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