.. | .. |
---|
| 1 | +/* SPDX-License-Identifier: GPL-2.0 */ |
---|
1 | 2 | /* |
---|
2 | 3 | * 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 | 4 | */ |
---|
13 | 5 | |
---|
14 | 6 | #ifndef _NVME_H |
---|
.. | .. |
---|
23 | 15 | #include <linux/sed-opal.h> |
---|
24 | 16 | #include <linux/fault-inject.h> |
---|
25 | 17 | #include <linux/rcupdate.h> |
---|
| 18 | +#include <linux/wait.h> |
---|
| 19 | +#include <linux/t10-pi.h> |
---|
| 20 | + |
---|
| 21 | +#include <trace/events/block.h> |
---|
26 | 22 | |
---|
27 | 23 | extern unsigned int nvme_io_timeout; |
---|
28 | 24 | #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ) |
---|
.. | .. |
---|
32 | 28 | |
---|
33 | 29 | #define NVME_DEFAULT_KATO 5 |
---|
34 | 30 | #define NVME_KATO_GRACE 10 |
---|
| 31 | + |
---|
| 32 | +#ifdef CONFIG_ARCH_NO_SG_CHAIN |
---|
| 33 | +#define NVME_INLINE_SG_CNT 0 |
---|
| 34 | +#define NVME_INLINE_METADATA_SG_CNT 0 |
---|
| 35 | +#else |
---|
| 36 | +#define NVME_INLINE_SG_CNT 2 |
---|
| 37 | +#define NVME_INLINE_METADATA_SG_CNT 1 |
---|
| 38 | +#endif |
---|
| 39 | + |
---|
| 40 | +/* |
---|
| 41 | + * Default to a 4K page size, with the intention to update this |
---|
| 42 | + * path in the future to accommodate architectures with differing |
---|
| 43 | + * kernel and IO page sizes. |
---|
| 44 | + */ |
---|
| 45 | +#define NVME_CTRL_PAGE_SHIFT 12 |
---|
| 46 | +#define NVME_CTRL_PAGE_SIZE (1 << NVME_CTRL_PAGE_SHIFT) |
---|
35 | 47 | |
---|
36 | 48 | extern struct workqueue_struct *nvme_wq; |
---|
37 | 49 | extern struct workqueue_struct *nvme_reset_wq; |
---|
.. | .. |
---|
90 | 102 | * Set MEDIUM priority on SQ creation |
---|
91 | 103 | */ |
---|
92 | 104 | NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7), |
---|
| 105 | + |
---|
| 106 | + /* |
---|
| 107 | + * Ignore device provided subnqn. |
---|
| 108 | + */ |
---|
| 109 | + NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8), |
---|
| 110 | + |
---|
| 111 | + /* |
---|
| 112 | + * Broken Write Zeroes. |
---|
| 113 | + */ |
---|
| 114 | + NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9), |
---|
| 115 | + |
---|
| 116 | + /* |
---|
| 117 | + * Force simple suspend/resume path. |
---|
| 118 | + */ |
---|
| 119 | + NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10), |
---|
| 120 | + |
---|
| 121 | + /* |
---|
| 122 | + * Use only one interrupt vector for all queues |
---|
| 123 | + */ |
---|
| 124 | + NVME_QUIRK_SINGLE_VECTOR = (1 << 11), |
---|
| 125 | + |
---|
| 126 | + /* |
---|
| 127 | + * Use non-standard 128 bytes SQEs. |
---|
| 128 | + */ |
---|
| 129 | + NVME_QUIRK_128_BYTES_SQES = (1 << 12), |
---|
| 130 | + |
---|
| 131 | + /* |
---|
| 132 | + * Prevent tag overlap between queues |
---|
| 133 | + */ |
---|
| 134 | + NVME_QUIRK_SHARED_TAGS = (1 << 13), |
---|
| 135 | + |
---|
| 136 | + /* |
---|
| 137 | + * Don't change the value of the temperature threshold feature |
---|
| 138 | + */ |
---|
| 139 | + NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14), |
---|
| 140 | + |
---|
| 141 | + /* |
---|
| 142 | + * The controller doesn't handle the Identify Namespace |
---|
| 143 | + * Identification Descriptor list subcommand despite claiming |
---|
| 144 | + * NVMe 1.3 compliance. |
---|
| 145 | + */ |
---|
| 146 | + NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15), |
---|
| 147 | + |
---|
| 148 | + /* |
---|
| 149 | + * The controller requires the command_id value be be limited, so skip |
---|
| 150 | + * encoding the generation sequence number. |
---|
| 151 | + */ |
---|
| 152 | + NVME_QUIRK_SKIP_CID_GEN = (1 << 17), |
---|
| 153 | + |
---|
| 154 | + /* |
---|
| 155 | + * Reports garbage in the namespace identifiers (eui64, nguid, uuid). |
---|
| 156 | + */ |
---|
| 157 | + NVME_QUIRK_BOGUS_NID = (1 << 18), |
---|
| 158 | + |
---|
| 159 | + /* |
---|
| 160 | + * Limit io queue depth to 32 |
---|
| 161 | + */ |
---|
| 162 | + NVME_QUIRK_LIMIT_IOQD32 = (1 << 31), |
---|
93 | 163 | }; |
---|
94 | 164 | |
---|
95 | 165 | /* |
---|
.. | .. |
---|
99 | 169 | struct nvme_request { |
---|
100 | 170 | struct nvme_command *cmd; |
---|
101 | 171 | union nvme_result result; |
---|
| 172 | + u8 genctr; |
---|
102 | 173 | u8 retries; |
---|
103 | 174 | u8 flags; |
---|
104 | 175 | u16 status; |
---|
.. | .. |
---|
122 | 193 | |
---|
123 | 194 | static inline u16 nvme_req_qid(struct request *req) |
---|
124 | 195 | { |
---|
125 | | - if (!req->rq_disk) |
