xref: /linux/drivers/nvme/host/nvme.h (revision 72ece656abd4de8645a421fb482d7a11e6533a45)
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/sed-opal.h>
15 #include <linux/fault-inject.h>
16 #include <linux/rcupdate.h>
17 #include <linux/wait.h>
18 #include <linux/t10-pi.h>
19 #include <linux/ratelimit_types.h>
20 
21 #include <trace/events/block.h>
22 
23 extern const struct pr_ops nvme_pr_ops;
24 
25 extern unsigned int nvme_io_timeout;
26 #define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
27 
28 extern unsigned int admin_timeout;
29 #define NVME_ADMIN_TIMEOUT	(admin_timeout * HZ)
30 
31 #define NVME_DEFAULT_KATO	5
32 
33 #ifdef CONFIG_ARCH_NO_SG_CHAIN
34 #define  NVME_INLINE_SG_CNT  0
35 #define  NVME_INLINE_METADATA_SG_CNT  0
36 #else
37 #define  NVME_INLINE_SG_CNT  2
38 #define  NVME_INLINE_METADATA_SG_CNT  1
39 #endif
40 
41 /*
42  * Default to a 4K page size, with the intention to update this
43  * path in the future to accommodate architectures with differing
44  * kernel and IO page sizes.
45  */
46 #define NVME_CTRL_PAGE_SHIFT	12
47 #define NVME_CTRL_PAGE_SIZE	(1 << NVME_CTRL_PAGE_SHIFT)
48 
49 extern struct workqueue_struct *nvme_wq;
50 extern struct workqueue_struct *nvme_reset_wq;
51 extern struct workqueue_struct *nvme_delete_wq;
52 extern struct mutex nvme_subsystems_lock;
53 
54 /*
55  * List of workarounds for devices that required behavior not specified in
56  * the standard.
57  */
58 enum nvme_quirks {
59 	/*
60 	 * Prefers I/O aligned to a stripe size specified in a vendor
61 	 * specific Identify field.
62 	 */
63 	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
64 
65 	/*
66 	 * The controller doesn't handle Identify value others than 0 or 1
67 	 * correctly.
68 	 */
69 	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
70 
71 	/*
72 	 * The controller deterministically returns 0's on reads to
73 	 * logical blocks that deallocate was called on.
74 	 */
75 	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
76 
77 	/*
78 	 * The controller needs a delay before starts checking the device
79 	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
80 	 */
81 	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
82 
83 	/*
84 	 * APST should not be used.
85 	 */
86 	NVME_QUIRK_NO_APST			= (1 << 4),
87 
88 	/*
89 	 * The deepest sleep state should not be used.
90 	 */
91 	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
92 
93 	/*
94 	 *  Problems seen with concurrent commands
95 	 */
96 	NVME_QUIRK_QDEPTH_ONE			= (1 << 6),
97 
98 	/*
99 	 * Set MEDIUM priority on SQ creation
100 	 */
101 	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
102 
103 	/*
104 	 * Ignore device provided subnqn.
105 	 */
106 	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
107 
108 	/*
109 	 * Broken Write Zeroes.
110 	 */
111 	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
112 
113 	/*
114 	 * Force simple suspend/resume path.
115 	 */
116 	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
117 
118 	/*
119 	 * Use only one interrupt vector for all queues
120 	 */
121 	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
122 
123 	/*
124 	 * Use non-standard 128 bytes SQEs.
125 	 */
126 	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
127 
128 	/*
129 	 * Prevent tag overlap between queues
130 	 */
131 	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
132 
133 	/*
134 	 * Don't change the value of the temperature threshold feature
135 	 */
136 	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
137 
138 	/*
139 	 * The controller doesn't handle the Identify Namespace
140 	 * Identification Descriptor list subcommand despite claiming
141 	 * NVMe 1.3 compliance.
142 	 */
143 	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
144 
145 	/*
146 	 * The controller does not properly handle DMA addresses over
147 	 * 48 bits.
148 	 */
149 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
150 
151 	/*
152 	 * The controller requires the command_id value be limited, so skip
153 	 * encoding the generation sequence number.
154 	 */
155 	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
156 
157 	/*
158 	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
159 	 */
160 	NVME_QUIRK_BOGUS_NID			= (1 << 18),
161 
162 	/*
163 	 * No temperature thresholds for channels other than 0 (Composite).
164 	 */
165 	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
166 
167 	/*
168 	 * Disables simple suspend/resume path.
169 	 */
170 	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
171 
172 	/*
173 	 * MSI (but not MSI-X) interrupts are broken and never fire.
