xref: /kernel/linux/linux-5.10/drivers/nvme/host/nvme.h (revision 8c2ecf20)
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/lightnvm.h>
15#include <linux/sed-opal.h>
16#include <linux/fault-inject.h>
17#include <linux/rcupdate.h>
18#include <linux/wait.h>
19#include <linux/t10-pi.h>
20
21#include <trace/events/block.h>
22
23extern unsigned int nvme_io_timeout;
24#define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
25
26extern unsigned int admin_timeout;
27#define ADMIN_TIMEOUT	(admin_timeout * HZ)
28
29#define NVME_DEFAULT_KATO	5
30#define NVME_KATO_GRACE		10
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
48extern struct workqueue_struct *nvme_wq;
49extern struct workqueue_struct *nvme_reset_wq;
50extern struct workqueue_struct *nvme_delete_wq;
51
52enum {
53	NVME_NS_LBA		= 0,
54	NVME_NS_LIGHTNVM	= 1,
55};
56
57/*
58 * List of workarounds for devices that required behavior not specified in
59 * the standard.
60 */
61enum nvme_quirks {
62	/*
63	 * Prefers I/O aligned to a stripe size specified in a vendor
64	 * specific Identify field.
65	 */
66	NVME_QUIRK_STRIPE_SIZE			= (1 << 0),
67
68	/*
69	 * The controller doesn't handle Identify value others than 0 or 1
70	 * correctly.
71	 */
72	NVME_QUIRK_IDENTIFY_CNS			= (1 << 1),
73
74	/*
75	 * The controller deterministically returns O's on reads to
76	 * logical blocks that deallocate was called on.
77	 */
78	NVME_QUIRK_DEALLOCATE_ZEROES		= (1 << 2),
79
80	/*
81	 * The controller needs a delay before starts checking the device
82	 * readiness, which is done by reading the NVME_CSTS_RDY bit.
83	 */
84	NVME_QUIRK_DELAY_BEFORE_CHK_RDY		= (1 << 3),
85
86	/*
87	 * APST should not be used.
88	 */
89	NVME_QUIRK_NO_APST			= (1 << 4),
90
91	/*
92	 * The deepest sleep state should not be used.
93	 */
94	NVME_QUIRK_NO_DEEPEST_PS		= (1 << 5),
95
96	/*
97	 * Supports the LighNVM command set if indicated in vs[1].
98	 */
99	NVME_QUIRK_LIGHTNVM			= (1 << 6),
100
101	/*
102	 * Set MEDIUM priority on SQ creation
103	 */
104	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
105
106	/*
107	 * Ignore device provided subnqn.
108	 */
109	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
110
111	/*
112	 * Broken Write Zeroes.
113	 */
114	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
115
116	/*
117	 * Force simple suspend/resume path.
118	 */
119	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
120
121	/*
122	 * Use only one interrupt vector for all queues
123	 */
124	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
125
126	/*
127	 * Use non-standard 128 bytes SQEs.
128	 */
129	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
130
131	/*
132	 * Prevent tag overlap between queues
133	 */
134	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
135
136	/*
137	 * Don't change the value of the temperature threshold feature
138	 */
139	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
140
141	/*
142	 * The controller doesn't handle the Identify Namespace
143	 * Identification Descriptor list subcommand despite claiming
144	 * NVMe 1.3 compliance.
145	 */
146	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
147
148	/*
149	 * The controller requires the command_id value be be limited, so skip
150	 * encoding the generation sequence number.
151	 */
152	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
153
154	/*
155	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
156	 */
157	NVME_QUIRK_BOGUS_NID			= (1 << 18),
158};
159
160/*
161 * Common request structure for NVMe passthrough.  All drivers must have
162 * this structure as the first member of their request-private data.
163 */
164struct nvme_request {
165	struct nvme_command	*cmd;
166	union nvme_result	result;
167	u8			genctr;
168	u8			retries;
169	u8			flags;
170	u16			status;
171	struct nvme_ctrl	*ctrl;
172};
173
174/*
175 * Mark a bio as coming in through the mpath node.