---|
| 196 | + if (!req->q->queuedata) |
---|
126 | 197 | return 0; |
---|
127 | 198 | return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1; |
---|
128 | 199 | } |
---|
.. | .. |
---|
134 | 205 | */ |
---|
135 | 206 | #define NVME_QUIRK_DELAY_AMOUNT 2300 |
---|
136 | 207 | |
---|
| 208 | +/* |
---|
| 209 | + * enum nvme_ctrl_state: Controller state |
---|
| 210 | + * |
---|
| 211 | + * @NVME_CTRL_NEW: New controller just allocated, initial state |
---|
| 212 | + * @NVME_CTRL_LIVE: Controller is connected and I/O capable |
---|
| 213 | + * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset) |
---|
| 214 | + * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the |
---|
| 215 | + * transport |
---|
| 216 | + * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion) |
---|
| 217 | + * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not |
---|
| 218 | + * disabled/failed immediately. This state comes |
---|
| 219 | + * after all async event processing took place and |
---|
| 220 | + * before ns removal and the controller deletion |
---|
| 221 | + * progress |
---|
| 222 | + * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during |
---|
| 223 | + * shutdown or removal. In this case we forcibly |
---|
| 224 | + * kill all inflight I/O as they have no chance to |
---|
| 225 | + * complete |
---|
| 226 | + */ |
---|
137 | 227 | enum nvme_ctrl_state { |
---|
138 | 228 | NVME_CTRL_NEW, |
---|
139 | 229 | NVME_CTRL_LIVE, |
---|
140 | | - NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */ |
---|
141 | 230 | NVME_CTRL_RESETTING, |
---|
142 | 231 | NVME_CTRL_CONNECTING, |
---|
143 | 232 | NVME_CTRL_DELETING, |
---|
| 233 | + NVME_CTRL_DELETING_NOIO, |
---|
144 | 234 | NVME_CTRL_DEAD, |
---|
145 | 235 | }; |
---|
146 | 236 | |
---|
| 237 | +struct nvme_fault_inject { |
---|
| 238 | +#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
---|
| 239 | + struct fault_attr attr; |
---|
| 240 | + struct dentry *parent; |
---|
| 241 | + bool dont_retry; /* DNR, do not retry */ |
---|
| 242 | + u16 status; /* status code */ |
---|
| 243 | +#endif |
---|
| 244 | +}; |
---|
| 245 | + |
---|
147 | 246 | struct nvme_ctrl { |
---|
| 247 | + bool comp_seen; |
---|
148 | 248 | enum nvme_ctrl_state state; |
---|
149 | 249 | bool identified; |
---|
150 | 250 | spinlock_t lock; |
---|
.. | .. |
---|
152 | 252 | const struct nvme_ctrl_ops *ops; |
---|
153 | 253 | struct request_queue *admin_q; |
---|
154 | 254 | struct request_queue *connect_q; |
---|
| 255 | + struct request_queue *fabrics_q; |
---|
155 | 256 | struct device *dev; |
---|
156 | 257 | int instance; |
---|
| 258 | + int numa_node; |
---|
157 | 259 | struct blk_mq_tag_set *tagset; |
---|
158 | 260 | struct blk_mq_tag_set *admin_tagset; |
---|
159 | 261 | struct list_head namespaces; |
---|
160 | 262 | struct rw_semaphore namespaces_rwsem; |
---|
161 | 263 | struct device ctrl_device; |
---|
162 | 264 | struct device *device; /* char device */ |
---|
| 265 | +#ifdef CONFIG_NVME_HWMON |
---|
| 266 | + struct device *hwmon_device; |
---|
| 267 | +#endif |
---|
163 | 268 | struct cdev cdev; |
---|
164 | 269 | struct work_struct reset_work; |
---|
165 | 270 | struct work_struct delete_work; |
---|
| 271 | + wait_queue_head_t state_wq; |
---|
166 | 272 | |
---|
167 | 273 | struct nvme_subsystem *subsys; |
---|
168 | 274 | struct list_head subsys_entry; |
---|
.. | .. |
---|
177 | 283 | u32 queue_count; |
---|
178 | 284 | |
---|
179 | 285 | u64 cap; |
---|
180 | | - u32 page_size; |
---|
181 | 286 | u32 max_hw_sectors; |
---|
182 | 287 | u32 max_segments; |
---|
| 288 | + u32 max_integrity_segments; |
---|
| 289 | +#ifdef CONFIG_BLK_DEV_ZONED |
---|
| 290 | + u32 max_zone_append; |
---|
| 291 | +#endif |
---|
| 292 | + u16 crdt[3]; |
---|
183 | 293 | u16 oncs; |
---|
184 | 294 | u16 oacs; |
---|
185 | 295 | u16 nssa; |
---|
186 | 296 | u16 nr_streams; |
---|
| 297 | + u16 sqsize; |
---|
187 | 298 | u32 max_namespaces; |
---|
188 | 299 | atomic_t abort_limit; |
---|
189 | 300 | u8 vwc; |
---|
.. | .. |
---|
192 | 303 | u16 kas; |
---|
193 | 304 | u8 npss; |
---|
194 | 305 | u8 apsta; |
---|
| 306 | + u16 wctemp; |
---|
| 307 | + u16 cctemp; |
---|
195 | 308 | u32 oaes; |
---|
196 | 309 | u32 aen_result; |
---|
| 310 | + u32 ctratt; |
---|
197 | 311 | unsigned int shutdown_timeout; |
---|
198 | 312 | unsigned int kato; |
---|
199 | 313 | bool subsystem; |
---|
