTENT_LOSS: 696 return true; 697 default: 698 return false; 699 } 700 } 701 702 static inline bool nvme_is_path_error(u16 status) 703 { 704 /* check for a status code type of 'path related status' */ 705 return (status & 0x700) == 0x300; 706 } 707 708 /* 709 * Fill in the status and result information from the CQE, and then figure out 710 * if blk-mq will need to use IPI magic to complete the request, and if yes do 711 * so. If not let the caller complete the request without an indirect function 712 * call. 713 */ 714 static inline bool nvme_try_complete_req(struct request *req, __le16 status, 715 union nvme_result result) 716 { 717 struct nvme_request *rq = nvme_req(req); 718 struct nvme_ctrl *ctrl = rq->ctrl; 719 720 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN)) 721 rq->genctr++; 722 723 rq->status = le16_to_cpu(status) >> 1; 724 rq->result = result; 725 /* inject error when permitted by fault injection framework */ 726 nvme_should_fail(req); 727 if (unlikely(blk_should_fake_timeout(req->q))) 728 return true; 729 return blk_mq_complete_request_remote(req); 730 } 731 732 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl) 733 { 734 get_device(ctrl->device); 735 } 736 737 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl) 738 { 739 put_device(ctrl->device); 740 } 741 742 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id) 743 { 744 return !qid && 745 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH; 746 } 747 748 /* 749 * Returns true for sink states that can't ever transition back to live. 750 */ 751 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl) 752 { 753 switch (nvme_ctrl_state(ctrl)) { 754 case NVME_CTRL_NEW: 755 case NVME_CTRL_LIVE: 756 case NVME_CTRL_RESETTING: 757 case NVME_CTRL_CONNECTING: 758 return false; 759 case NVME_CTRL_DELETING: 760 case NVME_CTRL_DELETING_NOIO: 761 case NVME_CTRL_DEAD: 762 return true; 763 default: 764 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state); 765 return true; 766 } 767 } 768 769 void nvme_end_req(struct request *req); 770 void nvme_complete_rq(struct request *req); 771 void nvme_complete_batch_req(struct request *req); 772 773 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob, 774 void (*fn)(struct request *rq)) 775 { 776 struct request *req; 777 778 rq_list_for_each(&iob->req_list, req) { 779 fn(req); 780 nvme_complete_batch_req(req); 781 } 782 blk_mq_end_request_batch(iob); 783 } 784 785 blk_status_t nvme_host_path_error(struct request *req); 786 bool nvme_cancel_request(struct request *req, void *data); 787 void nvme_cancel_tagset(struct nvme_ctrl *ctrl); 788 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl); 789 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, 790 enum nvme_ctrl_state new_state); 791 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown); 792 int nvme_enable_ctrl(struct nvme_ctrl *ctrl); 793 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, 794 const struct nvme_ctrl_ops *ops, unsigned long quirks); 795 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl); 796 void nvme_start_ctrl(struct nvme_ctrl *ctrl); 797 void nvme_stop_ctrl(struct nvme_ctrl *ctrl); 798 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended); 799 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 800 const struct blk_mq_ops *ops, unsigned int cmd_size); 801 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl); 802 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set, 803 const struct blk_mq_ops *ops, unsigned int nr_maps, 804 unsigned int cmd_size); 805 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl); 806 807 void nvme_remove_namespaces(struct nvme_ctrl *ctrl); 808 809 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, 810 volatile union nvme_result *res); 811 812 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl); 813 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl); 814 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl); 815 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl); 816 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl); 817 void nvme_sync_queues(struct nvme_ctrl *ctrl); 818 void nvme_sync_io_queues(struct nvme_ctrl *ctrl); 819 void nvme_unfreeze(struct nvme_ctrl *ctrl); 820 void nvme_wait_freeze(struct nvme_ctrl *ctrl); 821 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); 822 void nvme_start_freeze(struct nvme_ctrl *ctrl); 823 824 static inline enum req_op nvme_req_op(struct nvme_command *cmd) 825 { 826 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN; 