xref: /linux/drivers/nvme/host/nvme.h (revision 122333d6bd229af279cdb35d1b874b71b3b9ccfb)
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 unsigned int nvme_io_timeout;
23 #define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
24 
25 extern unsigned int admin_timeout;
26 #define NVME_ADMIN_TIMEOUT	(admin_timeout * HZ)
27 
28 #define NVME_DEFAULT_KATO	5
29 
30 #ifdef CONFIG_ARCH_NO_SG_CHAIN
31 #define  NVME_INLINE_SG_CNT  0
32 #define  NVME_INLINE_METADATA_SG_CNT  0
33 #else
34 #define  NVME_INLINE_SG_CNT  2
35 #define  NVME_INLINE_METADATA_SG_CNT  1
36 #endif
37 
38 /*
39  * Default to a 4K page size, with the intention to update this
40  * path in the future to accommodate architectures with differing
41  * kernel and IO page sizes.
42  */
43 #define NVME_CTRL_PAGE_SHIFT	12
44 #define NVME_CTRL_PAGE_SIZE	(1 << NVME_CTRL_PAGE_SHIFT)
45 
46 extern struct workqueue_struct *nvme_wq;
47 extern struct workqueue_struct *nvme_reset_wq;
48 extern struct workqueue_struct *nvme_delete_wq;
49 
50 /*
51  * List of workarounds for devices that required behavior not specified in
52  * the standard.
53  */
54 enum nvme_quirks {
55 	/*
56 	 * Prefers I/O aligned to a stripe size specified in a vendor
57 	 * specific Identify field.
58 	 */
59 	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
60 
61 	/*
62 	 * The controller doesn't handle Identify value others than 0 or 1
63 	 * correctly.
64 	 */
65 	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
66 
67 	/*
68 	 * The controller deterministically returns O's on reads to
69 	 * logical blocks that deallocate was called on.
70 	 */
71 	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
72 
73 	/*
74 	 * The controller needs a delay before starts checking the device
75 	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
76 	 */
77 	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
78 
79 	/*
80 	 * APST should not be used.
81 	 */
82 	NVME_QUIRK_NO_APST			= (1 << 4),
83 
84 	/*
85 	 * The deepest sleep state should not be used.
86 	 */
87 	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
88 
89 	/*
90 	 * Set MEDIUM priority on SQ creation
91 	 */
92 	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
93 
94 	/*
95 	 * Ignore device provided subnqn.
96 	 */
97 	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
98 
99 	/*
100 	 * Broken Write Zeroes.
101 	 */
102 	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
103 
104 	/*
105 	 * Force simple suspend/resume path.
106 	 */
107 	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
108 
109 	/*
110 	 * Use only one interrupt vector for all queues
111 	 */
112 	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
113 
114 	/*
115 	 * Use non-standard 128 bytes SQEs.
116 	 */
117 	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
118 
119 	/*
120 	 * Prevent tag overlap between queues
121 	 */
122 	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
123 
124 	/*
125 	 * Don't change the value of the temperature threshold feature
126 	 */
127 	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
128 
129 	/*
130 	 * The controller doesn't handle the Identify Namespace
131 	 * Identification Descriptor list subcommand despite claiming
132 	 * NVMe 1.3 compliance.
133 	 */
134 	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
135 
136 	/*
137 	 * The controller does not properly handle DMA addresses over
138 	 * 48 bits.
139 	 */
140 	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
141 
142 	/*
143 	 * The controller requires the command_id value be limited, so skip
144 	 * encoding the generation sequence number.
145 	 */
146 	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
147 
148 	/*
149 	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
150 	 */
151 	NVME_QUIRK_BOGUS_NID			= (1 << 18),
152 
153 	/*
154 	 * No temperature thresholds for channels other than 0 (Composite).
155 	 */
156 	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
157 };
158 
159 /*
160  * Common request structure for NVMe passthrough.  All drivers must have
161  * this structure as the first member of their request-private data.
162  */
163 struct nvme_request {
164 	struct nvme_command	*cmd;
165 	union nvme_result	result;
166 	u8			genctr;
167 	u8			retries;
168 	u8			flags;
169 	u16			status;
170 #ifdef CONFIG_NVME_MULTIPATH
171 	unsigned long		start_time;
172 #endif
173 	struct nvme_ctrl	*ctrl;
174 };
175 
176 /*
177  * Mark a bio as coming in through the mpath node.