174 	 */
175 	NVME_QUIRK_BROKEN_MSI			= (1 << 21),
176 
177 	/*
178 	 * Align dma pool segment size to 512 bytes
179 	 */
180 	NVME_QUIRK_DMAPOOL_ALIGN_512		= (1 << 22),
181 
182 	/*
183 	 * Admin queue DMA buffers must be page aligned
184 	 */
185 	NVME_QUIRK_ADMIN_PAGE_ALIGN		= (1 << 23),
186 };
187 
188 static inline char *nvme_quirk_name(enum nvme_quirks q)
189 {
190 	switch (q) {
191 	case NVME_QUIRK_STRIPE_SIZE:
192 		return "stripe_size";
193 	case NVME_QUIRK_IDENTIFY_CNS:
194 		return "identify_cns";
195 	case NVME_QUIRK_DEALLOCATE_ZEROES:
196 		return "deallocate_zeroes";
197 	case NVME_QUIRK_DELAY_BEFORE_CHK_RDY:
198 		return "delay_before_chk_rdy";
199 	case NVME_QUIRK_NO_APST:
200 		return "no_apst";
201 	case NVME_QUIRK_NO_DEEPEST_PS:
202 		return "no_deepest_ps";
203 	case NVME_QUIRK_QDEPTH_ONE:
204 		return "qdepth_one";
205 	case NVME_QUIRK_MEDIUM_PRIO_SQ:
206 		return "medium_prio_sq";
207 	case NVME_QUIRK_IGNORE_DEV_SUBNQN:
208 		return "ignore_dev_subnqn";
209 	case NVME_QUIRK_DISABLE_WRITE_ZEROES:
210 		return "disable_write_zeroes";
211 	case NVME_QUIRK_SIMPLE_SUSPEND:
212 		return "simple_suspend";
213 	case NVME_QUIRK_SINGLE_VECTOR:
214 		return "single_vector";
215 	case NVME_QUIRK_128_BYTES_SQES:
216 		return "128_bytes_sqes";
217 	case NVME_QUIRK_SHARED_TAGS:
218 		return "shared_tags";
219 	case NVME_QUIRK_NO_TEMP_THRESH_CHANGE:
220 		return "no_temp_thresh_change";
221 	case NVME_QUIRK_NO_NS_DESC_LIST:
222 		return "no_ns_desc_list";
223 	case NVME_QUIRK_DMA_ADDRESS_BITS_48:
224 		return "dma_address_bits_48";
225 	case NVME_QUIRK_SKIP_CID_GEN:
226 		return "skip_cid_gen";
227 	case NVME_QUIRK_BOGUS_NID:
228 		return "bogus_nid";
229 	case NVME_QUIRK_NO_SECONDARY_TEMP_THRESH:
230 		return "no_secondary_temp_thresh";
231 	case NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND:
232 		return "force_no_simple_suspend";
233 	case NVME_QUIRK_BROKEN_MSI:
234 		return "broken_msi";
235 	case NVME_QUIRK_DMAPOOL_ALIGN_512:
236 		return "dmapool_align_512";
237 	case NVME_QUIRK_ADMIN_PAGE_ALIGN:
238 		return "admin_page_align";
239 	}
240 
241 	return "unknown";
242 }
243 
244 /*
245  * Common request structure for NVMe passthrough.  All drivers must have
246  * this structure as the first member of their request-private data.
247  */
248 struct nvme_request {
249 	struct nvme_command	*cmd;
250 	union nvme_result	result;
251 	u8			genctr;
252 	u8			retries;
253 	u8			flags;
254 	u16			status;
255 #ifdef CONFIG_NVME_MULTIPATH
256 	unsigned long		start_time;
257 #endif
258 	struct nvme_ctrl	*ctrl;
259 };
260 
261 /*
262  * Mark a bio as coming in through the mpath node.
263  */
264 #define REQ_NVME_MPATH		REQ_DRV
265 
266 enum {
267 	NVME_REQ_CANCELLED		= (1 << 0),
268 	NVME_REQ_USERCMD		= (1 << 1),
269 	NVME_MPATH_IO_STATS		= (1 << 2),
270 	NVME_MPATH_CNT_ACTIVE		= (1 << 3),
271 };
272 
273 static inline struct nvme_request *nvme_req(struct request *req)
274 {
275 	return blk_mq_rq_to_pdu(req);
276 }
277 
278 static inline u16 nvme_req_qid(struct request *req)
279 {
280 	if (!req->q->queuedata)
281 		return 0;
282 
283 	return req->mq_hctx->queue_num + 1;
284 }
285 
286 /* The below value is the specific amount of delay needed before checking
287  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
288  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
289  * found empirically.