176 */
177#define REQ_NVME_MPATH		REQ_DRV
178
179enum {
180	NVME_REQ_CANCELLED		= (1 << 0),
181	NVME_REQ_USERCMD		= (1 << 1),
182};
183
184static inline struct nvme_request *nvme_req(struct request *req)
185{
186	return blk_mq_rq_to_pdu(req);
187}
188
189static inline u16 nvme_req_qid(struct request *req)
190{
191	if (!req->q->queuedata)
192		return 0;
193	return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
194}
195
196/* The below value is the specific amount of delay needed before checking
197 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
198 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
199 * found empirically.
200 */
201#define NVME_QUIRK_DELAY_AMOUNT		2300
202
203/*
204 * enum nvme_ctrl_state: Controller state
205 *
206 * @NVME_CTRL_NEW:		New controller just allocated, initial state
207 * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
208 * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
209 * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
210 *				transport
211 * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
212 * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
213 *				disabled/failed immediately. This state comes
214 * 				after all async event processing took place and
215 * 				before ns removal and the controller deletion
216 * 				progress
217 * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
218 *				shutdown or removal. In this case we forcibly
219 *				kill all inflight I/O as they have no chance to
220 *				complete
221 */
222enum nvme_ctrl_state {
223	NVME_CTRL_NEW,
224	NVME_CTRL_LIVE,
225	NVME_CTRL_RESETTING,
226	NVME_CTRL_CONNECTING,
227	NVME_CTRL_DELETING,
228	NVME_CTRL_DELETING_NOIO,
229	NVME_CTRL_DEAD,
230};
231
232struct nvme_fault_inject {
233#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
234	struct fault_attr attr;
235	struct dentry *parent;
236	bool dont_retry;	/* DNR, do not retry */
237	u16 status;		/* status code */
238#endif
239};
240
241struct nvme_ctrl {
242	bool comp_seen;
243	enum nvme_ctrl_state state;
244	bool identified;
245	spinlock_t lock;
246	struct mutex scan_lock;
247	const struct nvme_ctrl_ops *ops;
248	struct request_queue *admin_q;
249	struct request_queue *connect_q;
250	struct request_queue *fabrics_q;
251	struct device *dev;
252	int instance;
253	int numa_node;
254	struct blk_mq_tag_set *tagset;
255	struct blk_mq_tag_set *admin_tagset;
256	struct list_head namespaces;
257	struct rw_semaphore namespaces_rwsem;
258	struct device ctrl_device;
259	struct device *device;	/* char device */
260#ifdef CONFIG_NVME_HWMON
261	struct device *hwmon_device;
262#endif
263	struct cdev cdev;
264	struct work_struct reset_work;
265	struct work_struct delete_work;
266	wait_queue_head_t state_wq;
267
268	struct nvme_subsystem *subsys;
269	struct list_head subsys_entry;
270
271	struct opal_dev *opal_dev;
272
273	char name[12];
274	u16 cntlid;
275
276	u32 ctrl_config;
277	u16 mtfa;
278	u32 queue_count;
279
280	u64 cap;
281	u32 max_hw_sectors;
282	u32 max_segments;