178  */
179 #define REQ_NVME_MPATH		REQ_DRV
180 
181 enum {
182 	NVME_REQ_CANCELLED		= (1 << 0),
183 	NVME_REQ_USERCMD		= (1 << 1),
184 	NVME_MPATH_IO_STATS		= (1 << 2),
185 };
186 
187 static inline struct nvme_request *nvme_req(struct request *req)
188 {
189 	return blk_mq_rq_to_pdu(req);
190 }
191 
192 static inline u16 nvme_req_qid(struct request *req)
193 {
194 	if (!req->q->queuedata)
195 		return 0;
196 
197 	return req->mq_hctx->queue_num + 1;
198 }
199 
200 /* The below value is the specific amount of delay needed before checking
201  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
202  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
203  * found empirically.
204  */
205 #define NVME_QUIRK_DELAY_AMOUNT		2300
206 
207 /*
208  * enum nvme_ctrl_state: Controller state
209  *
210  * @NVME_CTRL_NEW:		New controller just allocated, initial state
211  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
212  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
213  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
214  *				transport
215  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
216  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
217  *				disabled/failed immediately. This state comes
218  * 				after all async event processing took place and
219  * 				before ns removal and the controller deletion
220  * 				progress
221  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
222  *				shutdown or removal. In this case we forcibly
223  *				kill all inflight I/O as they have no chance to
224  *				complete
225  */
226 enum nvme_ctrl_state {
227 	NVME_CTRL_NEW,
228 	NVME_CTRL_LIVE,
229 	NVME_CTRL_RESETTING,
230 	NVME_CTRL_CONNECTING,
231 	NVME_CTRL_DELETING,
232 	NVME_CTRL_DELETING_NOIO,
233 	NVME_CTRL_DEAD,
234 };
235 
236 struct nvme_fault_inject {
237 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
238 	struct fault_attr attr;
239 	struct dentry *parent;
240 	bool dont_retry;	/* DNR, do not retry */
241 	u16 status;		/* status code */
242 #endif
243 };
244 
245 enum nvme_ctrl_flags {
246 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
247 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
248 	NVME_CTRL_STARTED_ONCE		= 2,
249 	NVME_CTRL_STOPPED		= 3,
250 };
251 
252 struct nvme_ctrl {
253 	bool comp_seen;
254 	enum nvme_ctrl_state state;
255 	bool identified;
256 	spinlock_t lock;
257 	struct mutex scan_lock;
258 	const struct nvme_ctrl_ops *ops;
259 	struct request_queue *admin_q;
260 	struct request_queue *connect_q;
261 	struct request_queue *fabrics_q;
262 	struct device *dev;
263 	int instance;
264 	int numa_node;
265 	struct blk_mq_tag_set *tagset;
266 	struct blk_mq_tag_set *admin_tagset;
267 	struct list_head namespaces;
268 	struct rw_semaphore namespaces_rwsem;
269 	struct device ctrl_device;
270 	struct device *device;	/* char device */
271 #ifdef CONFIG_NVME_HWMON
272 	struct device *hwmon_device;
273 #endif
274 	struct cdev cdev;
275 	struct work_struct reset_work;
276 	struct work_struct delete_work;
277 	wait_queue_head_t state_wq;
278 
279 	struct nvme_subsystem *subsys;
280 	struct list_head subsys_entry;
281 
282 	struct opal_dev *opal_dev;
283 
284 	char name[12];
285 	u16 cntlid;
286 
287 	u32 ctrl_config;
288 	u16 mtfa;
289 	u32 queue_count;
290 
291 	u64 cap;
292 	u32 max_hw_sectors;
293 	u32 max_segments;
294 	u32 max_integrity_segments;
295 	u32 max_discard_sectors;