290  */
291 #define NVME_QUIRK_DELAY_AMOUNT		2300
292 
293 /*
294  * enum nvme_ctrl_state: Controller state
295  *
296  * @NVME_CTRL_NEW:		New controller just allocated, initial state
297  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
298  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
299  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
300  *				transport
301  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
302  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
303  *				disabled/failed immediately. This state comes
304  * 				after all async event processing took place and
305  * 				before ns removal and the controller deletion
306  * 				progress
307  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
308  *				shutdown or removal. In this case we forcibly
309  *				kill all inflight I/O as they have no chance to
310  *				complete
311  */
312 enum nvme_ctrl_state {
313 	NVME_CTRL_NEW,
314 	NVME_CTRL_LIVE,
315 	NVME_CTRL_RESETTING,
316 	NVME_CTRL_CONNECTING,
317 	NVME_CTRL_DELETING,
318 	NVME_CTRL_DELETING_NOIO,
319 	NVME_CTRL_DEAD,
320 };
321 
322 struct nvme_fault_inject {
323 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
324 	struct fault_attr attr;
325 	struct dentry *parent;
326 	u16 opcode;
327 	bool dont_retry;	/* DNR, do not retry */
328 	u16 status;		/* status code */
329 #endif
330 };
331 
332 enum nvme_ctrl_flags {
333 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
334 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
335 	NVME_CTRL_STARTED_ONCE		= 2,
336 	NVME_CTRL_STOPPED		= 3,
337 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
338 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
339 	NVME_CTRL_FROZEN		= 6,
340 };
341 
342 struct nvme_ctrl {
343 	bool comp_seen;
344 	bool identified;
345 	bool passthru_err_log_enabled;
346 	enum nvme_ctrl_state state;
347 	spinlock_t lock;
348 	struct mutex scan_lock;
349 	const struct nvme_ctrl_ops *ops;
350 	struct request_queue *admin_q;
351 	struct request_queue *connect_q;
352 	struct request_queue *fabrics_q;
353 	struct device *dev;
354 	int instance;
355 	int numa_node;
356 	struct blk_mq_tag_set *tagset;
357 	struct blk_mq_tag_set *admin_tagset;
358 	struct list_head namespaces;
359 	struct mutex namespaces_lock;
360 	struct srcu_struct srcu;
361 	struct device ctrl_device;
362 	struct device *device;	/* char device */
363 #ifdef CONFIG_NVME_HWMON
364 	struct device *hwmon_device;
365 #endif
366 	struct cdev cdev;
367 	struct work_struct reset_work;
368 	struct work_struct delete_work;
369 	wait_queue_head_t state_wq;
370 
371 	struct nvme_subsystem *subsys;
372 	struct list_head subsys_entry
373 		__guarded_by(&nvme_subsystems_lock);
374 
375 	struct opal_dev *opal_dev;
376 
377 	u16 cntlid;
378 
379 	u16 mtfa;
380 	u32 ctrl_config;
381 	u32 queue_count;
382 	u32 admin_timeout;
383 	u32 io_timeout;
384 
385 	u64 cap;
386 	u32 max_hw_sectors;
387 	u32 max_segments;
388 	u32 max_integrity_segments;
389 	u32 max_zeroes_sectors;
390 #ifdef CONFIG_BLK_DEV_ZONED
391 	u32 max_zone_append;
392 #endif
393 	u16 crdt[3];
394 	u16 oncs;
395 	u8 dmrl;
396 	u32 dmrsl;
397 	u16 oacs;
398 	u16 sqsize;
399 	u32 max_namespaces;
400 	atomic_t abort_limit;
401 	u8 vwc;
402 	u32 vs;
403 	u32 sgls;
404 	u16 kas;
405 	u8 npss;
406 	u8 apsta;
407 	u16 wctemp;
408 	u16 cctemp;
409 	u32 oaes;
410 	u32 aen_result;
411 	u32 ctratt;
412 	unsigned int shutdown_timeout;
413 	unsigned int kato;
414 	bool subsystem;
415 	unsigned long quirks;
416 	struct nvme_id_power_state psd[32];
417 	struct nvme_effects_log *effects;
418 	struct work_struct scan_work;
419 	struct work_struct async_event_work;
420 	struct delayed_work ka_work;
421 	struct delayed_work failfast_work;
422 	struct nvme_command ka_cmd;
423 	unsigned long ka_last_check_time;
424 	struct work_struct fw_act_work;
425 	unsigned long events;
426 	atomic_long_t errors;
427 	atomic_long_t nr_reset;
428 
429 #ifdef CONFIG_NVME_MULTIPATH
430 	/* asymmetric namespace access: */
431 	u8 anacap;
432 	u8 anatt;
433 	u32 anagrpmax;
434 	u32 nanagrpid;
435 	struct mutex ana_lock;
436 	struct nvme_ana_rsp_hdr *ana_log_buf;
437 	size_t ana_log_size;
438 	struct timer_list anatt_timer;
439 	struct work_struct ana_work;
440 	atomic_t nr_active;
441 #endif
442 
443 #ifdef CONFIG_NVME_HOST_AUTH
444 	struct work_struct dhchap_auth_work;
445 	struct mutex dhchap_auth_mutex;
446 	struct nvme_dhchap_queue_context *dhchap_ctxs;
447 	struct nvme_dhchap_key *host_key;
448 	struct nvme_dhchap_key *ctrl_key;
449 	u16 transaction;
450 #endif
451 	key_serial_t tls_pskid;
452 
453 	/* Power saving configuration */
454 	u64 ps_max_latency_us;
455 	bool apst_enabled;
456 
457 	/* PCIe only: */
458 	u16 hmmaxd;
459 	u32 hmpre;
460 	u32 hmmin;
461 	u32 hmminds;
462 
463 	/* Fabrics only */
464 	u32 ioccsz;
465 	u32 iorcsz;
466 	u16 icdoff;
467 	u16 maxcmd;
468 	int nr_reconnects;
469 	/* accumulate reconenct attempts, as nr_reconnects can reset to zero */
470 	atomic_long_t acc_reconnects;
471 	unsigned long flags;
472 	struct nvmf_ctrl_options *opts;
473 
474 	struct page *discard_page;
475 	unsigned long discard_page_busy;
476 
477 	struct nvme_fault_inject fault_inject;
478 
479 	enum nvme_ctrl_type cntrltype;
480 	enum nvme_dctype dctype;
481 
482 	u16			awupf; /* 0's based value. */
483 };
484 
485 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
486 {
487 	return READ_ONCE(ctrl->state);
488 }
489 
490 enum nvme_iopolicy {
491 	NVME_IOPOLICY_NUMA,
492 	NVME_IOPOLICY_RR,
493 	NVME_IOPOLICY_QD,
494 };
495 
496 struct nvme_subsystem {
497 	int			instance;
498 	struct device		dev;
499 	/*
500 	 * Because we unregister the device on the last put we need
501 	 * a separate refcount.
