xref: /kernel/linux/linux-6.6/drivers/nvme/host/nvme.h (revision 62306a36)
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
22extern const struct pr_ops nvme_pr_ops;
23
24extern unsigned int nvme_io_timeout;
25#define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
26
27extern 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
48extern struct workqueue_struct *nvme_wq;
49extern struct workqueue_struct *nvme_reset_wq;
50extern struct workqueue_struct *nvme_delete_wq;
51
52/*
53 * List of workarounds for devices that required behavior not specified in
54 * the standard.
55 */
56enum 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	 * Set MEDIUM priority on SQ creation
93	 */
94	NVME_QUIRK_MEDIUM_PRIO_SQ		= (1 << 7),
95
96	/*
97	 * Ignore device provided subnqn.
98	 */
99	NVME_QUIRK_IGNORE_DEV_SUBNQN		= (1 << 8),
100
101	/*
102	 * Broken Write Zeroes.
103	 */
104	NVME_QUIRK_DISABLE_WRITE_ZEROES		= (1 << 9),
105
106	/*
107	 * Force simple suspend/resume path.
108	 */
109	NVME_QUIRK_SIMPLE_SUSPEND		= (1 << 10),
110
111	/*
112	 * Use only one interrupt vector for all queues
113	 */
114	NVME_QUIRK_SINGLE_VECTOR		= (1 << 11),
115
116	/*
117	 * Use non-standard 128 bytes SQEs.
118	 */
119	NVME_QUIRK_128_BYTES_SQES		= (1 << 12),
120
121	/*
122	 * Prevent tag overlap between queues
123	 */
124	NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
125
126	/*
127	 * Don't change the value of the temperature threshold feature
128	 */
129	NVME_QUIRK_NO_TEMP_THRESH_CHANGE	= (1 << 14),
130
131	/*
132	 * The controller doesn't handle the Identify Namespace
133	 * Identification Descriptor list subcommand despite claiming
134	 * NVMe 1.3 compliance.
135	 */
136	NVME_QUIRK_NO_NS_DESC_LIST		= (1 << 15),
137
138	/*
139	 * The controller does not properly handle DMA addresses over
140	 * 48 bits.
141	 */
142	NVME_QUIRK_DMA_ADDRESS_BITS_48		= (1 << 16),
143
144	/*
145	 * The controller requires the command_id value be limited, so skip
146	 * encoding the generation sequence number.
147	 */
148	NVME_QUIRK_SKIP_CID_GEN			= (1 << 17),
149
150	/*
151	 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
152	 */
153	NVME_QUIRK_BOGUS_NID			= (1 << 18),
154
155	/*
156	 * No temperature thresholds for channels other than 0 (Composite).
157	 */
158	NVME_QUIRK_NO_SECONDARY_TEMP_THRESH	= (1 << 19),
159
160	/*
161	 * Disables simple suspend/resume path.
162	 */
163	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
164};
165
166/*
167 * Common request structure for NVMe passthrough.  All drivers must have
168 * this structure as the first member of their request-private data.
169 */
170struct nvme_request {
171	struct nvme_command	*cmd;
172	union nvme_result	result;
173	u8			genctr;
174	u8			retries;
175	u8			flags;
176	u16			status;
177#ifdef CONFIG_NVME_MULTIPATH
178	unsigned long		start_time;
179#endif
180	struct nvme_ctrl	*ctrl;
181};
182
183/*
184 * Mark a bio as coming in through the mpath node.
185 */
186#define REQ_NVME_MPATH		REQ_DRV
187
188enum {
189	NVME_REQ_CANCELLED		= (1 << 0),
190	NVME_REQ_USERCMD		= (1 << 1),
191	NVME_MPATH_IO_STATS		= (1 << 2),
192};
193
194static inline struct nvme_request *nvme_req(struct request *req)
195{
196	return blk_mq_rq_to_pdu(req);
197}
198
199static inline u16 nvme_req_qid(struct request *req)
200{
201	if (!req->q->queuedata)
202		return 0;
203
204	return req->mq_hctx->queue_num + 1;
205}
206
207/* The below value is the specific amount of delay needed before checking
208 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
209 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
210 * found empirically.