296 	u32 max_discard_segments;
297 	u32 max_zeroes_sectors;
298 #ifdef CONFIG_BLK_DEV_ZONED
299 	u32 max_zone_append;
300 #endif
301 	u16 crdt[3];
302 	u16 oncs;
303 	u32 dmrsl;
304 	u16 oacs;
305 	u16 sqsize;
306 	u32 max_namespaces;
307 	atomic_t abort_limit;
308 	u8 vwc;
309 	u32 vs;
310 	u32 sgls;
311 	u16 kas;
312 	u8 npss;
313 	u8 apsta;
314 	u16 wctemp;
315 	u16 cctemp;
316 	u32 oaes;
317 	u32 aen_result;
318 	u32 ctratt;
319 	unsigned int shutdown_timeout;
320 	unsigned int kato;
321 	bool subsystem;
322 	unsigned long quirks;
323 	struct nvme_id_power_state psd[32];
324 	struct nvme_effects_log *effects;
325 	struct xarray cels;
326 	struct work_struct scan_work;
327 	struct work_struct async_event_work;
328 	struct delayed_work ka_work;
329 	struct delayed_work failfast_work;
330 	struct nvme_command ka_cmd;
331 	struct work_struct fw_act_work;
332 	unsigned long events;
333 
334 #ifdef CONFIG_NVME_MULTIPATH
335 	/* asymmetric namespace access: */
336 	u8 anacap;
337 	u8 anatt;
338 	u32 anagrpmax;
339 	u32 nanagrpid;
340 	struct mutex ana_lock;
341 	struct nvme_ana_rsp_hdr *ana_log_buf;
342 	size_t ana_log_size;
343 	struct timer_list anatt_timer;
344 	struct work_struct ana_work;
345 #endif
346 
347 #ifdef CONFIG_NVME_AUTH
348 	struct work_struct dhchap_auth_work;
349 	struct mutex dhchap_auth_mutex;
350 	struct nvme_dhchap_queue_context *dhchap_ctxs;
351 	struct nvme_dhchap_key *host_key;
352 	struct nvme_dhchap_key *ctrl_key;
353 	u16 transaction;
354 #endif
355 
356 	/* Power saving configuration */
357 	u64 ps_max_latency_us;
358 	bool apst_enabled;
359 
360 	/* PCIe only: */
361 	u32 hmpre;
362 	u32 hmmin;
363 	u32 hmminds;
364 	u16 hmmaxd;
365 
366 	/* Fabrics only */
367 	u32 ioccsz;
368 	u32 iorcsz;
369 	u16 icdoff;
370 	u16 maxcmd;
371 	int nr_reconnects;
372 	unsigned long flags;
373 	struct nvmf_ctrl_options *opts;
374 
375 	struct page *discard_page;
376 	unsigned long discard_page_busy;
377 
378 	struct nvme_fault_inject fault_inject;
379 
380 	enum nvme_ctrl_type cntrltype;
381 	enum nvme_dctype dctype;
382 };
383 
384 enum nvme_iopolicy {
385 	NVME_IOPOLICY_NUMA,
386 	NVME_IOPOLICY_RR,
387 };
388 
389 struct nvme_subsystem {
390 	int			instance;
391 	struct device		dev;
392 	/*
393 	 * Because we unregister the device on the last put we need
394 	 * a separate refcount.
395 	 */
396 	struct kref		ref;
397 	struct list_head	entry;
398 	struct mutex		lock;
399 	struct list_head	ctrls;
400 	struct list_head	nsheads;
401 	char			subnqn[NVMF_NQN_SIZE];
402 	char			serial[20];
403 	char			model[40];
404 	char			firmware_rev[8];
405 	u8			cmic;
406 	enum nvme_subsys_type	subtype;
407 	u16			vendor_id;
408 	u16			awupf;	/* 0's based awupf value. */
409 	struct ida		ns_ida;
410 #ifdef CONFIG_NVME_MULTIPATH
411 	enum nvme_iopolicy	iopolicy;
412 #endif
413 };
414 
415 /*
416  * Container structure for uniqueue namespace identifiers.
417  */
418 struct nvme_ns_ids {
419 	u8	eui64[8];
420 	u8	nguid[16];
421 	uuid_t	uuid;
422 	u8	csi;
423 };
424 
425 /*
426  * Anchor structure for namespaces.  There is one for each namespace in a
427  * NVMe subsystem that any of our controllers can see, and the namespace
428  * structure for each controller is chained of it.  For private namespaces
429  * there is a 1:1 relation to our namespace structures, that is ->list
430  * only ever has a single entry for private namespaces.
