xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 0e6774ec)
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 
52 /*
53  * List of workarounds for devices that required behavior not specified in
54  * the standard.
55  */
56 enum nvme_quirks {
57 	/*
58 	 * Prefers I/O aligned to a stripe size specified in a vendor
59 	 * specific Identify field.
60 	 */
61 	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
62 
63 	/*
64 	 * The controller doesn't handle Identify value others than 0 or 1
65 	 * correctly.
66 	 */
67 	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
68 
69 	/*
70 	 * The controller deterministically returns O's on reads to
71 	 * logical blocks that deallocate was called on.
72 	 */
73 	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
74 
75 	/*
76 	 * The controller needs a delay before starts checking the device
77 	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
78 	 */
79 	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
80 
81 	/*
82 	 * APST should not be used.
83 	 */
84 	NVME_QUIRK_NO_APST			= (1 << 4),
85 
86 	/*
87 	 * The deepest sleep state should not be used.
88 	 */
89 	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
90 
91 	/*
92 	 *  Problems seen with concurrent commands
93 	 */
94 	NVME_QUIRK_QDEPTH_ONE			= (1 << 6),
95 
96 	/*
97 	 * Set MEDIUM priority on SQ creation
98 	 */
99 	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
100 
101 	/*
102 	 * Ignore device provided subnqn.
103 	 */
104 	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
105 
106 	/*
107 	 * Broken Write Zeroes.
108 	 */
109 	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
110 
111 	/*
112 	 * Force simple suspend/resume path.
113 	 */
114 	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
115 
116 	/*
117 	 * Use only one interrupt vector for all queues
118 	 */
119 	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
120 
121 	/*
122 	 * Use non-standard 128 bytes SQEs.
123 	 */
124 	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
125 
126 	/*
127 	 * Prevent tag overlap between queues
128 	 */
129 	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
130 
131 	/*
132 	 * Don't change the value of the temperature threshold feature
133 	 */
134 	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
135 
136 	/*
137 	 * The controller doesn't handle the Identify Namespace
138 	 * Identification Descriptor list subcommand despite claiming
139 	 * NVMe 1.3 compliance.
140 	 */
141 	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
142 
143 	/*
144 	 * The controller does not properly handle DMA addresses over
145 	 * 48 bits.
146 	 */
147 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
148 
149 	/*
150 	 * The controller requires the command_id value be limited, so skip
151 	 * encoding the generation sequence number.
152 	 */
153 	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
154 
155 	/*
156 	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
157 	 */
158 	NVME_QUIRK_BOGUS_NID			= (1 << 18),
159 
160 	/*
161 	 * No temperature thresholds for channels other than 0 (Composite).
162 	 */
163 	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
164 
165 	/*
166 	 * Disables simple suspend/resume path.
167 	 */
168 	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
169 
170 	/*
171 	 * MSI (but not MSI-X) interrupts are broken and never fire.
172 	 */
173 	NVME_QUIRK_BROKEN_MSI			= (1 << 21),
174 };
175 
176 /*
177  * Common request structure for NVMe passthrough.  All drivers must have
178  * this structure as the first member of their request-private data.
179  */
180 struct nvme_request {
181 	struct nvme_command	*cmd;
182 	union nvme_result	result;
183 	u8			genctr;
184 	u8			retries;
185 	u8			flags;
186 	u16			status;
187 #ifdef CONFIG_NVME_MULTIPATH
188 	unsigned long		start_time;
189 #endif
190 	struct nvme_ctrl	*ctrl;
191 };
192 
193 /*
194  * Mark a bio as coming in through the mpath node.
195  */
196 #define REQ_NVME_MPATH		REQ_DRV
197 
198 enum {
199 	NVME_REQ_CANCELLED		= (1 << 0),
200 	NVME_REQ_USERCMD		= (1 << 1),
201 	NVME_MPATH_IO_STATS		= (1 << 2),
202 };
203 
204 static inline struct nvme_request *nvme_req(struct request *req)
205 {
206 	return blk_mq_rq_to_pdu(req);
207 }
208 
209 static inline u16 nvme_req_qid(struct request *req)
210 {
211 	if (!req->q->queuedata)
212 		return 0;
213 
214 	return req->mq_hctx->queue_num + 1;
215 }
216 
217 /* The below value is the specific amount of delay needed before checking
218  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
219  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
220  * found empirically.