211 */
212#define NVME_QUIRK_DELAY_AMOUNT		2300
213
214/*
215 * enum nvme_ctrl_state: Controller state
216 *
217 * @NVME_CTRL_NEW:		New controller just allocated, initial state
218 * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
219 * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
220 * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
221 *				transport
222 * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
223 * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
224 *				disabled/failed immediately. This state comes
225 * 				after all async event processing took place and
226 * 				before ns removal and the controller deletion
227 * 				progress
228 * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
229 *				shutdown or removal. In this case we forcibly
230 *				kill all inflight I/O as they have no chance to
231 *				complete
232 */
233enum nvme_ctrl_state {
234	NVME_CTRL_NEW,
235	NVME_CTRL_LIVE,
236	NVME_CTRL_RESETTING,
237	NVME_CTRL_CONNECTING,
238	NVME_CTRL_DELETING,
239	NVME_CTRL_DELETING_NOIO,
240	NVME_CTRL_DEAD,
241};
242
243struct nvme_fault_inject {
244#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
245	struct fault_attr attr;
246	struct dentry *parent;
247	bool dont_retry;	/* DNR, do not retry */
248	u16 status;		/* status code */
249#endif
250};
251
252enum nvme_ctrl_flags {
253	NVME_CTRL_FAILFAST_EXPIRED	= 0,
254	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
255	NVME_CTRL_STARTED_ONCE		= 2,
256	NVME_CTRL_STOPPED		= 3,
257	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
258	NVME_CTRL_DIRTY_CAPABILITY	= 5,
259	NVME_CTRL_FROZEN		= 6,
260};
261
262struct nvme_ctrl {
263	bool comp_seen;
264	bool identified;
265	enum nvme_ctrl_state state;
266	spinlock_t lock;
267	struct mutex scan_lock;
268	const struct nvme_ctrl_ops *ops;
269	struct request_queue *admin_q;
270	struct request_queue *connect_q;
271	struct request_queue *fabrics_q;
272	struct device *dev;
273	int instance;
274	int numa_node;
275	struct blk_mq_tag_set *tagset;
276	struct blk_mq_tag_set *admin_tagset;
277	struct list_head namespaces;
278	struct rw_semaphore namespaces_rwsem;
279	struct device ctrl_device;
280	struct device *device;	/* char device */
281#ifdef CONFIG_NVME_HWMON
282	struct device *hwmon_device;
283#endif
284	struct cdev cdev;
285	struct work_struct reset_work;
286	struct work_struct delete_work;
287	wait_queue_head_t state_wq;
288
289	struct nvme_subsystem *subsys;
290	struct list_head subsys_entry;
291
292	struct opal_dev *opal_dev;
293
294	char name[12];
295	u16 cntlid;
296
297	u16 mtfa;
298	u32 ctrl_config;
299	u32 queue_count;
300
301	u64 cap;
302	u32 max_hw_sectors;
303	u32 max_segments;
304	u32 max_integrity_segments;
305	u32 max_discard_sectors;
306	u32 max_discard_segments;
307	u32 max_zeroes_sectors;
308#ifdef CONFIG_BLK_DEV_ZONED
309	u32 max_zone_append;
310#endif
311	u16 crdt[3];
312	u16 oncs;
313	u32 dmrsl;
314	u16 oacs;
315	u16 sqsize;
316	u32 max_namespaces;
317	atomic_t abort_limit;
318	u8 vwc;
319	u32 vs;
320	u32 sgls;
321	u16 kas;
322	u8 npss;
323	u8 apsta;