827 } 828 829 #define NVME_QID_ANY -1 830 void nvme_init_request(struct request *req, struct nvme_command *cmd); 831 void nvme_cleanup_cmd(struct request *req); 832 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req); 833 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl, 834 struct request *req); 835 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 836 bool queue_live, enum nvme_ctrl_state state); 837 838 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq, 839 bool queue_live) 840 { 841 enum nvme_ctrl_state state = nvme_ctrl_state(ctrl); 842 843 if (likely(state == NVME_CTRL_LIVE)) 844 return true; 845 if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING) 846 return queue_live; 847 return __nvme_check_ready(ctrl, rq, queue_live, state); 848 } 849 850 /* 851 * NSID shall be unique for all shared namespaces, or if at least one of the 852 * following conditions is met: 853 * 1. Namespace Management is supported by the controller 854 * 2. ANA is supported by the controller 855 * 3. NVM Set are supported by the controller 856 * 857 * In other case, private namespace are not required to report a unique NSID. 858 */ 859 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl, 860 struct nvme_ns_head *head) 861 { 862 return head->shared || 863 (ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) || 864 (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) || 865 (ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS); 866 } 867 868 /* 869 * Flags for __nvme_submit_sync_cmd() 870 */ 871 typedef __u32 __bitwise nvme_submit_flags_t; 872 873 enum { 874 /* Insert request at the head of the queue */ 875 NVME_SUBMIT_AT_HEAD = (__force nvme_submit_flags_t)(1 << 0), 876 /* Set BLK_MQ_REQ_NOWAIT when allocating request */ 877 NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1), 878 /* Set BLK_MQ_REQ_RESERVED when allocating request */ 879 NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2), 880 /* Retry command when NVME_SC_DNR is not set in the result */ 881 NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3), 882 }; 883 884 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 885 void *buf, unsigned bufflen); 886 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 887 union nvme_result *result, void *buffer, unsigned bufflen, 888 int qid, nvme_submit_flags_t flags); 889 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, 890 unsigned int dword11, void *buffer, size_t buflen, 891 u32 *result); 892 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, 893 unsigned int dword11, void *buffer, size_t buflen, 894 u32 *result); 895 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count); 896 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl); 897 int nvme_reset_ctrl(struct nvme_ctrl *ctrl); 898 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl); 899 int nvme_delete_ctrl(struct nvme_ctrl *ctrl); 900 void nvme_queue_scan(struct nvme_ctrl *ctrl); 901 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi, 902 void *log, size_t size, u64 offset); 903 bool nvme_tryget_ns_head(struct nvme_ns_head *head); 904 void nvme_put_ns_head(struct nvme_ns_head *head); 905 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device, 906 const struct file_operations *fops, struct module *owner); 907 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device); 908 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode, 909 unsigned int cmd, unsigned long arg); 910 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg); 911 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode, 912 unsigned int cmd, unsigned long arg); 913 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd, 914 unsigned long arg); 915 long nvme_dev_ioctl(struct file *file, unsigned int cmd, 916 unsigned long arg); 917 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd, 918 struct io_comp_batch *iob, unsigned int poll_flags); 919 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd, 920 unsigned int issue_flags); 921 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd, 922 unsigned int issue_flags); 923 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid, 924 struct nvme_id_ns **id); 925 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo); 926 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags); 927 928 extern const struct attribute_group *nvme_ns_attr_groups[]; 929 extern const struct pr_ops nvme_pr_ops; 930 extern const struct block_device_operations nvme_ns_head_ops; 931 extern const