221  */
222 #define NVME_QUIRK_DELAY_AMOUNT		2300
223 
224 /*
225  * enum nvme_ctrl_state: Controller state
226  *
227  * @NVME_CTRL_NEW:		New controller just allocated, initial state
228  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
229  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
230  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
231  *				transport
232  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
233  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
234  *				disabled/failed immediately. This state comes
235  * 				after all async event processing took place and
236  * 				before ns removal and the controller deletion
237  * 				progress
238  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
239  *				shutdown or removal. In this case we forcibly
240  *				kill all inflight I/O as they have no chance to
241  *				complete
242  */
243 enum nvme_ctrl_state {
244 	NVME_CTRL_NEW,
245 	NVME_CTRL_LIVE,
246 	NVME_CTRL_RESETTING,
247 	NVME_CTRL_CONNECTING,
248 	NVME_CTRL_DELETING,
249 	NVME_CTRL_DELETING_NOIO,
250 	NVME_CTRL_DEAD,
251 };
252 
253 struct nvme_fault_inject {
254 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
255 	struct fault_attr attr;
256 	struct dentry *parent;
257 	bool dont_retry;	/* DNR, do not retry */
258 	u16 status;		/* status code */
259 #endif
260 };
261 
262 enum nvme_ctrl_flags {
263 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
264 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
265 	NVME_CTRL_STARTED_ONCE		= 2,
266 	NVME_CTRL_STOPPED		= 3,
267 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
268 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
269 	NVME_CTRL_FROZEN		= 6,
270 };
271 
272 struct nvme_ctrl {
273 	bool comp_seen;
274 	bool identified;
275 	enum nvme_ctrl_state state;
276 	spinlock_t lock;
277 	struct mutex scan_lock;
278 	const struct nvme_ctrl_ops *ops;
279 	struct request_queue *admin_q;
280 	struct request_queue *connect_q;
281 	struct request_queue *fabrics_q;
282 	struct device *dev;
283 	int instance;
284 	int numa_node;
285 	struct blk_mq_tag_set *tagset;
286 	struct blk_mq_tag_set *admin_tagset;
287 	struct list_head namespaces;
288 	struct mutex namespaces_lock;
289 	struct srcu_struct srcu;
290 	struct device ctrl_device;
291 	struct device *device;	/* char device */
292 #ifdef CONFIG_NVME_HWMON
293 	struct device *hwmon_device;
294 #endif
295 	struct cdev cdev;
296 	struct work_struct reset_work;
297 	struct work_struct delete_work;
298 	wait_queue_head_t state_wq;
299 
300 	struct nvme_subsystem *subsys;
301 	struct list_head subsys_entry;
302 
303 	struct opal_dev *opal_dev;
304 
305 	char name[12];
306 	u16 cntlid;
307 
308 	u16 mtfa;
309 	u32 ctrl_config;
310 	u32 queue_count;
311 
312 	u64 cap;
313 	u32 max_hw_sectors;
314 	u32 max_segments;
315 	u32 max_integrity_segments;
316 	u32 max_discard_sectors;
317 	u32 max_discard_segments;
318 	u32 max_zeroes_sectors;
319 #ifdef CONFIG_BLK_DEV_ZONED
320 	u32 max_zone_append;
321 #endif
322 	u16 crdt[3];
323 	u16 oncs;
324 	u32 dmrsl;
325 	u16 oacs;
326 	u16 sqsize;
327 	u32 max_namespaces;
328 	atomic_t abort_limit;
329 	u8 vwc;
330 	u32 vs;
331 	u32 sgls;
332 	u16 kas;
333 	u8 npss;
