xref: /linux/drivers/nvme/host/nvme.h (revision 28f7d31c2aeb350df9a0116093fade0a840fb4ce)
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 	 * The device is extremely slow on FUA writes but fast on Flush;
189 	 * do not advertise FUA so the block layer emulates it.
190 	 */
191 	NVME_QUIRK_NO_FUA			= (1 << 24),
192 };
193 
194 static inline char *nvme_quirk_name(enum nvme_quirks q)
195 {
196 	switch (q) {
197 	case NVME_QUIRK_STRIPE_SIZE:
198 		return "stripe_size";
199 	case NVME_QUIRK_IDENTIFY_CNS:
200 		return "identify_cns";
201 	case NVME_QUIRK_DEALLOCATE_ZEROES:
202 		return "deallocate_zeroes";
203 	case NVME_QUIRK_DELAY_BEFORE_CHK_RDY:
204 		return "delay_before_chk_rdy";
205 	case NVME_QUIRK_NO_APST:
206 		return "no_apst";
207 	case NVME_QUIRK_NO_DEEPEST_PS:
208 		return "no_deepest_ps";
209 	case NVME_QUIRK_QDEPTH_ONE:
210 		return "qdepth_one";
211 	case NVME_QUIRK_MEDIUM_PRIO_SQ:
212 		return "medium_prio_sq";
213 	case NVME_QUIRK_IGNORE_DEV_SUBNQN:
214 		return "ignore_dev_subnqn";
215 	case NVME_QUIRK_DISABLE_WRITE_ZEROES:
216 		return "disable_write_zeroes";
217 	case NVME_QUIRK_SIMPLE_SUSPEND:
218 		return "simple_suspend";
219 	case NVME_QUIRK_SINGLE_VECTOR:
220 		return "single_vector";
221 	case NVME_QUIRK_128_BYTES_SQES:
222 		return "128_bytes_sqes";
223 	case NVME_QUIRK_SHARED_TAGS:
224 		return "shared_tags";
225 	case NVME_QUIRK_NO_TEMP_THRESH_CHANGE:
226 		return "no_temp_thresh_change";
227 	case NVME_QUIRK_NO_NS_DESC_LIST:
228 		return "no_ns_desc_list";
229 	case NVME_QUIRK_DMA_ADDRESS_BITS_48:
230 		return "dma_address_bits_48";
231 	case NVME_QUIRK_SKIP_CID_GEN:
232 		return "skip_cid_gen";
233 	case NVME_QUIRK_BOGUS_NID:
234 		return "bogus_nid";
235 	case NVME_QUIRK_NO_SECONDARY_TEMP_THRESH:
236 		return "no_secondary_temp_thresh";
237 	case NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND:
238 		return "force_no_simple_suspend";
239 	case NVME_QUIRK_BROKEN_MSI:
240 		return "broken_msi";
241 	case NVME_QUIRK_DMAPOOL_ALIGN_512:
242 		return "dmapool_align_512";
243 	case NVME_QUIRK_ADMIN_PAGE_ALIGN:
244 		return "admin_page_align";
245 	case NVME_QUIRK_NO_FUA:
246 		return "no_fua";
247 	}
248 
249 	return "unknown";
250 }
251 
252 /*
253  * Common request structure for NVMe passthrough.  All drivers must have
254  * this structure as the first member of their request-private data.
255  */
256 struct nvme_request {
257 	struct nvme_command	*cmd;
258 	union nvme_result	result;
259 	u8			genctr;
260 	u8			retries;
261 	u8			flags;
262 	u16			status;
263 #ifdef CONFIG_NVME_MULTIPATH
264 	unsigned long		start_time;
265 #endif
266 	struct nvme_ctrl	*ctrl;
267 };
268 
269 /*
270  * Mark a bio as coming in through the mpath node.
271  */
272 #define REQ_NVME_MPATH		REQ_DRV
273 
274 enum {
275 	NVME_REQ_CANCELLED		= (1 << 0),
276 	NVME_REQ_USERCMD		= (1 << 1),
277 	NVME_MPATH_IO_STATS		= (1 << 2),
278 	NVME_MPATH_CNT_ACTIVE		= (1 << 3),
279 };
280 
281 static inline struct nvme_request *nvme_req(struct request *req)
282 {
283 	return blk_mq_rq_to_pdu(req);
284 }
285 
286 static inline u16 nvme_req_qid(struct request *req)
287 {
288 	if (!req->q->queuedata)
289 		return 0;
290 
291 	return req->mq_hctx->queue_num + 1;
292 }
293 
294 /* The below value is the specific amount of delay needed before checking
295  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
296  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
297  * found empirically.