200 | 314 | unsigned long quirks; |
---|
201 | 315 | struct nvme_id_power_state psd[32]; |
---|
202 | 316 | struct nvme_effects_log *effects; |
---|
| 317 | + struct xarray cels; |
---|
203 | 318 | struct work_struct scan_work; |
---|
204 | 319 | struct work_struct async_event_work; |
---|
205 | 320 | struct delayed_work ka_work; |
---|
206 | 321 | struct nvme_command ka_cmd; |
---|
207 | 322 | struct work_struct fw_act_work; |
---|
208 | 323 | unsigned long events; |
---|
209 | | - bool created; |
---|
210 | 324 | |
---|
211 | 325 | #ifdef CONFIG_NVME_MULTIPATH |
---|
212 | 326 | /* asymmetric namespace access: */ |
---|
.. | .. |
---|
232 | 346 | u16 hmmaxd; |
---|
233 | 347 | |
---|
234 | 348 | /* Fabrics only */ |
---|
235 | | - u16 sqsize; |
---|
236 | 349 | u32 ioccsz; |
---|
237 | 350 | u32 iorcsz; |
---|
238 | 351 | u16 icdoff; |
---|
.. | .. |
---|
242 | 355 | |
---|
243 | 356 | struct page *discard_page; |
---|
244 | 357 | unsigned long discard_page_busy; |
---|
| 358 | + |
---|
| 359 | + struct nvme_fault_inject fault_inject; |
---|
| 360 | +}; |
---|
| 361 | + |
---|
| 362 | +enum nvme_iopolicy { |
---|
| 363 | + NVME_IOPOLICY_NUMA, |
---|
| 364 | + NVME_IOPOLICY_RR, |
---|
245 | 365 | }; |
---|
246 | 366 | |
---|
247 | 367 | struct nvme_subsystem { |
---|
.. | .. |
---|
262 | 382 | char firmware_rev[8]; |
---|
263 | 383 | u8 cmic; |
---|
264 | 384 | u16 vendor_id; |
---|
| 385 | + u16 awupf; /* 0's based awupf value. */ |
---|
265 | 386 | struct ida ns_ida; |
---|
| 387 | +#ifdef CONFIG_NVME_MULTIPATH |
---|
| 388 | + enum nvme_iopolicy iopolicy; |
---|
| 389 | +#endif |
---|
266 | 390 | }; |
---|
267 | 391 | |
---|
268 | 392 | /* |
---|
.. | .. |
---|
272 | 396 | u8 eui64[8]; |
---|
273 | 397 | u8 nguid[16]; |
---|
274 | 398 | uuid_t uuid; |
---|
| 399 | + u8 csi; |
---|
275 | 400 | }; |
---|
276 | 401 | |
---|
277 | 402 | /* |
---|
.. | .. |
---|
282 | 407 | * only ever has a single entry for private namespaces. |
---|
283 | 408 | */ |
---|
284 | 409 | struct nvme_ns_head { |
---|
285 | | -#ifdef CONFIG_NVME_MULTIPATH |
---|
286 | | - struct gendisk *disk; |
---|
287 | | - struct nvme_ns __rcu *current_path; |
---|
288 | | - struct bio_list requeue_list; |
---|
289 | | - spinlock_t requeue_lock; |
---|
290 | | - struct work_struct requeue_work; |
---|
291 | | - struct mutex lock; |
---|
292 | | -#endif |
---|
293 | 410 | struct list_head list; |
---|
294 | 411 | struct srcu_struct srcu; |
---|
295 | 412 | struct nvme_subsystem *subsys; |
---|
.. | .. |
---|
297 | 414 | struct nvme_ns_ids ids; |
---|
298 | 415 | struct list_head entry; |
---|
299 | 416 | struct kref ref; |
---|
| 417 | + bool shared; |
---|
300 | 418 | int instance; |
---|
| 419 | + struct nvme_effects_log *effects; |
---|
| 420 | +#ifdef CONFIG_NVME_MULTIPATH |
---|
| 421 | + struct gendisk *disk; |
---|
| 422 | + struct bio_list requeue_list; |
---|
| 423 | + spinlock_t requeue_lock; |
---|
| 424 | + struct work_struct requeue_work; |
---|
| 425 | + struct mutex lock; |
---|
| 426 | + unsigned long flags; |
---|
| 427 | +#define NVME_NSHEAD_DISK_LIVE 0 |
---|
| 428 | + struct nvme_ns __rcu *current_path[]; |
---|
| 429 | +#endif |
---|
301 | 430 | }; |
---|
302 | 431 | |
---|
303 | | -#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
---|
304 | | -struct nvme_fault_inject { |
---|
305 | | - struct fault_attr attr; |
---|
306 | | - struct dentry *parent; |
---|
307 | | - bool dont_retry; /* DNR, do not retry */ |
---|
308 | | - u16 status; /* status code */ |
---|
| 432 | +enum nvme_ns_features { |
---|
| 433 | + NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */ |
---|
| 434 | + NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */ |
---|
309 | 435 | }; |
---|
310 | | -#endif |
---|
311 | 436 | |
---|
312 | 437 | struct nvme_ns { |
---|
313 | 438 | struct list_head list; |
---|
.. | .. |
---|
328 | 453 | u16 ms; |
---|
329 | 454 | u16 sgs; |
---|
330 | 455 | u32 sws; |
---|
331 | | - bool ext; |
---|
332 | 456 | u8 pi_type; |
---|
| 457 | +#ifdef CONFIG_BLK_DEV_ZONED |
---|
| 458 | + u64 zsze; |
---|
| 459 | +#endif |
---|
| 460 | + unsigned long features; |
---|
333 | 461 | unsigned long flags; |
---|
334 | 462 | #define NVME_NS_REMOVING 0 |
---|
335 | 463 | #define NVME_NS_DEAD 1 |
---|
336 | 464 | #define NVME_NS_ANA_PENDING 2 |
---|
337 | | - u16 noiob; |
---|
338 | 465 | |
---|
339 | | -#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
---|