324	u16 wctemp;
325	u16 cctemp;
326	u32 oaes;
327	u32 aen_result;
328	u32 ctratt;
329	unsigned int shutdown_timeout;
330	unsigned int kato;
331	bool subsystem;
332	unsigned long quirks;
333	struct nvme_id_power_state psd[32];
334	struct nvme_effects_log *effects;
335	struct xarray cels;
336	struct work_struct scan_work;
337	struct work_struct async_event_work;
338	struct delayed_work ka_work;
339	struct delayed_work failfast_work;
340	struct nvme_command ka_cmd;
341	unsigned long ka_last_check_time;
342	struct work_struct fw_act_work;
343	unsigned long events;
344
345#ifdef CONFIG_NVME_MULTIPATH
346	/* asymmetric namespace access: */
347	u8 anacap;
348	u8 anatt;
349	u32 anagrpmax;
350	u32 nanagrpid;
351	struct mutex ana_lock;
352	struct nvme_ana_rsp_hdr *ana_log_buf;
353	size_t ana_log_size;
354	struct timer_list anatt_timer;
355	struct work_struct ana_work;
356#endif
357
358#ifdef CONFIG_NVME_AUTH
359	struct work_struct dhchap_auth_work;
360	struct mutex dhchap_auth_mutex;
361	struct nvme_dhchap_queue_context *dhchap_ctxs;
362	struct nvme_dhchap_key *host_key;
363	struct nvme_dhchap_key *ctrl_key;
364	u16 transaction;
365#endif
366
367	/* Power saving configuration */
368	u64 ps_max_latency_us;
369	bool apst_enabled;
370
371	/* PCIe only: */
372	u16 hmmaxd;
373	u32 hmpre;
374	u32 hmmin;
375	u32 hmminds;
376
377	/* Fabrics only */
378	u32 ioccsz;
379	u32 iorcsz;
380	u16 icdoff;
381	u16 maxcmd;
382	int nr_reconnects;
383	unsigned long flags;
384	struct nvmf_ctrl_options *opts;
385
386	struct page *discard_page;
387	unsigned long discard_page_busy;
388
389	struct nvme_fault_inject fault_inject;
390
391	enum nvme_ctrl_type cntrltype;
392	enum nvme_dctype dctype;
393};
394
395static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
396{
397	return READ_ONCE(ctrl->state);
398}
399
400enum nvme_iopolicy {
401	NVME_IOPOLICY_NUMA,
402	NVME_IOPOLICY_RR,
403};
404
405struct nvme_subsystem {
406	int			instance;
407	struct device		dev;
408	/*
409	 * Because we unregister the device on the last put we need
410	 * a separate refcount.
411	 */
412	struct kref		ref;
413	struct list_head	entry;
414	struct mutex		lock;
415	struct list_head	ctrls;
416	struct list_head	nsheads;
417	char			subnqn[NVMF_NQN_SIZE];
418	char			serial[20];
419	char			model[40];
420	char			firmware_rev[8];
421	u8			cmic;
422	enum nvme_subsys_type	subtype;
423	u16			vendor_id;
424	u16			awupf;	/* 0's based awupf value. */
425	struct ida		ns_ida;
426#ifdef CONFIG_NVME_MULTIPATH
427	enum nvme_iopolicy	iopolicy;
428#endif
429};
430
431/*
432 * Container structure for uniqueue namespace identifiers.
433 */
434struct nvme_ns_ids {
435	u8	eui64[8];
436	u8	nguid[16];
437	uuid_t	uuid;
438	u8	csi;
439};
440
441/*
442 * Anchor structure for namespaces.  There is one for each namespace in a
443 * NVMe subsystem that any of our controllers can see, and the namespace
444 * structure for each controller is chained of it.  For private namespaces
445 * there is a 1:1 relation to our namespace structures, that is ->list
446 * only ever has a single entry for private namespaces.