298  */
299 #define NVME_QUIRK_DELAY_AMOUNT		2300
300 
301 /*
302  * enum nvme_ctrl_state: Controller state
303  *
304  * @NVME_CTRL_NEW:		New controller just allocated, initial state
305  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
306  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
307  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
308  *				transport
309  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
310  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
311  *				disabled/failed immediately. This state comes
312  * 				after all async event processing took place and
313  * 				before ns removal and the controller deletion
314  * 				progress
315  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
316  *				shutdown or removal. In this case we forcibly
317  *				kill all inflight I/O as they have no chance to
318  *				complete
319  */
320 enum nvme_ctrl_state {
321 	NVME_CTRL_NEW,
322 	NVME_CTRL_LIVE,
323 	NVME_CTRL_RESETTING,
324 	NVME_CTRL_CONNECTING,
325 	NVME_CTRL_DELETING,
326 	NVME_CTRL_DELETING_NOIO,
327 	NVME_CTRL_DEAD,
328 };
329 
330 struct nvme_fault_inject {
331 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
332 	struct fault_attr attr;
333 	struct dentry *parent;
334 	u16 opcode;
335 	bool dont_retry;	/* DNR, do not retry */
336 	u16 status;		/* status code */
337 #endif
338 };
339 
340 enum nvme_ctrl_flags {
341 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
342 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
343 	NVME_CTRL_STARTED_ONCE		= 2,
344 	NVME_CTRL_STOPPED		= 3,
345 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
346 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
347 	NVME_CTRL_FROZEN		= 6,
348 };
349 
350 struct nvme_ctrl {
351 	bool comp_seen;
352 	bool identified;
353 	bool passthru_err_log_enabled;
354 	enum nvme_ctrl_state state;
355 	spinlock_t lock;
356 	struct mutex scan_lock;
357 	const struct nvme_ctrl_ops *ops;
358 	struct request_queue *admin_q;
359 	struct request_queue *connect_q;
360 	struct request_queue *fabrics_q;
361 	struct device *dev;
362 	int instance;
363 	int numa_node;
364 	struct blk_mq_tag_set *tagset;
365 	struct blk_mq_tag_set *admin_tagset;
366 	struct list_head namespaces;
367 	struct mutex namespaces_lock;
368 	struct srcu_struct srcu;
369 	struct device ctrl_device;
370 	struct device *device;	/* char device */
371 #ifdef CONFIG_NVME_HWMON
372 	struct device *hwmon_device;
373 #endif
374 	struct cdev cdev;
375 	struct work_struct reset_work;
376 	struct work_struct delete_work;
377 	wait_queue_head_t state_wq;
378 
379 	struct nvme_subsystem *subsys;
380 	struct list_head subsys_entry
381 		__guarded_by(&nvme_subsystems_lock);
382 
383 	struct opal_dev *opal_dev;
384 
385 	u16 cntlid;
386 
387 	u16 mtfa;
388 	u32 ctrl_config;
389 	u32 queue_count;
390 	u32 admin_timeout;
391 	u32 io_timeout;
392 
393 	u64 cap;
394 	u32 max_hw_sectors;
395 	u32 max_segments;
396 	u32 max_integrity_segments;
397 	u32 max_zeroes_sectors;
398 #ifdef CONFIG_BLK_DEV_ZONED
399 	u32 max_zone_append;
400 #endif
401 	u16 crdt[3];
402 	u16 oncs;
403 	u8 dmrl;
404 	u32 dmrsl;
405 	u16 oacs;
406 	u16 sqsize;
407 	u32 max_namespaces;
408 	atomic_t abort_limit;
409 	u8 vwc;
410 	u32 vs;
411 	u32 sgls;
412 	u16 kas;
413 	u8 npss;
414 	u8 apsta;
415 	u16 wctemp;
416 	u16 cctemp;
417 	u32 oaes;
418 	u32 aen_result;
419 	u32 ctratt;
420 	unsigned int shutdown_timeout;
421 	unsigned int kato;
422 	bool subsystem;
