xref: /linux/drivers/nvme/host/nvme.h (revision b58b13f156c00c2457035b7071eaaac105fe6836)
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 
53 /*
54  * List of workarounds for devices that required behavior not specified in
55  * the standard.
56  */
57 enum nvme_quirks {
58 	/*
59 	 * Prefers I/O aligned to a stripe size specified in a vendor
60 	 * specific Identify field.
61 	 */
62 	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
63 
64 	/*
65 	 * The controller doesn't handle Identify value others than 0 or 1
66 	 * correctly.
67 	 */
68 	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
69 
70 	/*
71 	 * The controller deterministically returns O's on reads to
72 	 * logical blocks that deallocate was called on.
73 	 */
74 	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
75 
76 	/*
77 	 * The controller needs a delay before starts checking the device
78 	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
79 	 */
80 	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
81 
82 	/*
83 	 * APST should not be used.
84 	 */
85 	NVME_QUIRK_NO_APST			= (1 << 4),
86 
87 	/*
88 	 * The deepest sleep state should not be used.
89 	 */
90 	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
91 
92 	/*
93 	 * Set MEDIUM priority on SQ creation
94 	 */
95 	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
96 
97 	/*
98 	 * Ignore device provided subnqn.
99 	 */
100 	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
101 
102 	/*
103 	 * Broken Write Zeroes.
104 	 */
105 	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
106 
107 	/*
108 	 * Force simple suspend/resume path.
109 	 */
110 	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
111 
112 	/*
113 	 * Use only one interrupt vector for all queues
114 	 */
115 	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
116 
117 	/*
118 	 * Use non-standard 128 bytes SQEs.
119 	 */
120 	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
121 
122 	/*
123 	 * Prevent tag overlap between queues
124 	 */
125 	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
126 
127 	/*
128 	 * Don't change the value of the temperature threshold feature
129 	 */
130 	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
131 
132 	/*
133 	 * The controller doesn't handle the Identify Namespace
134 	 * Identification Descriptor list subcommand despite claiming
135 	 * NVMe 1.3 compliance.
136 	 */
137 	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
138 
139 	/*
140 	 * The controller does not properly handle DMA addresses over
141 	 * 48 bits.
142 	 */
143 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
144 
145 	/*
146 	 * The controller requires the command_id value be limited, so skip
147 	 * encoding the generation sequence number.
148 	 */
149 	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
150 
151 	/*
152 	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
153 	 */
154 	NVME_QUIRK_BOGUS_NID			= (1 << 18),
155 
156 	/*
157 	 * No temperature thresholds for channels other than 0 (Composite).
158 	 */
159 	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
160 
161 	/*
162 	 * Disables simple suspend/resume path.
163 	 */
164 	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
165 };
166 
167 /*
168  * Common request structure for NVMe passthrough.  All drivers must have
169  * this structure as the first member of their request-private data.
170  */
171 struct nvme_request {
172 	struct nvme_command	*cmd;
173 	union nvme_result	result;
174 	u8			genctr;
175 	u8			retries;
176 	u8			flags;
177 	u16			status;
178 #ifdef CONFIG_NVME_MULTIPATH
179 	unsigned long		start_time;
180 #endif
181 	struct nvme_ctrl	*ctrl;
182 };
183 
184 /*
185  * Mark a bio as coming in through the mpath node.
186  */
187 #define REQ_NVME_MPATH		REQ_DRV
188 
189 enum {
190 	NVME_REQ_CANCELLED		= (1 << 0),
191 	NVME_REQ_USERCMD		= (1 << 1),
192 	NVME_MPATH_IO_STATS		= (1 << 2),
193 };
194 
195 static inline struct nvme_request *nvme_req(struct request *req)
196 {
197 	return blk_mq_rq_to_pdu(req);
198 }
199 
200 static inline u16 nvme_req_qid(struct request *req)
201 {
202 	if (!req->q->queuedata)
203 		return 0;
204 
205 	return req->mq_hctx->queue_num + 1;
206 }
207 
208 /* The below value is the specific amount of delay needed before checking
209  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
210  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
211  * found empirically.