283	u32 max_integrity_segments;
284#ifdef CONFIG_BLK_DEV_ZONED
285	u32 max_zone_append;
286#endif
287	u16 crdt[3];
288	u16 oncs;
289	u16 oacs;
290	u16 nssa;
291	u16 nr_streams;
292	u16 sqsize;
293	u32 max_namespaces;
294	atomic_t abort_limit;
295	u8 vwc;
296	u32 vs;
297	u32 sgls;
298	u16 kas;
299	u8 npss;
300	u8 apsta;
301	u16 wctemp;
302	u16 cctemp;
303	u32 oaes;
304	u32 aen_result;
305	u32 ctratt;
306	unsigned int shutdown_timeout;
307	unsigned int kato;
308	bool subsystem;
309	unsigned long quirks;
310	struct nvme_id_power_state psd[32];
311	struct nvme_effects_log *effects;
312	struct xarray cels;
313	struct work_struct scan_work;
314	struct work_struct async_event_work;
315	struct delayed_work ka_work;
316	struct nvme_command ka_cmd;
317	struct work_struct fw_act_work;
318	unsigned long events;
319
320#ifdef CONFIG_NVME_MULTIPATH
321	/* asymmetric namespace access: */
322	u8 anacap;
323	u8 anatt;
324	u32 anagrpmax;
325	u32 nanagrpid;
326	struct mutex ana_lock;
327	struct nvme_ana_rsp_hdr *ana_log_buf;
328	size_t ana_log_size;
329	struct timer_list anatt_timer;
330	struct work_struct ana_work;
331#endif
332
333	/* Power saving configuration */
334	u64 ps_max_latency_us;
335	bool apst_enabled;
336
337	/* PCIe only: */
338	u32 hmpre;
339	u32 hmmin;
340	u32 hmminds;
341	u16 hmmaxd;
342
343	/* Fabrics only */
344	u32 ioccsz;
345	u32 iorcsz;
346	u16 icdoff;
347	u16 maxcmd;
348	int nr_reconnects;
349	unsigned long flags;
350#define NVME_CTRL_ADMIN_Q_STOPPED	0
351	struct nvmf_ctrl_options *opts;
352
353	struct page *discard_page;
354	unsigned long discard_page_busy;
355
356	struct nvme_fault_inject fault_inject;
357};
358
359static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
360{
361	return READ_ONCE(ctrl->state);
362}
363
364enum nvme_iopolicy {
365	NVME_IOPOLICY_NUMA,
366	NVME_IOPOLICY_RR,
367};
368
369struct nvme_subsystem {
370	int			instance;
371	struct device		dev;
372	/*
373	 * Because we unregister the device on the last put we need
374	 * a separate refcount.
375	 */
376	struct kref		ref;
377	struct list_head	entry;
378	struct mutex		lock;
379	struct list_head	ctrls;
380	struct list_head	nsheads;
381	char			subnqn[NVMF_NQN_SIZE];
382	char			serial[20];
383	char			model[40];
384	char			firmware_rev[8];
385	u8			cmic;
386	u16			vendor_id;
387	u16			awupf;	/* 0's based awupf value. */
388	struct ida		ns_ida;
389#ifdef CONFIG_NVME_MULTIPATH
390	enum nvme_iopolicy	iopolicy;
391#endif
392};
393
394/*
395 * Container structure for uniqueue namespace identifiers.
396 */
397struct nvme_ns_ids {
398	u8	eui64[8];
399	u8	nguid[16];
400	uuid_t	uuid;
401	u8	csi;
402};
403
404/*
405 * Anchor structure for namespaces.  There is one for each namespace in a
406 * NVMe subsystem that any of our controllers can see, and the namespace
407 * structure for each controller is chained of it.  For private namespaces
408 * there is a 1:1 relation to our namespace structures, that is ->list
409 * only ever has a single entry for private namespaces.