447 */
448struct nvme_ns_head {
449	struct list_head	list;
450	struct srcu_struct      srcu;
451	struct nvme_subsystem	*subsys;
452	unsigned		ns_id;
453	struct nvme_ns_ids	ids;
454	struct list_head	entry;
455	struct kref		ref;
456	bool			shared;
457	int			instance;
458	struct nvme_effects_log *effects;
459
460	struct cdev		cdev;
461	struct device		cdev_device;
462
463	struct gendisk		*disk;
464#ifdef CONFIG_NVME_MULTIPATH
465	struct bio_list		requeue_list;
466	spinlock_t		requeue_lock;
467	struct work_struct	requeue_work;
468	struct mutex		lock;
469	unsigned long		flags;
470#define NVME_NSHEAD_DISK_LIVE	0
471	struct nvme_ns __rcu	*current_path[];
472#endif
473};
474
475static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
476{
477	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
478}
479
480enum nvme_ns_features {
481	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
482	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
483	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
484};
485
486struct nvme_ns {
487	struct list_head list;
488
489	struct nvme_ctrl *ctrl;
490	struct request_queue *queue;
491	struct gendisk *disk;
492#ifdef CONFIG_NVME_MULTIPATH
493	enum nvme_ana_state ana_state;
494	u32 ana_grpid;
495#endif
496	struct list_head siblings;
497	struct kref kref;
498	struct nvme_ns_head *head;
499
500	int lba_shift;
501	u16 ms;
502	u16 pi_size;
503	u16 sgs;
504	u32 sws;
505	u8 pi_type;
506	u8 guard_type;
507#ifdef CONFIG_BLK_DEV_ZONED
508	u64 zsze;
509#endif
510	unsigned long features;
511	unsigned long flags;
512#define NVME_NS_REMOVING	0
513#define NVME_NS_ANA_PENDING	2
514#define NVME_NS_FORCE_RO	3
515#define NVME_NS_READY		4
516
517	struct cdev		cdev;
518	struct device		cdev_device;
519
520	struct nvme_fault_inject fault_inject;
521
522};
523
524/* NVMe ns supports metadata actions by the controller (generate/strip) */
525static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
526{
527	return ns->pi_type && ns->ms == ns->pi_size;
528}
529
530struct nvme_ctrl_ops {
531	const char *name;
532	struct module *module;
533	unsigned int flags;
534#define NVME_F_FABRICS			(1 << 0)
535#define NVME_F_METADATA_SUPPORTED	(1 << 1)
536#define NVME_F_BLOCKING			(1 << 2)
537
538	const struct attribute_group **dev_attr_groups;
539	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
540	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
541	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
542	void (*free_ctrl)(struct nvme_ctrl *ctrl);
543	void (*submit_async_event)(struct nvme_ctrl *ctrl);
544	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
545	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
546	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
547	void (*print_device_info)(struct nvme_ctrl *ctrl);
548	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
549};
550
551/*
552 * nvme command_id is constructed as such:
553 * | xxxx | xxxxxxxxxxxx |
554 *   gen    request tag
555 */
556#define nvme_genctr_mask(gen)			(gen & 0xf)
557#define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
558#define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
559#define nvme_tag_from_cid(cid)			(cid & 0xfff)
560
561static inline u16 nvme_cid(struct request *rq)
562{
563	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
564}
565
566static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
567		u16 command_id)
568{
569	u8 genctr = nvme_genctr_from_cid(command_id);
570	u16 tag = nvme_tag_from_cid(command_id);
571	struct request *rq;
572
573	rq = blk_mq_tag_to_rq(tags, tag);
574	if (unlikely(!rq)) {
575		pr_err("could not locate request for tag %#x\n",
576			tag);
577		return NULL;
578	}
579	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
580		dev_err(nvme_req(rq)->ctrl->device,
581			"request %#x genctr mismatch (got %#x expected %#x)\n",
582			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
583		return NULL;
584	}
585	return rq;
586}
587
588static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
589                u16 command_id)
590{
591	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
592}
593
594/*
595 * Return the length of the string without the space padding
596 */
597static inline int nvme_strlen(char *s, int len)
598{
599	while (s[len - 1] == ' ')
600		len--;
601	return len;
602}
603
604static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
605{
606	struct nvme_subsystem *subsys = ctrl->subsys;
607
608	if (ctrl->ops->print_device_info) {
609		ctrl->ops->print_device_info(ctrl);
610		return;
611	}
612
613	dev_err(ctrl->device,
614		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
615		nvme_strlen(subsys->model, sizeof(subsys->model)),
616		subsys->model, nvme_strlen(subsys->firmware_rev,
617					   sizeof(subsys->firmware_rev)),
618		subsys->firmware_rev);
619}
620
621#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
622void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
623			    const char *dev_name);
624void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
625void nvme_should_fail(struct request *req);
626#else
627static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
628					  const char *dev_name)
629{
630}
631static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
632{
633}
634static inline void nvme_should_fail(struct request *req) {}
635#endif
636
637bool nvme_wait_reset(struct nvme_ctrl *ctrl);
638int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
639
640static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
641{
642	int ret;
643
644	if (!ctrl->subsystem)
645		return -ENOTTY;
646	if (!nvme_wait_reset(ctrl))
647		return -EBUSY;
648
649	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
650	if (ret)
651		return ret;
652
653	return nvme_try_sched_reset(ctrl);
654}
655
656/*
657 * Convert a 512B sector number to a device logical block number.