212  */
213 #define NVME_QUIRK_DELAY_AMOUNT		2300
214 
215 /*
216  * enum nvme_ctrl_state: Controller state
217  *
218  * @NVME_CTRL_NEW:		New controller just allocated, initial state
219  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
220  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
221  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
222  *				transport
223  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
224  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
225  *				disabled/failed immediately. This state comes
226  * 				after all async event processing took place and
227  * 				before ns removal and the controller deletion
228  * 				progress
229  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
230  *				shutdown or removal. In this case we forcibly
231  *				kill all inflight I/O as they have no chance to
232  *				complete
233  */
234 enum nvme_ctrl_state {
235 	NVME_CTRL_NEW,
236 	NVME_CTRL_LIVE,
237 	NVME_CTRL_RESETTING,
238 	NVME_CTRL_CONNECTING,
239 	NVME_CTRL_DELETING,
240 	NVME_CTRL_DELETING_NOIO,
241 	NVME_CTRL_DEAD,
242 };
243 
244 struct nvme_fault_inject {
245 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
246 	struct fault_attr attr;
247 	struct dentry *parent;
248 	bool dont_retry;	/* DNR, do not retry */
249 	u16 status;		/* status code */
250 #endif
251 };
252 
253 enum nvme_ctrl_flags {
254 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
255 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
256 	NVME_CTRL_STARTED_ONCE		= 2,
257 	NVME_CTRL_STOPPED		= 3,
258 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
259 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
260 	NVME_CTRL_FROZEN		= 6,
261 };
262 
263 struct nvme_ctrl {
264 	bool comp_seen;
265 	bool identified;
266 	enum nvme_ctrl_state state;
267 	spinlock_t lock;
268 	struct mutex scan_lock;
269 	const struct nvme_ctrl_ops *ops;
270 	struct request_queue *admin_q;
271 	struct request_queue *connect_q;
272 	struct request_queue *fabrics_q;
273 	struct device *dev;
274 	int instance;
275 	int numa_node;
276 	struct blk_mq_tag_set *tagset;
277 	struct blk_mq_tag_set *admin_tagset;
278 	struct list_head namespaces;
279 	struct rw_semaphore namespaces_rwsem;
280 	struct device ctrl_device;
281 	struct device *device;	/* char device */
282 #ifdef CONFIG_NVME_HWMON
283 	struct device *hwmon_device;
284 #endif
285 	struct cdev cdev;
286 	struct work_struct reset_work;
287 	struct work_struct delete_work;
288 	wait_queue_head_t state_wq;
289 
290 	struct nvme_subsystem *subsys;
291 	struct list_head subsys_entry;
292 
293 	struct opal_dev *opal_dev;
294 
295 	char name[12];
296 	u16 cntlid;
297 
298 	u16 mtfa;
299 	u32 ctrl_config;
300 	u32 queue_count;
301 
302 	u64 cap;
303 	u32 max_hw_sectors;
304 	u32 max_segments;
305 	u32 max_integrity_segments;
306 	u32 max_zeroes_sectors;
307 #ifdef CONFIG_BLK_DEV_ZONED
308 	u32 max_zone_append;
309 #endif
310 	u16 crdt[3];
311 	u16 oncs;
312 	u8 dmrl;
313 	u32 dmrsl;
314 	u16 oacs;
315 	u16 sqsize;
316 	u32 max_namespaces;
317 	atomic_t abort_limit;
318 	u8 vwc;
319 	u32 vs;
320 	u32 sgls;
321 	u16 kas;
322 	u8 npss;
323 	u8 apsta;
324 	u16 wctemp;
325 	u16 cctemp;
