xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 8938c48f)
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/lightnvm.h>
15 #include <linux/sed-opal.h>
16 #include <linux/fault-inject.h>
17 #include <linux/rcupdate.h>
18 #include <linux/wait.h>
19 #include <linux/t10-pi.h>
20 
21 #include <trace/events/block.h>
22 
23 extern unsigned int nvme_io_timeout;
24 #define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
25 
26 extern unsigned int admin_timeout;
27 #define NVME_ADMIN_TIMEOUT	(admin_timeout * HZ)
28 
29 #define NVME_DEFAULT_KATO	5
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)
47 
48 extern struct workqueue_struct *nvme_wq;
49 extern struct workqueue_struct *nvme_reset_wq;
50 extern struct workqueue_struct *nvme_delete_wq;
51 
52 enum {
53 	NVME_NS_LBA		= 0,
54 	NVME_NS_LIGHTNVM	= 1,
55 };
56 
57 /*
58  * List of workarounds for devices that required behavior not specified in
59  * the standard.
60  */
61 enum nvme_quirks {
62 	/*
63 	 * Prefers I/O aligned to a stripe size specified in a vendor
64 	 * specific Identify field.
65 	 */
66 	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
67 
68 	/*
69 	 * The controller doesn't handle Identify value others than 0 or 1
70 	 * correctly.
71 	 */
72 	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
73 
74 	/*
75 	 * The controller deterministically returns O's on reads to
76 	 * logical blocks that deallocate was called on.
77 	 */
78 	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
79 
80 	/*
81 	 * The controller needs a delay before starts checking the device
82 	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
83 	 */
84 	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
85 
86 	/*
87 	 * APST should not be used.
88 	 */
89 	NVME_QUIRK_NO_APST			= (1 << 4),
90 
91 	/*
92 	 * The deepest sleep state should not be used.
93 	 */
94 	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
95 
96 	/*
97 	 * Supports the LighNVM command set if indicated in vs[1].
98 	 */
99 	NVME_QUIRK_LIGHTNVM			= (1 << 6),
100 
101 	/*
102 	 * Set MEDIUM priority on SQ creation
103 	 */
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 does not properly handle DMA addresses over
150 	 * 48 bits.
151 	 */
152 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
153 };
154 
155 /*
156  * Common request structure for NVMe passthrough.  All drivers must have
157  * this structure as the first member of their request-private data.
158  */
159 struct nvme_request {
160 	struct nvme_command	*cmd;
161 	union nvme_result	result;
162 	u8			retries;
163 	u8			flags;
164 	u16			status;
165 	struct nvme_ctrl	*ctrl;
166 };
167 
168 /*
169  * Mark a bio as coming in through the mpath node.
170  */
171 #define REQ_NVME_MPATH		REQ_DRV
172 
173 enum {
174 	NVME_REQ_CANCELLED		= (1 << 0),
175 	NVME_REQ_USERCMD		= (1 << 1),
176 };
177 
178 static inline struct nvme_request *nvme_req(struct request *req)
179 {
180 	return blk_mq_rq_to_pdu(req);
181 }
182 
183 static inline u16 nvme_req_qid(struct request *req)
184 {
185 	if (!req->q->queuedata)
186 		return 0;
187 
188 	return req->mq_hctx->queue_num + 1;
189 }
190 
191 /* The below value is the specific amount of delay needed before checking
192  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
193  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
194  * found empirically.