431  */
432 struct nvme_ns_head {
433 	struct list_head	list;
434 	struct srcu_struct      srcu;
435 	struct nvme_subsystem	*subsys;
436 	unsigned		ns_id;
437 	struct nvme_ns_ids	ids;
438 	struct list_head	entry;
439 	struct kref		ref;
440 	bool			shared;
441 	int			instance;
442 	struct nvme_effects_log *effects;
443 
444 	struct cdev		cdev;
445 	struct device		cdev_device;
446 
447 	struct gendisk		*disk;
448 #ifdef CONFIG_NVME_MULTIPATH
449 	struct bio_list		requeue_list;
450 	spinlock_t		requeue_lock;
451 	struct work_struct	requeue_work;
452 	struct mutex		lock;
453 	unsigned long		flags;
454 #define NVME_NSHEAD_DISK_LIVE	0
455 	struct nvme_ns __rcu	*current_path[];
456 #endif
457 };
458 
459 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
460 {
461 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
462 }
463 
464 enum nvme_ns_features {
465 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
466 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
467 	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
468 };
469 
470 struct nvme_ns {
471 	struct list_head list;
472 
473 	struct nvme_ctrl *ctrl;
474 	struct request_queue *queue;
475 	struct gendisk *disk;
476 #ifdef CONFIG_NVME_MULTIPATH
477 	enum nvme_ana_state ana_state;
478 	u32 ana_grpid;
479 #endif
480 	struct list_head siblings;
481 	struct kref kref;
482 	struct nvme_ns_head *head;
483 
484 	int lba_shift;
485 	u16 ms;
486 	u16 pi_size;
487 	u16 sgs;
488 	u32 sws;
489 	u8 pi_type;
490 	u8 guard_type;
491 #ifdef CONFIG_BLK_DEV_ZONED
492 	u64 zsze;
493 #endif
494 	unsigned long features;
495 	unsigned long flags;
496 #define NVME_NS_REMOVING	0
497 #define NVME_NS_ANA_PENDING	2
498 #define NVME_NS_FORCE_RO	3
499 #define NVME_NS_READY		4
500 
501 	struct cdev		cdev;
502 	struct device		cdev_device;
503 
504 	struct nvme_fault_inject fault_inject;
505 
506 };
507 
508 /* NVMe ns supports metadata actions by the controller (generate/strip) */
509 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
510 {
511 	return ns->pi_type && ns->ms == ns->pi_size;
512 }
513 
514 struct nvme_ctrl_ops {
515 	const char *name;
516 	struct module *module;
517 	unsigned int flags;
518 #define NVME_F_FABRICS			(1 << 0)
519 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
520 #define NVME_F_BLOCKING			(1 << 2)
521 
522 	const struct attribute_group **dev_attr_groups;
523 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
524 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
525 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
526 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
527 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
528 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
529 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
530 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
531 	void (*print_device_info)(struct nvme_ctrl *ctrl);
532 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
533 };
534 
535 /*
536  * nvme command_id is constructed as such:
537  * | xxxx | xxxxxxxxxxxx |
538  *   gen    request tag
539  */
540 #define nvme_genctr_mask(gen)			(gen & 0xf)
541 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
542 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
543 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
544 
545 static inline u16 nvme_cid(struct request *rq)
546 {
547 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
548 }
549 
550 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
551 		u16 command_id)
552 {
553 	u8 genctr = nvme_genctr_from_cid(command_id);
554 	u16 tag = nvme_tag_from_cid(command_id);
555 	struct request *rq;
556 
557 	rq = blk_mq_tag_to_rq(tags, tag);
558 	if (unlikely(!rq)) {
559 		pr_err("could not locate request for tag %#x\n",
560 			tag);
561 		return NULL;
562 	}
563 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
564 		dev_err(nvme_req(rq)->ctrl->device,
565 			"request %#x genctr mismatch (got %#x expected %#x)\n",
566 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
567 		return NULL;
568 	}
569 	return rq;
570 }
571 
572 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
573                 u16 command_id)
574 {
575 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
576 }
577 
578 /*
579  * Return the length of the string without the space padding
580  */
581 static inline int nvme_strlen(char *s, int len)
582 {
583 	while (s[len - 1] == ' ')
584 		len--;
585 	return len;
586 }
587 
588 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
589 {
590 	struct nvme_subsystem *subsys = ctrl->subsys;
591 
592 	if (ctrl->ops->print_device_info) {
593 		ctrl->ops->print_device_info(ctrl);
594 		return;
595 	}
596 
597 	dev_err(ctrl->device,
598 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
599 		nvme_strlen(subsys->model, sizeof(subsys->model)),
600 		subsys->model, nvme_strlen(subsys->firmware_rev,
601 					   sizeof(subsys->firmware_rev)),
602 		subsys->firmware_rev);
603 }
604 
605 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
606 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
607 			    const char *dev_name);
608 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
609 void nvme_should_fail(struct request *req);
610 #else
611 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
612 					  const char *dev_name)
613 {
614 }
615 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
616 {
617 }
618 static inline void nvme_should_fail(struct request *req) {}
619 #endif
620 
621 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
622 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
623 
624 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
625 {
626 	int ret;
627 
628 	if (!ctrl->subsystem)
629 		return -ENOTTY;
630 	if (!nvme_wait_reset(ctrl))
631 		return -EBUSY;
632 
633 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
634 	if (ret)
635 		return ret;
636 
637 	return nvme_try_sched_reset(ctrl);
638 }
639 
640 /*
641  * Convert a 512B sector number to a device logical block number.
