1 /* 2 * Copyright (c) 2011-2014, Intel Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 */ 13 14 #ifndef _NVME_H 15 #define _NVME_H 16 17 #include <linux/nvme.h> 18 #include <linux/cdev.h> 19 #include <linux/pci.h> 20 #include <linux/kref.h> 21 #include <linux/blk-mq.h> 22 #include <linux/lightnvm.h> 23 #include <linux/sed-opal.h> 24 #include <linux/fault-inject.h> 25 #include <linux/rcupdate.h> 26 27 extern unsigned int nvme_io_timeout; 28 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ) 29 30 extern unsigned int admin_timeout; 31 #define ADMIN_TIMEOUT (admin_timeout * HZ) 32 33 #define NVME_DEFAULT_KATO 5 34 #define NVME_KATO_GRACE 10 35 36 extern struct workqueue_struct *nvme_wq; 37 extern struct workqueue_struct *nvme_reset_wq; 38 extern struct workqueue_struct *nvme_delete_wq; 39 40 enum { 41 NVME_NS_LBA = 0, 42 NVME_NS_LIGHTNVM = 1, 43 }; 44 45 /* 46 * List of workarounds for devices that required behavior not specified in 47 * the standard. 48 */ 49 enum nvme_quirks { 50 /* 51 * Prefers I/O aligned to a stripe size specified in a vendor 52 * specific Identify field. 53 */ 54 NVME_QUIRK_STRIPE_SIZE = (1 << 0), 55 56 /* 57 * The controller doesn't handle Identify value others than 0 or 1 58 * correctly. 59 */ 60 NVME_QUIRK_IDENTIFY_CNS = (1 << 1), 61 62 /* 63 * The controller deterministically returns O's on reads to 64 * logical blocks that deallocate was called on. 65 */ 66 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2), 67 68 /* 69 * The controller needs a delay before starts checking the device 70 * readiness, which is done by reading the NVME_CSTS_RDY bit. 71 */ 72 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3), 73 74 /* 75 * APST should not be used. 76 */ 77 NVME_QUIRK_NO_APST = (1 << 4), 78 79 /* 80 * The deepest sleep state should not be used. 81 */ 82 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5), 83 84 /* 85 * Supports the LighNVM command set if indicated in vs[1]. 86 */ 87 NVME_QUIRK_LIGHTNVM = (1 << 6), 88 89 /* 90 * Set MEDIUM priority on SQ creation 91 */ 92 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7), 93 }; 94 95 /* 96 * Common request structure for NVMe passthrough. All drivers must have 97 * this structure as the first member of their request-private data. 98 */ 99 struct nvme_request { 100 struct nvme_command *cmd; 101 union nvme_result result; 102 u8 retries; 103 u8 flags; 104 u16 status; 105 struct nvme_ctrl *ctrl; 106 }; 107 108 /* 109 * Mark a bio as coming in through the mpath node. 110 */ 111 #define REQ_NVME_MPATH REQ_DRV 112 113 enum { 114 NVME_REQ_CANCELLED = (1 << 0), 115 NVME_REQ_USERCMD = (1 << 1), 116 }; 117 118 static inline struct nvme_request *nvme_req(struct request *req) 119 { 120 return blk_mq_rq_to_pdu(req); 121 } 122 123 static inline u16 nvme_req_qid(struct request *req) 124 { 125 if (!req->rq_disk) 126 return 0; 127 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1; 128 } 129 130 /* The below value is the specific amount of delay needed before checking 131 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the 132 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was 133 * found empirically. 134 */ 135 #define NVME_QUIRK_DELAY_AMOUNT 2300 136 137 enum nvme_ctrl_state { 138 NVME_CTRL_NEW, 139 NVME_CTRL_LIVE, 140 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */ 141 NVME_CTRL_RESETTING, 142 NVME_CTRL_CONNECTING, 143 NVME_CTRL_DELETING, 144 NVME_CTRL_DEAD, 145 }; 146 147 struct nvme_ctrl { 148 enum nvme_ctrl_state state; 149 bool identified; 150 spinlock_t lock; 151 const struct nvme_ctrl_ops *ops; 152 struct request_queue *admin_q; 153 struct request_queue *connect_q; 154 struct device *dev; 155 int instance; 156 struct blk_mq_tag_set *tagset; 157 struct blk_mq_tag_set *admin_tagset; 158 struct list_head namespaces; 159 struct rw_semaphore namespaces_rwsem; 160 struct device ctrl_device; 161 struct device *device; /* char device */ 162 struct cdev cdev; 163 struct work_struct reset_work; 164 struct work_struct delete_work; 165 166 struct nvme_subsystem *subsys; 167 struct list_head subsys_entry; 168 169 struct opal_dev *opal_dev; 170 171 char