658 */
659static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
660{
661	return sector >> (ns->lba_shift - SECTOR_SHIFT);
662}
663
664/*
665 * Convert a device logical block number to a 512B sector number.
666 */
667static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
668{
669	return lba << (ns->lba_shift - SECTOR_SHIFT);
670}
671
672/*
673 * Convert byte length to nvme's 0-based num dwords
674 */
675static inline u32 nvme_bytes_to_numd(size_t len)
676{
677	return (len >> 2) - 1;
678}
679
680static inline bool nvme_is_ana_error(u16 status)
681{
682	switch (status & 0x7ff) {
683	case NVME_SC_ANA_TRANSITION:
684	case NVME_SC_ANA_INACCESSIBLE:
685	case NVME_SC_ANA_PERSISTENT_LOSS:
686		return true;
687	default:
688		return false;
689	}
690}
691
692static inline bool nvme_is_path_error(u16 status)
693{
694	/* check for a status code type of 'path related status' */
695	return (status & 0x700) == 0x300;
696}
697
698/*
699 * Fill in the status and result information from the CQE, and then figure out
700 * if blk-mq will need to use IPI magic to complete the request, and if yes do
701 * so.  If not let the caller complete the request without an indirect function
702 * call.
703 */
704static inline bool nvme_try_complete_req(struct request *req, __le16 status,
705		union nvme_result result)
706{
707	struct nvme_request *rq = nvme_req(req);
708	struct nvme_ctrl *ctrl = rq->ctrl;
709
710	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
711		rq->genctr++;
712
713	rq->status = le16_to_cpu(status) >> 1;
714	rq->result = result;
715	/* inject error when permitted by fault injection framework */
716	nvme_should_fail(req);
717	if (unlikely(blk_should_fake_timeout(req->q)))
718		return true;
719	return blk_mq_complete_request_remote(req);
720}
721
722static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
723{
724	get_device(ctrl->device);
725}
726
727static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
728{
729	put_device(ctrl->device);
730}
731
732static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
733{
734	return !qid &&
735		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
736}
737
738void nvme_complete_rq(struct request *req);
739void nvme_complete_batch_req(struct request *req);
740
741static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
742						void (*fn)(struct request *rq))
743{
744	struct request *req;
745
746	rq_list_for_each(&iob->req_list, req) {
747		fn(req);
748		nvme_complete_batch_req(req);
749	}
750	blk_mq_end_request_batch(iob);
751}
752
753blk_status_t nvme_host_path_error(struct request *req);
754bool nvme_cancel_request(struct request *req, void *data);
755void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
756void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
757bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
758		enum nvme_ctrl_state new_state);
759int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
760int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
761int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
762		const struct nvme_ctrl_ops *ops, unsigned long quirks);
763void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
764void nvme_start_ctrl(struct nvme_ctrl *ctrl);
765void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
766int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
767int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
768		const struct blk_mq_ops *ops, unsigned int cmd_size);
769void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
770int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
771		const struct blk_mq_ops *ops, unsigned int nr_maps,
772		unsigned int cmd_size);
773void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
774
775void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
776
777void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
778		volatile union nvme_result *res);
779
780void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
781void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
782void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
783void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
784void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
785void nvme_sync_queues(struct nvme_ctrl *ctrl);
786void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
787void nvme_unfreeze(struct nvme_ctrl *ctrl);
788void nvme_wait_freeze(struct nvme_ctrl *ctrl);
789int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
790void nvme_start_freeze(struct nvme_ctrl *ctrl);
791
792static inline enum req_op nvme_req_op(struct nvme_command *cmd)
793{
794	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
795}
796
797#define NVME_QID_ANY -1
798void nvme_init_request(struct request *req, struct nvme_command *cmd);
799void nvme_cleanup_cmd(struct request *req);
800blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
801blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
802		struct request *req);
803bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
804		bool queue_live);
805
806static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
807		bool queue_live)
808{
809	if (likely(ctrl->state == NVME_CTRL_LIVE))
810		return true;
811	if (ctrl->ops->flags & NVME_F_FABRICS &&
812	    ctrl->state == NVME_CTRL_DELETING)
813		return queue_live;
814	return __nvme_check_ready(ctrl, rq, queue_live);
815}
816
817/*
818 * NSID shall be unique for all shared namespaces, or if at least one of the
819 * following conditions is met:
820 *   1. Namespace Management is supported by the controller
821 *   2. ANA is supported by the controller
822 *   3. NVM Set are supported by the controller
823 *
824 * In other case, private namespace are not required to report a unique NSID.
