xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision e5242c5f)
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 
20 #include <trace/events/block.h>
21 
22 extern const struct pr_ops nvme_pr_ops;
23 
24 extern unsigned int nvme_io_timeout;
25 #define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
26 
27 extern unsigned int admin_timeout;
28 #define NVME_ADMIN_TIMEOUT	(admin_timeout * HZ)
29 
30 #define NVME_DEFAULT_KATO	5
31 
32 #ifdef CONFIG_ARCH_NO_SG_CHAIN
33 #define  NVME_INLINE_SG_CNT  0
34 #define  NVME_INLINE_METADATA_SG_CNT  0
35 #else
36 #define  NVME_INLINE_SG_CNT  2
37 #define  NVME_INLINE_METADATA_SG_CNT  1
38 #endif
39 
40 /*
41  * Default to a 4K page size, with the intention to update this
42  * path in the future to accommodate architectures with differing
43  * kernel and IO page sizes.
44  */
45 #define NVME_CTRL_PAGE_SHIFT	12
46 #define NVME_CTRL_PAGE_SIZE	(1 << NVME_CTRL_PAGE_SHIFT)
47 
48 extern struct workqueue_struct *nvme_wq;
49 extern struct workqueue_struct *nvme_reset_wq;
50 extern struct workqueue_struct *nvme_delete_wq;
51 extern struct mutex nvme_subsystems_lock;
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 	 *  Problems seen with concurrent commands
94 	 */
95 	NVME_QUIRK_QDEPTH_ONE			= (1 << 6),
96 
97 	/*
98 	 * Set MEDIUM priority on SQ creation
99 	 */
100 	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
101 
102 	/*
103 	 * Ignore device provided subnqn.
104 	 */
105 	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
106 
107 	/*
108 	 * Broken Write Zeroes.
109 	 */
110 	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
111 
112 	/*
113 	 * Force simple suspend/resume path.
114 	 */
115 	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
116 
117 	/*
118 	 * Use only one interrupt vector for all queues
119 	 */
120 	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
121 
122 	/*
123 	 * Use non-standard 128 bytes SQEs.
124 	 */
125 	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
126 
127 	/*
128 	 * Prevent tag overlap between queues
129 	 */
130 	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
131 
132 	/*
133 	 * Don't change the value of the temperature threshold feature
134 	 */
135 	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
136 
137 	/*
138 	 * The controller doesn't handle the Identify Namespace
139 	 * Identification Descriptor list subcommand despite claiming
140 	 * NVMe 1.3 compliance.
141 	 */
142 	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
143 
144 	/*
145 	 * The controller does not properly handle DMA addresses over
146 	 * 48 bits.
147 	 */
148 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
149 
150 	/*
151 	 * The controller requires the command_id value be limited, so skip
152 	 * encoding the generation sequence number.
153 	 */
154 	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
155 
156 	/*
157 	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
158 	 */
159 	NVME_QUIRK_BOGUS_NID			= (1 << 18),
160 
161 	/*
162 	 * No temperature thresholds for channels other than 0 (Composite).
163 	 */
164 	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
165 
166 	/*
167 	 * Disables simple suspend/resume path.
168 	 */
169 	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
170 
171 	/*
172 	 * MSI (but not MSI-X) interrupts are broken and never fire.
173 	 */
174 	NVME_QUIRK_BROKEN_MSI			= (1 << 21),
175 };
176 
177 /*
178  * Common request structure for NVMe passthrough.  All drivers must have
179  * this structure as the first member of their request-private data.
180  */
181 struct nvme_request {
182 	struct nvme_command	*cmd;
183 	union nvme_result	result;
184 	u8			genctr;
185 	u8			retries;
186 	u8			flags;
187 	u16			status;
188 #ifdef CONFIG_NVME_MULTIPATH
189 	unsigned long		start_time;
190 #endif
191 	struct nvme_ctrl	*ctrl;
192 };
193 
194 /*
195  * Mark a bio as coming in through the mpath node.