326 	u32 oaes;
327 	u32 aen_result;
328 	u32 ctratt;
329 	unsigned int shutdown_timeout;
330 	unsigned int kato;
331 	bool subsystem;
332 	unsigned long quirks;
333 	struct nvme_id_power_state psd[32];
334 	struct nvme_effects_log *effects;
335 	struct xarray cels;
336 	struct work_struct scan_work;
337 	struct work_struct async_event_work;
338 	struct delayed_work ka_work;
339 	struct delayed_work failfast_work;
340 	struct nvme_command ka_cmd;
341 	unsigned long ka_last_check_time;
342 	struct work_struct fw_act_work;
343 	unsigned long events;
344 
345 #ifdef CONFIG_NVME_MULTIPATH
346 	/* asymmetric namespace access: */
347 	u8 anacap;
348 	u8 anatt;
349 	u32 anagrpmax;
350 	u32 nanagrpid;
351 	struct mutex ana_lock;
352 	struct nvme_ana_rsp_hdr *ana_log_buf;
353 	size_t ana_log_size;
354 	struct timer_list anatt_timer;
355 	struct work_struct ana_work;
356 #endif
357 
358 #ifdef CONFIG_NVME_HOST_AUTH
359 	struct work_struct dhchap_auth_work;
360 	struct mutex dhchap_auth_mutex;
361 	struct nvme_dhchap_queue_context *dhchap_ctxs;
362 	struct nvme_dhchap_key *host_key;
363 	struct nvme_dhchap_key *ctrl_key;
364 	u16 transaction;
365 #endif
366 	struct key *tls_key;
367 
368 	/* Power saving configuration */
369 	u64 ps_max_latency_us;
370 	bool apst_enabled;
371 
372 	/* PCIe only: */
373 	u16 hmmaxd;
374 	u32 hmpre;
375 	u32 hmmin;
376 	u32 hmminds;
377 
378 	/* Fabrics only */
379 	u32 ioccsz;
380 	u32 iorcsz;
381 	u16 icdoff;
382 	u16 maxcmd;
383 	int nr_reconnects;
384 	unsigned long flags;
385 	struct nvmf_ctrl_options *opts;
386 
387 	struct page *discard_page;
388 	unsigned long discard_page_busy;
389 
390 	struct nvme_fault_inject fault_inject;
391 
392 	enum nvme_ctrl_type cntrltype;
393 	enum nvme_dctype dctype;
394 };
395 
396 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
397 {
398 	return READ_ONCE(ctrl->state);
399 }
400 
401 enum nvme_iopolicy {
402 	NVME_IOPOLICY_NUMA,
403 	NVME_IOPOLICY_RR,
404 };
405 
406 struct nvme_subsystem {
407 	int			instance;
408 	struct device		dev;
409 	/*
410 	 * Because we unregister the device on the last put we need
411 	 * a separate refcount.
412 	 */
413 	struct kref		ref;
414 	struct list_head	entry;
415 	struct mutex		lock;
416 	struct list_head	ctrls;
417 	struct list_head	nsheads;
418 	char			subnqn[NVMF_NQN_SIZE];
419 	char			serial[20];
420 	char			model[40];
421 	char			firmware_rev[8];
422 	u8			cmic;
423 	enum nvme_subsys_type	subtype;
424 	u16			vendor_id;
425 	u16			awupf;	/* 0's based awupf value. */
426 	struct ida		ns_ida;
427 #ifdef CONFIG_NVME_MULTIPATH
428 	enum nvme_iopolicy	iopolicy;
429 #endif
430 };
431 
432 /*
433  * Container structure for uniqueue namespace identifiers.
434  */
435 struct nvme_ns_ids {
436 	u8	eui64[8];
437 	u8	nguid[16];
438 	uuid_t	uuid;
439 	u8	csi;
440 };
441 
442 /*
443  * Anchor structure for namespaces.  There is one for each namespace in a
444  * NVMe subsystem that any of our controllers can see, and the namespace
445  * structure for each controller is chained of it.  For private namespaces
446  * there is a 1:1 relation to our namespace structures, that is ->list
447  * only ever has a single entry for private namespaces.