423 	unsigned long quirks;
424 	struct nvme_id_power_state psd[32];
425 	struct nvme_effects_log *effects;
426 	struct work_struct scan_work;
427 	struct work_struct async_event_work;
428 	struct delayed_work ka_work;
429 	struct delayed_work failfast_work;
430 	struct nvme_command ka_cmd;
431 	unsigned long ka_last_check_time;
432 	struct work_struct fw_act_work;
433 	unsigned long events;
434 	atomic_long_t errors;
435 	atomic_long_t nr_reset;
436 
437 #ifdef CONFIG_NVME_MULTIPATH
438 	/* asymmetric namespace access: */
439 	u8 anacap;
440 	u8 anatt;
441 	u32 anagrpmax;
442 	u32 nanagrpid;
443 	struct mutex ana_lock;
444 	struct nvme_ana_rsp_hdr *ana_log_buf;
445 	size_t ana_log_size;
446 	struct timer_list anatt_timer;
447 	struct work_struct ana_work;
448 	atomic_t nr_active;
449 #endif
450 
451 #ifdef CONFIG_NVME_HOST_AUTH
452 	struct work_struct dhchap_auth_work;
453 	struct mutex dhchap_auth_mutex;
454 	struct nvme_dhchap_queue_context *dhchap_ctxs;
455 	struct nvme_dhchap_key *host_key;
456 	struct nvme_dhchap_key *ctrl_key;
457 	u16 transaction;
458 #endif
459 	key_serial_t tls_pskid;
460 
461 	/* Power saving configuration */
462 	u64 ps_max_latency_us;
463 	bool apst_enabled;
464 
465 	/* PCIe only: */
466 	u16 hmmaxd;
467 	u32 hmpre;
468 	u32 hmmin;
469 	u32 hmminds;
470 
471 	/* Fabrics only */
472 	u32 ioccsz;
473 	u32 iorcsz;
474 	u16 icdoff;
475 	u16 maxcmd;
476 	int nr_reconnects;
477 	/* accumulate reconenct attempts, as nr_reconnects can reset to zero */
478 	atomic_long_t acc_reconnects;
479 	unsigned long flags;
480 	struct nvmf_ctrl_options *opts;
481 
482 	struct page *discard_page;
483 	unsigned long discard_page_busy;
484 
485 	struct nvme_fault_inject fault_inject;
486 
487 	enum nvme_ctrl_type cntrltype;
488 	enum nvme_dctype dctype;
489 
490 	u16			awupf; /* 0's based value. */
491 };
492 
493 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
494 {
495 	return READ_ONCE(ctrl->state);
496 }
497 
498 enum nvme_iopolicy {
499 	NVME_IOPOLICY_NUMA,
500 	NVME_IOPOLICY_RR,
501 	NVME_IOPOLICY_QD,
502 };
503 
504 struct nvme_subsystem {
505 	int			instance;
506 	struct device		dev;
507 	/*
508 	 * Because we unregister the device on the last put we need
509 	 * a separate refcount.
510 	 */
511 	struct kref		ref;
512 	struct list_head	entry
513 		__guarded_by(&nvme_subsystems_lock);
514 	struct mutex		lock;
515 	struct list_head	ctrls
516 		__guarded_by(&nvme_subsystems_lock);
517 	struct list_head	nsheads
518 		__guarded_by(&lock);
519 	/* Command effects logs, indexed by CSI and protected by lock. */
520 	struct xarray		cels;
521 	char			subnqn[NVMF_NQN_SIZE];
522 	char			serial[20];
523 	char			model[40];
524 	char			firmware_rev[8];
525 	u8			cmic;
526 	enum nvme_subsys_type	subtype;
527 	u16			vendor_id;
528 	struct ida		ns_ida;
529 #ifdef CONFIG_NVME_MULTIPATH
530 	enum nvme_iopolicy	iopolicy;
531 #endif
532 };
533 
534 /*
535  * Container structure for uniqueue namespace identifiers.
536  */
537 struct nvme_ns_ids {
538 	u8	eui64[8];
539 	u8	nguid[16];
540 	uuid_t	uuid;
541 	u8	csi;
542 };
543 
544 /*
545  * Anchor structure for namespaces.  There is one for each namespace in a
546  * NVMe subsystem that any of our controllers can see, and the namespace
547  * structure for each controller is chained of it.  For private namespaces
548  * there is a 1:1 relation to our namespace structures, that is ->list
549  * only ever has a single entry for private namespaces.