334 	u8 apsta;
335 	u16 wctemp;
336 	u16 cctemp;
337 	u32 oaes;
338 	u32 aen_result;
339 	u32 ctratt;
340 	unsigned int shutdown_timeout;
341 	unsigned int kato;
342 	bool subsystem;
343 	unsigned long quirks;
344 	struct nvme_id_power_state psd[32];
345 	struct nvme_effects_log *effects;
346 	struct xarray cels;
347 	struct work_struct scan_work;
348 	struct work_struct async_event_work;
349 	struct delayed_work ka_work;
350 	struct delayed_work failfast_work;
351 	struct nvme_command ka_cmd;
352 	unsigned long ka_last_check_time;
353 	struct work_struct fw_act_work;
354 	unsigned long events;
355 
356 #ifdef CONFIG_NVME_MULTIPATH
357 	/* asymmetric namespace access: */
358 	u8 anacap;
359 	u8 anatt;
360 	u32 anagrpmax;
361 	u32 nanagrpid;
362 	struct mutex ana_lock;
363 	struct nvme_ana_rsp_hdr *ana_log_buf;
364 	size_t ana_log_size;
365 	struct timer_list anatt_timer;
366 	struct work_struct ana_work;
367 #endif
368 
369 #ifdef CONFIG_NVME_AUTH
370 	struct work_struct dhchap_auth_work;
371 	struct mutex dhchap_auth_mutex;
372 	struct nvme_dhchap_queue_context *dhchap_ctxs;
373 	struct nvme_dhchap_key *host_key;
374 	struct nvme_dhchap_key *ctrl_key;
375 	u16 transaction;
376 #endif
377 
378 	/* Power saving configuration */
379 	u64 ps_max_latency_us;
380 	bool apst_enabled;
381 
382 	/* PCIe only: */
383 	u16 hmmaxd;
384 	u32 hmpre;
385 	u32 hmmin;
386 	u32 hmminds;
387 
388 	/* Fabrics only */
389 	u32 ioccsz;
390 	u32 iorcsz;
391 	u16 icdoff;
392 	u16 maxcmd;
393 	int nr_reconnects;
394 	unsigned long flags;
395 	struct nvmf_ctrl_options *opts;
396 
397 	struct page *discard_page;
398 	unsigned long discard_page_busy;
399 
400 	struct nvme_fault_inject fault_inject;
401 
402 	enum nvme_ctrl_type cntrltype;
403 	enum nvme_dctype dctype;
404 };
405 
406 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
407 {
408 	return READ_ONCE(ctrl->state);
409 }
410 
411 enum nvme_iopolicy {
412 	NVME_IOPOLICY_NUMA,
413 	NVME_IOPOLICY_RR,
414 };
415 
416 struct nvme_subsystem {
417 	int			instance;
418 	struct device		dev;
419 	/*
420 	 * Because we unregister the device on the last put we need
421 	 * a separate refcount.
422 	 */
423 	struct kref		ref;
424 	struct list_head	entry;
425 	struct mutex		lock;
426 	struct list_head	ctrls;
427 	struct list_head	nsheads;
428 	char			subnqn[NVMF_NQN_SIZE];
429 	char			serial[20];
430 	char			model[40];
431 	char			firmware_rev[8];
432 	u8			cmic;
433 	enum nvme_subsys_type	subtype;
434 	u16			vendor_id;
435 	u16			awupf;	/* 0's based awupf value. */
436 	struct ida		ns_ida;
437 #ifdef CONFIG_NVME_MULTIPATH
438 	enum nvme_iopolicy	iopolicy;
439 #endif
440 };
441 
442 /*
443  * Container structure for uniqueue namespace identifiers.
444  */
445 struct nvme_ns_ids {
446 	u8	eui64[8];
447 	u8	nguid[16];
448 	uuid_t	uuid;
449 	u8	csi;
450 };
451 
452 /*
453  * Anchor structure for namespaces.  There is one for each namespace in a
454  * NVMe subsystem that any of our controllers can see, and the namespace
455  * structure for each controller is chained of it.  For private namespaces
456  * there is a 1:1 relation to our namespace structures, that is ->list
457  * only ever has a single entry for private namespaces.