642  */
643 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
644 {
645 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
646 }
647 
648 /*
649  * Convert a device logical block number to a 512B sector number.
650  */
651 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
652 {
653 	return lba << (ns->lba_shift - SECTOR_SHIFT);
654 }
655 
656 /*
657  * Convert byte length to nvme's 0-based num dwords
658  */
659 static inline u32 nvme_bytes_to_numd(size_t len)
660 {
661 	return (len >> 2) - 1;
662 }
663 
664 static inline bool nvme_is_ana_error(u16 status)
665 {
666 	switch (status & 0x7ff) {
667 	case NVME_SC_ANA_TRANSITION:
668 	case NVME_SC_ANA_INACCESSIBLE:
669 	case NVME_SC_ANA_PERSISTENT_LOSS:
670 		return true;
671 	default:
672 		return false;
673 	}
674 }
675 
676 static inline bool nvme_is_path_error(u16 status)
677 {
678 	/* check for a status code type of 'path related status' */
679 	return (status & 0x700) == 0x300;
680 }
681 
682 /*
683  * Fill in the status and result information from the CQE, and then figure out
684  * if blk-mq will need to use IPI magic to complete the request, and if yes do
685  * so.  If not let the caller complete the request without an indirect function
686  * call.
687  */
688 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
689 		union nvme_result result)
690 {
691 	struct nvme_request *rq = nvme_req(req);
692 	struct nvme_ctrl *ctrl = rq->ctrl;
693 
694 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
695 		rq->genctr++;
696 
697 	rq->status = le16_to_cpu(status) >> 1;
698 	rq->result = result;
699 	/* inject error when permitted by fault injection framework */
700 	nvme_should_fail(req);
701 	if (unlikely(blk_should_fake_timeout(req->q)))
702 		return true;
703 	return blk_mq_complete_request_remote(req);
704 }
705 
706 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
707 {
708 	get_device(ctrl->device);
709 }
710 
711 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
712 {
713 	put_device(ctrl->device);
714 }
715 
716 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
717 {
718 	return !qid &&
719 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
720 }
721 
722 void nvme_complete_rq(struct request *req);
723 void nvme_complete_batch_req(struct request *req);
724 
725 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
726 						void (*fn)(struct request *rq))
727 {
728 	struct request *req;
729 
730 	rq_list_for_each(&iob->req_list, req) {
731 		fn(req);
732 		nvme_complete_batch_req(req);
733 	}
734 	blk_mq_end_request_batch(iob);
735 }
736 
737 blk_status_t nvme_host_path_error(struct request *req);
738 bool nvme_cancel_request(struct request *req, void *data);
739 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
740 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
741 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
742 		enum nvme_ctrl_state new_state);
743 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
744 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
745 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
746 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
747 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
748 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
749 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
750 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
751 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
752 		const struct blk_mq_ops *ops, unsigned int cmd_size);
753 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
754 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
755 		const struct blk_mq_ops *ops, unsigned int nr_maps,
756 		unsigned int cmd_size);
757 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
758 
759 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
760 
761 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
762 		volatile union nvme_result *res);
763 
764 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
765 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
766 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
767 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
768 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
769 void nvme_sync_queues(struct nvme_ctrl *ctrl);
770 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
771 void nvme_unfreeze(struct nvme_ctrl *ctrl);
772 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
773 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
774 void nvme_start_freeze(struct nvme_ctrl *ctrl);
775 
776 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
777 {
778 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
779 }
780 
781 #define NVME_QID_ANY -1
782 void nvme_init_request(struct request *req, struct nvme_command *cmd);
783 void nvme_cleanup_cmd(struct request *req);
784 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
785 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
786 		struct request *req);
787 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
788 		bool queue_live);
789 
790 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
791 		bool queue_live)
792 {
793 	if (likely(ctrl->state == NVME_CTRL_LIVE))
794 		return true;
795 	if (ctrl->ops->flags & NVME_F_FABRICS &&
796 	    ctrl->state == NVME_CTRL_DELETING)
797 		return queue_live;
798 	return __nvme_check_ready(ctrl, rq, queue_live);
799 }
800 
801 /*
802  * NSID shall be unique for all shared namespaces, or if at least one of the
803  * following conditions is met:
804  *   1. Namespace Management is supported by the controller