550  */
551 struct nvme_ns_head {
552 	struct list_head	list;
553 	struct srcu_struct      srcu;
554 	struct nvme_subsystem	*subsys;
555 	struct nvme_ns_ids	ids;
556 	u8			lba_shift;
557 	u16			ms;
558 	u16			pi_size;
559 	u8			pi_type;
560 	u8			guard_type;
561 	struct list_head	entry;
562 	struct kref		ref;
563 	bool			shared;
564 	bool			rotational;
565 	bool			passthru_err_log_enabled;
566 	struct nvme_effects_log *effects;
567 	u64			nuse;
568 	unsigned		ns_id;
569 	int			instance;
570 #ifdef CONFIG_BLK_DEV_ZONED
571 	u64			zsze;
572 #endif
573 	unsigned long		features;
574 
575 	struct ratelimit_state	rs_nuse;
576 
577 	struct cdev		cdev;
578 	struct device		cdev_device;
579 
580 	struct gendisk		*disk;
581 
582 	u16			nr_plids;
583 	u16			*plids;
584 	u32			write_stream_granularity;
585 #ifdef CONFIG_NVME_MULTIPATH
586 	struct bio_list		requeue_list
587 		__guarded_by(&requeue_lock);
588 	spinlock_t		requeue_lock;
589 	struct work_struct	requeue_work;
590 	struct work_struct	partition_scan_work;
591 	struct mutex		lock;
592 	unsigned long		flags;
593 	struct delayed_work	remove_work;
594 	unsigned int		delayed_removal_secs
595 		__guarded_by(&subsys->lock);
596 	atomic_long_t		io_requeue_no_usable_path_count;
597 	atomic_long_t		io_fail_no_available_path_count;
598 #define NVME_NSHEAD_DISK_LIVE		0
599 #define NVME_NSHEAD_QUEUE_IF_NO_PATH	1
600 #define NVME_NSHEAD_CDEV_LIVE		2
601 #define NVME_NSHEAD_FAIL_ON_CTRL_LOSS	3
602 	struct nvme_ns __rcu_guarded	*current_path[];
603 #endif
604 };
605 
606 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
607 {
608 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
609 }
610 
611 enum nvme_ns_features {
612 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
613 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
614 	NVME_NS_DEAC = 1 << 2,		/* DEAC bit in Write Zeroes supported */
615 };
616 
617 struct nvme_ns {
618 	struct list_head list;
619 
620 	struct nvme_ctrl *ctrl;
621 	struct request_queue *queue;
622 	struct gendisk *disk;
623 #ifdef CONFIG_NVME_MULTIPATH
624 	enum nvme_ana_state ana_state;
625 	u32 ana_grpid;
626 	atomic_long_t failover;
627 #endif
628 	atomic_long_t retries;
629 	atomic_long_t errors;
630 	struct list_head siblings;
631 	struct kref kref;
632 	struct nvme_ns_head *head;
633 
634 	unsigned long flags;
635 #define NVME_NS_REMOVING		0
636 #define NVME_NS_ANA_PENDING		2
637 #define NVME_NS_FORCE_RO		3
638 #define NVME_NS_READY			4
639 #define NVME_NS_SYSFS_ATTR_LINK	5
640 #define NVME_NS_CDEV_LIVE		6
641 
642 	struct cdev		cdev;
643 	struct device		cdev_device;
644 
645 	struct nvme_fault_inject fault_inject;
646 };
647 
648 /* NVMe ns supports metadata actions by the controller (generate/strip) */
649 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head)
650 {
651 	return head->pi_type && head->ms == head->pi_size;
652 }
653 
654 static inline unsigned long nvme_get_virt_boundary(struct nvme_ctrl *ctrl,
655 						   bool is_admin)
656 {
657 	return NVME_CTRL_PAGE_SIZE - 1;
658 }
659 
660 struct nvme_ctrl_ops {
661 	const char *name;
662 	struct module *module;
663 	unsigned int flags;
664 #define NVME_F_FABRICS			(1 << 0)
665 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
666 #define NVME_F_BLOCKING			(1 << 2)
667 
668 	const struct attribute_group **dev_attr_groups;
669 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
670 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
671 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
672 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
673 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
674 	int (*subsystem_reset)(struct nvme_ctrl *ctrl);
675 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
676 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
677 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
678 	void (*print_device_info)(struct nvme_ctrl *ctrl);
679 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
680 	int (*init_dma_buf_io_ctx)(struct nvme_ctrl *ctrl,
681 				   struct dma_buf_io_ctx *ctx);
682 	unsigned long (*get_virt_boundary)(struct nvme_ctrl *ctrl, bool is_admin);
683 };
684 
685 /*
686  * nvme command_id is constructed as such:
687  * | xxxx | xxxxxxxxxxxx |
688  *   gen    request tag
689  */
690 #define nvme_genctr_mask(gen)			(gen & 0xf)
691 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
692 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
693 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
694 
695 static inline u16 nvme_cid(struct request *rq)
696 {
697 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
698 }
699 
700 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
701 		u16 command_id)
702 {
703 	u8 genctr = nvme_genctr_from_cid(command_id);