448  */
449 struct nvme_ns_head {
450 	struct list_head	list;
451 	struct srcu_struct      srcu;
452 	struct nvme_subsystem	*subsys;
453 	struct nvme_ns_ids	ids;
454 	struct list_head	entry;
455 	struct kref		ref;
456 	bool			shared;
457 	int			instance;
458 	struct nvme_effects_log *effects;
459 	u64			nuse;
460 	unsigned		ns_id;
461 	int			lba_shift;
462 	u16			ms;
463 	u16			pi_size;
464 	u8			pi_type;
465 	u8			guard_type;
466 	u16			sgs;
467 	u32			sws;
468 #ifdef CONFIG_BLK_DEV_ZONED
469 	u64			zsze;
470 #endif
471 	unsigned long		features;
472 
473 	struct ratelimit_state	rs_nuse;
474 
475 	struct cdev		cdev;
476 	struct device		cdev_device;
477 
478 	struct gendisk		*disk;
479 #ifdef CONFIG_NVME_MULTIPATH
480 	struct bio_list		requeue_list;
481 	spinlock_t		requeue_lock;
482 	struct work_struct	requeue_work;
483 	struct mutex		lock;
484 	unsigned long		flags;
485 #define NVME_NSHEAD_DISK_LIVE	0
486 	struct nvme_ns __rcu	*current_path[];
487 #endif
488 };
489 
490 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
491 {
492 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
493 }
494 
495 enum nvme_ns_features {
496 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
497 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
498 	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
499 };
500 
501 struct nvme_ns {
502 	struct list_head list;
503 
504 	struct nvme_ctrl *ctrl;
505 	struct request_queue *queue;
506 	struct gendisk *disk;
507 #ifdef CONFIG_NVME_MULTIPATH
508 	enum nvme_ana_state ana_state;
509 	u32 ana_grpid;
510 #endif
511 	struct list_head siblings;
512 	struct kref kref;
513 	struct nvme_ns_head *head;
514 
515 	unsigned long flags;
516 #define NVME_NS_REMOVING	0
517 #define NVME_NS_ANA_PENDING	2
518 #define NVME_NS_FORCE_RO	3
519 #define NVME_NS_READY		4
520 
521 	struct cdev		cdev;
522 	struct device		cdev_device;
523 
524 	struct nvme_fault_inject fault_inject;
525 
526 };
527 
528 /* NVMe ns supports metadata actions by the controller (generate/strip) */
529 static inline bool nvme_ns_has_pi(struct nvme_ns_head *head)
530 {
531 	return head->pi_type && head->ms == head->pi_size;
532 }
533 
534 struct nvme_ctrl_ops {
535 	const char *name;
536 	struct module *module;
537 	unsigned int flags;
538 #define NVME_F_FABRICS			(1 << 0)
539 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
540 #define NVME_F_BLOCKING			(1 << 2)
541 
542 	const struct attribute_group **dev_attr_groups;
543 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
544 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
545 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
546 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
547 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
548 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
549 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
550 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
551 	void (*print_device_info)(struct nvme_ctrl *ctrl);
552 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
553 };
554 
555 /*
556  * nvme command_id is constructed as such:
557  * | xxxx | xxxxxxxxxxxx |
558  *   gen    request tag
559  */
560 #define nvme_genctr_mask(gen)			(gen & 0xf)
561 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
562 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
563 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
564 
565 static inline u16 nvme_cid(struct request *rq)
566 {
567 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
568 }
569 
570 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
571 		u16 command_id)
572 {
573 	u8 genctr = nvme_genctr_from_cid(command_id);
574 	u16 tag = nvme_tag_from_cid(command_id);
575 	struct request *rq;
576 
577 	rq = blk_mq_tag_to_rq(tags, tag);
578 	if (unlikely(!rq)) {
579 		pr_err("could not locate request for tag %#x\n",
580 			tag);
581 		return NULL;
582 	}
583 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
584 		dev_err(nvme_req(rq)->ctrl->device,
585 			"request %#x genctr mismatch (got %#x expected %#x)\n",
586 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
587 		return NULL;
588 	}
589 	return rq;
590 }
591 
592 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
593                 u16 command_id)
594 {
595 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
596 }
597 
598 /*
599  * Return the length of the string without the space padding
600  */
601 static inline int nvme_strlen(char *s, int len)
602 {
603 	while (s[len - 1] == ' ')
604 		len--;
605 	return len;
606 }
607 
608 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
609 {
610 	struct nvme_subsystem *subsys = ctrl->subsys;
611 
612 	if (ctrl->ops->print_device_info) {
613 		ctrl->ops->print_device_info(ctrl);
614 		return;
615 	}
616 
617 	dev_err(ctrl->device,
618 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
619 		nvme_strlen(subsys->model, sizeof(subsys->model)),
620 		subsys->model, nvme_strlen(subsys->firmware_rev,
621 					   sizeof(subsys->firmware_rev)),
622 		subsys->firmware_rev);
623 }
624 
625 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
626 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
627 			    const char *dev_name);
628 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
629 void nvme_should_fail(struct request *req);
630 #else
631 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
632 					  const char *dev_name)
633 {
634 }
635 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
636 {
637 }
638 static inline void nvme_should_fail(struct request *req) {}
639 #endif
640 
641 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
642 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
643 
644 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
645 {
646 	int ret;
647 
648 	if (!ctrl->subsystem)
649 		return -ENOTTY;
650 	if (!nvme_wait_reset(ctrl))
651 		return -EBUSY;
652 
653 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
654 	if (ret)
655 		return ret;
656 
657 	return nvme_try_sched_reset(ctrl);
658 }
659 
660 /*
661  * Convert a 512B sector number to a device logical block number.