name[12]; 172 u16 cntlid; 173 174 u32 ctrl_config; 175 u16 mtfa; 176 u32 queue_count; 177 178 u64 cap; 179 u32 page_size; 180 u32 max_hw_sectors; 181 u32 max_segments; 182 u16 oncs; 183 u16 oacs; 184 u16 nssa; 185 u16 nr_streams; 186 u32 max_namespaces; 187 atomic_t abort_limit; 188 u8 vwc; 189 u32 vs; 190 u32 sgls; 191 u16 kas; 192 u8 npss; 193 u8 apsta; 194 u32 oaes; 195 u32 aen_result; 196 unsigned int shutdown_timeout; 197 unsigned int kato; 198 bool subsystem; 199 unsigned long quirks; 200 struct nvme_id_power_state psd[32]; 201 struct nvme_effects_log *effects; 202 struct work_struct scan_work; 203 struct work_struct async_event_work; 204 struct delayed_work ka_work; 205 struct nvme_command ka_cmd; 206 struct work_struct fw_act_work; 207 unsigned long events; 208 209 #ifdef CONFIG_NVME_MULTIPATH 210 /* asymmetric namespace access: */ 211 u8 anacap; 212 u8 anatt; 213 u32 anagrpmax; 214 u32 nanagrpid; 215 struct mutex ana_lock; 216 struct nvme_ana_rsp_hdr *ana_log_buf; 217 size_t ana_log_size; 218 struct timer_list anatt_timer; 219 struct work_struct ana_work; 220 #endif 221 222 /* Power saving configuration */ 223 u64 ps_max_latency_us; 224 bool apst_enabled; 225 226 /* PCIe only: */ 227 u32 hmpre; 228 u32 hmmin; 229 u32 hmminds; 230 u16 hmmaxd; 231 232 /* Fabrics only */ 233 u16 sqsize; 234 u32 ioccsz; 235 u32 iorcsz; 236 u16 icdoff; 237 u16 maxcmd; 238 int nr_reconnects; 239 struct nvmf_ctrl_options *opts; 240 }; 241 242 struct nvme_subsystem { 243 int instance; 244 struct device dev; 245 /* 246 * Because we unregister the device on the last put we need 247 * a separate refcount. 248 */ 249 struct kref ref; 250 struct list_head entry; 251 struct mutex lock; 252 struct list_head ctrls; 253 struct list_head nsheads; 254 char subnqn[NVMF_NQN_SIZE]; 255 char serial[20]; 256 char model[40]; 257 char firmware_rev[8]; 258 u8 cmic; 259 u16 vendor_id; 260 struct ida ns_ida; 261 }; 262 263 /* 264 * Container structure for uniqueue namespace identifiers. 265 */ 266 struct nvme_ns_ids { 267 u8 eui64[8]; 268 u8 nguid[16]; 269 uuid_t uuid; 270 }; 271 272 /* 273 * Anchor structure for namespaces. There is one for each namespace in a 274 * NVMe subsystem that any of our controllers can see, and the namespace 275 * structure for each controller is chained of it. For private namespaces 276 * there is a 1:1 relation to our namespace structures, that is ->list 277 * only ever has a single entry for private namespaces. 278 */ 279 struct nvme_ns_head { 280 #ifdef CONFIG_NVME_MULTIPATH 281 struct gendisk *disk; 282 struct nvme_ns __rcu *current_path; 283 struct bio_list requeue_list; 284 spinlock_t requeue_lock; 285 struct work_struct requeue_work; 286 struct mutex lock; 287 #endif 288 struct list_head list; 289 struct srcu_struct srcu; 290 struct nvme_subsystem *subsys; 291 unsigned ns_id; 292 struct nvme_ns_ids ids; 293 struct list_head entry; 294 struct kref ref; 295 int instance; 296 }; 297 298 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 299 struct nvme_fault_inject { 300 struct fault_attr attr; 301 struct dentry *parent; 302 bool dont_retry; /* DNR, do not retry */ 303 u16 status; /* status code */ 304 }; 305 #endif 306 307 struct nvme_ns { 308 struct list_head list; 309 310 struct nvme_ctrl *ctrl; 311 struct request_queue *queue; 312 struct gendisk *disk; 313 #ifdef CONFIG_NVME_MULTIPATH 314 enum nvme_ana_state ana_state; 315 u32 ana_grpid; 316 #endif 317 struct list_head siblings; 318 struct nvm_dev *ndev; 319 struct kref kref; 320 struct nvme_ns_head *head; 321 322 int lba_shift; 323 u16 ms; 324 u16 sgs; 325 u32 sws; 326 bool ext; 327 u8 pi_type; 328 unsigned long flags; 329 #define NVME_NS_REMOVING 0 330 #define NVME_NS_DEAD 1 331 #define NVME_NS_ANA_PENDING 2 332 u16 noiob; 333 334 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 335 struct nvme_fault_inject fault_inject; 336 #endif 337 338 }; 339 340 struct nvme_ctrl_ops { 341 const char *name; 342 struct module *module; 343 unsigned int flags; 344 #define NVME_F_FABRICS (1 << 0) 345 #define NVME_F_METADATA_SUPPORTED (1 << 1) 346 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val); 347 