410 */
411struct nvme_ns_head {
412	struct list_head	list;
413	struct srcu_struct      srcu;
414	struct nvme_subsystem	*subsys;
415	unsigned		ns_id;
416	struct nvme_ns_ids	ids;
417	struct list_head	entry;
418	struct kref		ref;
419	bool			shared;
420	int			instance;
421	struct nvme_effects_log *effects;
422#ifdef CONFIG_NVME_MULTIPATH
423	struct gendisk		*disk;
424	struct bio_list		requeue_list;
425	spinlock_t		requeue_lock;
426	struct work_struct	requeue_work;
427	struct mutex		lock;
428	unsigned long		flags;
429#define NVME_NSHEAD_DISK_LIVE	0
430	struct nvme_ns __rcu	*current_path[];
431#endif
432};
433
434enum nvme_ns_features {
435	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
436	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
437};
438
439struct nvme_ns {
440	struct list_head list;
441
442	struct nvme_ctrl *ctrl;
443	struct request_queue *queue;
444	struct gendisk *disk;
445#ifdef CONFIG_NVME_MULTIPATH
446	enum nvme_ana_state ana_state;
447	u32 ana_grpid;
448#endif
449	struct list_head siblings;
450	struct nvm_dev *ndev;
451	struct kref kref;
452	struct nvme_ns_head *head;
453
454	int lba_shift;
455	u16 ms;
456	u16 sgs;
457	u32 sws;
458	u8 pi_type;
459#ifdef CONFIG_BLK_DEV_ZONED
460	u64 zsze;
461#endif
462	unsigned long features;
463	unsigned long flags;
464#define NVME_NS_REMOVING	0
465#define NVME_NS_DEAD     	1
466#define NVME_NS_ANA_PENDING	2
467#define NVME_NS_STOPPED		3
468
469	struct nvme_fault_inject fault_inject;
470
471};
472
473/* NVMe ns supports metadata actions by the controller (generate/strip) */
474static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
475{
476	return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
477}
478
479struct nvme_ctrl_ops {
480	const char *name;
481	struct module *module;
482	unsigned int flags;
483#define NVME_F_FABRICS			(1 << 0)
484#define NVME_F_METADATA_SUPPORTED	(1 << 1)
485#define NVME_F_PCI_P2PDMA		(1 << 2)
486	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
487	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
488	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
489	void (*free_ctrl)(struct nvme_ctrl *ctrl);
490	void (*submit_async_event)(struct nvme_ctrl *ctrl);
491	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
492	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
493	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
494};
495
496/*
497 * nvme command_id is constructed as such:
498 * | xxxx | xxxxxxxxxxxx |
499 *   gen    request tag
500 */
501#define nvme_genctr_mask(gen)			(gen & 0xf)
502#define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
503#define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
504#define nvme_tag_from_cid(cid)			(cid & 0xfff)
505
506static inline u16 nvme_cid(struct request *rq)
507{
508	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
509}
510
511static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
512		u16 command_id)
513{
514	u8 genctr = nvme_genctr_from_cid(command_id);
515	u16 tag = nvme_tag_from_cid(command_id);
516	struct request *rq;
517
518	rq = blk_mq_tag_to_rq(tags, tag);
519	if (unlikely(!rq)) {
520		pr_err("could not locate request for tag %#x\n",
521			tag);
522		return NULL;
523	}
524	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
525		dev_err(nvme_req(rq)->ctrl->device,
526			"request %#x genctr mismatch (got %#x expected %#x)\n",
527			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
528		return NULL;
529	}
530	return rq;
531}
532
533static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
534                u16 command_id)
535{
536	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
537}
538
539#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
540void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
541			    const char *dev_name);
542void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
543void nvme_should_fail(struct request *req);
544#else
545static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
546					  const char *dev_name)
547{
548}
549static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
550{
551}
552static inline void nvme_should_fail(struct request *req) {}
553#endif
554
555bool nvme_wait_reset(struct nvme_ctrl *ctrl);
556int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
557
558static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
559{
560	int ret;
561
562	if (!ctrl->subsystem)
563		return -ENOTTY;
564	if (!nvme_wait_reset(ctrl))
565		return -EBUSY;
566
567	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
568	if (ret)
569		return ret;
570
571	return nvme_try_sched_reset(ctrl);
572}
573
574/*
575 * Convert a 512B sector number to a device logical block number.
576 */
577static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
578{
579	return sector >> (ns->lba_shift - SECTOR_SHIFT);
580}
581
582/*
583 * Convert a device logical block number to a 512B sector number.