662  */
663 static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector)
664 {
665 	return sector >> (head->lba_shift - SECTOR_SHIFT);
666 }
667 
668 /*
669  * Convert a device logical block number to a 512B sector number.
670  */
671 static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba)
672 {
673 	return lba << (head->lba_shift - SECTOR_SHIFT);
674 }
675 
676 /*
677  * Convert byte length to nvme's 0-based num dwords
678  */
679 static inline u32 nvme_bytes_to_numd(size_t len)
680 {
681 	return (len >> 2) - 1;
682 }
683 
684 static inline bool nvme_is_ana_error(u16 status)
685 {
686 	switch (status & 0x7ff) {
687 	case NVME_SC_ANA_TRANSITION:
688 	case NVME_SC_ANA_INACCESSIBLE:
689 	case NVME_SC_ANA_PERSISTENT_LOSS:
690 		return true;
691 	default:
692 		return false;
693 	}
694 }
695 
696 static inline bool nvme_is_path_error(u16 status)
697 {
698 	/* check for a status code type of 'path related status' */
699 	return (status & 0x700) == 0x300;
700 }
701 
702 /*
703  * Fill in the status and result information from the CQE, and then figure out
704  * if blk-mq will need to use IPI magic to complete the request, and if yes do
705  * so.  If not let the caller complete the request without an indirect function
706  * call.
707  */
708 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
709 		union nvme_result result)
710 {
711 	struct nvme_request *rq = nvme_req(req);
712 	struct nvme_ctrl *ctrl = rq->ctrl;
713 
714 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
715 		rq->genctr++;
716 
717 	rq->status = le16_to_cpu(status) >> 1;
718 	rq->result = result;
719 	/* inject error when permitted by fault injection framework */
720 	nvme_should_fail(req);
721 	if (unlikely(blk_should_fake_timeout(req->q)))
722 		return true;
723 	return blk_mq_complete_request_remote(req);
724 }
725 
726 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
727 {
728 	get_device(ctrl->device);
729 }
730 
731 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
732 {
733 	put_device(ctrl->device);
734 }
735 
736 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
737 {
738 	return !qid &&
739 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
740 }
741 
742 void nvme_complete_rq(struct request *req);
743 void nvme_complete_batch_req(struct request *req);
744 
745 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
746 						void (*fn)(struct request *rq))
747 {
748 	struct request *req;
749 
750 	rq_list_for_each(&iob->req_list, req) {
751 		fn(req);
752 		nvme_complete_batch_req(req);
753 	}
754 	blk_mq_end_request_batch(iob);
755 }
756 
757 blk_status_t nvme_host_path_error(struct request *req);
758 bool nvme_cancel_request(struct request *req, void *data);
759 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
760 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
761 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
762 		enum nvme_ctrl_state new_state);
763 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
764 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
765 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
766 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
767 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
768 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
769 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
770 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
771 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
772 		const struct blk_mq_ops *ops, unsigned int cmd_size);
773 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
774 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
775 		const struct blk_mq_ops *ops, unsigned int nr_maps,
776 		unsigned int cmd_size);
777 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
778 
779 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
780 
781 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
782 		volatile union nvme_result *res);
783 
784 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
785 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
786 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
787 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
788 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
789 void nvme_sync_queues(struct nvme_ctrl *ctrl);
790 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
791 void nvme_unfreeze(struct nvme_ctrl *ctrl);
792 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
793 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
794 void nvme_start_freeze(struct nvme_ctrl *ctrl);
795 
796 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
797 {
798 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
799 }
800 
801 #define NVME_QID_ANY -1
802 void nvme_init_request(struct request *req, struct nvme_command *cmd);
803 void nvme_cleanup_cmd(struct request *req);
804 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
805 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
806 		struct request *req);
807 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
808 		bool queue_live);
809 
810 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
811 		bool queue_live)
812 {
813 	if (likely(ctrl->state == NVME_CTRL_LIVE))
814 		return true;
815 	if (ctrl->ops->flags & NVME_F_FABRICS &&
816 	    ctrl->state == NVME_CTRL_DELETING)
817 		return queue_live;
818 	return __nvme_check_ready(ctrl, rq, queue_live);
819 }
820 
821 /*
822  * NSID shall be unique for all shared namespaces, or if at least one of the
823  * following conditions is met:
824  *   1. Namespace Management is supported by the controller
825  *   2. ANA is supported by the controller
826  *   3. NVM Set are supported by the controller
827  *
828  * In other case, private namespace are not required to report a unique NSID.