195  */
196 #define NVME_QUIRK_DELAY_AMOUNT		2300
197 
198 /*
199  * enum nvme_ctrl_state: Controller state
200  *
201  * @NVME_CTRL_NEW:		New controller just allocated, initial state
202  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
203  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
204  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
205  *				transport
206  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
207  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
208  *				disabled/failed immediately. This state comes
209  * 				after all async event processing took place and
210  * 				before ns removal and the controller deletion
211  * 				progress
212  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
213  *				shutdown or removal. In this case we forcibly
214  *				kill all inflight I/O as they have no chance to
215  *				complete
216  */
217 enum nvme_ctrl_state {
218 	NVME_CTRL_NEW,
219 	NVME_CTRL_LIVE,
220 	NVME_CTRL_RESETTING,
221 	NVME_CTRL_CONNECTING,
222 	NVME_CTRL_DELETING,
223 	NVME_CTRL_DELETING_NOIO,
224 	NVME_CTRL_DEAD,
225 };
226 
227 struct nvme_fault_inject {
228 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
229 	struct fault_attr attr;
230 	struct dentry *parent;
231 	bool dont_retry;	/* DNR, do not retry */
232 	u16 status;		/* status code */
233 #endif
234 };
235 
236 struct nvme_ctrl {
237 	bool comp_seen;
238 	enum nvme_ctrl_state state;
239 	bool identified;
240 	spinlock_t lock;
241 	struct mutex scan_lock;
242 	const struct nvme_ctrl_ops *ops;
243 	struct request_queue *admin_q;
244 	struct request_queue *connect_q;
245 	struct request_queue *fabrics_q;
246 	struct device *dev;
247 	int instance;
248 	int numa_node;
249 	struct blk_mq_tag_set *tagset;
250 	struct blk_mq_tag_set *admin_tagset;
251 	struct list_head namespaces;
252 	struct rw_semaphore namespaces_rwsem;
253 	struct device ctrl_device;
254 	struct device *device;	/* char device */
255 #ifdef CONFIG_NVME_HWMON
256 	struct device *hwmon_device;
257 #endif
258 	struct cdev cdev;
259 	struct work_struct reset_work;
260 	struct work_struct delete_work;
261 	wait_queue_head_t state_wq;
262 
263 	struct nvme_subsystem *subsys;
264 	struct list_head subsys_entry;
265 
266 	struct opal_dev *opal_dev;
267 
268 	char name[12];
269 	u16 cntlid;
270 
271 	u32 ctrl_config;
272 	u16 mtfa;
273 	u32 queue_count;
274 
275 	u64 cap;
276 	u32 max_hw_sectors;
277 	u32 max_segments;
278 	u32 max_integrity_segments;
279 #ifdef CONFIG_BLK_DEV_ZONED
280 	u32 max_zone_append;
281 #endif
282 	u16 crdt[3];
283 	u16 oncs;
284 	u16 oacs;
285 	u16 nssa;
286 	u16 nr_streams;
287 	u16 sqsize;
288 	u32 max_namespaces;
289 	atomic_t abort_limit;
290 	u8 vwc;
291 	u32 vs;
292 	u32 sgls;
293 	u16 kas;
294 	u8 npss;
295 	u8 apsta;
296 	u16 wctemp;
297 	u16 cctemp;
298 	u32 oaes;
299 	u32 aen_result;
300 	u32 ctratt;
301 	unsigned int shutdown_timeout;
302 	unsigned int kato;
303 	bool subsystem;
304 	unsigned long quirks;
305 	struct nvme_id_power_state psd[32];
306 	struct nvme_effects_log *effects;
307 	struct xarray cels;
308 	struct work_struct scan_work;
309 	struct work_struct async_event_work;
310 	struct delayed_work ka_work;
311 	struct delayed_work failfast_work;
312 	struct nvme_command ka_cmd;
313 	struct work_struct fw_act_work;
314 	unsigned long events;
315 
316 #ifdef CONFIG_NVME_MULTIPATH
317 	/* asymmetric namespace access: */
318 	u8 anacap;
319 	u8 anatt;
320 	u32 anagrpmax;
321 	u32 nanagrpid;
322 	struct mutex ana_lock;
323 	struct nvme_ana_rsp_hdr *ana_log_buf;
324 	size_t ana_log_size;
325 	struct timer_list anatt_timer;
326 	struct work_struct ana_work;
327 #endif
328 
329 	/* Power saving configuration */
330 	u64 ps_max_latency_us;
331 	bool apst_enabled;
332 
333 	/* PCIe only: */
334 	u32 hmpre;
335 	u32 hmmin;
336 	u32 hmminds;
337 	u16 hmmaxd;
338 
339 	/* Fabrics only */
340 	u32 ioccsz;
341 	u32 iorcsz;
342 	u16 icdoff;
343 	u16 maxcmd;
344 	int nr_reconnects;
345 	unsigned long flags;
346 #define NVME_CTRL_FAILFAST_EXPIRED	0
347 	struct nvmf_ctrl_options *opts;
348 
349 	struct page *discard_page;
350 	unsigned long discard_page_busy;
351 
352 	struct nvme_fault_inject fault_inject;
353 };
354 
355 enum nvme_iopolicy {
356 	NVME_IOPOLICY_NUMA,
357 	NVME_IOPOLICY_RR,
358 };
359 
360 struct nvme_subsystem {
361 	int			instance;
362 	struct device		dev;
363 	/*
364 	 * Because we unregister the device on the last put we need
365 	 * a separate refcount.