340 | 466 | struct nvme_fault_inject fault_inject; |
---|
341 | | -#endif |
---|
342 | 467 | |
---|
343 | 468 | }; |
---|
| 469 | + |
---|
| 470 | +/* NVMe ns supports metadata actions by the controller (generate/strip) */ |
---|
| 471 | +static inline bool nvme_ns_has_pi(struct nvme_ns *ns) |
---|
| 472 | +{ |
---|
| 473 | + return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple); |
---|
| 474 | +} |
---|
344 | 475 | |
---|
345 | 476 | struct nvme_ctrl_ops { |
---|
346 | 477 | const char *name; |
---|
.. | .. |
---|
348 | 479 | unsigned int flags; |
---|
349 | 480 | #define NVME_F_FABRICS (1 << 0) |
---|
350 | 481 | #define NVME_F_METADATA_SUPPORTED (1 << 1) |
---|
| 482 | +#define NVME_F_PCI_P2PDMA (1 << 2) |
---|
351 | 483 | int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val); |
---|
352 | 484 | int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val); |
---|
353 | 485 | int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val); |
---|
354 | 486 | void (*free_ctrl)(struct nvme_ctrl *ctrl); |
---|
355 | 487 | void (*submit_async_event)(struct nvme_ctrl *ctrl); |
---|
356 | 488 | void (*delete_ctrl)(struct nvme_ctrl *ctrl); |
---|
357 | | - int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); |
---|
358 | 489 | void (*stop_ctrl)(struct nvme_ctrl *ctrl); |
---|
| 490 | + int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size); |
---|
359 | 491 | }; |
---|
360 | 492 | |
---|
| 493 | +/* |
---|
| 494 | + * nvme command_id is constructed as such: |
---|
| 495 | + * | xxxx | xxxxxxxxxxxx | |
---|
| 496 | + * gen request tag |
---|
| 497 | + */ |
---|
| 498 | +#define nvme_genctr_mask(gen) (gen & 0xf) |
---|
| 499 | +#define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12) |
---|
| 500 | +#define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12) |
---|
| 501 | +#define nvme_tag_from_cid(cid) (cid & 0xfff) |
---|
| 502 | + |
---|
| 503 | +static inline u16 nvme_cid(struct request *rq) |
---|
| 504 | +{ |
---|
| 505 | + return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag; |
---|
| 506 | +} |
---|
| 507 | + |
---|
| 508 | +static inline struct request *nvme_find_rq(struct blk_mq_tags *tags, |
---|
| 509 | + u16 command_id) |
---|
| 510 | +{ |
---|
| 511 | + u8 genctr = nvme_genctr_from_cid(command_id); |
---|
| 512 | + u16 tag = nvme_tag_from_cid(command_id); |
---|
| 513 | + struct request *rq; |
---|
| 514 | + |
---|
| 515 | + rq = blk_mq_tag_to_rq(tags, tag); |
---|
| 516 | + if (unlikely(!rq)) { |
---|
| 517 | + pr_err("could not locate request for tag %#x\n", |
---|
| 518 | + tag); |
---|
| 519 | + return NULL; |
---|
| 520 | + } |
---|
| 521 | + if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) { |
---|
| 522 | + dev_err(nvme_req(rq)->ctrl->device, |
---|
| 523 | + "request %#x genctr mismatch (got %#x expected %#x)\n", |
---|
| 524 | + tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr)); |
---|
| 525 | + return NULL; |
---|
| 526 | + } |
---|
| 527 | + return rq; |
---|
| 528 | +} |
---|
| 529 | + |
---|
| 530 | +static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags, |
---|
| 531 | + u16 command_id) |
---|
| 532 | +{ |
---|
| 533 | + return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id)); |
---|
| 534 | +} |
---|
| 535 | + |
---|
361 | 536 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS |
---|
362 | | -void nvme_fault_inject_init(struct nvme_ns *ns); |
---|
363 | | -void nvme_fault_inject_fini(struct nvme_ns *ns); |
---|
| 537 | +void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, |
---|
| 538 | + const char *dev_name); |
---|
| 539 | +void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject); |
---|
364 | 540 | void nvme_should_fail(struct request *req); |
---|
365 | 541 | #else |
---|
366 | | -static inline void nvme_fault_inject_init(struct nvme_ns *ns) {} |
---|
367 | | -static inline void nvme_fault_inject_fini(struct nvme_ns *ns) {} |
---|
| 542 | +static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj, |
---|
| 543 | + const char *dev_name) |
---|
| 544 | +{ |
---|
| 545 | +} |
---|
| 546 | +static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj) |
---|
| 547 | +{ |
---|
| 548 | +} |
---|
368 | 549 | static inline void nvme_should_fail(struct request *req) {} |
---|
369 | 550 | #endif |
---|
370 | 551 | |
---|
371 | | -static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl) |
---|
372 | | -{ |
---|
373 | | - u32 val = 0; |