458  */
459 struct nvme_ns_head {
460 	struct list_head	list;
461 	struct srcu_struct      srcu;
462 	struct nvme_subsystem	*subsys;
463 	unsigned		ns_id;
464 	struct nvme_ns_ids	ids;
465 	struct list_head	entry;
466 	struct kref		ref;
467 	bool			shared;
468 	int			instance;
469 	struct nvme_effects_log *effects;
470 
471 	struct cdev		cdev;
472 	struct device		cdev_device;
473 
474 	struct gendisk		*disk;
475 #ifdef CONFIG_NVME_MULTIPATH
476 	struct bio_list		requeue_list;
477 	spinlock_t		requeue_lock;
478 	struct work_struct	requeue_work;
479 	struct mutex		lock;
480 	unsigned long		flags;
481 #define NVME_NSHEAD_DISK_LIVE	0
482 	struct nvme_ns __rcu	*current_path[];
483 #endif
484 };
485 
486 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
487 {
488 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
489 }
490 
491 enum nvme_ns_features {
492 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
493 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
494 	NVME_NS_DEAC = 1 << 2,		/* DEAC bit in Write Zeores supported */
495 };
496 
497 struct nvme_ns {
498 	struct list_head list;
499 
500 	struct nvme_ctrl *ctrl;
501 	struct request_queue *queue;
502 	struct gendisk *disk;
503 #ifdef CONFIG_NVME_MULTIPATH
504 	enum nvme_ana_state ana_state;
505 	u32 ana_grpid;
506 #endif
507 	struct list_head siblings;
508 	struct kref kref;
509 	struct nvme_ns_head *head;
510 
511 	int lba_shift;
512 	u16 ms;
513 	u16 pi_size;
514 	u16 sgs;
515 	u32 sws;
516 	u8 pi_type;
517 	u8 guard_type;
518 #ifdef CONFIG_BLK_DEV_ZONED
519 	u64 zsze;
520 #endif
521 	unsigned long features;
522 	unsigned long flags;
523 #define NVME_NS_REMOVING	0
524 #define NVME_NS_ANA_PENDING	2
525 #define NVME_NS_FORCE_RO	3
526 #define NVME_NS_READY		4
527 
528 	struct cdev		cdev;
529 	struct device		cdev_device;
530 
531 	struct nvme_fault_inject fault_inject;
532 
533 };
534 
535 /* NVMe ns supports metadata actions by the controller (generate/strip) */
536 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
537 {
538 	return ns->pi_type && ns->ms == ns->pi_size;
539 }
540 
541 struct nvme_ctrl_ops {
542 	const char *name;
543 	struct module *module;
544 	unsigned int flags;
545 #define NVME_F_FABRICS			(1 << 0)
546 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
547 #define NVME_F_BLOCKING			(1 << 2)
548 
549 	const struct attribute_group **dev_attr_groups;
550 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
551 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
552 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
553 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
554 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
555 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
556 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
557 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
558 	void (*print_device_info)(struct nvme_ctrl *ctrl);
559 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
560 };
561 
562 /*
563  * nvme command_id is constructed as such:
564  * | xxxx | xxxxxxxxxxxx |
565  *   gen    request tag
566  */
567 #define nvme_genctr_mask(gen)			(gen & 0xf)
568 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
569 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
570 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
571 
572 static inline u16 nvme_cid(struct request *rq)
573 {
574 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
575 }
576 
577 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
578 		u16 command_id)
579 {
580 	u8 genctr = nvme_genctr_from_cid(command_id);
581 	u16 tag = nvme_tag_from_cid(command_id);
582 	struct request *rq;
583 
584 	rq = blk_mq_tag_to_rq(tags, tag);
585 	if (unlikely(!rq)) {
586 		pr_err("could not locate request for tag %#x\n",
587 			tag);
588 		return NULL;
589 	}
590 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
591 		dev_err(nvme_req(rq)->ctrl->device,
592 			"request %#x genctr mismatch (got %#x expected %#x)\n",
593 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
594 		return NULL;
595 	}
596 	return rq;
597 }
598 
599 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
600                 u16 command_id)
601 {
602 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
603 }
604 
605 /*
606  * Return the length of the string without the space padding
607  */
608 static inline int nvme_strlen(char *s, int len)
609 {
610 	while (s[len - 1] == ' ')
611 		len--;
612 	return len;
613 }
614 
615 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
616 {
617 	struct nvme_subsystem *subsys = ctrl->subsys;
618 
619 	if (ctrl->ops->print_device_info) {
620 		ctrl->ops->print_device_info(ctrl);
621 		return;
622 	}
623 
624 	dev_err(ctrl->device,
625 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
626 		nvme_strlen(subsys->model, sizeof(subsys->model)),
627 		subsys->model, nvme_strlen(subsys->firmware_rev,
628 					   sizeof(subsys->firmware_rev)),
629 		subsys->firmware_rev);
630 }
631 
632 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
633 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
634 			    const char *dev_name);
635 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
636 void nvme_should_fail(struct request *req);
637 #else
638 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
639 					  const char *dev_name)
640 {
641 }
642 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
643 {
644 }
645 static inline void nvme_should_fail(struct request *req) {}
646 #endif
647 
648 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
649 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
650 
651 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
652 {
653 	int ret;
654 
655 	if (!ctrl->subsystem)
656 		return -ENOTTY;
657 	if (!nvme_wait_reset(ctrl))
658 		return -EBUSY;
659 
660 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
661 	if (ret)
662 		return ret;
663 
664 	return nvme_try_sched_reset(ctrl);
665 }
666 
667 /*
668  * Convert a 512B sector number to a device logical block number.