805  *   2. ANA is supported by the controller
806  *   3. NVM Set are supported by the controller
807  *
808  * In other case, private namespace are not required to report a unique NSID.
809  */
810 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
811 		struct nvme_ns_head *head)
812 {
813 	return head->shared ||
814 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
815 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
816 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
817 }
818 
819 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
820 		void *buf, unsigned bufflen);
821 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
822 		union nvme_result *result, void *buffer, unsigned bufflen,
823 		int qid, int at_head,
824 		blk_mq_req_flags_t flags);
825 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
826 		      unsigned int dword11, void *buffer, size_t buflen,
827 		      u32 *result);
828 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
829 		      unsigned int dword11, void *buffer, size_t buflen,
830 		      u32 *result);
831 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
832 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
833 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
834 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
835 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
836 void nvme_queue_scan(struct nvme_ctrl *ctrl);
837 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
838 		void *log, size_t size, u64 offset);
839 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
840 void nvme_put_ns_head(struct nvme_ns_head *head);
841 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
842 		const struct file_operations *fops, struct module *owner);
843 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
844 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
845 		unsigned int cmd, unsigned long arg);
846 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
847 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
848 		unsigned int cmd, unsigned long arg);
849 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
850 		unsigned long arg);
851 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
852 		unsigned long arg);
853 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
854 		struct io_comp_batch *iob, unsigned int poll_flags);
855 int nvme_ns_head_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
856 		struct io_comp_batch *iob, unsigned int poll_flags);
857 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
858 		unsigned int issue_flags);
859 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
860 		unsigned int issue_flags);
861 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
862 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
863 
864 extern const struct attribute_group *nvme_ns_id_attr_groups[];
865 extern const struct pr_ops nvme_pr_ops;
866 extern const struct block_device_operations nvme_ns_head_ops;
867 extern const struct attribute_group nvme_dev_attrs_group;
868 
869 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
870 #ifdef CONFIG_NVME_MULTIPATH
871 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
872 {
873 	return ctrl->ana_log_buf != NULL;
874 }
875 
876 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
877 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
878 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
879 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
880 void nvme_failover_req(struct request *req);
881 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
882 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
883 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
884 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
885 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
886 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
887 void nvme_mpath_update(struct nvme_ctrl *ctrl);
888 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
889 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
890 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
891 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
892 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
893 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
894 void nvme_mpath_start_request(struct request *rq);
895 void nvme_mpath_end_request(struct request *rq);
896 
897 static inline void nvme_trace_bio_complete(struct request *req)
898 {
899 	struct nvme_ns *ns = req->q->queuedata;
900 
901 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
902 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
903 }
904 
905 extern bool multipath;
906 extern struct device_attribute dev_attr_ana_grpid;
907 extern struct device_attribute dev_attr_ana_state;
908 extern struct device_attribute subsys_attr_iopolicy;
909 
910 #else
911 #define multipath false
912 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
913 {
914 	return false;
915 }
916 static inline void nvme_failover_req(struct request *req)
917 {
918 }
919 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
920 {
921 }
922 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
923 		struct nvme_ns_head *head)
924 {
925 	return 0;
926 }
927 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
928 {
929 }
930 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
931 {
932 }
933 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
934 {
935 	return false;
936 }
937 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
938 {
939 }