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val); 348 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val); 349 void (*free_ctrl)(struct nvme_ctrl *ctrl); 350 void (*submit_async_event)(struct nvme_ctrl *ctrl); 351 void (*delete_ctrl)(struct nvme_ctrl *ctrl); 352 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); 353 void (*stop_ctrl)(struct nvme_ctrl *ctrl); 354 }; 355 356 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS 357 void nvme_fault_inject_init(struct nvme_ns *ns); 358 void nvme_fault_inject_fini(struct nvme_ns *ns); 359 void nvme_should_fail(struct request *req); 360 #else 361 static inline void nvme_fault_inject_init(struct nvme_ns *ns) {} 362 static inline void nvme_fault_inject_fini(struct nvme_ns *ns) {} 363 static inline void nvme_should_fail(struct request *req) {} 364 #endif 365 366 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl) 367 { 368 u32 val = 0; 369 370 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val)) 371 return false; 372 return val & NVME_CSTS_RDY; 373 } 374 375 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl) 376 { 377 if (!ctrl->subsystem) 378 return -ENOTTY; 379 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65); 380 } 381 382 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector) 383 { 384 return (sector >> (ns->lba_shift - 9)); 385 } 386 387 static inline void nvme_end_request(struct request *req, __le16 status, 388 union nvme_result result) 389 { 390 struct nvme_request *rq = nvme_req(req); 391 392 rq->status = le16_to_cpu(status) >> 1; 393 rq->result = result; 394 /* inject error when permitted by fault injection framework */ 395 nvme_should_fail(req); 396 blk_mq_complete_request(req); 397 } 398 399 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl) 400 { 401 get_device(ctrl->device); 402 } 403 404 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl) 405 { 406 put_device(ctrl->device); 407 } 408 409 void nvme_complete_rq(struct request *req); 410 void nvme_cancel_request(struct request *req, void *data, bool reserved); 411 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, 412 enum nvme_ctrl_state new_state); 413 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap); 414 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap); 415 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl); 416 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, 417 const struct nvme_ctrl_ops *ops, unsigned long quirks); 418 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl); 419 void nvme_start_ctrl(struct nvme_ctrl *ctrl); 420 void nvme_stop_ctrl(struct nvme_ctrl *ctrl); 421 void nvme_put_ctrl(struct nvme_ctrl *ctrl); 422 int nvme_init_identify(struct nvme_ctrl *ctrl); 423 424 void nvme_remove_namespaces(struct nvme_ctrl *ctrl); 425 426 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len, 427 bool send); 428 429 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status, 430 volatile union nvme_result *res); 431 432 void nvme_stop_queues(struct nvme_ctrl *ctrl); 433 void nvme_start_queues(struct nvme_ctrl *ctrl); 434 void nvme_kill_queues(struct nvme_ctrl *ctrl); 435 void nvme_unfreeze(struct nvme_ctrl *ctrl); 436 void nvme_wait_freeze(struct nvme_ctrl *ctrl); 437 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); 438 void nvme_start_freeze(struct nvme_ctrl *ctrl); 439 440 #define NVME_QID_ANY -1 441 struct request *nvme_alloc_request(struct request_queue *q, 442 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid); 443 void nvme_cleanup_cmd(struct request *req); 444 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req, 445 struct nvme_command *cmd); 446 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 447 void *buf, unsigned bufflen); 448 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, 449 union nvme_result *result, void *buffer, unsigned bufflen, 450 unsigned timeout, int qid, int at_head, 451 blk_mq_req_flags_t flags); 452 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count); 453 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl); 454 int nvme_reset_ctrl(struct nvme_ctrl *ctrl); 455 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl); 