584 */
585static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
586{
587	return lba << (ns->lba_shift - SECTOR_SHIFT);
588}
589
590/*
591 * Convert byte length to nvme's 0-based num dwords
592 */
593static inline u32 nvme_bytes_to_numd(size_t len)
594{
595	return (len >> 2) - 1;
596}
597
598static inline bool nvme_is_ana_error(u16 status)
599{
600	switch (status & 0x7ff) {
601	case NVME_SC_ANA_TRANSITION:
602	case NVME_SC_ANA_INACCESSIBLE:
603	case NVME_SC_ANA_PERSISTENT_LOSS:
604		return true;
605	default:
606		return false;
607	}
608}
609
610static inline bool nvme_is_path_error(u16 status)
611{
612	/* check for a status code type of 'path related status' */
613	return (status & 0x700) == 0x300;
614}
615
616/*
617 * Fill in the status and result information from the CQE, and then figure out
618 * if blk-mq will need to use IPI magic to complete the request, and if yes do
619 * so.  If not let the caller complete the request without an indirect function
620 * call.
621 */
622static inline bool nvme_try_complete_req(struct request *req, __le16 status,
623		union nvme_result result)
624{
625	struct nvme_request *rq = nvme_req(req);
626
627	rq->status = le16_to_cpu(status) >> 1;
628	rq->result = result;
629	/* inject error when permitted by fault injection framework */
630	nvme_should_fail(req);
631	if (unlikely(blk_should_fake_timeout(req->q)))
632		return true;
633	return blk_mq_complete_request_remote(req);
634}
635
636static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
637{
638	get_device(ctrl->device);
639}
640
641static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
642{
643	put_device(ctrl->device);
644}
645
646static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
647{
648	return !qid &&
649		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
650}
651
652void nvme_complete_rq(struct request *req);
653bool nvme_cancel_request(struct request *req, void *data, bool reserved);
654void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
655void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
656bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
657		enum nvme_ctrl_state new_state);
658int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
659int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
660int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
661int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
662		const struct nvme_ctrl_ops *ops, unsigned long quirks);
663void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
664void nvme_start_ctrl(struct nvme_ctrl *ctrl);
665void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
666int nvme_init_identify(struct nvme_ctrl *ctrl);
667
668void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
669
670int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
671		bool send);
672
673void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
674		volatile union nvme_result *res);
675
676void nvme_stop_queues(struct nvme_ctrl *ctrl);
677void nvme_start_queues(struct nvme_ctrl *ctrl);
678void nvme_stop_admin_queue(struct nvme_ctrl *ctrl);
679void nvme_start_admin_queue(struct nvme_ctrl *ctrl);
680void nvme_kill_queues(struct nvme_ctrl *ctrl);
681void nvme_sync_queues(struct nvme_ctrl *ctrl);
682void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
683void nvme_unfreeze(struct nvme_ctrl *ctrl);
684void nvme_wait_freeze(struct nvme_ctrl *ctrl);
685int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
686void nvme_start_freeze(struct nvme_ctrl *ctrl);
687
688#define NVME_QID_ANY -1
689struct request *nvme_alloc_request(struct request_queue *q,
690		struct nvme_command *cmd, blk_mq_req_flags_t flags);
691struct request *nvme_alloc_request_qid(struct request_queue *q,
692		struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
693void nvme_cleanup_cmd(struct request *req);
694blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
695		struct nvme_command *cmd);
696int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
697		void *buf, unsigned bufflen);
698int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
699		union nvme_result *result, void *buffer, unsigned bufflen,
700		unsigned timeout, int qid, int at_head,
701		blk_mq_req_flags_t flags, bool poll);
702int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
703		      unsigned int dword11, void *buffer, size_t buflen,
704		      u32 *result);
705int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
706		      unsigned int dword11, void *buffer, size_t buflen,
707		      u32 *result);
708int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
709void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
710int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
711int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
712int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
713
714int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
715		void *log, size_t size, u64 offset);
716struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
717		struct nvme_ns_head **head, int *srcu_idx);
718void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx);
719
720extern const struct attribute_group *nvme_ns_id_attr_groups[];
721extern const struct block_device_operations nvme_ns_head_ops;
722
723#ifdef CONFIG_NVME_MULTIPATH
724static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
725{
726	return ctrl->ana_log_buf != NULL;
727}
728
729void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
730void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
731void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
732void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
733			struct nvme_ctrl *ctrl, int *flags);
734void nvme_failover_req(struct request *req);