---|
374 | | - |
---|
375 | | - if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val)) |
---|
376 | | - return false; |
---|
377 | | - return val & NVME_CSTS_RDY; |
---|
378 | | -} |
---|
| 552 | +bool nvme_wait_reset(struct nvme_ctrl *ctrl); |
---|
| 553 | +int nvme_try_sched_reset(struct nvme_ctrl *ctrl); |
---|
379 | 554 | |
---|
380 | 555 | static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl) |
---|
381 | 556 | { |
---|
| 557 | + int ret; |
---|
| 558 | + |
---|
382 | 559 | if (!ctrl->subsystem) |
---|
383 | 560 | return -ENOTTY; |
---|
384 | | - return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65); |
---|
| 561 | + if (!nvme_wait_reset(ctrl)) |
---|
| 562 | + return -EBUSY; |
---|
| 563 | + |
---|
| 564 | + ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65); |
---|
| 565 | + if (ret) |
---|
| 566 | + return ret; |
---|
| 567 | + |
---|
| 568 | + return nvme_try_sched_reset(ctrl); |
---|
385 | 569 | } |
---|
386 | 570 | |
---|
387 | | -static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector) |
---|
| 571 | +/* |
---|
| 572 | + * Convert a 512B sector number to a device logical block number. |
---|
| 573 | + */ |
---|
| 574 | +static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector) |
---|
388 | 575 | { |
---|
389 | | - return (sector >> (ns->lba_shift - 9)); |
---|
| 576 | + return sector >> (ns->lba_shift - SECTOR_SHIFT); |
---|
390 | 577 | } |
---|
391 | 578 | |
---|
392 | | -static inline void nvme_end_request(struct request *req, __le16 status, |
---|
| 579 | +/* |
---|
| 580 | + * Convert a device logical block number to a 512B sector number. |
---|
| 581 | + */ |
---|
| 582 | +static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba) |
---|
| 583 | +{ |
---|
| 584 | + return lba << (ns->lba_shift - SECTOR_SHIFT); |
---|
| 585 | +} |
---|
| 586 | + |
---|
| 587 | +/* |
---|
| 588 | + * Convert byte length to nvme's 0-based num dwords |
---|
| 589 | + */ |
---|
| 590 | +static inline u32 nvme_bytes_to_numd(size_t len) |
---|
| 591 | +{ |
---|
| 592 | + return (len >> 2) - 1; |
---|
| 593 | +} |
---|
| 594 | + |
---|
| 595 | +static inline bool nvme_is_ana_error(u16 status) |
---|
| 596 | +{ |
---|
| 597 | + switch (status & 0x7ff) { |
---|
| 598 | + case NVME_SC_ANA_TRANSITION: |
---|
| 599 | + case NVME_SC_ANA_INACCESSIBLE: |
---|
| 600 | + case NVME_SC_ANA_PERSISTENT_LOSS: |
---|
| 601 | + return true; |
---|
| 602 | + default: |
---|
| 603 | + return false; |
---|
| 604 | + } |
---|
| 605 | +} |
---|
| 606 | + |
---|
| 607 | +static inline bool nvme_is_path_error(u16 status) |
---|
| 608 | +{ |
---|
| 609 | + /* check for a status code type of 'path related status' */ |
---|
| 610 | + return (status & 0x700) == 0x300; |
---|
| 611 | +} |
---|
| 612 | + |
---|
| 613 | +/* |
---|
| 614 | + * Fill in the status and result information from the CQE, and then figure out |
---|
| 615 | + * if blk-mq will need to use IPI magic to complete the request, and if yes do |
---|
| 616 | + * so. If not let the caller complete the request without an indirect function |
---|
| 617 | + * call. |
---|
| 618 | + */ |
---|
| 619 | +static inline bool nvme_try_complete_req(struct request *req, __le16 status, |
---|
393 | 620 | union nvme_result result) |
---|
394 | 621 | { |
---|
395 | 622 | struct nvme_request *rq = nvme_req(req); |
---|
.. | .. |
---|
398 | 625 | rq->result = result; |
---|
399 | 626 | /* inject error when permitted by fault injection framework */ |
---|
400 | 627 | nvme_should_fail(req); |
---|
401 | | - blk_mq_complete_request(req); |
---|
| 628 | + if (unlikely(blk_should_fake_timeout(req->q))) |
---|
| 629 | + return true; |
---|
| 630 | + return blk_mq_complete_request_remote(req); |
---|
402 | 631 | } |
---|
403 | 632 | |
---|
404 | 633 | static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl) |
---|
.. | .. |
---|
411 | 640 | put_device(ctrl->device); |
---|
412 | 641 | } |
---|
413 | 642 | |
---|
| 643 | +static inline bool nvme_is_aen_req(u16 qid, __u16 command_id) |
---|
| 644 | +{ |
---|
| 645 | + return !qid && |
---|
| 646 | + nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH; |
---|
| 647 | +} |
---|
| 648 | + |
---|
414 | 649 | void nvme_complete_rq(struct request *req); |
---|
415 | | -void nvme_cancel_request(struct request *req, void *data, bool reserved); |
---|
| 650 | +bool nvme_cancel_request(struct request *req, void *data, bool reserved); |