669  */
670 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
671 {
672 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
673 }
674 
675 /*
676  * Convert a device logical block number to a 512B sector number.
677  */
678 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
679 {
680 	return lba << (ns->lba_shift - SECTOR_SHIFT);
681 }
682 
683 /*
684  * Convert byte length to nvme's 0-based num dwords
685  */
686 static inline u32 nvme_bytes_to_numd(size_t len)
687 {
688 	return (len >> 2) - 1;
689 }
690 
691 static inline bool nvme_is_ana_error(u16 status)
692 {
693 	switch (status & 0x7ff) {
694 	case NVME_SC_ANA_TRANSITION:
695 	case NVME_SC_ANA_INACCESSIBLE:
696 	case NVME_SC_ANA_PERSISTENT_LOSS:
697 		return true;
698 	default:
699 		return false;
700 	}
701 }
702 
703 static inline bool nvme_is_path_error(u16 status)
704 {
705 	/* check for a status code type of 'path related status' */
706 	return (status & 0x700) == 0x300;
707 }
708 
709 /*
710  * Fill in the status and result information from the CQE, and then figure out
711  * if blk-mq will need to use IPI magic to complete the request, and if yes do
712  * so.  If not let the caller complete the request without an indirect function
713  * call.
714  */
715 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
716 		union nvme_result result)
717 {
718 	struct nvme_request *rq = nvme_req(req);
719 	struct nvme_ctrl *ctrl = rq->ctrl;
720 
721 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
722 		rq->genctr++;
723 
724 	rq->status = le16_to_cpu(status) >> 1;
725 	rq->result = result;
726 	/* inject error when permitted by fault injection framework */
727 	nvme_should_fail(req);
728 	if (unlikely(blk_should_fake_timeout(req->q)))
729 		return true;
730 	return blk_mq_complete_request_remote(req);
731 }
732 
733 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
734 {
735 	get_device(ctrl->device);
736 }
737 
738 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
739 {
740 	put_device(ctrl->device);
741 }
742 
743 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
744 {
745 	return !qid &&
746 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
747 }
748 
749 /*
750  * Returns true for sink states that can't ever transition back to live.
751  */
752 static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
753 {
754 	switch (nvme_ctrl_state(ctrl)) {
755 	case NVME_CTRL_NEW:
756 	case NVME_CTRL_LIVE:
757 	case NVME_CTRL_RESETTING:
758 	case NVME_CTRL_CONNECTING:
759 		return false;
760 	case NVME_CTRL_DELETING:
761 	case NVME_CTRL_DELETING_NOIO:
762 	case NVME_CTRL_DEAD:
763 		return true;
764 	default:
765 		WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
766 		return true;
767 	}
768 }
769 
770 void nvme_end_req(struct request *req);
771 void nvme_complete_rq(struct request *req);
772 void nvme_complete_batch_req(struct request *req);
773 
774 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
775 						void (*fn)(struct request *rq))
776 {
777 	struct request *req;
778 
779 	rq_list_for_each(&iob->req_list, req) {
780 		fn(req);
781 		nvme_complete_batch_req(req);
782 	}
783 	blk_mq_end_request_batch(iob);
784 }
785 
786 blk_status_t nvme_host_path_error(struct request *req);
787 bool nvme_cancel_request(struct request *req, void *data);
788 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
789 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
790 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
791 		enum nvme_ctrl_state new_state);
792 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
793 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
794 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
795 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
796 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
797 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
798 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
799 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
800 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
801 		const struct blk_mq_ops *ops, unsigned int cmd_size);
802 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
803 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
804 		const struct blk_mq_ops *ops, unsigned int nr_maps,
805 		unsigned int cmd_size);
806 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
807 
808 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
809 
810 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
811 		volatile union nvme_result *res);
812 
813 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
814 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
815 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
816 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
817 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
818 void nvme_sync_queues(struct nvme_ctrl *ctrl);
819 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
820 void nvme_unfreeze(struct nvme_ctrl *ctrl);
821 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
822 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
823 void nvme_start_freeze(struct nvme_ctrl *ctrl);
824 
825 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
826 {
827 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
828 }
829 
830 #define NVME_QID_ANY -1
831 void nvme_init_request(struct request *req, struct nvme_command *cmd);
832 void nvme_cleanup_cmd(struct request *req);
833 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
834 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
835 		struct request *req);
836 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
837 		bool queue_live);
838 
839 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
840 		bool queue_live)
841 {
842 	if (likely(ctrl->state == NVME_CTRL_LIVE))
843 		return true;
844 	if (ctrl->ops->flags & NVME_F_FABRICS &&
845 	    ctrl->state == NVME_CTRL_DELETING)
846 		return queue_live;
847 	return __nvme_check_ready(ctrl, rq, queue_live);
848 }
849 
850 /*
851  * NSID shall be unique for all shared namespaces, or if at least one of the
852  * following conditions is met:
853  *   1. Namespace Management is supported by the controller
854  *   2. ANA is supported by the controller
855  *   3. NVM Set are supported by the controller
856  *
857  * In other case, private namespace are not required to report a unique NSID.