502 	 */
503 	struct kref		ref;
504 	struct list_head	entry
505 		__guarded_by(&nvme_subsystems_lock);
506 	struct mutex		lock;
507 	struct list_head	ctrls
508 		__guarded_by(&nvme_subsystems_lock);
509 	struct list_head	nsheads
510 		__guarded_by(&lock);
511 	/* Command effects logs, indexed by CSI and protected by lock. */
512 	struct xarray		cels;
513 	char			subnqn[NVMF_NQN_SIZE];
514 	char			serial[20];
515 	char			model[40];
516 	char			firmware_rev[8];
517 	u8			cmic;
518 	enum nvme_subsys_type	subtype;
519 	u16			vendor_id;
520 	struct ida		ns_ida;
521 #ifdef CONFIG_NVME_MULTIPATH
522 	enum nvme_iopolicy	iopolicy;
523 #endif
524 };
525 
526 /*
527  * Container structure for uniqueue namespace identifiers.
528  */
529 struct nvme_ns_ids {
530 	u8	eui64[8];
531 	u8	nguid[16];
532 	uuid_t	uuid;
533 	u8	csi;
534 };
535 
536 /*
537  * Anchor structure for namespaces.  There is one for each namespace in a
538  * NVMe subsystem that any of our controllers can see, and the namespace
539  * structure for each controller is chained of it.  For private namespaces
540  * there is a 1:1 relation to our namespace structures, that is ->list
541  * only ever has a single entry for private namespaces.
542  */
543 struct nvme_ns_head {
544 	struct list_head	list;
545 	struct srcu_struct      srcu;
546 	struct nvme_subsystem	*subsys;
547 	struct nvme_ns_ids	ids;
548 	u8			lba_shift;
549 	u16			ms;
550 	u16			pi_size;
551 	u8			pi_type;
552 	u8			guard_type;
553 	struct list_head	entry;
554 	struct kref		ref;
555 	bool			shared;
556 	bool			rotational;
557 	bool			passthru_err_log_enabled;
558 	struct nvme_effects_log *effects;
559 	u64			nuse;
560 	unsigned		ns_id;
561 	int			instance;
562 #ifdef CONFIG_BLK_DEV_ZONED
563 	u64			zsze;
564 #endif
565 	unsigned long		features;
566 
567 	struct ratelimit_state	rs_nuse;
568 
569 	struct cdev		cdev;
570 	struct device		cdev_device;
571 
572 	struct gendisk		*disk;
573 
574 	u16			nr_plids;
575 	u16			*plids;
576 	u32			write_stream_granularity;
577 #ifdef CONFIG_NVME_MULTIPATH
578 	struct bio_list		requeue_list
579 		__guarded_by(&requeue_lock);
580 	spinlock_t		requeue_lock;
581 	struct work_struct	requeue_work;
582 	struct work_struct	partition_scan_work;
583 	struct mutex		lock;
584 	unsigned long		flags;
585 	struct delayed_work	remove_work;
586 	unsigned int		delayed_removal_secs
587 		__guarded_by(&subsys->lock);
588 	atomic_long_t		io_requeue_no_usable_path_count;
589 	atomic_long_t		io_fail_no_available_path_count;
590 #define NVME_NSHEAD_DISK_LIVE		0
591 #define NVME_NSHEAD_QUEUE_IF_NO_PATH	1
592 #define NVME_NSHEAD_CDEV_LIVE		2
593 	struct nvme_ns __rcu_guarded	*current_path[];
594 #endif
595 };
596 
597 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
598 {
599 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
600 }
601 
602 enum nvme_ns_features {
603 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
604 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
605 	NVME_NS_DEAC = 1 << 2,		/* DEAC bit in Write Zeroes supported */
606 };
607 
608 struct nvme_ns {
609 	struct list_head list;
610 
611 	struct nvme_ctrl *ctrl;
612 	struct request_queue *queue;
613 	struct gendisk *disk;
614 #ifdef CONFIG_NVME_MULTIPATH
615 	enum nvme_ana_state ana_state;
616 	u32 ana_grpid;
617 	atomic_long_t failover;
618 #endif
619 	atomic_long_t retries;
620 	atomic_long_t errors;
621 	struct list_head siblings;
622 	struct kref kref;
623 	struct nvme_ns_head *head;
624 
625 	unsigned long flags;
626 #define NVME_NS_REMOVING		0
627 #define NVME_NS_ANA_PENDING		2
628 #define NVME_NS_FORCE_RO		3
629 #define NVME_NS_READY			4
630 #define NVME_NS_SYSFS_ATTR_LINK	5
631 #define NVME_NS_CDEV_LIVE		6
632 
633 	struct cdev		cdev;
634 	struct device		cdev_device;
635 
636 	struct nvme_fault_inject fault_inject;
637 };
638 
639 /* NVMe ns supports metadata actions by the controller (generate/strip) */
640 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head)
641 {
642 	return head->pi_type && head->ms == head->pi_size;
643 }
644 
645 static inline unsigned long nvme_get_virt_boundary(struct nvme_ctrl *ctrl,
646 						   bool is_admin)
647 {
648 	return NVME_CTRL_PAGE_SIZE - 1;
649 }
650 
651 struct nvme_ctrl_ops {
652 	const char *name;
653 	struct module *module;
654 	unsigned int flags;
655 #define NVME_F_FABRICS			(1 << 0)
656 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
657 #define NVME_F_BLOCKING			(1 << 2)
658 
659 	const struct attribute_group **dev_attr_groups;
660 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
661 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
662 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
663 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
664 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
