dm writecache: add cond_resched to loop in persistent_memory_claim()
[linux/fpc-iii.git] / drivers / nvme / host / nvme.h
blob2e04a36296d952ae3527f16e1cd34e2efc5ffd71
1 /* SPDX-License-Identifier: GPL-2.0 */
2 /*
3 * Copyright (c) 2011-2014, Intel Corporation.
4 */
6 #ifndef _NVME_H
7 #define _NVME_H
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/lightnvm.h>
15 #include <linux/sed-opal.h>
16 #include <linux/fault-inject.h>
17 #include <linux/rcupdate.h>
18 #include <linux/wait.h>
20 #include <trace/events/block.h>
22 extern unsigned int nvme_io_timeout;
23 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
25 extern unsigned int admin_timeout;
26 #define ADMIN_TIMEOUT (admin_timeout * HZ)
28 #define NVME_DEFAULT_KATO 5
29 #define NVME_KATO_GRACE 10
31 #ifdef CONFIG_ARCH_NO_SG_CHAIN
32 #define NVME_INLINE_SG_CNT 0
33 #else
34 #define NVME_INLINE_SG_CNT 2
35 #endif
37 extern struct workqueue_struct *nvme_wq;
38 extern struct workqueue_struct *nvme_reset_wq;
39 extern struct workqueue_struct *nvme_delete_wq;
41 enum {
42 NVME_NS_LBA = 0,
43 NVME_NS_LIGHTNVM = 1,
47 * List of workarounds for devices that required behavior not specified in
48 * the standard.
50 enum nvme_quirks {
52 * Prefers I/O aligned to a stripe size specified in a vendor
53 * specific Identify field.
55 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
58 * The controller doesn't handle Identify value others than 0 or 1
59 * correctly.
61 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
64 * The controller deterministically returns O's on reads to
65 * logical blocks that deallocate was called on.
67 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
70 * The controller needs a delay before starts checking the device
71 * readiness, which is done by reading the NVME_CSTS_RDY bit.
73 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
76 * APST should not be used.
78 NVME_QUIRK_NO_APST = (1 << 4),
81 * The deepest sleep state should not be used.
83 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
86 * Supports the LighNVM command set if indicated in vs[1].
88 NVME_QUIRK_LIGHTNVM = (1 << 6),
91 * Set MEDIUM priority on SQ creation
93 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
96 * Ignore device provided subnqn.
98 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
101 * Broken Write Zeroes.
103 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
106 * Force simple suspend/resume path.
108 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
111 * Use only one interrupt vector for all queues
113 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
116 * Use non-standard 128 bytes SQEs.
118 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
121 * Prevent tag overlap between queues
123 NVME_QUIRK_SHARED_TAGS = (1 << 13),
126 * Don't change the value of the temperature threshold feature
128 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14),
132 * Common request structure for NVMe passthrough. All drivers must have
133 * this structure as the first member of their request-private data.
135 struct nvme_request {
136 struct nvme_command *cmd;
137 union nvme_result result;
138 u8 retries;
139 u8 flags;
140 u16 status;
141 struct nvme_ctrl *ctrl;
145 * Mark a bio as coming in through the mpath node.
147 #define REQ_NVME_MPATH REQ_DRV
149 enum {
150 NVME_REQ_CANCELLED = (1 << 0),
151 NVME_REQ_USERCMD = (1 << 1),
154 static inline struct nvme_request *nvme_req(struct request *req)
156 return blk_mq_rq_to_pdu(req);
159 static inline u16 nvme_req_qid(struct request *req)
161 if (!req->rq_disk)
162 return 0;
163 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
166 /* The below value is the specific amount of delay needed before checking
167 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
168 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
169 * found empirically.