---|
| 651 | +void nvme_cancel_tagset(struct nvme_ctrl *ctrl); |
---|
| 652 | +void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl); |
---|
416 | 653 | bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl, |
---|
417 | 654 | enum nvme_ctrl_state new_state); |
---|
418 | | -int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap); |
---|
419 | | -int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap); |
---|
| 655 | +int nvme_disable_ctrl(struct nvme_ctrl *ctrl); |
---|
| 656 | +int nvme_enable_ctrl(struct nvme_ctrl *ctrl); |
---|
420 | 657 | int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl); |
---|
421 | 658 | int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev, |
---|
422 | 659 | const struct nvme_ctrl_ops *ops, unsigned long quirks); |
---|
423 | 660 | void nvme_uninit_ctrl(struct nvme_ctrl *ctrl); |
---|
424 | 661 | void nvme_start_ctrl(struct nvme_ctrl *ctrl); |
---|
425 | 662 | void nvme_stop_ctrl(struct nvme_ctrl *ctrl); |
---|
426 | | -void nvme_put_ctrl(struct nvme_ctrl *ctrl); |
---|
427 | 663 | int nvme_init_identify(struct nvme_ctrl *ctrl); |
---|
428 | 664 | |
---|
429 | 665 | void nvme_remove_namespaces(struct nvme_ctrl *ctrl); |
---|
.. | .. |
---|
437 | 673 | void nvme_stop_queues(struct nvme_ctrl *ctrl); |
---|
438 | 674 | void nvme_start_queues(struct nvme_ctrl *ctrl); |
---|
439 | 675 | void nvme_kill_queues(struct nvme_ctrl *ctrl); |
---|
| 676 | +void nvme_sync_queues(struct nvme_ctrl *ctrl); |
---|
| 677 | +void nvme_sync_io_queues(struct nvme_ctrl *ctrl); |
---|
440 | 678 | void nvme_unfreeze(struct nvme_ctrl *ctrl); |
---|
441 | 679 | void nvme_wait_freeze(struct nvme_ctrl *ctrl); |
---|
442 | | -void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); |
---|
| 680 | +int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout); |
---|
443 | 681 | void nvme_start_freeze(struct nvme_ctrl *ctrl); |
---|
444 | 682 | |
---|
445 | 683 | #define NVME_QID_ANY -1 |
---|
446 | 684 | struct request *nvme_alloc_request(struct request_queue *q, |
---|
| 685 | + struct nvme_command *cmd, blk_mq_req_flags_t flags); |
---|
| 686 | +struct request *nvme_alloc_request_qid(struct request_queue *q, |
---|
447 | 687 | struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid); |
---|
448 | 688 | void nvme_cleanup_cmd(struct request *req); |
---|
449 | 689 | blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req, |
---|
.. | .. |
---|
453 | 693 | int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd, |
---|
454 | 694 | union nvme_result *result, void *buffer, unsigned bufflen, |
---|
455 | 695 | unsigned timeout, int qid, int at_head, |
---|
456 | | - blk_mq_req_flags_t flags); |
---|
| 696 | + blk_mq_req_flags_t flags, bool poll); |
---|
| 697 | +int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid, |
---|
| 698 | + unsigned int dword11, void *buffer, size_t buflen, |
---|
| 699 | + u32 *result); |
---|
| 700 | +int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid, |
---|
| 701 | + unsigned int dword11, void *buffer, size_t buflen, |
---|
| 702 | + u32 *result); |
---|
457 | 703 | int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count); |
---|
458 | 704 | void nvme_stop_keep_alive(struct nvme_ctrl *ctrl); |
---|
459 | 705 | int nvme_reset_ctrl(struct nvme_ctrl *ctrl); |
---|
460 | 706 | int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl); |
---|
461 | 707 | int nvme_delete_ctrl(struct nvme_ctrl *ctrl); |
---|
462 | | -int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl); |
---|
463 | 708 | |
---|
464 | | -int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, |
---|
| 709 | +int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi, |
---|
465 | 710 | void *log, size_t size, u64 offset); |
---|
| 711 | +struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk, |
---|
| 712 | + struct nvme_ns_head **head, int *srcu_idx); |
---|
| 713 | +void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx); |
---|
466 | 714 | |
---|
467 | | -extern const struct attribute_group nvme_ns_id_attr_group; |
---|
| 715 | +extern const struct attribute_group *nvme_ns_id_attr_groups[]; |
---|
468 | 716 | extern const struct block_device_operations nvme_ns_head_ops; |
---|
469 | 717 | |
---|
470 | 718 | #ifdef CONFIG_NVME_MULTIPATH |
---|
.. | .. |
---|
478 | 726 | void nvme_mpath_start_freeze(struct nvme_subsystem *subsys); |
---|