940 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
941 {
942 }
943 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
944 {
945 }
946 static inline void nvme_trace_bio_complete(struct request *req)
947 {
948 }
949 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
950 {
951 }
952 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
953 		struct nvme_id_ctrl *id)
954 {
955 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
956 		dev_warn(ctrl->device,
957 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
958 	return 0;
959 }
960 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
961 {
962 }
963 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
964 {
965 }
966 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
967 {
968 }
969 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
970 {
971 }
972 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
973 {
974 }
975 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
976 {
977 }
978 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
979 {
980 }
981 static inline void nvme_mpath_start_request(struct request *rq)
982 {
983 }
984 static inline void nvme_mpath_end_request(struct request *rq)
985 {
986 }
987 #endif /* CONFIG_NVME_MULTIPATH */
988 
989 int nvme_revalidate_zones(struct nvme_ns *ns);
990 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
991 		unsigned int nr_zones, report_zones_cb cb, void *data);
992 #ifdef CONFIG_BLK_DEV_ZONED
993 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
994 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
995 				       struct nvme_command *cmnd,
996 				       enum nvme_zone_mgmt_action action);
997 #else
998 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
999 		struct request *req, struct nvme_command *cmnd,
1000 		enum nvme_zone_mgmt_action action)
1001 {
1002 	return BLK_STS_NOTSUPP;
1003 }
1004 
1005 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1006 {
1007 	dev_warn(ns->ctrl->device,
1008 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1009 	return -EPROTONOSUPPORT;
1010 }
1011 #endif
1012 
1013 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1014 {
1015 	return dev_to_disk(dev)->private_data;
1016 }
1017 
1018 #ifdef CONFIG_NVME_HWMON
1019 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1020 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1021 #else
1022 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1023 {
1024 	return 0;
1025 }
1026 
1027 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1028 {
1029 }
1030 #endif
1031 
1032 static inline void nvme_start_request(struct request *rq)
1033 {
1034 	if (rq->cmd_flags & REQ_NVME_MPATH)
1035 		nvme_mpath_start_request(rq);
1036 	blk_mq_start_request(rq);
1037 }
1038 
1039 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1040 {
1041 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1042 }
1043 
1044 #ifdef CONFIG_NVME_AUTH
1045 int __init nvme_init_auth(void);
1046 void __exit nvme_exit_auth(void);
1047 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1048 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1049 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1050 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1051 void nvme_auth_free(struct nvme_ctrl *ctrl);
1052 #else
1053 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1054 {
1055 	return 0;
1056 }
1057 static inline int __init nvme_init_auth(void)
1058 {
1059 	return 0;
1060 }
1061 static inline void __exit nvme_exit_auth(void)
1062 {
1063 }
1064 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1065 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1066 {
1067 	return -EPROTONOSUPPORT;
1068 }
1069 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1070 {
1071 	return NVME_SC_AUTH_REQUIRED;
1072 }
1073 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1074 #endif
1075 
1076 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1077 			 u8 opcode);
1078 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1079 int nvme_execute_rq(struct request *rq, bool at_head);
1080 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1081 		       struct nvme_command *cmd, int status);
1082 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1083 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1084 void nvme_put_ns(struct nvme_ns *ns);
1085 
1086 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1087 {
1088 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1089 }
1090 
1091 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1092 const unsigned char *nvme_get_error_status_str(u16 status);
1093 const unsigned char *nvme_get_opcode_str(u8 opcode);
1094 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1095 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1096 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1097 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1098 {
1099 	return "I/O Error";
1100 }
1101 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1102 {
1103 	return "I/O Cmd";
1104 }
1105 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1106 {
1107 	return "Admin Cmd";
1108 }
1109 
1110 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1111 {
1112 	return "Fabrics Cmd";
1113 }
1114 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1115 
1116 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1117 {
1118 	if (opcode == nvme_fabrics_command)
1119 		return nvme_get_fabrics_opcode_str(fctype);
1120 	return qid ? nvme_get_opcode_str(opcode) :
1121 		nvme_get_admin_opcode_str(opcode);
1122 }
1123 #endif /* _NVME_H */
1124