171 #define NVME_QUIRK_DELAY_AMOUNT 2300
173 enum nvme_ctrl_state {
174 NVME_CTRL_NEW,
175 NVME_CTRL_LIVE,
176 NVME_CTRL_RESETTING,
177 NVME_CTRL_CONNECTING,
178 NVME_CTRL_DELETING,
179 NVME_CTRL_DEAD,
182 struct nvme_fault_inject {
183 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
184 struct fault_attr attr;
185 struct dentry *parent;
186 bool dont_retry; /* DNR, do not retry */
187 u16 status; /* status code */
188 #endif
191 struct nvme_ctrl {
192 bool comp_seen;
193 enum nvme_ctrl_state state;
194 bool identified;
195 spinlock_t lock;
196 struct mutex scan_lock;
197 const struct nvme_ctrl_ops *ops;
198 struct request_queue *admin_q;
199 struct request_queue *connect_q;
200 struct request_queue *fabrics_q;
201 struct device *dev;
202 int instance;
203 int numa_node;
204 struct blk_mq_tag_set *tagset;
205 struct blk_mq_tag_set *admin_tagset;
206 struct list_head namespaces;
207 struct rw_semaphore namespaces_rwsem;
208 struct device ctrl_device;
209 struct device *device; /* char device */
210 struct cdev cdev;
211 struct work_struct reset_work;
212 struct work_struct delete_work;
213 wait_queue_head_t state_wq;
215 struct nvme_subsystem *subsys;
216 struct list_head subsys_entry;
218 struct opal_dev *opal_dev;
220 char name[12];
221 u16 cntlid;
223 u32 ctrl_config;
224 u16 mtfa;
225 u32 queue_count;
227 u64 cap;
228 u32 page_size;
229 u32 max_hw_sectors;
230 u32 max_segments;
231 u16 crdt[3];
232 u16 oncs;
233 u16 oacs;
234 u16 nssa;
235 u16 nr_streams;
236 u16 sqsize;
237 u32 max_namespaces;
238 atomic_t abort_limit;
239 u8 vwc;
240 u32 vs;
241 u32 sgls;
242 u16 kas;
243 u8 npss;
244 u8 apsta;
245 u16 wctemp;
246 u16 cctemp;
247 u32 oaes;
248 u32 aen_result;
249 u32 ctratt;
250 unsigned int shutdown_timeout;
251 unsigned int kato;
252 bool subsystem;
253 unsigned long quirks;
254 struct nvme_id_power_state psd[32];
255 struct nvme_effects_log *effects;
256 struct work_struct scan_work;
257 struct work_struct async_event_work;
258 struct delayed_work ka_work;
259 struct nvme_command ka_cmd;
260 struct work_struct fw_act_work;
261 unsigned long events;
262 bool created;
264 #ifdef CONFIG_NVME_MULTIPATH
265 /* asymmetric namespace access: */
266 u8 anacap;
267 u8 anatt;
268 u32 anagrpmax;
269 u32 nanagrpid;
270 struct mutex ana_lock;
271 struct nvme_ana_rsp_hdr *ana_log_buf;
272 size_t ana_log_size;
273 struct timer_list anatt_timer;
274 struct work_struct ana_work;
275 #endif
277 /* Power saving configuration */
278 u64 ps_max_latency_us;
279 bool apst_enabled;
281 /* PCIe only: */
282 u32 hmpre;
283 u32 hmmin;
284 u32 hmminds;
285 u16 hmmaxd;
287 /* Fabrics only */
288 u32 ioccsz;
289 u32 iorcsz;
290 u16 icdoff;
291 u16 maxcmd;
292 int nr_reconnects;
293 struct nvmf_ctrl_options *opts;
295 struct page *discard_page;
296 unsigned long discard_page_busy;
298 struct nvme_fault_inject fault_inject;
301 enum nvme_iopolicy {
302 NVME_IOPOLICY_NUMA,
303 NVME_IOPOLICY_RR,
306 struct nvme_subsystem {
307 int instance;
308 struct device dev;
310 * Because we unregister the device on the last put we need
311 * a separate refcount.
313 struct kref ref;
314 struct list_head entry;
315 struct mutex lock;
316 struct list_head ctrls;
317 struct list_head nsheads;
318 char subnqn[NVMF_NQN_SIZE];
319 char serial[20];
320 char model[40];
321 char firmware_rev[8];
322 u8 cmic;
323 u16 vendor_id;
324 u16 awupf; /* 0's based awupf value. */
325 struct ida ns_ida;
326 #ifdef CONFIG_NVME_MULTIPATH
327 enum nvme_iopolicy iopolicy;
328 #endif
332 * Container structure for uniqueue namespace identifiers.