366 	 */
367 	struct kref		ref;
368 	struct list_head	entry;
369 	struct mutex		lock;
370 	struct list_head	ctrls;
371 	struct list_head	nsheads;
372 	char			subnqn[NVMF_NQN_SIZE];
373 	char			serial[20];
374 	char			model[40];
375 	char			firmware_rev[8];
376 	u8			cmic;
377 	u16			vendor_id;
378 	u16			awupf;	/* 0's based awupf value. */
379 	struct ida		ns_ida;
380 #ifdef CONFIG_NVME_MULTIPATH
381 	enum nvme_iopolicy	iopolicy;
382 #endif
383 };
384 
385 /*
386  * Container structure for uniqueue namespace identifiers.
387  */
388 struct nvme_ns_ids {
389 	u8	eui64[8];
390 	u8	nguid[16];
391 	uuid_t	uuid;
392 	u8	csi;
393 };
394 
395 /*
396  * Anchor structure for namespaces.  There is one for each namespace in a
397  * NVMe subsystem that any of our controllers can see, and the namespace
398  * structure for each controller is chained of it.  For private namespaces
399  * there is a 1:1 relation to our namespace structures, that is ->list
400  * only ever has a single entry for private namespaces.
401  */
402 struct nvme_ns_head {
403 	struct list_head	list;
404 	struct srcu_struct      srcu;
405 	struct nvme_subsystem	*subsys;
406 	unsigned		ns_id;
407 	struct nvme_ns_ids	ids;
408 	struct list_head	entry;
409 	struct kref		ref;
410 	bool			shared;
411 	int			instance;
412 	struct nvme_effects_log *effects;
413 #ifdef CONFIG_NVME_MULTIPATH
414 	struct gendisk		*disk;
415 	struct bio_list		requeue_list;
416 	spinlock_t		requeue_lock;
417 	struct work_struct	requeue_work;
418 	struct mutex		lock;
419 	unsigned long		flags;
420 #define NVME_NSHEAD_DISK_LIVE	0
421 	struct nvme_ns __rcu	*current_path[];
422 #endif
423 };
424 
425 enum nvme_ns_features {
426 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
427 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
428 };
429 
430 struct nvme_ns {
431 	struct list_head list;
432 
433 	struct nvme_ctrl *ctrl;
434 	struct request_queue *queue;
435 	struct gendisk *disk;
436 #ifdef CONFIG_NVME_MULTIPATH
437 	enum nvme_ana_state ana_state;
438 	u32 ana_grpid;
439 #endif
440 	struct list_head siblings;
441 	struct nvm_dev *ndev;
442 	struct kref kref;
443 	struct nvme_ns_head *head;
444 
445 	int lba_shift;
446 	u16 ms;
447 	u16 sgs;
448 	u32 sws;
449 	u8 pi_type;
450 #ifdef CONFIG_BLK_DEV_ZONED
451 	u64 zsze;
452 #endif
453 	unsigned long features;
454 	unsigned long flags;
455 #define NVME_NS_REMOVING	0
456 #define NVME_NS_DEAD     	1
457 #define NVME_NS_ANA_PENDING	2
458 #define NVME_NS_FORCE_RO	3
459 
460 	struct nvme_fault_inject fault_inject;
461 
462 };
463 
464 /* NVMe ns supports metadata actions by the controller (generate/strip) */
465 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
466 {
467 	return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
468 }
469 
470 struct nvme_ctrl_ops {
471 	const char *name;
472 	struct module *module;
473 	unsigned int flags;
474 #define NVME_F_FABRICS			(1 << 0)
475 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
476 #define NVME_F_PCI_P2PDMA		(1 << 2)
477 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
478 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
479 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
480 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
481 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
482 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
483 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
484 };
485 
486 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
487 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
488 			    const char *dev_name);
489 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
490 void nvme_should_fail(struct request *req);
491 #else
492 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
493 					  const char *dev_name)
494 {
495 }
496 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
497 {
498 }
499 static inline void nvme_should_fail(struct request *req) {}
500 #endif
501 
502 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
503 {
504 	if (!ctrl->subsystem)
505 		return -ENOTTY;
506 	return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
507 }
508 
509 /*
510  * Convert a 512B sector number to a device logical block number.