196  */
197 #define REQ_NVME_MPATH		REQ_DRV
198 
199 enum {
200 	NVME_REQ_CANCELLED		= (1 << 0),
201 	NVME_REQ_USERCMD		= (1 << 1),
202 	NVME_MPATH_IO_STATS		= (1 << 2),
203 	NVME_MPATH_CNT_ACTIVE		= (1 << 3),
204 };
205 
206 static inline struct nvme_request *nvme_req(struct request *req)
207 {
208 	return blk_mq_rq_to_pdu(req);
209 }
210 
211 static inline u16 nvme_req_qid(struct request *req)
212 {
213 	if (!req->q->queuedata)
214 		return 0;
215 
216 	return req->mq_hctx->queue_num + 1;
217 }
218 
219 /* The below value is the specific amount of delay needed before checking
220  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
221  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
222  * found empirically.
223  */
224 #define NVME_QUIRK_DELAY_AMOUNT		2300
225 
226 /*
227  * enum nvme_ctrl_state: Controller state
228  *
229  * @NVME_CTRL_NEW:		New controller just allocated, initial state
230  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
231  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
232  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
233  *				transport
234  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
235  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
236  *				disabled/failed immediately. This state comes
237  * 				after all async event processing took place and
238  * 				before ns removal and the controller deletion
239  * 				progress
240  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
241  *				shutdown or removal. In this case we forcibly
242  *				kill all inflight I/O as they have no chance to
243  *				complete
244  */
245 enum nvme_ctrl_state {
246 	NVME_CTRL_NEW,
247 	NVME_CTRL_LIVE,
248 	NVME_CTRL_RESETTING,
249 	NVME_CTRL_CONNECTING,
250 	NVME_CTRL_DELETING,
251 	NVME_CTRL_DELETING_NOIO,
252 	NVME_CTRL_DEAD,
253 };
254 
255 struct nvme_fault_inject {
256 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
257 	struct fault_attr attr;
258 	struct dentry *parent;
259 	bool dont_retry;	/* DNR, do not retry */
260 	u16 status;		/* status code */
261 #endif
262 };
263 
264 enum nvme_ctrl_flags {
265 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
266 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
267 	NVME_CTRL_STARTED_ONCE		= 2,
268 	NVME_CTRL_STOPPED		= 3,
269 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
270 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
271 	NVME_CTRL_FROZEN		= 6,
272 };
273 
274 struct nvme_ctrl {
275 	bool comp_seen;
276 	bool identified;
277 	enum nvme_ctrl_state state;
278 	spinlock_t lock;
279 	struct mutex scan_lock;
280 	const struct nvme_ctrl_ops *ops;
281 	struct request_queue *admin_q;
282 	struct request_queue *connect_q;
283 	struct request_queue *fabrics_q;
284 	struct device *dev;
285 	int instance;
286 	int numa_node;
287 	struct blk_mq_tag_set *tagset;
288 	struct blk_mq_tag_set *admin_tagset;
289 	struct list_head namespaces;
290 	struct mutex namespaces_lock;
291 	struct srcu_struct srcu;
292 	struct device ctrl_device;
293 	struct device *device;	/* char device */
294 #ifdef CONFIG_NVME_HWMON
295 	struct device *hwmon_device;
296 #endif
297 	struct cdev cdev;
298 	struct work_struct reset_work;
299 	struct work_struct delete_work;
300 	wait_queue_head_t state_wq;
301 
302 	struct nvme_subsystem *subsys;
303 	struct list_head subsys_entry;
304 
305 	struct opal_dev *opal_dev;
306 
307 	char name[12];
308 	u16 cntlid;
309 
310 	u16 mtfa;
311 	u32 ctrl_config;
312 	u32 queue_count;
313 
314 	u64 cap;
315 	u32 max_hw_sectors;
316 	u32 max_segments;
317 	u32 max_integrity_segments;
318 	u32 max_discard_sectors;