704 	u16 tag = nvme_tag_from_cid(command_id);
705 	struct request *rq;
706 
707 	rq = blk_mq_tag_to_rq(tags, tag);
708 	if (unlikely(!rq)) {
709 		pr_err_ratelimited("could not locate request for tag %#x\n",
710 				   tag);
711 		return NULL;
712 	}
713 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
714 		dev_err_ratelimited(nvme_req(rq)->ctrl->device,
715 			"request %#x genctr mismatch (got %#x expected %#x)\n",
716 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
717 		return NULL;
718 	}
719 	return rq;
720 }
721 
722 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
723                 u16 command_id)
724 {
725 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
726 }
727 
728 /*
729  * Return the length of the string without the space padding
730  */
731 static inline int nvme_strlen(char *s, int len)
732 {
733 	while (s[len - 1] == ' ')
734 		len--;
735 	return len;
736 }
737 
738 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
739 {
740 	struct nvme_subsystem *subsys = ctrl->subsys;
741 
742 	if (ctrl->ops->print_device_info) {
743 		ctrl->ops->print_device_info(ctrl);
744 		return;
745 	}
746 
747 	dev_err(ctrl->device,
748 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
749 		nvme_strlen(subsys->model, sizeof(subsys->model)),
750 		subsys->model, nvme_strlen(subsys->firmware_rev,
751 					   sizeof(subsys->firmware_rev)),
752 		subsys->firmware_rev);
753 }
754 
755 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
756 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
757 			    const char *dev_name);
758 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
759 void nvme_should_fail(struct request *req);
760 #else
761 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
762 					  const char *dev_name)
763 {
764 }
765 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
766 {
767 }
768 static inline void nvme_should_fail(struct request *req) {}
769 #endif
770 
771 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
772 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
773 
774 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
775 {
776 	if (!ctrl->subsystem || !ctrl->ops->subsystem_reset)
777 		return -ENOTTY;
778 	return ctrl->ops->subsystem_reset(ctrl);
779 }
780 
781 /*
782  * Convert a 512B sector number to a device logical block number.
783  */
784 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector)
785 {
786 	return sector >> (head->lba_shift - SECTOR_SHIFT);
787 }
788 
789 /*
790  * Convert a device logical block number to a 512B sector number.
791  */
792 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba)
793 {
794 	return lba << (head->lba_shift - SECTOR_SHIFT);
795 }
796 
797 /*
798  * Convert byte length to nvme's 0-based num dwords
799  */
800 static inline u32 nvme_bytes_to_numd(size_t len)
801 {
802 	return (len >> 2) - 1;
803 }
804 
805 /* Decode a 2-byte "0's based"/"0-based" field */
806 static inline u32 from0based(__le16 value)
807 {
808 	return (u32)le16_to_cpu(value) + 1;
809 }
810 
811 static inline bool nvme_is_ana_error(u16 status)
812 {
813 	switch (status & NVME_SCT_SC_MASK) {
814 	case NVME_SC_ANA_TRANSITION:
815 	case NVME_SC_ANA_INACCESSIBLE:
816 	case NVME_SC_ANA_PERSISTENT_LOSS:
817 		return true;
818 	default:
819 		return false;
820 	}
821 }
822 
823 static inline bool nvme_is_path_error(u16 status)
824 {
825 	/* check for a status code type of 'path related status' */
826 	return (status & NVME_SCT_MASK) == NVME_SCT_PATH;
827 }
828 
829 /*
830  * Fill in the status and result information from the CQE, and then figure out
831  * if blk-mq will need to use IPI magic to complete the request, and if yes do
832  * so.  If not let the caller complete the request without an indirect function
833  * call.
834  */
835 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
836 		union nvme_result result)
837 {
838 	struct nvme_request *rq = nvme_req(req);
839 	struct nvme_ctrl *ctrl = rq->ctrl;
840 
841 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
842 		rq->genctr++;
843 
844 	rq->status = le16_to_cpu(status) >> 1;
845 	rq->result = result;
846 	/* inject error when permitted by fault injection framework */
847 	nvme_should_fail(req);
848 	if (unlikely(blk_should_fake_timeout(req->q)))
849 		return true;
850 	return blk_mq_complete_request_remote(req);
851 }
852 
853 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
854 {
855 	get_device(ctrl->device);
856 }
857 
858 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
859 {
860 	put_device(ctrl->device);
861 }
862 
863 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
864 {
865 	return !qid &&
866 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
867 }
868 
869 /*
870  * Returns true for sink states that can't ever transition back to live.