479 | 727 | void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns, |
---|
480 | 728 | struct nvme_ctrl *ctrl, int *flags); |
---|
481 | | -bool nvme_failover_req(struct request *req); |
---|
| 729 | +void nvme_failover_req(struct request *req); |
---|
482 | 730 | void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl); |
---|
483 | 731 | int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head); |
---|
484 | 732 | void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id); |
---|
485 | 733 | void nvme_mpath_remove_disk(struct nvme_ns_head *head); |
---|
486 | | -int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); |
---|
| 734 | +int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id); |
---|
| 735 | +void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl); |
---|
| 736 | +void nvme_mpath_update(struct nvme_ctrl *ctrl); |
---|
487 | 737 | void nvme_mpath_uninit(struct nvme_ctrl *ctrl); |
---|
488 | 738 | void nvme_mpath_stop(struct nvme_ctrl *ctrl); |
---|
489 | | - |
---|
490 | | -static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns) |
---|
491 | | -{ |
---|
492 | | - struct nvme_ns_head *head = ns->head; |
---|
493 | | - |
---|
494 | | - if (head && ns == rcu_access_pointer(head->current_path)) |
---|
495 | | - rcu_assign_pointer(head->current_path, NULL); |
---|
496 | | -} |
---|
| 739 | +bool nvme_mpath_clear_current_path(struct nvme_ns *ns); |
---|
| 740 | +void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl); |
---|
497 | 741 | struct nvme_ns *nvme_find_path(struct nvme_ns_head *head); |
---|
| 742 | +blk_qc_t nvme_ns_head_submit_bio(struct bio *bio); |
---|
498 | 743 | |
---|
499 | 744 | static inline void nvme_mpath_check_last_path(struct nvme_ns *ns) |
---|
500 | 745 | { |
---|
.. | .. |
---|
504 | 749 | kblockd_schedule_work(&head->requeue_work); |
---|
505 | 750 | } |
---|
506 | 751 | |
---|
507 | | -static inline void nvme_mpath_update_disk_size(struct gendisk *disk) |
---|
| 752 | +static inline void nvme_trace_bio_complete(struct request *req, |
---|
| 753 | + blk_status_t status) |
---|
508 | 754 | { |
---|
509 | | - struct block_device *bdev = bdget_disk(disk, 0); |
---|
| 755 | + struct nvme_ns *ns = req->q->queuedata; |
---|
510 | 756 | |
---|
511 | | - if (bdev) { |
---|
512 | | - bd_set_size(bdev, get_capacity(disk) << SECTOR_SHIFT); |
---|
513 | | - bdput(bdev); |
---|
514 | | - } |
---|
| 757 | + if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio) |
---|
| 758 | + trace_block_bio_complete(ns->head->disk->queue, req->bio); |
---|
515 | 759 | } |
---|
516 | 760 | |
---|
517 | 761 | extern struct device_attribute dev_attr_ana_grpid; |
---|
518 | 762 | extern struct device_attribute dev_attr_ana_state; |
---|
| 763 | +extern struct device_attribute subsys_attr_iopolicy; |
---|
519 | 764 | |
---|
520 | 765 | #else |
---|
521 | 766 | static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl) |
---|
.. | .. |
---|
532 | 777 | sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance); |
---|
533 | 778 | } |
---|
534 | 779 | |
---|
535 | | -static inline bool nvme_failover_req(struct request *req) |
---|
| 780 | +static inline void nvme_failover_req(struct request *req) |
---|
536 | 781 | { |
---|
537 | | - return false; |
---|
538 | 782 | } |
---|
539 | 783 | static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl) |
---|
540 | 784 | { |
---|
.. | .. |
---|
551 | 795 | static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head) |
---|
552 | 796 | { |
---|
553 | 797 | } |
---|
554 | | -static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns) |
---|
| 798 | +static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns) |
---|
| 799 | +{ |
---|
| 800 | + return false; |
---|
| 801 | +} |
---|
| 802 | +static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl) |
---|
555 | 803 | { |
---|
556 | 804 | } |
---|
557 | 805 | static inline void nvme_mpath_check_last_path(struct nvme_ns *ns) |
---|
558 | 806 | { |
---|
559 | 807 | } |
---|
560 | | -static inline int nvme_mpath_init(struct nvme_ctrl *ctrl, |
---|
| 808 | +static inline void nvme_trace_bio_complete(struct request *req, |
---|
| 809 | + blk_status_t status) |
---|
| 810 | +{ |
---|
| 811 | +} |
---|
| 812 | +static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl) |
---|
| 813 | +{ |
---|
| 814 | +} |
---|
| 815 | +static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, |
---|
561 | 816 | struct nvme_id_ctrl *id) |
---|