319 	u32 max_discard_segments;
320 	u32 max_zeroes_sectors;
321 #ifdef CONFIG_BLK_DEV_ZONED
322 	u32 max_zone_append;
323 #endif
324 	u16 crdt[3];
325 	u16 oncs;
326 	u32 dmrsl;
327 	u16 oacs;
328 	u16 sqsize;
329 	u32 max_namespaces;
330 	atomic_t abort_limit;
331 	u8 vwc;
332 	u32 vs;
333 	u32 sgls;
334 	u16 kas;
335 	u8 npss;
336 	u8 apsta;
337 	u16 wctemp;
338 	u16 cctemp;
339 	u32 oaes;
340 	u32 aen_result;
341 	u32 ctratt;
342 	unsigned int shutdown_timeout;
343 	unsigned int kato;
344 	bool subsystem;
345 	unsigned long quirks;
346 	struct nvme_id_power_state psd[32];
347 	struct nvme_effects_log *effects;
348 	struct xarray cels;
349 	struct work_struct scan_work;
350 	struct work_struct async_event_work;
351 	struct delayed_work ka_work;
352 	struct delayed_work failfast_work;
353 	struct nvme_command ka_cmd;
354 	unsigned long ka_last_check_time;
355 	struct work_struct fw_act_work;
356 	unsigned long events;
357 
358 #ifdef CONFIG_NVME_MULTIPATH
359 	/* asymmetric namespace access: */
360 	u8 anacap;
361 	u8 anatt;
362 	u32 anagrpmax;
363 	u32 nanagrpid;
364 	struct mutex ana_lock;
365 	struct nvme_ana_rsp_hdr *ana_log_buf;
366 	size_t ana_log_size;
367 	struct timer_list anatt_timer;
368 	struct work_struct ana_work;
369 	atomic_t nr_active;
370 #endif
371 
372 #ifdef CONFIG_NVME_AUTH
373 	struct work_struct dhchap_auth_work;
374 	struct mutex dhchap_auth_mutex;
375 	struct nvme_dhchap_queue_context *dhchap_ctxs;
376 	struct nvme_dhchap_key *host_key;
377 	struct nvme_dhchap_key *ctrl_key;
378 	u16 transaction;
379 #endif
380 
381 	/* Power saving configuration */
382 	u64 ps_max_latency_us;
383 	bool apst_enabled;
384 
385 	/* PCIe only: */
386 	u16 hmmaxd;
387 	u32 hmpre;
388 	u32 hmmin;
389 	u32 hmminds;
390 
391 	/* Fabrics only */
392 	u32 ioccsz;
393 	u32 iorcsz;
394 	u16 icdoff;
395 	u16 maxcmd;
396 	int nr_reconnects;
397 	unsigned long flags;
398 	struct nvmf_ctrl_options *opts;
399 
400 	struct page *discard_page;
401 	unsigned long discard_page_busy;
402 
403 	struct nvme_fault_inject fault_inject;
404 
405 	enum nvme_ctrl_type cntrltype;
406 	enum nvme_dctype dctype;
407 };
408 
409 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
410 {
411 	return READ_ONCE(ctrl->state);
412 }
413 
414 enum nvme_iopolicy {
415 	NVME_IOPOLICY_NUMA,
416 	NVME_IOPOLICY_RR,
417 	NVME_IOPOLICY_QD,
418 };
419 
420 struct nvme_subsystem {
421 	int			instance;
422 	struct device		dev;
423 	/*
424 	 * Because we unregister the device on the last put we need
425 	 * a separate refcount.
426 	 */
427 	struct kref		ref;
428 	struct list_head	entry;
429 	struct mutex		lock;
430 	struct list_head	ctrls;
431 	struct list_head	nsheads;
432 	char			subnqn[NVMF_NQN_SIZE];
433 	char			serial[20];
434 	char			model[40];
435 	char			firmware_rev[8];
436 	u8			cmic;
437 	enum nvme_subsys_type	subtype;
438 	u16			vendor_id;
439 	u16			awupf;	/* 0's based awupf value. */
440 	struct ida		ns_ida;
441 #ifdef CONFIG_NVME_MULTIPATH
442 	enum nvme_iopolicy	iopolicy;
443 #endif
444 };
445 
446 /*
447  * Container structure for uniqueue namespace identifiers.
448  */
449 struct nvme_ns_ids {
450 	u8	eui64[8];
451 	u8	nguid[16];
452 	uuid_t	uuid;
453 	u8	csi;
454 };
455 
456 /*
457  * Anchor structure for namespaces.  There is one for each namespace in a
458  * NVMe subsystem that any of our controllers can see, and the namespace
459  * structure for each controller is chained of it.  For private namespaces
460  * there is a 1:1 relation to our namespace structures, that is ->list
461  * only ever has a single entry for private namespaces.