871  */
872 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
873 {
874 	switch (nvme_ctrl_state(ctrl)) {
875 	case NVME_CTRL_NEW:
876 	case NVME_CTRL_LIVE:
877 	case NVME_CTRL_RESETTING:
878 	case NVME_CTRL_CONNECTING:
879 		return false;
880 	case NVME_CTRL_DELETING:
881 	case NVME_CTRL_DELETING_NOIO:
882 	case NVME_CTRL_DEAD:
883 		return true;
884 	default:
885 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
886 		return true;
887 	}
888 }
889 
890 void nvme_end_req(struct request *req);
891 void nvme_complete_rq(struct request *req);
892 void nvme_complete_batch_req(struct request *req);
893 
894 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
895 						void (*fn)(struct request *rq))
896 {
897 	struct request *req;
898 
899 	rq_list_for_each(&iob->req_list, req) {
900 		fn(req);
901 		nvme_complete_batch_req(req);
902 	}
903 	blk_mq_end_request_batch(iob);
904 }
905 
906 blk_status_t nvme_host_path_error(struct request *req);
907 bool nvme_cancel_request(struct request *req, void *data);
908 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
909 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
910 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
911 		enum nvme_ctrl_state new_state);
912 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
913 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
914 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
915 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
916 int nvme_add_ctrl(struct nvme_ctrl *ctrl);
917 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
918 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
919 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
920 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
921 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
922 		const struct blk_mq_ops *ops, unsigned int cmd_size);
923 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
924 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
925 		const struct blk_mq_ops *ops, unsigned int nr_maps,
926 		unsigned int cmd_size);
927 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
928 
929 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
930 
931 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
932 		volatile union nvme_result *res);
933 
934 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
935 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
936 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
937 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
938 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
939 void nvme_sync_queues(struct nvme_ctrl *ctrl);
940 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
941 void nvme_unfreeze(struct nvme_ctrl *ctrl);
942 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
943 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl);
944 void nvme_start_freeze(struct nvme_ctrl *ctrl);
945 
946 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
947 {
948 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
949 }
950 
951 #define NVME_QID_ANY -1
952 void nvme_init_request(struct request *req, struct nvme_command *cmd);
953 void nvme_cleanup_cmd(struct request *req);
954 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
955 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
956 		struct request *req);
957 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
958 		bool queue_live, enum nvme_ctrl_state state);
959 
960 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
961 		bool queue_live)
962 {
963 	enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
964 
965 	if (likely(state == NVME_CTRL_LIVE))
966 		return true;
967 	if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING)
968 		return queue_live;
969 	return __nvme_check_ready(ctrl, rq, queue_live, state);
970 }
971 
972 /*
973  * NSID shall be unique for all shared namespaces, or if at least one of the
974  * following conditions is met:
975  *   1. Namespace Management is supported by the controller
976  *   2. ANA is supported by the controller
977  *   3. NVM Set are supported by the controller
978  *
979  * In other case, private namespace are not required to report a unique NSID.
980  */
981 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
982 		struct nvme_ns_head *head)
983 {
984 	return head->shared ||
985 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
986 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
987 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
988 }
989 
990 /*
991  * Flags for __nvme_submit_sync_cmd()
992  */
993 typedef __u32 __bitwise nvme_submit_flags_t;
994 
995 enum {
996 	/* Insert request at the head of the queue */
997 	NVME_SUBMIT_AT_HEAD  = (__force nvme_submit_flags_t)(1 << 0),
998 	/* Set BLK_MQ_REQ_NOWAIT when allocating request */
999 	NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1),
1000 	/* Set BLK_MQ_REQ_RESERVED when allocating request */
1001 	NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2),
1002 	/* Retry command when NVME_STATUS_DNR is not set in the result */
1003 	NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3),
1004 };
1005 
1006 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1007 		void *buf, unsigned bufflen);
1008 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1009 		union nvme_result *result, void *buffer, unsigned bufflen,
1010 		int qid, nvme_submit_flags_t flags);
1011 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1012 		      unsigned int dword11, void *buffer, size_t buflen,
1013 		      void *result);
1014 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1015 		      unsigned int dword11, void *buffer, size_t buflen,
1016 		      void *result);
1017 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
1018 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
1019 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
1020 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
1021 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