829  */
830 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
831 		struct nvme_ns_head *head)
832 {
833 	return head->shared ||
834 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
835 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
836 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
837 }
838 
839 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
840 		void *buf, unsigned bufflen);
841 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
842 		union nvme_result *result, void *buffer, unsigned bufflen,
843 		int qid, int at_head,
844 		blk_mq_req_flags_t flags);
845 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
846 		      unsigned int dword11, void *buffer, size_t buflen,
847 		      u32 *result);
848 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
849 		      unsigned int dword11, void *buffer, size_t buflen,
850 		      u32 *result);
851 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
852 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
853 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
854 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
855 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
856 void nvme_queue_scan(struct nvme_ctrl *ctrl);
857 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
858 		void *log, size_t size, u64 offset);
859 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
860 void nvme_put_ns_head(struct nvme_ns_head *head);
861 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
862 		const struct file_operations *fops, struct module *owner);
863 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
864 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
865 		unsigned int cmd, unsigned long arg);
866 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
867 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
868 		unsigned int cmd, unsigned long arg);
869 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
870 		unsigned long arg);
871 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
872 		unsigned long arg);
873 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
874 		struct io_comp_batch *iob, unsigned int poll_flags);
875 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
876 		unsigned int issue_flags);
877 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
878 		unsigned int issue_flags);
879 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
880 		struct nvme_id_ns **id);
881 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
882 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
883 
884 extern const struct attribute_group *nvme_ns_attr_groups[];
885 extern const struct pr_ops nvme_pr_ops;
886 extern const struct block_device_operations nvme_ns_head_ops;
887 extern const struct attribute_group nvme_dev_attrs_group;
888 extern const struct attribute_group *nvme_subsys_attrs_groups[];
889 extern const struct attribute_group *nvme_dev_attr_groups[];
890 extern const struct block_device_operations nvme_bdev_ops;
891 
892 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
893 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
894 #ifdef CONFIG_NVME_MULTIPATH
895 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
896 {
897 	return ctrl->ana_log_buf != NULL;
898 }
899 
900 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
901 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
902 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
903 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
904 void nvme_failover_req(struct request *req);
905 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
906 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
907 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
908 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
909 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
910 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
911 void nvme_mpath_update(struct nvme_ctrl *ctrl);
912 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
913 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
914 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
915 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
916 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
917 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
918 void nvme_mpath_start_request(struct request *rq);
919 void nvme_mpath_end_request(struct request *rq);
920 
921 static inline void nvme_trace_bio_complete(struct request *req)
922 {
923 	struct nvme_ns *ns = req->q->queuedata;
924 
925 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
926 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
927 }
928 
929 extern bool multipath;
930 extern struct device_attribute dev_attr_ana_grpid;
931 extern struct device_attribute dev_attr_ana_state;
932 extern struct device_attribute subsys_attr_iopolicy;
933 
934 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
935 {
936 	return disk->fops == &nvme_ns_head_ops;
937 }
938 #else
939 #define multipath false
940 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
941 {
942 	return false;
943 }
944 static inline void nvme_failover_req(struct request *req)
945 {
946 }
947 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
948 {
949 }
950 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
951 		struct nvme_ns_head *head)
952 {
953 	return 0;
954 }
955 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
956 {
957 }
958 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
959 {
960 }
961 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
962 {
963 	return false;
964 }
965 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
966 {
967 }
968 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