665 	int (*subsystem_reset)(struct nvme_ctrl *ctrl);
666 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
667 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
668 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
669 	void (*print_device_info)(struct nvme_ctrl *ctrl);
670 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
671 	unsigned long (*get_virt_boundary)(struct nvme_ctrl *ctrl, bool is_admin);
672 };
673 
674 /*
675  * nvme command_id is constructed as such:
676  * | xxxx | xxxxxxxxxxxx |
677  *   gen    request tag
678  */
679 #define nvme_genctr_mask(gen)			(gen & 0xf)
680 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
681 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
682 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
683 
684 static inline u16 nvme_cid(struct request *rq)
685 {
686 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
687 }
688 
689 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
690 		u16 command_id)
691 {
692 	u8 genctr = nvme_genctr_from_cid(command_id);
693 	u16 tag = nvme_tag_from_cid(command_id);
694 	struct request *rq;
695 
696 	rq = blk_mq_tag_to_rq(tags, tag);
697 	if (unlikely(!rq)) {
698 		pr_err_ratelimited("could not locate request for tag %#x\n",
699 				   tag);
700 		return NULL;
701 	}
702 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
703 		dev_err_ratelimited(nvme_req(rq)->ctrl->device,
704 			"request %#x genctr mismatch (got %#x expected %#x)\n",
705 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
706 		return NULL;
707 	}
708 	return rq;
709 }
710 
711 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
712                 u16 command_id)
713 {
714 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
715 }
716 
717 /*
718  * Return the length of the string without the space padding
719  */
720 static inline int nvme_strlen(char *s, int len)
721 {
722 	while (s[len - 1] == ' ')
723 		len--;
724 	return len;
725 }
726 
727 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
728 {
729 	struct nvme_subsystem *subsys = ctrl->subsys;
730 
731 	if (ctrl->ops->print_device_info) {
732 		ctrl->ops->print_device_info(ctrl);
733 		return;
734 	}
735 
736 	dev_err(ctrl->device,
737 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
738 		nvme_strlen(subsys->model, sizeof(subsys->model)),
739 		subsys->model, nvme_strlen(subsys->firmware_rev,
740 					   sizeof(subsys->firmware_rev)),
741 		subsys->firmware_rev);
742 }
743 
744 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
745 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
746 			    const char *dev_name);
747 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
748 void nvme_should_fail(struct request *req);
749 #else
750 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
751 					  const char *dev_name)
752 {
753 }
754 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
755 {
756 }
757 static inline void nvme_should_fail(struct request *req) {}
758 #endif
759 
760 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
761 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
762 
763 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
764 {
765 	if (!ctrl->subsystem || !ctrl->ops->subsystem_reset)
766 		return -ENOTTY;
767 	return ctrl->ops->subsystem_reset(ctrl);
768 }
769 
770 /*
771  * Convert a 512B sector number to a device logical block number.
772  */
773 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector)
774 {
775 	return sector >> (head->lba_shift - SECTOR_SHIFT);
776 }
777 
778 /*
779  * Convert a device logical block number to a 512B sector number.
780  */
781 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba)
782 {
783 	return lba << (head->lba_shift - SECTOR_SHIFT);
784 }
785 
786 /*
787  * Convert byte length to nvme's 0-based num dwords
788  */
789 static inline u32 nvme_bytes_to_numd(size_t len)
790 {
791 	return (len >> 2) - 1;
792 }
793 
794 /* Decode a 2-byte "0's based"/"0-based" field */
795 static inline u32 from0based(__le16 value)
796 {
797 	return (u32)le16_to_cpu(value) + 1;
798 }
799 
800 static inline bool nvme_is_ana_error(u16 status)
801 {
802 	switch (status & NVME_SCT_SC_MASK) {
803 	case NVME_SC_ANA_TRANSITION:
804 	case NVME_SC_ANA_INACCESSIBLE:
805 	case NVME_SC_ANA_PERSISTENT_LOSS:
806 		return true;
807 	default:
808 		return false;
809 	}
810 }
811 
812 static inline bool nvme_is_path_error(u16 status)
813 {
814 	/* check for a status code type of 'path related status' */
815 	return (status & NVME_SCT_MASK) == NVME_SCT_PATH;
816 }
817 
818 /*
819  * Fill in the status and result information from the CQE, and then figure out
820  * if blk-mq will need to use IPI magic to complete the request, and if yes do
821  * so.  If not let the caller complete the request without an indirect function
822  * call.