511  */
512 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
513 {
514 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
515 }
516 
517 /*
518  * Convert a device logical block number to a 512B sector number.
519  */
520 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
521 {
522 	return lba << (ns->lba_shift - SECTOR_SHIFT);
523 }
524 
525 /*
526  * Convert byte length to nvme's 0-based num dwords
527  */
528 static inline u32 nvme_bytes_to_numd(size_t len)
529 {
530 	return (len >> 2) - 1;
531 }
532 
533 static inline bool nvme_is_ana_error(u16 status)
534 {
535 	switch (status & 0x7ff) {
536 	case NVME_SC_ANA_TRANSITION:
537 	case NVME_SC_ANA_INACCESSIBLE:
538 	case NVME_SC_ANA_PERSISTENT_LOSS:
539 		return true;
540 	default:
541 		return false;
542 	}
543 }
544 
545 static inline bool nvme_is_path_error(u16 status)
546 {
547 	/* check for a status code type of 'path related status' */
548 	return (status & 0x700) == 0x300;
549 }
550 
551 /*
552  * Fill in the status and result information from the CQE, and then figure out
553  * if blk-mq will need to use IPI magic to complete the request, and if yes do
554  * so.  If not let the caller complete the request without an indirect function
555  * call.
556  */
557 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
558 		union nvme_result result)
559 {
560 	struct nvme_request *rq = nvme_req(req);
561 
562 	rq->status = le16_to_cpu(status) >> 1;
563 	rq->result = result;
564 	/* inject error when permitted by fault injection framework */
565 	nvme_should_fail(req);
566 	if (unlikely(blk_should_fake_timeout(req->q)))
567 		return true;
568 	return blk_mq_complete_request_remote(req);
569 }
570 
571 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
572 {
573 	get_device(ctrl->device);
574 }
575 
576 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
577 {
578 	put_device(ctrl->device);
579 }
580 
581 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
582 {
583 	return !qid && command_id >= NVME_AQ_BLK_MQ_DEPTH;
584 }
585 
586 void nvme_complete_rq(struct request *req);
587 blk_status_t nvme_host_path_error(struct request *req);
588 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
589 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
590 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
591 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
592 		enum nvme_ctrl_state new_state);
593 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
594 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
595 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
596 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
597 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
598 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
599 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
600 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
601 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
602 int nvme_init_identify(struct nvme_ctrl *ctrl);
603 
604 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
605 
606 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
607 		bool send);
608 
609 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
610 		volatile union nvme_result *res);
611 
612 void nvme_stop_queues(struct nvme_ctrl *ctrl);
613 void nvme_start_queues(struct nvme_ctrl *ctrl);
614 void nvme_kill_queues(struct nvme_ctrl *ctrl);
615 void nvme_sync_queues(struct nvme_ctrl *ctrl);
616 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
617 void nvme_unfreeze(struct nvme_ctrl *ctrl);
618 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
619 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
620 void nvme_start_freeze(struct nvme_ctrl *ctrl);
621 
622 #define NVME_QID_ANY -1
623 struct request *nvme_alloc_request(struct request_queue *q,
624 		struct nvme_command *cmd, blk_mq_req_flags_t flags);
625 void nvme_cleanup_cmd(struct request *req);
626 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
627 		struct nvme_command *cmd);
628 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
629 		void *buf, unsigned bufflen);
630 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
631 		union nvme_result *result, void *buffer, unsigned bufflen,
632 		unsigned timeout, int qid, int at_head,
633 		blk_mq_req_flags_t flags, bool poll);
634 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
635 		      unsigned int dword11, void *buffer, size_t buflen,
636 		      u32 *result);
637 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
638 		      unsigned int dword11, void *buffer, size_t buflen,
639 		      u32 *result);
640 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
641 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
642 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
643 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
644 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
645 
646 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
647 		void *log, size_t size, u64 offset);
648 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
649 		struct nvme_ns_head **head, int *srcu_idx);
650 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx);
651 
652 extern const struct attribute_group *nvme_ns_id_attr_groups[];
653 extern const struct block_device_operations nvme_ns_head_ops;
654 
655 #ifdef CONFIG_NVME_MULTIPATH
656 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
657 {
658 	return ctrl->ana_log_buf != NULL;
659 }
660 
661 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
662 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