334 struct nvme_ns_ids {
335 u8 eui64[8];
336 u8 nguid[16];
337 uuid_t uuid;
341 * Anchor structure for namespaces. There is one for each namespace in a
342 * NVMe subsystem that any of our controllers can see, and the namespace
343 * structure for each controller is chained of it. For private namespaces
344 * there is a 1:1 relation to our namespace structures, that is ->list
345 * only ever has a single entry for private namespaces.
347 struct nvme_ns_head {
348 struct list_head list;
349 struct srcu_struct srcu;
350 struct nvme_subsystem *subsys;
351 unsigned ns_id;
352 struct nvme_ns_ids ids;
353 struct list_head entry;
354 struct kref ref;
355 int instance;
356 #ifdef CONFIG_NVME_MULTIPATH
357 struct gendisk *disk;
358 struct bio_list requeue_list;
359 spinlock_t requeue_lock;
360 struct work_struct requeue_work;
361 struct mutex lock;
362 struct nvme_ns __rcu *current_path[];
363 #endif
366 struct nvme_ns {
367 struct list_head list;
369 struct nvme_ctrl *ctrl;
370 struct request_queue *queue;
371 struct gendisk *disk;
372 #ifdef CONFIG_NVME_MULTIPATH
373 enum nvme_ana_state ana_state;
374 u32 ana_grpid;
375 #endif
376 struct list_head siblings;
377 struct nvm_dev *ndev;
378 struct kref kref;
379 struct nvme_ns_head *head;
381 int lba_shift;
382 u16 ms;
383 u16 sgs;
384 u32 sws;
385 bool ext;
386 u8 pi_type;
387 unsigned long flags;
388 #define NVME_NS_REMOVING 0
389 #define NVME_NS_DEAD 1
390 #define NVME_NS_ANA_PENDING 2
391 u16 noiob;
393 struct nvme_fault_inject fault_inject;
397 struct nvme_ctrl_ops {
398 const char *name;
399 struct module *module;
400 unsigned int flags;
401 #define NVME_F_FABRICS (1 << 0)
402 #define NVME_F_METADATA_SUPPORTED (1 << 1)
403 #define NVME_F_PCI_P2PDMA (1 << 2)
404 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
405 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
406 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
407 void (*free_ctrl)(struct nvme_ctrl *ctrl);
408 void (*submit_async_event)(struct nvme_ctrl *ctrl);
409 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
410 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
413 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
414 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
415 const char *dev_name);
416 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
417 void nvme_should_fail(struct request *req);
418 #else
419 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
420 const char *dev_name)
423 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
426 static inline void nvme_should_fail(struct request *req) {}
427 #endif
429 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
431 if (!ctrl->subsystem)
432 return -ENOTTY;
433 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
437 * Convert a 512B sector number to a device logical block number.
439 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
441 return sector >> (ns->lba_shift - SECTOR_SHIFT);
445 * Convert a device logical block number to a 512B sector number.