823  */
824 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
825 		union nvme_result result)
826 {
827 	struct nvme_request *rq = nvme_req(req);
828 	struct nvme_ctrl *ctrl = rq->ctrl;
829 
830 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
831 		rq->genctr++;
832 
833 	rq->status = le16_to_cpu(status) >> 1;
834 	rq->result = result;
835 	/* inject error when permitted by fault injection framework */
836 	nvme_should_fail(req);
837 	if (unlikely(blk_should_fake_timeout(req->q)))
838 		return true;
839 	return blk_mq_complete_request_remote(req);
840 }
841 
842 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
843 {
844 	get_device(ctrl->device);
845 }
846 
847 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
848 {
849 	put_device(ctrl->device);
850 }
851 
852 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
853 {
854 	return !qid &&
855 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
856 }
857 
858 /*
859  * Returns true for sink states that can't ever transition back to live.
860  */
861 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
862 {
863 	switch (nvme_ctrl_state(ctrl)) {
864 	case NVME_CTRL_NEW:
865 	case NVME_CTRL_LIVE:
866 	case NVME_CTRL_RESETTING:
867 	case NVME_CTRL_CONNECTING:
868 		return false;
869 	case NVME_CTRL_DELETING:
870 	case NVME_CTRL_DELETING_NOIO:
871 	case NVME_CTRL_DEAD:
872 		return true;
873 	default:
874 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
875 		return true;
876 	}
877 }
878 
879 void nvme_end_req(struct request *req);
880 void nvme_complete_rq(struct request *req);
881 void nvme_complete_batch_req(struct request *req);
882 
883 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
884 						void (*fn)(struct request *rq))
885 {
886 	struct request *req;
887 
888 	rq_list_for_each(&iob->req_list, req) {
889 		fn(req);
890 		nvme_complete_batch_req(req);
891 	}
892 	blk_mq_end_request_batch(iob);
893 }
894 
895 blk_status_t nvme_host_path_error(struct request *req);
896 bool nvme_cancel_request(struct request *req, void *data);
897 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
898 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
899 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
900 		enum nvme_ctrl_state new_state);
901 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
902 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
903 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
904 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
905 int nvme_add_ctrl(struct nvme_ctrl *ctrl);
906 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
907 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
908 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
909 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
910 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
911 		const struct blk_mq_ops *ops, unsigned int cmd_size);
912 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
913 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
914 		const struct blk_mq_ops *ops, unsigned int nr_maps,
915 		unsigned int cmd_size);
916 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
917 
918 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
919 
920 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
921 		volatile union nvme_result *res);
922 
923 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
924 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
925 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
926 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
927 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
928 void nvme_sync_queues(struct nvme_ctrl *ctrl);
929 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
930 void nvme_unfreeze(struct nvme_ctrl *ctrl);
931 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
932 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl);
933 void nvme_start_freeze(struct nvme_ctrl *ctrl);
934 
935 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
936 {
937 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
938 }
939 
940 #define NVME_QID_ANY -1
941 void nvme_init_request(struct request *req, struct nvme_command *cmd);
942 void nvme_cleanup_cmd(struct request *req);
943 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
944 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
945 		struct request *req);
946 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
947 		bool queue_live, enum nvme_ctrl_state state);
948 
949 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
950 		bool queue_live)
951 {
952 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
953 
954 	if (likely(state == NVME_CTRL_LIVE))
955 		return true;
956 	if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING)
957 		return queue_live;
958 	return __nvme_check_ready(ctrl, rq, queue_live, state);
959 }
960 
961 /*
962  * NSID shall be unique for all shared namespaces, or if at least one of the
963  * following conditions is met:
964  *   1. Namespace Management is supported by the controller
965  *   2. ANA is supported by the controller
966  *   3. NVM Set are supported by the controller
967  *
968  * In other case, private namespace are not required to report a unique NSID.