858  */
859 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
860 		struct nvme_ns_head *head)
861 {
862 	return head->shared ||
863 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
864 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
865 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
866 }
867 
868 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
869 		void *buf, unsigned bufflen);
870 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
871 		union nvme_result *result, void *buffer, unsigned bufflen,
872 		int qid, int at_head,
873 		blk_mq_req_flags_t flags);
874 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
875 		      unsigned int dword11, void *buffer, size_t buflen,
876 		      u32 *result);
877 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
878 		      unsigned int dword11, void *buffer, size_t buflen,
879 		      u32 *result);
880 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
881 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
882 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
883 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
884 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
885 void nvme_queue_scan(struct nvme_ctrl *ctrl);
886 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
887 		void *log, size_t size, u64 offset);
888 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
889 void nvme_put_ns_head(struct nvme_ns_head *head);
890 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
891 		const struct file_operations *fops, struct module *owner);
892 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
893 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
894 		unsigned int cmd, unsigned long arg);
895 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
896 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
897 		unsigned int cmd, unsigned long arg);
898 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
899 		unsigned long arg);
900 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
901 		unsigned long arg);
902 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
903 		struct io_comp_batch *iob, unsigned int poll_flags);
904 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
905 		unsigned int issue_flags);
906 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
907 		unsigned int issue_flags);
908 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
909 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
910 
911 extern const struct attribute_group *nvme_ns_id_attr_groups[];
912 extern const struct pr_ops nvme_pr_ops;
913 extern const struct block_device_operations nvme_ns_head_ops;
914 extern const struct attribute_group nvme_dev_attrs_group;
915 extern const struct attribute_group *nvme_subsys_attrs_groups[];
916 extern const struct attribute_group *nvme_dev_attr_groups[];
917 extern const struct block_device_operations nvme_bdev_ops;
918 
919 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
920 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
921 #ifdef CONFIG_NVME_MULTIPATH
922 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
923 {
924 	return ctrl->ana_log_buf != NULL;
925 }
926 
927 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
928 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
929 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
930 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
931 void nvme_failover_req(struct request *req);
932 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
933 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
934 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
935 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
936 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
937 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
938 void nvme_mpath_update(struct nvme_ctrl *ctrl);
939 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
940 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
941 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
942 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
943 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
944 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
945 void nvme_mpath_start_request(struct request *rq);
946 void nvme_mpath_end_request(struct request *rq);
947 
948 static inline void nvme_trace_bio_complete(struct request *req)
949 {
950 	struct nvme_ns *ns = req->q->queuedata;
951 
952 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
953 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
954 }
955 
956 extern bool multipath;
957 extern struct device_attribute dev_attr_ana_grpid;
958 extern struct device_attribute dev_attr_ana_state;
959 extern struct device_attribute subsys_attr_iopolicy;
960 
961 #else
962 #define multipath false
963 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
964 {
965 	return false;
966 }
967 static inline void nvme_failover_req(struct request *req)
968 {
969 }
970 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
971 {
972 }
973 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
974 		struct nvme_ns_head *head)
975 {
976 	return 0;
977 }
978 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
979 {
980 }
981 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
982 {
983 }
984 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
985 {
986 	return false;
987 }
988 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
989 {
990 }
991 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
992 {
993 }