447 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
449 return lba << (ns->lba_shift - SECTOR_SHIFT);
452 static inline void nvme_end_request(struct request *req, __le16 status,
453 union nvme_result result)
455 struct nvme_request *rq = nvme_req(req);
457 rq->status = le16_to_cpu(status) >> 1;
458 rq->result = result;
459 /* inject error when permitted by fault injection framework */
460 nvme_should_fail(req);
461 blk_mq_complete_request(req);
464 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
466 get_device(ctrl->device);
469 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
471 put_device(ctrl->device);
474 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
476 return !qid && command_id >= NVME_AQ_BLK_MQ_DEPTH;
479 void nvme_complete_rq(struct request *req);
480 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
481 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
482 enum nvme_ctrl_state new_state);
483 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
484 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
485 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
486 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
487 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
488 const struct nvme_ctrl_ops *ops, unsigned long quirks);
489 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
490 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
491 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
492 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
493 int nvme_init_identify(struct nvme_ctrl *ctrl);
495 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
497 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
498 bool send);
500 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
501 volatile union nvme_result *res);
503 void nvme_stop_queues(struct nvme_ctrl *ctrl);
504 void nvme_start_queues(struct nvme_ctrl *ctrl);
505 void nvme_kill_queues(struct nvme_ctrl *ctrl);
506 void nvme_sync_queues(struct nvme_ctrl *ctrl);
507 void nvme_unfreeze(struct nvme_ctrl *ctrl);
508 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
509 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
510 void nvme_start_freeze(struct nvme_ctrl *ctrl);
512 #define NVME_QID_ANY -1
513 struct request *nvme_alloc_request(struct request_queue *q,
514 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
515 void nvme_cleanup_cmd(struct request *req);
516 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
517 struct nvme_command *cmd);
518 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
519 void *buf, unsigned bufflen);
520 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
521 union nvme_result *result, void *buffer, unsigned bufflen,
522 unsigned timeout, int qid, int at_head,
523 blk_mq_req_flags_t flags, bool poll);
524 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
525 unsigned int dword11, void *buffer, size_t buflen,
526 u32 *result);
527 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
528 unsigned int dword11, void *buffer, size_t buflen,
529 u32 *result);
530 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
531 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
532 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
533 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
534 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
535 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
537 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
538 void *log, size_t size, u64 offset);
540 extern const struct attribute_group *nvme_ns_id_attr_groups[];
541 extern const struct block_device_operations nvme_ns_head_ops;
543 #ifdef CONFIG_NVME_MULTIPATH
544 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
546 return ctrl->ana_log_buf != NULL;
549 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
550 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
551 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
552 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
553 struct nvme_ctrl *ctrl, int *flags);
554 bool nvme_failover_req(struct request *req);
555 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
556 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
557 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
558 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
559 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
560 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
561 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
562 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
563 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
564 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
566 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
568 struct nvme_ns_head *head = ns->head;
570 if (head->disk && list_empty(&head->list))
571 kblockd_schedule_work(&head->requeue_work);
574 static inline void nvme_trace_bio_complete(struct request *req,
575 blk_status_t status)
577 struct nvme_ns *ns = req->q->queuedata;
579 if (req->cmd_flags & REQ_NVME_MPATH)
580 trace_block_bio_complete(ns->head->disk->queue,
581 req->bio, status);
584 extern struct device_attribute dev_attr_ana_grpid;
585 extern struct device_attribute dev_attr_ana_state;
586 extern struct device_attribute subsys_attr_iopolicy;
588 #else
589 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
591 return false;
594 * Without the multipath code enabled, multiple controller per subsystems are
595 * visible as devices and thus we cannot use the subsystem instance.
597 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
598 struct nvme_ctrl *ctrl, int *flags)
600 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
603 static inline bool nvme_failover_req(struct request *req)
605 return false;
607 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
610 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
611 struct nvme_ns_head *head)
613 return 0;
615 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
616 struct nvme_id_ns *id)
619 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
622 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
624 return false;
626 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
629 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
632 static inline void nvme_trace_bio_complete(struct request *req,
633 blk_status_t status)
636 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
637 struct nvme_id_ctrl *id)
639 if (ctrl->subsys->cmic & (1 << 3))
640 dev_warn(ctrl->device,
641 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
642 return 0;
644 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
647 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
650 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
653 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
656 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
659 #endif /* CONFIG_NVME_MULTIPATH */
661 #ifdef CONFIG_NVM
662 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
663 void nvme_nvm_unregister(struct nvme_ns *ns);
664 extern const struct attribute_group nvme_nvm_attr_group;
665 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
666 #else
667 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
668 int node)
670 return 0;
673 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
674 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
675 unsigned long arg)
677 return -ENOTTY;
679 #endif /* CONFIG_NVM */
681 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
683 return dev_to_disk(dev)->private_data;
686 #ifdef CONFIG_NVME_HWMON
687 void nvme_hwmon_init(struct nvme_ctrl *ctrl);
688 #else
689 static inline void nvme_hwmon_init(struct nvme_ctrl *ctrl) { }
690 #endif
692 #endif /* _NVME_H */