struct attribute_group nvme_dev_attrs_group; 932 extern const struct attribute_group *nvme_subsys_attrs_groups[]; 933 extern const struct attribute_group *nvme_dev_attr_groups[]; 934 extern const struct block_device_operations nvme_bdev_ops; 935 936 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl); 937 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head); 938 #ifdef CONFIG_NVME_MULTIPATH 939 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 940 { 941 return ctrl->ana_log_buf != NULL; 942 } 943 944 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys); 945 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys); 946 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys); 947 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys); 948 void nvme_failover_req(struct request *req); 949 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl); 950 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head); 951 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid); 952 void nvme_mpath_remove_disk(struct nvme_ns_head *head); 953 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); 954 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl); 955 void nvme_mpath_update(struct nvme_ctrl *ctrl); 956 void nvme_mpath_uninit(struct nvme_ctrl *ctrl); 957 void nvme_mpath_stop(struct nvme_ctrl *ctrl); 958 bool nvme_mpath_clear_current_path(struct nvme_ns *ns); 959 void nvme_mpath_revalidate_paths(struct nvme_ns *ns); 960 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl); 961 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head); 962 void nvme_mpath_start_request(struct request *rq); 963 void nvme_mpath_end_request(struct request *rq); 964 965 static inline void nvme_trace_bio_complete(struct request *req) 966 { 967 struct nvme_ns *ns = req->q->queuedata; 968 969 if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio) 970 trace_block_bio_complete(ns->head->disk->queue, req->bio); 971 } 972 973 extern bool multipath; 974 extern struct device_attribute dev_attr_ana_grpid; 975 extern struct device_attribute dev_attr_ana_state; 976 extern struct device_attribute subsys_attr_iopolicy; 977 978 static inline bool nvme_disk_is_ns_head(struct gendisk *disk) 979 { 980 return disk->fops == &nvme_ns_head_ops; 981 } 982 #else 983 #define multipath false 984 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 985 { 986 return false; 987 } 988 static inline void nvme_failover_req(struct request *req) 989 { 990 } 991 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl) 992 { 993 } 994 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, 995 struct nvme_ns_head *head) 996 { 997 return 0; 998 } 999 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid) 1000 { 1001 } 1002 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head) 1003 { 1004 } 1005 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns) 1006 { 1007 return false; 1008 } 1009 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns) 1010 { 1011 } 1012 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl) 1013 { 1014 } 1015 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head) 1016 { 1017 } 1018 static inline void nvme_trace_bio_complete(struct request *req) 1019 { 1020 } 1021 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl) 1022 { 1023 } 1024 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, 1025 struct nvme_id_ctrl *id) 1026 { 1027 if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) 1028 dev_warn(ctrl->device, 1029 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n"); 1030 return 0; 1031 } 1032 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl) 1033 { 1034 } 1035 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl) 1036 { 1037 } 1038 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl) 1039 { 1040 } 1041 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys) 1042 { 1043 } 1044 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys) 1045 { 1046 } 1047 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys) 1048 { 1049 } 1050 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys) 1051 { 1052 } 1053 static inline void nvme_mpath_start_request(struct request *rq) 1054 { 1055 } 1056 static inline void nvme_mpath_end_request(struct request *rq) 1057 { 1058 } 1059 static inline bool nvme_disk_is_ns_head(struct gendisk *disk) 1060 { 1061 return false; 1062 } 1063 #endif /* CONFIG_NVME_MULTIPATH */ 1064 1065 struct nvme_zone_info { 1066 u64 zone_size; 1067 unsigned int max_open_zones; 1068 unsigned int