456 int nvme_delete_ctrl(struct nvme_ctrl *ctrl); 457 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl); 458 459 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, 460 void *log, size_t size, u64 offset); 461 462 extern const struct attribute_group nvme_ns_id_attr_group; 463 extern const struct block_device_operations nvme_ns_head_ops; 464 465 #ifdef CONFIG_NVME_MULTIPATH 466 bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl); 467 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns, 468 struct nvme_ctrl *ctrl, int *flags); 469 void nvme_failover_req(struct request *req); 470 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl); 471 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head); 472 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id); 473 void nvme_mpath_remove_disk(struct nvme_ns_head *head); 474 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); 475 void nvme_mpath_uninit(struct nvme_ctrl *ctrl); 476 void nvme_mpath_stop(struct nvme_ctrl *ctrl); 477 478 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns) 479 { 480 struct nvme_ns_head *head = ns->head; 481 482 if (head && ns == rcu_access_pointer(head->current_path)) 483 rcu_assign_pointer(head->current_path, NULL); 484 } 485 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head); 486 487 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns) 488 { 489 struct nvme_ns_head *head = ns->head; 490 491 if (head->disk && list_empty(&head->list)) 492 kblockd_schedule_work(&head->requeue_work); 493 } 494 495 extern struct device_attribute dev_attr_ana_grpid; 496 extern struct device_attribute dev_attr_ana_state; 497 498 #else 499 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) 500 { 501 return false; 502 } 503 /* 504 * Without the multipath code enabled, multiple controller per subsystems are 505 * visible as devices and thus we cannot use the subsystem instance. 506 */ 507 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns, 508 struct nvme_ctrl *ctrl, int *flags) 509 { 510 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance); 511 } 512 513 static inline void nvme_failover_req(struct request *req) 514 { 515 } 516 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl) 517 { 518 } 519 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl, 520 struct nvme_ns_head *head) 521 { 522 return 0; 523 } 524 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, 525 struct nvme_id_ns *id) 526 { 527 } 528 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head) 529 { 530 } 531 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns) 532 { 533 } 534 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns) 535 { 536 } 537 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl, 538 struct nvme_id_ctrl *id) 539 { 540 return 0; 541 } 542 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl) 543 { 544 } 545 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl) 546 { 547 } 548 #endif /* CONFIG_NVME_MULTIPATH */ 549 550 #ifdef CONFIG_NVM 551 void nvme_nvm_update_nvm_info(struct nvme_ns *ns); 552 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node); 553 void nvme_nvm_unregister(struct nvme_ns *ns); 554 int nvme_nvm_register_sysfs(struct nvme_ns *ns); 555 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns); 556 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg); 557 #else 558 static inline void nvme_nvm_update_nvm_info(struct nvme_ns *ns) {}; 559 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, 560 int node) 561 { 562 return 0; 563 } 564 565 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {}; 566 static inline int nvme_nvm_register_sysfs(struct nvme_ns *ns) 567 { 568 return 0; 569 } 570 static inline void nvme_nvm_unregister_sysfs(struct nvme_ns *ns) {}; 571 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, 572 unsigned long arg) 573 { 574 return -ENOTTY; 575 } 576 #endif /* CONFIG_NVM */ 577 578 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev) 579 { 580 return dev_to_disk(dev)->private_data; 581 } 582 583 int __init nvme_core_init(void); 584 void nvme_core_exit(void); 585 586 #endif /* _NVME_H */ 587