1022 void nvme_queue_scan(struct nvme_ctrl *ctrl);
1023 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
1024 		void *log, size_t size, u64 offset);
1025 void nvme_get_ns_head(struct nvme_ns_head *head);
1026 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
1027 void nvme_put_ns_head(struct nvme_ns_head *head);
1028 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
1029 		const struct file_operations *fops, struct module *owner,
1030 		int ctrl, int head);
1031 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
1032 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
1033 		unsigned int cmd, unsigned long arg);
1034 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
1035 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
1036 		unsigned int cmd, unsigned long arg);
1037 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
1038 		unsigned long arg);
1039 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1040 		unsigned long arg);
1041 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
1042 		struct io_comp_batch *iob, unsigned int poll_flags);
1043 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
1044 		unsigned int issue_flags);
1045 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
1046 		unsigned int issue_flags);
1047 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1048 		struct nvme_id_ns **id);
1049 int nvme_getgeo(struct gendisk *disk, struct hd_geometry *geo);
1050 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
1051 
1052 extern const struct attribute_group *nvme_ns_attr_groups[];
1053 extern const struct attribute_group nvme_ns_mpath_attr_group;
1054 extern const struct pr_ops nvme_pr_ops;
1055 extern const struct block_device_operations nvme_ns_head_ops;
1056 extern const struct attribute_group nvme_dev_attrs_group;
1057 extern const struct attribute_group nvme_dev_diag_attrs_group;
1058 extern const struct attribute_group *nvme_subsys_attrs_groups[];
1059 extern const struct attribute_group *nvme_dev_attr_groups[];
1060 extern const struct block_device_operations nvme_bdev_ops;
1061 
1062 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
1063 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head)
1064 	__must_hold_shared(&head->srcu);
1065 #ifdef CONFIG_NVME_MULTIPATH
1066 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1067 {
1068 	return ctrl->ana_log_buf != NULL;
1069 }
1070 
1071 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1072 	__must_hold(&subsys->lock);
1073 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1074 	__must_hold(&subsys->lock);
1075 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1076 	__must_hold(&subsys->lock);
1077 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
1078 void nvme_failover_req(struct request *req);
1079 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
1080 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
1081 void nvme_mpath_add_sysfs_link(struct nvme_ns_head *ns);
1082 void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns);
1083 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
1084 void nvme_mpath_put_disk(struct nvme_ns_head *head);
1085 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
1086 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
1087 void nvme_mpath_update(struct nvme_ctrl *ctrl);
1088 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
1089 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
1090 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
1091 void nvme_mpath_revalidate_paths(struct nvme_ns_head *head);
1092 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
1093 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
1094 void nvme_mpath_start_request(struct request *rq);
1095 void nvme_mpath_end_request(struct request *rq);
1096 
1097 static inline void nvme_trace_bio_complete(struct request *req)
1098 {
1099 	struct nvme_ns *ns = req->q->queuedata;
1100 
1101 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
1102 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
1103 }
1104 
1105 extern bool multipath;
1106 extern struct device_attribute dev_attr_ana_grpid;
1107 extern struct device_attribute dev_attr_ana_state;
1108 extern struct device_attribute dev_attr_queue_depth;
1109 extern struct device_attribute dev_attr_numa_nodes;
1110 extern struct device_attribute dev_attr_path_state;
1111 extern struct device_attribute dev_attr_delayed_removal_secs;
1112 extern struct device_attribute dev_attr_fail_on_ctrl_loss;
1113 extern struct device_attribute dev_attr_multipath_failover_count;
1114 extern struct device_attribute dev_attr_io_requeue_no_usable_path_count;
1115 extern struct device_attribute dev_attr_io_fail_no_available_path_count;
1116 extern struct device_attribute subsys_attr_iopolicy;
1117 extern struct device_attribute subsys_attr_iopolicies;
1118 
1119 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1120 {
1121 	return disk->fops == &nvme_ns_head_ops;
1122 }
1123 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1124 {
1125 	if (test_bit(NVME_NSHEAD_QUEUE_IF_NO_PATH, &head->flags))
1126 		return true;
1127 	return false;
1128 }
1129 #else
1130 #define multipath false
1131 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1132 {
1133 	return false;
1134 }
1135 static inline void nvme_failover_req(struct request *req)
1136 {
1137 }
1138 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
1139 {
1140 }
1141 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
1142 		struct nvme_ns_head *head)
1143 {
1144 	return 0;
1145 }
1146 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
1147 {
1148 }
1149 static inline void nvme_mpath_put_disk(struct nvme_ns_head *head)
1150 {
1151 }
1152 static inline void nvme_mpath_add_sysfs_link(struct nvme_ns *ns)
1153 {
1154 }
1155 static inline void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns)
1156 {
1157 }
1158 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
1159 {
1160 	return false;