max_active_zones; 1069 }; 1070 1071 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector, 1072 unsigned int nr_zones, report_zones_cb cb, void *data); 1073 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf, 1074 struct nvme_zone_info *zi); 1075 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim, 1076 struct nvme_zone_info *zi); 1077 #ifdef CONFIG_BLK_DEV_ZONED 1078 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req, 1079 struct nvme_command *cmnd, 1080 enum nvme_zone_mgmt_action action); 1081 #else 1082 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, 1083 struct request *req, struct nvme_command *cmnd, 1084 enum nvme_zone_mgmt_action action) 1085 { 1086 return BLK_STS_NOTSUPP; 1087 } 1088 #endif 1089 1090 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev) 1091 { 1092 struct gendisk *disk = dev_to_disk(dev); 1093 1094 WARN_ON(nvme_disk_is_ns_head(disk)); 1095 return disk->private_data; 1096 } 1097 1098 #ifdef CONFIG_NVME_HWMON 1099 int nvme_hwmon_init(struct nvme_ctrl *ctrl); 1100 void nvme_hwmon_exit(struct nvme_ctrl *ctrl); 1101 #else 1102 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl) 1103 { 1104 return 0; 1105 } 1106 1107 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl) 1108 { 1109 } 1110 #endif 1111 1112 static inline void nvme_start_request(struct request *rq) 1113 { 1114 if (rq->cmd_flags & REQ_NVME_MPATH) 1115 nvme_mpath_start_request(rq); 1116 blk_mq_start_request(rq); 1117 } 1118 1119 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl) 1120 { 1121 return ctrl->sgls & ((1 << 0) | (1 << 1)); 1122 } 1123 1124 #ifdef CONFIG_NVME_HOST_AUTH 1125 int __init nvme_init_auth(void); 1126 void __exit nvme_exit_auth(void); 1127 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl); 1128 void nvme_auth_stop(struct nvme_ctrl *ctrl); 1129 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid); 1130 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid); 1131 void nvme_auth_free(struct nvme_ctrl *ctrl); 1132 #else 1133 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl) 1134 { 1135 return 0; 1136 } 1137 static inline int __init nvme_init_auth(void) 1138 { 1139 return 0; 1140 } 1141 static inline void __exit nvme_exit_auth(void) 1142 { 1143 } 1144 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {}; 1145 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid) 1146 { 1147 return -EPROTONOSUPPORT; 1148 } 1149 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid) 1150 { 1151 return -EPROTONOSUPPORT; 1152 } 1153 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {}; 1154 #endif 1155 1156 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, 1157 u8 opcode); 1158 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode); 1159 int nvme_execute_rq(struct request *rq, bool at_head); 1160 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects, 1161 struct nvme_command *cmd, int status); 1162 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file); 1163 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid); 1164 bool nvme_get_ns(struct nvme_ns *ns); 1165 void nvme_put_ns(struct nvme_ns *ns); 1166 1167 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl) 1168 { 1169 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI; 1170 } 1171 1172 #ifdef CONFIG_NVME_VERBOSE_ERRORS 1173 const char *nvme_get_error_status_str(u16 status); 1174 const char *nvme_get_opcode_str(u8 opcode); 1175 const char *nvme_get_admin_opcode_str(u8 opcode); 1176 const char *nvme_get_fabrics_opcode_str(u8 opcode); 1177 #else /* CONFIG_NVME_VERBOSE_ERRORS */ 1178 static inline const char *nvme_get_error_status_str(u16 status) 1179 { 1180 return "I/O Error"; 1181 } 1182 static inline const char *nvme_get_opcode_str(u8 opcode) 1183 { 1184 return "I/O Cmd"; 1185 } 1186 static inline const char *nvme_get_admin_opcode_str(u8 opcode) 1187 { 1188 return "Admin Cmd"; 1189 } 1190 1191 static inline const char *nvme_get_fabrics_opcode_str(u8 opcode) 1192 { 1193 return "Fabrics Cmd"; 1194 } 1195 #endif /* CONFIG_NVME_VERBOSE_ERRORS */ 1196 1197 static inline const char *nvme_opcode_str(int qid, u8 opcode) 1198 { 1199 return qid ? nvme_get_opcode_str(opcode) : 1200 nvme_get_admin_opcode_str(opcode); 1201 } 1202 1203 static inline const char *nvme_fabrics_opcode_str( 1204 int qid, const struct nvme_command *cmd) 1205 { 1206 if (nvme_is_fabrics(cmd)) 1207 return nvme_get_fabrics_opcode_str(cmd->fabrics.fctype); 1208 1209 return nvme_opcode_str(qid, cmd->common.opcode); 1210 } 1211 #endif /* _NVME_H */ 1212