663 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
664 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
665 			struct nvme_ctrl *ctrl, int *flags);
666 void nvme_failover_req(struct request *req);
667 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
668 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
669 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
670 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
671 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
672 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
673 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
674 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
675 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
676 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
677 blk_qc_t nvme_ns_head_submit_bio(struct bio *bio);
678 
679 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
680 {
681 	struct nvme_ns_head *head = ns->head;
682 
683 	if (head->disk && list_empty(&head->list))
684 		kblockd_schedule_work(&head->requeue_work);
685 }
686 
687 static inline void nvme_trace_bio_complete(struct request *req)
688 {
689 	struct nvme_ns *ns = req->q->queuedata;
690 
691 	if (req->cmd_flags & REQ_NVME_MPATH)
692 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
693 }
694 
695 extern struct device_attribute dev_attr_ana_grpid;
696 extern struct device_attribute dev_attr_ana_state;
697 extern struct device_attribute subsys_attr_iopolicy;
698 
699 #else
700 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
701 {
702 	return false;
703 }
704 /*
705  * Without the multipath code enabled, multiple controller per subsystems are
706  * visible as devices and thus we cannot use the subsystem instance.
707  */
708 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
709 				      struct nvme_ctrl *ctrl, int *flags)
710 {
711 	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
712 }
713 
714 static inline void nvme_failover_req(struct request *req)
715 {
716 }
717 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
718 {
719 }
720 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
721 		struct nvme_ns_head *head)
722 {
723 	return 0;
724 }
725 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
726 		struct nvme_id_ns *id)
727 {
728 }
729 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
730 {
731 }
732 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
733 {
734 	return false;
735 }
736 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
737 {
738 }
739 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
740 {
741 }
742 static inline void nvme_trace_bio_complete(struct request *req)
743 {
744 }
745 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
746 		struct nvme_id_ctrl *id)
747 {
748 	if (ctrl->subsys->cmic & (1 << 3))
749 		dev_warn(ctrl->device,
750 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
751 	return 0;
752 }
753 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
754 {
755 }
756 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
757 {
758 }
759 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
760 {
761 }
762 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
763 {
764 }
765 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
766 {
767 }
768 #endif /* CONFIG_NVME_MULTIPATH */
769 
770 int nvme_revalidate_zones(struct nvme_ns *ns);
771 #ifdef CONFIG_BLK_DEV_ZONED
772 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
773 int nvme_report_zones(struct gendisk *disk, sector_t sector,
774 		      unsigned int nr_zones, report_zones_cb cb, void *data);
775 
776 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
777 				       struct nvme_command *cmnd,
778 				       enum nvme_zone_mgmt_action action);
779 #else
780 #define nvme_report_zones NULL
781 
782 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
783 		struct request *req, struct nvme_command *cmnd,
784 		enum nvme_zone_mgmt_action action)
785 {
786 	return BLK_STS_NOTSUPP;
787 }
788 
789 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
790 {
791 	dev_warn(ns->ctrl->device,
792 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
793 	return -EPROTONOSUPPORT;
794 }
795 #endif
796 
797 #ifdef CONFIG_NVM
798 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
799 void nvme_nvm_unregister(struct nvme_ns *ns);
800 extern const struct attribute_group nvme_nvm_attr_group;
801 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
802 #else
803 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
804 				    int node)
805 {
806 	return 0;
807 }
808 
809 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
810 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
811 							unsigned long arg)
812 {
813 	return -ENOTTY;
814 }
815 #endif /* CONFIG_NVM */
816 
817 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
818 {
819 	return dev_to_disk(dev)->private_data;
820 }
821 
822 #ifdef CONFIG_NVME_HWMON
823 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
824 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
825 #else
826 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
827 {
828 	return 0;
829 }
830 
831 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
832 {
833 }
834 #endif
835 
836 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
837 			 u8 opcode);
838 void nvme_execute_passthru_rq(struct request *rq);
839 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
840 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
841 void nvme_put_ns(struct nvme_ns *ns);
842 
843 #endif /* _NVME_H */
844