825 */
826static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
827		struct nvme_ns_head *head)
828{
829	return head->shared ||
830		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
831		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
832		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
833}
834
835int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
836		void *buf, unsigned bufflen);
837int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
838		union nvme_result *result, void *buffer, unsigned bufflen,
839		int qid, int at_head,
840		blk_mq_req_flags_t flags);
841int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
842		      unsigned int dword11, void *buffer, size_t buflen,
843		      u32 *result);
844int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
845		      unsigned int dword11, void *buffer, size_t buflen,
846		      u32 *result);
847int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
848void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
849int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
850int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
851int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
852void nvme_queue_scan(struct nvme_ctrl *ctrl);
853int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
854		void *log, size_t size, u64 offset);
855bool nvme_tryget_ns_head(struct nvme_ns_head *head);
856void nvme_put_ns_head(struct nvme_ns_head *head);
857int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
858		const struct file_operations *fops, struct module *owner);
859void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
860int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
861		unsigned int cmd, unsigned long arg);
862long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
863int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
864		unsigned int cmd, unsigned long arg);
865long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
866		unsigned long arg);
867long nvme_dev_ioctl(struct file *file, unsigned int cmd,
868		unsigned long arg);
869int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
870		struct io_comp_batch *iob, unsigned int poll_flags);
871int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
872		unsigned int issue_flags);
873int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
874		unsigned int issue_flags);
875int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
876int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
877
878extern const struct attribute_group *nvme_ns_id_attr_groups[];
879extern const struct pr_ops nvme_pr_ops;
880extern const struct block_device_operations nvme_ns_head_ops;
881extern const struct attribute_group nvme_dev_attrs_group;
882extern const struct attribute_group *nvme_subsys_attrs_groups[];
883extern const struct attribute_group *nvme_dev_attr_groups[];
884extern const struct block_device_operations nvme_bdev_ops;
885
886void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
887struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
888#ifdef CONFIG_NVME_MULTIPATH
889static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
890{
891	return ctrl->ana_log_buf != NULL;
892}
893
894void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
895void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
896void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
897void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
898void nvme_failover_req(struct request *req);
899void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
900int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
901void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
902void nvme_mpath_remove_disk(struct nvme_ns_head *head);
903int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
904void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
905void nvme_mpath_update(struct nvme_ctrl *ctrl);
906void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
907void nvme_mpath_stop(struct nvme_ctrl *ctrl);
908bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
909void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
910void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
911void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
912void nvme_mpath_start_request(struct request *rq);
913void nvme_mpath_end_request(struct request *rq);
914
915static inline void nvme_trace_bio_complete(struct request *req)
916{
917	struct nvme_ns *ns = req->q->queuedata;
918
919	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
920		trace_block_bio_complete(ns->head->disk->queue, req->bio);
921}
922
923extern bool multipath;
924extern struct device_attribute dev_attr_ana_grpid;
925extern struct device_attribute dev_attr_ana_state;
926extern struct device_attribute subsys_attr_iopolicy;
927
928#else
929#define multipath false
930static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
931{
932	return false;
933}
934static inline void nvme_failover_req(struct request *req)
935{
936}
937static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
938{
939}
940static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
941		struct nvme_ns_head *head)
942{
943	return 0;
944}
945static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
946{
947}
948static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
949{
950}
951static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
952{
953	return false;
954}
955static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
956{
957}