994 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
995 {
996 }
997 static inline void nvme_trace_bio_complete(struct request *req)
998 {
999 }
1000 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1001 {
1002 }
1003 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1004 		struct nvme_id_ctrl *id)
1005 {
1006 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1007 		dev_warn(ctrl->device,
1008 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1009 	return 0;
1010 }
1011 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1012 {
1013 }
1014 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1015 {
1016 }
1017 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1018 {
1019 }
1020 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1021 {
1022 }
1023 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1024 {
1025 }
1026 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1027 {
1028 }
1029 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1030 {
1031 }
1032 static inline void nvme_mpath_start_request(struct request *rq)
1033 {
1034 }
1035 static inline void nvme_mpath_end_request(struct request *rq)
1036 {
1037 }
1038 #endif /* CONFIG_NVME_MULTIPATH */
1039 
1040 int nvme_revalidate_zones(struct nvme_ns *ns);
1041 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1042 		unsigned int nr_zones, report_zones_cb cb, void *data);
1043 #ifdef CONFIG_BLK_DEV_ZONED
1044 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1045 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1046 				       struct nvme_command *cmnd,
1047 				       enum nvme_zone_mgmt_action action);
1048 #else
1049 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1050 		struct request *req, struct nvme_command *cmnd,
1051 		enum nvme_zone_mgmt_action action)
1052 {
1053 	return BLK_STS_NOTSUPP;
1054 }
1055 
1056 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1057 {
1058 	dev_warn(ns->ctrl->device,
1059 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1060 	return -EPROTONOSUPPORT;
1061 }
1062 #endif
1063 
1064 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1065 {
1066 	return dev_to_disk(dev)->private_data;
1067 }
1068 
1069 #ifdef CONFIG_NVME_HWMON
1070 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1071 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1072 #else
1073 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1074 {
1075 	return 0;
1076 }
1077 
1078 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1079 {
1080 }
1081 #endif
1082 
1083 static inline void nvme_start_request(struct request *rq)
1084 {
1085 	if (rq->cmd_flags & REQ_NVME_MPATH)
1086 		nvme_mpath_start_request(rq);
1087 	blk_mq_start_request(rq);
1088 }
1089 
1090 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1091 {
1092 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1093 }
1094 
1095 #ifdef CONFIG_NVME_AUTH
1096 int __init nvme_init_auth(void);
1097 void __exit nvme_exit_auth(void);
1098 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1099 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1100 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1101 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1102 void nvme_auth_free(struct nvme_ctrl *ctrl);
1103 #else
1104 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1105 {
1106 	return 0;
1107 }
1108 static inline int __init nvme_init_auth(void)
1109 {
1110 	return 0;
1111 }
1112 static inline void __exit nvme_exit_auth(void)
1113 {
1114 }
1115 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1116 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1117 {
1118 	return -EPROTONOSUPPORT;
1119 }
1120 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1121 {
1122 	return NVME_SC_AUTH_REQUIRED;
1123 }
1124 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1125 #endif
1126 
1127 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1128 			 u8 opcode);
1129 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1130 int nvme_execute_rq(struct request *rq, bool at_head);
1131 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1132 		       struct nvme_command *cmd, int status);
1133 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1134 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1135 bool nvme_get_ns(struct nvme_ns *ns);
1136 void nvme_put_ns(struct nvme_ns *ns);
1137 
1138 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1139 {
1140 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1141 }
1142 
1143 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1144 const unsigned char *nvme_get_error_status_str(u16 status);
1145 const unsigned char *nvme_get_opcode_str(u8 opcode);
1146 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1147 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1148 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1149 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1150 {
1151 	return "I/O Error";
1152 }
1153 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1154 {
1155 	return "I/O Cmd";
1156 }
1157 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1158 {
1159 	return "Admin Cmd";
1160 }
1161 
1162 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1163 {
1164 	return "Fabrics Cmd";
1165 }
1166 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1167 
1168 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1169 {
1170 	if (opcode == nvme_fabrics_command)
1171 		return nvme_get_fabrics_opcode_str(fctype);
1172 	return qid ? nvme_get_opcode_str(opcode) :
1173 		nvme_get_admin_opcode_str(opcode);
1174 }
1175 #endif /* _NVME_H */
1176