969  */
970 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
971 		struct nvme_ns_head *head)
972 {
973 	return head->shared ||
974 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
975 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
976 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
977 }
978 
979 /*
980  * Flags for __nvme_submit_sync_cmd()
981  */
982 typedef __u32 __bitwise nvme_submit_flags_t;
983 
984 enum {
985 	/* Insert request at the head of the queue */
986 	NVME_SUBMIT_AT_HEAD  = (__force nvme_submit_flags_t)(1 << 0),
987 	/* Set BLK_MQ_REQ_NOWAIT when allocating request */
988 	NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1),
989 	/* Set BLK_MQ_REQ_RESERVED when allocating request */
990 	NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2),
991 	/* Retry command when NVME_STATUS_DNR is not set in the result */
992 	NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3),
993 };
994 
995 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
996 		void *buf, unsigned bufflen);
997 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
998 		union nvme_result *result, void *buffer, unsigned bufflen,
999 		int qid, nvme_submit_flags_t flags);
1000 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1001 		      unsigned int dword11, void *buffer, size_t buflen,
1002 		      void *result);
1003 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1004 		      unsigned int dword11, void *buffer, size_t buflen,
1005 		      void *result);
1006 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
1007 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
1008 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
1009 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
1010 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
1011 void nvme_queue_scan(struct nvme_ctrl *ctrl);
1012 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
1013 		void *log, size_t size, u64 offset);
1014 void nvme_get_ns_head(struct nvme_ns_head *head);
1015 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
1016 void nvme_put_ns_head(struct nvme_ns_head *head);
1017 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
1018 		const struct file_operations *fops, struct module *owner,
1019 		int ctrl, int head);
1020 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
1021 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
1022 		unsigned int cmd, unsigned long arg);
1023 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1024 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
1025 		unsigned int cmd, unsigned long arg);
1026 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
1027 		unsigned long arg);
1028 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1029 		unsigned long arg);
1030 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
1031 		struct io_comp_batch *iob, unsigned int poll_flags);
1032 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
1033 		unsigned int issue_flags);
1034 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
1035 		unsigned int issue_flags);
1036 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1037 		struct nvme_id_ns **id);
1038 int nvme_getgeo(struct gendisk *disk, struct hd_geometry *geo);
1039 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
1040 
1041 extern const struct attribute_group *nvme_ns_attr_groups[];
1042 extern const struct attribute_group nvme_ns_mpath_attr_group;
1043 extern const struct pr_ops nvme_pr_ops;
1044 extern const struct block_device_operations nvme_ns_head_ops;
1045 extern const struct attribute_group nvme_dev_attrs_group;
1046 extern const struct attribute_group nvme_dev_diag_attrs_group;
1047 extern const struct attribute_group *nvme_subsys_attrs_groups[];
1048 extern const struct attribute_group *nvme_dev_attr_groups[];
1049 extern const struct block_device_operations nvme_bdev_ops;
1050 
1051 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
1052 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head)
1053 	__must_hold_shared(&head->srcu);
1054 #ifdef CONFIG_NVME_MULTIPATH
1055 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1056 {
1057 	return ctrl->ana_log_buf != NULL;
1058 }
1059 
1060 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1061 	__must_hold(&subsys->lock);
1062 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1063 	__must_hold(&subsys->lock);
1064 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1065 	__must_hold(&subsys->lock);
1066 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
1067 void nvme_failover_req(struct request *req);
1068 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
1069 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
1070 void nvme_mpath_add_sysfs_link(struct nvme_ns_head *ns);
1071 void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns);
1072 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
1073 void nvme_mpath_put_disk(struct nvme_ns_head *head);
1074 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
1075 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
1076 void nvme_mpath_update(struct nvme_ctrl *ctrl);
1077 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
1078 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
1079 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
1080 void nvme_mpath_revalidate_paths(struct nvme_ns_head *head);
1081 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
1082 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
1083 void nvme_mpath_start_request(struct request *rq);
1084 void nvme_mpath_end_request(struct request *rq);
1085 
1086 static inline void nvme_trace_bio_complete(struct request *req)
1087 {
1088 	struct nvme_ns *ns = req->q->queuedata;
1089 
1090 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
1091 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
1092 }
1093 
1094 extern bool multipath;
1095 extern struct device_attribute dev_attr_ana_grpid;
1096 extern struct device_attribute dev_attr_ana_state;
1097 extern struct device_attribute dev_attr_queue_depth;
1098 extern struct device_attribute dev_attr_numa_nodes;
1099 extern struct device_attribute dev_attr_delayed_removal_secs;
1100 extern struct device_attribute dev_attr_multipath_failover_count;
1101 extern struct device_attribute dev_attr_io_requeue_no_usable_path_count;
1102 extern struct device_attribute dev_attr_io_fail_no_available_path_count;
1103 extern struct device_attribute subsys_attr_iopolicy;
1104 
1105 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1106 {
1107 	return disk->fops == &nvme_ns_head_ops;
1108 }
1109 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1110 {
1111 	if (test_bit(NVME_NSHEAD_QUEUE_IF_NO_PATH, &head->flags))
1112 		return true;
1113 	return false;
1114 }
1115 #else
1116 #define multipath false