462  */
463 struct nvme_ns_head {
464 	struct list_head	list;
465 	struct srcu_struct      srcu;
466 	struct nvme_subsystem	*subsys;
467 	unsigned		ns_id;
468 	struct nvme_ns_ids	ids;
469 	struct list_head	entry;
470 	struct kref		ref;
471 	bool			shared;
472 	int			instance;
473 	struct nvme_effects_log *effects;
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 work_struct	partition_scan_work;
484 	struct mutex		lock;
485 	unsigned long		flags;
486 #define NVME_NSHEAD_DISK_LIVE	0
487 	struct nvme_ns __rcu	*current_path[];
488 #endif
489 };
490 
491 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
492 {
493 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
494 }
495 
496 enum nvme_ns_features {
497 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
498 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
499 	NVME_NS_DEAC = 1 << 2,		/* DEAC bit in Write Zeores supported */
500 };
501 
502 struct nvme_ns {
503 	struct list_head list;
504 
505 	struct nvme_ctrl *ctrl;
506 	struct request_queue *queue;
507 	struct gendisk *disk;
508 #ifdef CONFIG_NVME_MULTIPATH
509 	enum nvme_ana_state ana_state;
510 	u32 ana_grpid;
511 #endif
512 	struct list_head siblings;
513 	struct kref kref;
514 	struct nvme_ns_head *head;
515 
516 	int lba_shift;
517 	u16 ms;
518 	u16 pi_size;
519 	u16 sgs;
520 	u32 sws;
521 	u8 pi_type;
522 	u8 guard_type;
523 #ifdef CONFIG_BLK_DEV_ZONED
524 	u64 zsze;
525 #endif
526 	unsigned long features;
527 	unsigned long flags;
528 #define NVME_NS_REMOVING	0
529 #define NVME_NS_ANA_PENDING	2
530 #define NVME_NS_FORCE_RO	3
531 #define NVME_NS_READY		4
532 
533 	struct cdev		cdev;
534 	struct device		cdev_device;
535 
536 	struct nvme_fault_inject fault_inject;
537 
538 };
539 
540 /* NVMe ns supports metadata actions by the controller (generate/strip) */
541 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
542 {
543 	return ns->pi_type && ns->ms == ns->pi_size;
544 }
545 
546 struct nvme_ctrl_ops {
547 	const char *name;
548 	struct module *module;
549 	unsigned int flags;
550 #define NVME_F_FABRICS			(1 << 0)
551 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
552 #define NVME_F_BLOCKING			(1 << 2)
553 
554 	const struct attribute_group **dev_attr_groups;
555 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
556 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
557 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
558 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
559 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
560 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
561 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
562 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
563 	void (*print_device_info)(struct nvme_ctrl *ctrl);
564 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
565 };
566 
567 /*
568  * nvme command_id is constructed as such:
569  * | xxxx | xxxxxxxxxxxx |
570  *   gen    request tag
571  */
572 #define nvme_genctr_mask(gen)			(gen & 0xf)
573 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
574 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
575 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
576 
577 static inline u16 nvme_cid(struct request *rq)
578 {
579 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
580 }
581 
582 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
583 		u16 command_id)
584 {
585 	u8 genctr = nvme_genctr_from_cid(command_id);
586 	u16 tag = nvme_tag_from_cid(command_id);
587 	struct request *rq;
588 
589 	rq = blk_mq_tag_to_rq(tags, tag);
590 	if (unlikely(!rq)) {
591 		pr_err("could not locate request for tag %#x\n",
592 			tag);
593 		return NULL;
594 	}
595 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
596 		dev_err(nvme_req(rq)->ctrl->device,
597 			"request %#x genctr mismatch (got %#x expected %#x)\n",
598 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
599 		return NULL;
600 	}
601 	return rq;
602 }
603 
604 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
605                 u16 command_id)
606 {
607 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
608 }
609 
610 /*
611  * Return the length of the string without the space padding
612  */
613 static inline int nvme_strlen(char *s, int len)
614 {
615 	while (s[len - 1] == ' ')
616 		len--;
617 	return len;
618 }
619 
620 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
621 {
622 	struct nvme_subsystem *subsys = ctrl->subsys;
623 
624 	if (ctrl->ops->print_device_info) {
625 		ctrl->ops->print_device_info(ctrl);
626 		return;
627 	}
628 
629 	dev_err(ctrl->device,
630 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
631 		nvme_strlen(subsys->model, sizeof(subsys->model)),
632 		subsys->model, nvme_strlen(subsys->firmware_rev,
633 					   sizeof(subsys->firmware_rev)),
634 		subsys->firmware_rev);
635 }
636 
637 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
638 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
639 			    const char *dev_name);
640 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
641 void nvme_should_fail(struct request *req);
642 #else
643 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
644 					  const char *dev_name)
645 {
646 }
647 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
648 {
649 }
650 static inline void nvme_should_fail(struct request *req) {}
651 #endif
652 
653 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
654 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
655 
656 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
657 {
658 	int ret;
659 
660 	if (!ctrl->subsystem)
661 		return -ENOTTY;
662 	if (!nvme_wait_reset(ctrl))
663 		return -EBUSY;
664 
665 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
666 	if (ret)
667 		return ret;
668 
669 	return nvme_try_sched_reset(ctrl);
670 }
671 
672 /*
673  * Convert a 512B sector number to a device logical block number.