1161 }
1162 static inline void nvme_mpath_revalidate_paths(struct nvme_ns_head *head)
1163 {
1164 }
1165 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
1166 {
1167 }
1168 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
1169 {
1170 }
1171 static inline void nvme_trace_bio_complete(struct request *req)
1172 {
1173 }
1174 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1175 {
1176 }
1177 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1178 		struct nvme_id_ctrl *id)
1179 {
1180 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1181 		dev_warn(ctrl->device,
1182 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1183 	return 0;
1184 }
1185 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1186 {
1187 }
1188 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1189 {
1190 }
1191 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1192 {
1193 }
1194 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1195 {
1196 }
1197 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1198 {
1199 }
1200 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1201 {
1202 }
1203 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1204 {
1205 }
1206 static inline void nvme_mpath_start_request(struct request *rq)
1207 {
1208 }
1209 static inline void nvme_mpath_end_request(struct request *rq)
1210 {
1211 }
1212 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1213 {
1214 	return false;
1215 }
1216 static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1217 {
1218 	return false;
1219 }
1220 #endif /* CONFIG_NVME_MULTIPATH */
1221 
1222 #if defined(CONFIG_NVME_MULTIPATH) && defined(CONFIG_BLK_DEV_ZONED)
1223 int nvme_mpath_revalidate_zones(struct nvme_ns_head *head);
1224 #else
1225 static inline int nvme_mpath_revalidate_zones(struct nvme_ns_head *head)
1226 {
1227 	return 0;
1228 }
1229 #endif
1230 
1231 int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16],
1232 		enum blk_unique_id type);
1233 
1234 struct nvme_zone_info {
1235 	u64 zone_size;
1236 	unsigned int max_open_zones;
1237 	unsigned int max_active_zones;
1238 };
1239 
1240 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1241 		unsigned int nr_zones, struct blk_report_zones_args *args);
1242 int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf,
1243 		struct nvme_zone_info *zi);
1244 void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim,
1245 		struct nvme_zone_info *zi);
1246 #ifdef CONFIG_BLK_DEV_ZONED
1247 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1248 				       struct nvme_command *cmnd,
1249 				       enum nvme_zone_mgmt_action action);
1250 #else
1251 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1252 		struct request *req, struct nvme_command *cmnd,
1253 		enum nvme_zone_mgmt_action action)
1254 {
1255 	return BLK_STS_NOTSUPP;
1256 }
1257 #endif
1258 
1259 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1260 {
1261 	struct gendisk *disk = dev_to_disk(dev);
1262 
1263 	WARN_ON(nvme_disk_is_ns_head(disk));
1264 	return disk->private_data;
1265 }
1266 
1267 #ifdef CONFIG_NVME_HWMON
1268 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1269 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1270 #else
1271 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1272 {
1273 	return 0;
1274 }
1275 
1276 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1277 {
1278 }
1279 #endif
1280 
1281 static inline void nvme_start_request(struct request *rq)
1282 {
1283 	if (rq->cmd_flags & REQ_NVME_MPATH)
1284 		nvme_mpath_start_request(rq);
1285 	blk_mq_start_request(rq);
1286 }
1287 
1288 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1289 {
1290 	return ctrl->sgls & (NVME_CTRL_SGLS_BYTE_ALIGNED |
1291 			     NVME_CTRL_SGLS_DWORD_ALIGNED);
1292 }
1293 
1294 static inline bool nvme_ctrl_meta_sgl_supported(struct nvme_ctrl *ctrl)
1295 {
1296 	if (ctrl->ops->flags & NVME_F_FABRICS)
1297 		return true;
1298 	return ctrl->sgls & NVME_CTRL_SGLS_MSDS;
1299 }
1300 
1301 #ifdef CONFIG_NVME_HOST_AUTH
1302 int __init nvme_init_auth(void);
1303 void __exit nvme_exit_auth(void);
1304 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1305 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1306 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1307 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1308 void nvme_auth_free(struct nvme_ctrl *ctrl);
1309 void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl);
1310 #else
1311 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1312 {
1313 	return 0;
1314 }
1315 static inline int __init nvme_init_auth(void)
1316 {
1317 	return 0;
1318 }
1319 static inline void __exit nvme_exit_auth(void)
1320 {
1321 }
1322 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1323 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1324 {
1325 	return -EPROTONOSUPPORT;
1326 }
1327 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1328 {
1329 	return -EPROTONOSUPPORT;
1330 }
1331 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1332 static inline void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl) {};
1333 #endif
1334 
1335 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1336 			 u8 opcode);
1337 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1338 	__cond_acquires(nonzero, &ctrl->subsys->lock)
1339 	__cond_acquires(nonzero, &ctrl->scan_lock);
1340 
1341 int nvme_execute_rq(struct request *rq, bool at_head);
1342 u32 nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1343 		       struct nvme_command *cmd, int status)
1344 		       __cond_releases(nonzero, &ctrl->scan_lock)
1345 		       __cond_releases(nonzero, &ctrl->subsys->lock);
1346 
1347 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1348 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1349 bool nvme_get_ns(struct nvme_ns *ns);
1350 void nvme_put_ns(struct nvme_ns *ns);
1351 
1352 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1353 {
1354 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1355 }
1356 
1357 #endif /* _NVME_H */
1358