958static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
959{
960}
961static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
962{
963}
964static inline void nvme_trace_bio_complete(struct request *req)
965{
966}
967static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
968{
969}
970static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
971		struct nvme_id_ctrl *id)
972{
973	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
974		dev_warn(ctrl->device,
975"Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
976	return 0;
977}
978static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
979{
980}
981static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
982{
983}
984static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
985{
986}
987static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
988{
989}
990static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
991{
992}
993static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
994{
995}
996static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
997{
998}
999static inline void nvme_mpath_start_request(struct request *rq)
1000{
1001}
1002static inline void nvme_mpath_end_request(struct request *rq)
1003{
1004}
1005#endif /* CONFIG_NVME_MULTIPATH */
1006
1007int nvme_revalidate_zones(struct nvme_ns *ns);
1008int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1009		unsigned int nr_zones, report_zones_cb cb, void *data);
1010#ifdef CONFIG_BLK_DEV_ZONED
1011int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1012blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1013				       struct nvme_command *cmnd,
1014				       enum nvme_zone_mgmt_action action);
1015#else
1016static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1017		struct request *req, struct nvme_command *cmnd,
1018		enum nvme_zone_mgmt_action action)
1019{
1020	return BLK_STS_NOTSUPP;
1021}
1022
1023static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1024{
1025	dev_warn(ns->ctrl->device,
1026		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1027	return -EPROTONOSUPPORT;
1028}
1029#endif
1030
1031static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1032{
1033	return dev_to_disk(dev)->private_data;
1034}
1035
1036#ifdef CONFIG_NVME_HWMON
1037int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1038void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1039#else
1040static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1041{
1042	return 0;
1043}
1044
1045static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1046{
1047}
1048#endif
1049
1050static inline void nvme_start_request(struct request *rq)
1051{
1052	if (rq->cmd_flags & REQ_NVME_MPATH)
1053		nvme_mpath_start_request(rq);
1054	blk_mq_start_request(rq);
1055}
1056
1057static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1058{
1059	return ctrl->sgls & ((1 << 0) | (1 << 1));
1060}
1061
1062#ifdef CONFIG_NVME_AUTH
1063int __init nvme_init_auth(void);
1064void __exit nvme_exit_auth(void);
1065int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1066void nvme_auth_stop(struct nvme_ctrl *ctrl);
1067int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1068int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1069void nvme_auth_free(struct nvme_ctrl *ctrl);
1070#else
1071static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1072{
1073	return 0;
1074}
1075static inline int __init nvme_init_auth(void)
1076{
1077	return 0;
1078}
1079static inline void __exit nvme_exit_auth(void)
1080{
1081}
1082static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1083static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1084{
1085	return -EPROTONOSUPPORT;
1086}
1087static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1088{
1089	return NVME_SC_AUTH_REQUIRED;
1090}
1091static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1092#endif
1093
1094u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1095			 u8 opcode);
1096u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1097int nvme_execute_rq(struct request *rq, bool at_head);
1098void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1099		       struct nvme_command *cmd, int status);
1100struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1101struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1102void nvme_put_ns(struct nvme_ns *ns);
1103
1104static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1105{
1106	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1107}
1108
1109#ifdef CONFIG_NVME_VERBOSE_ERRORS
1110const unsigned char *nvme_get_error_status_str(u16 status);
1111const unsigned char *nvme_get_opcode_str(u8 opcode);
1112const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1113const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1114#else /* CONFIG_NVME_VERBOSE_ERRORS */
1115static inline const unsigned char *nvme_get_error_status_str(u16 status)
1116{
1117	return "I/O Error";
1118}
1119static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1120{
1121	return "I/O Cmd";
1122}
1123static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1124{
1125	return "Admin Cmd";
1126}
1127
1128static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1129{
1130	return "Fabrics Cmd";
1131}
1132#endif /* CONFIG_NVME_VERBOSE_ERRORS */
1133
1134static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1135{
1136	if (opcode == nvme_fabrics_command)
1137		return nvme_get_fabrics_opcode_str(fctype);
1138	return qid ? nvme_get_opcode_str(opcode) :
1139		nvme_get_admin_opcode_str(opcode);
1140}
1141#endif /* _NVME_H */
1142