969 {
970 }
971 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
972 {
973 }
974 static inline void nvme_trace_bio_complete(struct request *req)
975 {
976 }
977 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
978 {
979 }
980 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
981 		struct nvme_id_ctrl *id)
982 {
983 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
984 		dev_warn(ctrl->device,
985 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
986 	return 0;
987 }
988 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
989 {
990 }
991 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
992 {
993 }
994 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
995 {
996 }
997 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
998 {
999 }
1000 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1001 {
1002 }
1003 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1004 {
1005 }
1006 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1007 {
1008 }
1009 static inline void nvme_mpath_start_request(struct request *rq)
1010 {
1011 }
1012 static inline void nvme_mpath_end_request(struct request *rq)
1013 {
1014 }
1015 static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1016 {
1017 	return false;
1018 }
1019 #endif /* CONFIG_NVME_MULTIPATH */
1020 
1021 int nvme_revalidate_zones(struct nvme_ns *ns);
1022 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1023 		unsigned int nr_zones, report_zones_cb cb, void *data);
1024 #ifdef CONFIG_BLK_DEV_ZONED
1025 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1026 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1027 				       struct nvme_command *cmnd,
1028 				       enum nvme_zone_mgmt_action action);
1029 #else
1030 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1031 		struct request *req, struct nvme_command *cmnd,
1032 		enum nvme_zone_mgmt_action action)
1033 {
1034 	return BLK_STS_NOTSUPP;
1035 }
1036 
1037 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1038 {
1039 	dev_warn(ns->ctrl->device,
1040 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1041 	return -EPROTONOSUPPORT;
1042 }
1043 #endif
1044 
1045 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1046 {
1047 	struct gendisk *disk = dev_to_disk(dev);
1048 
1049 	WARN_ON(nvme_disk_is_ns_head(disk));
1050 	return disk->private_data;
1051 }
1052 
1053 #ifdef CONFIG_NVME_HWMON
1054 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1055 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1056 #else
1057 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1058 {
1059 	return 0;
1060 }
1061 
1062 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1063 {
1064 }
1065 #endif
1066 
1067 static inline void nvme_start_request(struct request *rq)
1068 {
1069 	if (rq->cmd_flags & REQ_NVME_MPATH)
1070 		nvme_mpath_start_request(rq);
1071 	blk_mq_start_request(rq);
1072 }
1073 
1074 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1075 {
1076 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1077 }
1078 
1079 #ifdef CONFIG_NVME_HOST_AUTH
1080 int __init nvme_init_auth(void);
1081 void __exit nvme_exit_auth(void);
1082 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1083 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1084 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1085 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1086 void nvme_auth_free(struct nvme_ctrl *ctrl);
1087 #else
1088 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1089 {
1090 	return 0;
1091 }
1092 static inline int __init nvme_init_auth(void)
1093 {
1094 	return 0;
1095 }
1096 static inline void __exit nvme_exit_auth(void)
1097 {
1098 }
1099 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1100 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1101 {
1102 	return -EPROTONOSUPPORT;
1103 }
1104 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1105 {
1106 	return NVME_SC_AUTH_REQUIRED;
1107 }
1108 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1109 #endif
1110 
1111 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1112 			 u8 opcode);
1113 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1114 int nvme_execute_rq(struct request *rq, bool at_head);
1115 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1116 		       struct nvme_command *cmd, int status);
1117 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1118 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1119 void nvme_put_ns(struct nvme_ns *ns);
1120 
1121 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1122 {
1123 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1124 }
1125 
1126 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1127 const unsigned char *nvme_get_error_status_str(u16 status);
1128 const unsigned char *nvme_get_opcode_str(u8 opcode);
1129 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1130 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1131 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1132 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1133 {
1134 	return "I/O Error";
1135 }
1136 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1137 {
1138 	return "I/O Cmd";
1139 }
1140 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1141 {
1142 	return "Admin Cmd";
1143 }
1144 
1145 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1146 {
1147 	return "Fabrics Cmd";
1148 }
1149 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1150 
1151 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1152 {
1153 	if (opcode == nvme_fabrics_command)
1154 		return nvme_get_fabrics_opcode_str(fctype);
1155 	return qid ? nvme_get_opcode_str(opcode) :
1156 		nvme_get_admin_opcode_str(opcode);
1157 }
1158 #endif /* _NVME_H */
1159