562 | 817 | { |
---|
563 | 818 | if (ctrl->subsys->cmic & (1 << 3)) |
---|
564 | 819 | dev_warn(ctrl->device, |
---|
565 | 820 | "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n"); |
---|
566 | 821 | return 0; |
---|
| 822 | +} |
---|
| 823 | +static inline void nvme_mpath_update(struct nvme_ctrl *ctrl) |
---|
| 824 | +{ |
---|
567 | 825 | } |
---|
568 | 826 | static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl) |
---|
569 | 827 | { |
---|
.. | .. |
---|
580 | 838 | static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys) |
---|
581 | 839 | { |
---|
582 | 840 | } |
---|
583 | | -static inline void nvme_mpath_update_disk_size(struct gendisk *disk) |
---|
584 | | -{ |
---|
585 | | -} |
---|
586 | 841 | #endif /* CONFIG_NVME_MULTIPATH */ |
---|
587 | 842 | |
---|
| 843 | +int nvme_revalidate_zones(struct nvme_ns *ns); |
---|
| 844 | +#ifdef CONFIG_BLK_DEV_ZONED |
---|
| 845 | +int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf); |
---|
| 846 | +int nvme_report_zones(struct gendisk *disk, sector_t sector, |
---|
| 847 | + unsigned int nr_zones, report_zones_cb cb, void *data); |
---|
| 848 | + |
---|
| 849 | +blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req, |
---|
| 850 | + struct nvme_command *cmnd, |
---|
| 851 | + enum nvme_zone_mgmt_action action); |
---|
| 852 | +#else |
---|
| 853 | +#define nvme_report_zones NULL |
---|
| 854 | + |
---|
| 855 | +static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, |
---|
| 856 | + struct request *req, struct nvme_command *cmnd, |
---|
| 857 | + enum nvme_zone_mgmt_action action) |
---|
| 858 | +{ |
---|
| 859 | + return BLK_STS_NOTSUPP; |
---|
| 860 | +} |
---|
| 861 | + |
---|
| 862 | +static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf) |
---|
| 863 | +{ |
---|
| 864 | + dev_warn(ns->ctrl->device, |
---|
| 865 | + "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n"); |
---|
| 866 | + return -EPROTONOSUPPORT; |
---|
| 867 | +} |
---|
| 868 | +#endif |
---|
| 869 | + |
---|
588 | 870 | #ifdef CONFIG_NVM |
---|
589 | | -void nvme_nvm_update_nvm_info(struct nvme_ns *ns); |
---|
590 | 871 | int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node); |
---|
591 | 872 | void nvme_nvm_unregister(struct nvme_ns *ns); |
---|
592 | | -int nvme_nvm_register_sysfs(struct nvme_ns *ns); |
---|
593 | | -void nvme_nvm_unregister_sysfs(struct nvme_ns *ns); |
---|
| 873 | +extern const struct attribute_group nvme_nvm_attr_group; |
---|
594 | 874 | int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg); |
---|
595 | 875 | #else |
---|
596 | | -static inline void nvme_nvm_update_nvm_info(struct nvme_ns *ns) {}; |
---|
597 | 876 | static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, |
---|
598 | 877 | int node) |
---|
599 | 878 | { |
---|
.. | .. |
---|
601 | 880 | } |
---|
602 | 881 | |
---|
603 | 882 | static inline void nvme_nvm_unregister(struct nvme_ns *ns) {}; |
---|
604 | | -static inline int nvme_nvm_register_sysfs(struct nvme_ns *ns) |
---|
605 | | -{ |
---|
606 | | - return 0; |
---|
607 | | -} |
---|
608 | | -static inline void nvme_nvm_unregister_sysfs(struct nvme_ns *ns) {}; |
---|
609 | 883 | static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, |
---|
610 | 884 | unsigned long arg) |
---|
611 | 885 | { |
---|
.. | .. |
---|
618 | 892 | return dev_to_disk(dev)->private_data; |
---|
619 | 893 | } |
---|
620 | 894 | |
---|
621 | | -int __init nvme_core_init(void); |
---|
622 | | -void nvme_core_exit(void); |
---|
| 895 | +#ifdef CONFIG_NVME_HWMON |
---|
| 896 | +int nvme_hwmon_init(struct nvme_ctrl *ctrl); |
---|
| 897 | +void nvme_hwmon_exit(struct nvme_ctrl *ctrl); |
---|
| 898 | +#else |
---|
| 899 | +static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl) |
---|
| 900 | +{ |
---|
| 901 | + return 0; |
---|
| 902 | +} |
---|
| 903 | + |
---|
| 904 | +static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl) |
---|
| 905 | +{ |
---|
| 906 | +} |
---|
| 907 | +#endif |
---|
| 908 | + |
---|
| 909 | +u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, |
---|
| 910 | + u8 opcode); |
---|
| 911 | +void nvme_execute_passthru_rq(struct request *rq); |
---|
| 912 | +struct nvme_ctrl *nvme_ctrl_from_file(struct file *file); |
---|
| 913 | +struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid); |
---|
| 914 | +void nvme_put_ns(struct nvme_ns *ns); |
---|
623 | 915 | |
---|
624 | 916 | #endif /* _NVME_H */ |
---|