1117 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1118 {
1119 	return false;
1120 }
1121 static inline void nvme_failover_req(struct request *req)
1122 {
1123 }
1124 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
1125 {
1126 }
1127 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
1128 		struct nvme_ns_head *head)
1129 {
1130 	return 0;
1131 }
1132 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
1133 {
1134 }
1135 static inline void nvme_mpath_put_disk(struct nvme_ns_head *head)
1136 {
1137 }
1138 static inline void nvme_mpath_add_sysfs_link(struct nvme_ns *ns)
1139 {
1140 }
1141 static inline void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns)
1142 {
1143 }
1144 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
1145 {
1146 	return false;
1147 }
1148 static inline void nvme_mpath_revalidate_paths(struct nvme_ns_head *head)
1149 {
1150 }
1151 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
1152 {
1153 }
1154 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
1155 {
1156 }
1157 static inline void nvme_trace_bio_complete(struct request *req)
1158 {
1159 }
1160 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1161 {
1162 }
1163 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1164 		struct nvme_id_ctrl *id)
1165 {
1166 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1167 		dev_warn(ctrl->device,
1168 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1169 	return 0;
1170 }
1171 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1172 {
1173 }
1174 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1175 {
1176 }
1177 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1178 {
1179 }
1180 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1181 {
1182 }
1183 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1184 {
1185 }
1186 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1187 {
1188 }
1189 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1190 {
1191 }
1192 static inline void nvme_mpath_start_request(struct request *rq)
1193 {
1194 }
1195 static inline void nvme_mpath_end_request(struct request *rq)
1196 {
1197 }
1198 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1199 {
1200 	return false;
1201 }
1202 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1203 {
1204 	return false;
1205 }
1206 #endif /* CONFIG_NVME_MULTIPATH */
1207 
1208 #if defined(CONFIG_NVME_MULTIPATH) && defined(CONFIG_BLK_DEV_ZONED)
1209 int nvme_mpath_revalidate_zones(struct nvme_ns_head *head);
1210 #else
1211 static inline int nvme_mpath_revalidate_zones(struct nvme_ns_head *head)
1212 {
1213 	return 0;
1214 }
1215 #endif
1216 
1217 int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16],
1218 		enum blk_unique_id type);
1219 
1220 struct nvme_zone_info {
1221 	u64 zone_size;
1222 	unsigned int max_open_zones;
1223 	unsigned int max_active_zones;
1224 };
1225 
1226 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1227 		unsigned int nr_zones, struct blk_report_zones_args *args);
1228 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf,
1229 		struct nvme_zone_info *zi);
1230 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim,
1231 		struct nvme_zone_info *zi);
1232 #ifdef CONFIG_BLK_DEV_ZONED
1233 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1234 				       struct nvme_command *cmnd,
1235 				       enum nvme_zone_mgmt_action action);
1236 #else
1237 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1238 		struct request *req, struct nvme_command *cmnd,
1239 		enum nvme_zone_mgmt_action action)
1240 {
1241 	return BLK_STS_NOTSUPP;
1242 }
1243 #endif
1244 
1245 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1246 {
1247 	struct gendisk *disk = dev_to_disk(dev);
1248 
1249 	WARN_ON(nvme_disk_is_ns_head(disk));
1250 	return disk->private_data;
1251 }
1252 
1253 #ifdef CONFIG_NVME_HWMON
1254 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1255 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1256 #else
1257 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1258 {
1259 	return 0;
1260 }
1261 
1262 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1263 {
1264 }
1265 #endif
1266 
1267 static inline void nvme_start_request(struct request *rq)
1268 {
1269 	if (rq->cmd_flags & REQ_NVME_MPATH)
1270 		nvme_mpath_start_request(rq);
1271 	blk_mq_start_request(rq);
1272 }
1273 
1274 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1275 {
1276 	return ctrl->sgls & (NVME_CTRL_SGLS_BYTE_ALIGNED |
1277 			     NVME_CTRL_SGLS_DWORD_ALIGNED);
1278 }
1279 
1280 static inline bool nvme_ctrl_meta_sgl_supported(struct nvme_ctrl *ctrl)
1281 {
1282 	if (ctrl->ops->flags & NVME_F_FABRICS)
1283 		return true;
1284 	return ctrl->sgls & NVME_CTRL_SGLS_MSDS;
1285 }
1286 
1287 #ifdef CONFIG_NVME_HOST_AUTH
1288 int __init nvme_init_auth(void);
1289 void __exit nvme_exit_auth(void);
1290 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1291 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1292 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1293 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1294 void nvme_auth_free(struct nvme_ctrl *ctrl);
1295 void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl);
1296 #else
1297 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1298 {
1299 	return 0;
1300 }
1301 static inline int __init nvme_init_auth(void)
1302 {
1303 	return 0;
1304 }
1305 static inline void __exit nvme_exit_auth(void)
1306 {
1307 }
1308 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1309 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1310 {
1311 	return -EPROTONOSUPPORT;
1312 }
1313 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1314 {
1315 	return -EPROTONOSUPPORT;
1316 }
1317 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1318 static inline void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl) {};
1319 #endif
1320 
1321 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1322 			 u8 opcode);
1323 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1324 	__cond_acquires(nonzero, &ctrl->subsys->lock)
1325 	__cond_acquires(nonzero, &ctrl->scan_lock);
1326 
1327 int nvme_execute_rq(struct request *rq, bool at_head);
1328 u32 nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1329 		       struct nvme_command *cmd, int status)
1330 		       __cond_releases(nonzero, &ctrl->scan_lock)
1331 		       __cond_releases(nonzero, &ctrl->subsys->lock);
1332 
1333 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1334 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1335 bool nvme_get_ns(struct nvme_ns *ns);
1336 void nvme_put_ns(struct nvme_ns *ns);
1337 
1338 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1339 {
1340 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1341 }
1342 
1343 #endif /* _NVME_H */
1344