674  */
675 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
676 {
677 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
678 }
679 
680 /*
681  * Convert a device logical block number to a 512B sector number.
682  */
683 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
684 {
685 	return lba << (ns->lba_shift - SECTOR_SHIFT);
686 }
687 
688 /*
689  * Convert byte length to nvme's 0-based num dwords
690  */
691 static inline u32 nvme_bytes_to_numd(size_t len)
692 {
693 	return (len >> 2) - 1;
694 }
695 
696 static inline bool nvme_is_ana_error(u16 status)
697 {
698 	switch (status & 0x7ff) {
699 	case NVME_SC_ANA_TRANSITION:
700 	case NVME_SC_ANA_INACCESSIBLE:
701 	case NVME_SC_ANA_PERSISTENT_LOSS:
702 		return true;
703 	default:
704 		return false;
705 	}
706 }
707 
708 static inline bool nvme_is_path_error(u16 status)
709 {
710 	/* check for a status code type of 'path related status' */
711 	return (status & 0x700) == 0x300;
712 }
713 
714 /*
715  * Fill in the status and result information from the CQE, and then figure out
716  * if blk-mq will need to use IPI magic to complete the request, and if yes do
717  * so.  If not let the caller complete the request without an indirect function
718  * call.
719  */
720 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
721 		union nvme_result result)
722 {
723 	struct nvme_request *rq = nvme_req(req);
724 	struct nvme_ctrl *ctrl = rq->ctrl;
725 
726 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
727 		rq->genctr++;
728 
729 	rq->status = le16_to_cpu(status) >> 1;
730 	rq->result = result;
731 	/* inject error when permitted by fault injection framework */
732 	nvme_should_fail(req);
733 	if (unlikely(blk_should_fake_timeout(req->q)))
734 		return true;
735 	return blk_mq_complete_request_remote(req);
736 }
737 
738 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
739 {
740 	get_device(ctrl->device);
741 }
742 
743 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
744 {
745 	put_device(ctrl->device);
746 }
747 
748 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
749 {
750 	return !qid &&
751 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
752 }
753 
754 /*
755  * Returns true for sink states that can't ever transition back to live.
756  */
757 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
758 {
759 	switch (nvme_ctrl_state(ctrl)) {
760 	case NVME_CTRL_NEW:
761 	case NVME_CTRL_LIVE:
762 	case NVME_CTRL_RESETTING:
763 	case NVME_CTRL_CONNECTING:
764 		return false;
765 	case NVME_CTRL_DELETING:
766 	case NVME_CTRL_DELETING_NOIO:
767 	case NVME_CTRL_DEAD:
768 		return true;
769 	default:
770 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
771 		return true;
772 	}
773 }
774 
775 void nvme_end_req(struct request *req);
776 void nvme_complete_rq(struct request *req);
777 void nvme_complete_batch_req(struct request *req);
778 
779 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
780 						void (*fn)(struct request *rq))
781 {
782 	struct request *req;
783 
784 	rq_list_for_each(&iob->req_list, req) {
785 		fn(req);
786 		nvme_complete_batch_req(req);
787 	}
788 	blk_mq_end_request_batch(iob);
789 }
790 
791 blk_status_t nvme_host_path_error(struct request *req);
792 bool nvme_cancel_request(struct request *req, void *data);
793 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
794 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
795 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
796 		enum nvme_ctrl_state new_state);
797 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
798 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
799 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
800 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
801 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
802 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
803 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
804 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
805 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
806 		const struct blk_mq_ops *ops, unsigned int cmd_size);
807 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
808 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
809 		const struct blk_mq_ops *ops, unsigned int nr_maps,
810 		unsigned int cmd_size);
811 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
812 
813 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
814 
815 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
816 		volatile union nvme_result *res);
817 
818 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
819 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
820 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
821 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
822 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
823 void nvme_sync_queues(struct nvme_ctrl *ctrl);
824 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
825 void nvme_unfreeze(struct nvme_ctrl *ctrl);
826 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
827 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
828 void nvme_start_freeze(struct nvme_ctrl *ctrl);
829 
830 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
831 {
832 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
833 }
834 
835 #define NVME_QID_ANY -1
836 void nvme_init_request(struct request *req, struct nvme_command *cmd);
837 void nvme_cleanup_cmd(struct request *req);
838 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
839 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
840 		struct request *req);
841 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
842 		bool queue_live);
843 
844 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
845 		bool queue_live)
846 {
847 	if (likely(ctrl->state == NVME_CTRL_LIVE))
848 		return true;
849 	if (ctrl->ops->flags & NVME_F_FABRICS &&
850 	    ctrl->state == NVME_CTRL_DELETING)
851 		return queue_live;
852 	return __nvme_check_ready(ctrl, rq, queue_live);
853 }
854 
855 /*
856  * NSID shall be unique for all shared namespaces, or if at least one of the
857  * following conditions is met:
858  *   1. Namespace Management is supported by the controller
859  *   2. ANA is supported by the controller
860  *   3. NVM Set are supported by the controller
861  *
862  * In other case, private namespace are not required to report a unique NSID.