735void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
736int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
737void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
738void nvme_mpath_remove_disk(struct nvme_ns_head *head);
739int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
740void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
741void nvme_mpath_update(struct nvme_ctrl *ctrl);
742void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
743void nvme_mpath_stop(struct nvme_ctrl *ctrl);
744bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
745void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
746struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
747blk_qc_t nvme_ns_head_submit_bio(struct bio *bio);
748
749static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
750{
751	struct nvme_ns_head *head = ns->head;
752
753	if (head->disk && list_empty(&head->list))
754		kblockd_schedule_work(&head->requeue_work);
755}
756
757static inline void nvme_trace_bio_complete(struct request *req,
758        blk_status_t status)
759{
760	struct nvme_ns *ns = req->q->queuedata;
761
762	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
763		trace_block_bio_complete(ns->head->disk->queue, req->bio);
764}
765
766extern struct device_attribute dev_attr_ana_grpid;
767extern struct device_attribute dev_attr_ana_state;
768extern struct device_attribute subsys_attr_iopolicy;
769
770#else
771static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
772{
773	return false;
774}
775/*
776 * Without the multipath code enabled, multiple controller per subsystems are
777 * visible as devices and thus we cannot use the subsystem instance.
778 */
779static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
780				      struct nvme_ctrl *ctrl, int *flags)
781{
782	sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
783}
784
785static inline void nvme_failover_req(struct request *req)
786{
787}
788static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
789{
790}
791static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
792		struct nvme_ns_head *head)
793{
794	return 0;
795}
796static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
797		struct nvme_id_ns *id)
798{
799}
800static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
801{
802}
803static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
804{
805	return false;
806}
807static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
808{
809}
810static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
811{
812}
813static inline void nvme_trace_bio_complete(struct request *req,
814        blk_status_t status)
815{
816}
817static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
818{
819}
820static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
821		struct nvme_id_ctrl *id)
822{
823	if (ctrl->subsys->cmic & (1 << 3))
824		dev_warn(ctrl->device,
825"Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
826	return 0;
827}
828static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
829{
830}
831static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
832{
833}
834static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
835{
836}
837static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
838{
839}
840static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
841{
842}
843static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
844{
845}
846#endif /* CONFIG_NVME_MULTIPATH */
847
848int nvme_revalidate_zones(struct nvme_ns *ns);
849#ifdef CONFIG_BLK_DEV_ZONED
850int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
851int nvme_report_zones(struct gendisk *disk, sector_t sector,
852		      unsigned int nr_zones, report_zones_cb cb, void *data);
853
854blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
855				       struct nvme_command *cmnd,
856				       enum nvme_zone_mgmt_action action);
857#else
858#define nvme_report_zones NULL
859
860static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
861		struct request *req, struct nvme_command *cmnd,
862		enum nvme_zone_mgmt_action action)
863{
864	return BLK_STS_NOTSUPP;
865}
866
867static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
868{
869	dev_warn(ns->ctrl->device,
870		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
871	return -EPROTONOSUPPORT;
872}
873#endif
874
875#ifdef CONFIG_NVM
876int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
877void nvme_nvm_unregister(struct nvme_ns *ns);
878extern const struct attribute_group nvme_nvm_attr_group;
879int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
880#else
881static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
882				    int node)
883{
884	return 0;
885}
886
887static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
888static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
889							unsigned long arg)
890{
891	return -ENOTTY;
892}
893#endif /* CONFIG_NVM */
894
895static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
896{
897	return dev_to_disk(dev)->private_data;
898}
899
900#ifdef CONFIG_NVME_HWMON
901int nvme_hwmon_init(struct nvme_ctrl *ctrl);
902void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
903#else
904static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
905{
906	return 0;
907}
908
909static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
910{
911}
912#endif
913
914u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
915			 u8 opcode);
916void nvme_execute_passthru_rq(struct request *rq);
917struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
918struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
919void nvme_put_ns(struct nvme_ns *ns);
920
921#endif /* _NVME_H */
922