863  */
864 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
865 		struct nvme_ns_head *head)
866 {
867 	return head->shared ||
868 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
869 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
870 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
871 }
872 
873 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
874 		void *buf, unsigned bufflen);
875 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
876 		union nvme_result *result, void *buffer, unsigned bufflen,
877 		int qid, int at_head,
878 		blk_mq_req_flags_t flags);
879 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
880 		      unsigned int dword11, void *buffer, size_t buflen,
881 		      u32 *result);
882 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
883 		      unsigned int dword11, void *buffer, size_t buflen,
884 		      u32 *result);
885 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
886 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
887 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
888 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
889 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
890 void nvme_queue_scan(struct nvme_ctrl *ctrl);
891 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
892 		void *log, size_t size, u64 offset);
893 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
894 void nvme_put_ns_head(struct nvme_ns_head *head);
895 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
896 		const struct file_operations *fops, struct module *owner);
897 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
898 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
899 		unsigned int cmd, unsigned long arg);
900 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
901 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
902 		unsigned int cmd, unsigned long arg);
903 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
904 		unsigned long arg);
905 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
906 		unsigned long arg);
907 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
908 		struct io_comp_batch *iob, unsigned int poll_flags);
909 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
910 		unsigned int issue_flags);
911 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
912 		unsigned int issue_flags);
913 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
914 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
915 
916 extern const struct attribute_group *nvme_ns_id_attr_groups[];
917 extern const struct pr_ops nvme_pr_ops;
918 extern const struct block_device_operations nvme_ns_head_ops;
919 extern const struct attribute_group nvme_dev_attrs_group;
920 extern const struct attribute_group *nvme_subsys_attrs_groups[];
921 extern const struct attribute_group *nvme_dev_attr_groups[];
922 extern const struct block_device_operations nvme_bdev_ops;
923 
924 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
925 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
926 #ifdef CONFIG_NVME_MULTIPATH
927 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
928 {
929 	return ctrl->ana_log_buf != NULL;
930 }
931 
932 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
933 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
934 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
935 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
936 void nvme_failover_req(struct request *req);
937 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
938 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
939 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
940 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
941 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
942 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
943 void nvme_mpath_update(struct nvme_ctrl *ctrl);
944 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
945 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
946 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
947 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
948 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
949 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
950 void nvme_mpath_start_request(struct request *rq);
951 void nvme_mpath_end_request(struct request *rq);
952 
953 static inline void nvme_trace_bio_complete(struct request *req)
954 {
955 	struct nvme_ns *ns = req->q->queuedata;
956 
957 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
958 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
959 }
960 
961 extern bool multipath;
962 extern struct device_attribute dev_attr_ana_grpid;
963 extern struct device_attribute dev_attr_ana_state;
964 extern struct device_attribute subsys_attr_iopolicy;
965 
966 #else
967 #define multipath false
968 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
969 {
970 	return false;
971 }
972 static inline void nvme_failover_req(struct request *req)
973 {
974 }
975 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
976 {
977 }
978 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
979 		struct nvme_ns_head *head)
980 {
981 	return 0;
982 }
983 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
984 {
985 }
986 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
987 {
988 }
989 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
990 {
991 	return false;
992 }
993 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
994 {
995 }
996 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
997 {
998 }
999 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
1000 {
1001 }
1002 static inline void nvme_trace_bio_complete(struct request *req)
1003 {
1004 }
1005 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1006 {
1007 }
1008 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1009 		struct nvme_id_ctrl *id)
1010 {
1011 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1012 		dev_warn(ctrl->device,
1013 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1014 	return 0;
1015 }
1016 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1017 {
1018 }
1019 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1020 {
1021 }
1022 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1023 {
1024 }
1025 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1026 {
1027 }
1028 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1029 {
1030 }
1031 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1032 {
1033 }
1034 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1035 {
1036 }
1037 static inline void nvme_mpath_start_request(struct request *rq)
1038 {
1039 }
1040 static inline void nvme_mpath_end_request(struct request *rq)
1041 {
1042 }
1043 #endif /* CONFIG_NVME_MULTIPATH */
1044 
1045 int nvme_revalidate_zones(struct nvme_ns *ns);
1046 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1047 		unsigned int nr_zones, report_zones_cb cb, void *data);
1048 #ifdef CONFIG_BLK_DEV_ZONED
1049 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1050 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1051 				       struct nvme_command *cmnd,
1052 				       enum nvme_zone_mgmt_action action);
1053 #else
1054 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1055 		struct request *req, struct nvme_command *cmnd,
1056 		enum nvme_zone_mgmt_action action)
1057 {
1058 	return BLK_STS_NOTSUPP;
1059 }
1060 
1061 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1062 {
1063 	dev_warn(ns->ctrl->device,
1064 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1065 	return -EPROTONOSUPPORT;
1066 }
1067 #endif
1068 
1069 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1070 {
1071 	return dev_to_disk(dev)->private_data;
1072 }
1073 
1074 #ifdef CONFIG_NVME_HWMON
1075 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1076 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1077 #else
1078 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1079 {
1080 	return 0;
1081 }
1082 
1083 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1084 {
1085 }
1086 #endif
1087 
1088 static inline void nvme_start_request(struct request *rq)
1089 {
1090 	if (rq->cmd_flags & REQ_NVME_MPATH)
1091 		nvme_mpath_start_request(rq);
1092 	blk_mq_start_request(rq);
1093 }
1094 
1095 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1096 {
1097 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1098 }
1099 
1100 #ifdef CONFIG_NVME_AUTH
1101 int __init nvme_init_auth(void);
1102 void __exit nvme_exit_auth(void);
1103 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1104 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1105 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1106 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1107 void nvme_auth_free(struct nvme_ctrl *ctrl);
1108 #else
1109 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1110 {
1111 	return 0;
1112 }
1113 static inline int __init nvme_init_auth(void)
1114 {
1115 	return 0;
1116 }
1117 static inline void __exit nvme_exit_auth(void)
1118 {
1119 }
1120 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1121 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1122 {
1123 	return -EPROTONOSUPPORT;
1124 }
1125 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1126 {
1127 	return NVME_SC_AUTH_REQUIRED;
1128 }
1129 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1130 #endif
1131 
1132 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1133 			 u8 opcode);
1134 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1135 int nvme_execute_rq(struct request *rq, bool at_head);
1136 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1137 		       struct nvme_command *cmd, int status);
1138 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1139 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1140 bool nvme_get_ns(struct nvme_ns *ns);
1141 void nvme_put_ns(struct nvme_ns *ns);
1142 
1143 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1144 {
1145 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1146 }
1147 
1148 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1149 const unsigned char *nvme_get_error_status_str(u16 status);
1150 const unsigned char *nvme_get_opcode_str(u8 opcode);
1151 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1152 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1153 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1154 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1155 {
1156 	return "I/O Error";
1157 }
1158 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1159 {
1160 	return "I/O Cmd";
1161 }
1162 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1163 {
1164 	return "Admin Cmd";
1165 }
1166 
1167 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1168 {
1169 	return "Fabrics Cmd";
1170 }
1171 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1172 
1173 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1174 {
1175 	if (opcode == nvme_fabrics_command)
1176 		return nvme_get_fabrics_opcode_str(fctype);
1177 	return qid ? nvme_get_opcode_str(opcode) :
1178 		nvme_get_admin_opcode_str(opcode);
1179 }
1180 #endif /* _NVME_H */
1181