xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 25b892b5)
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 /*
144  * Common request structure for NVMe passthrough.  All drivers must have
145  * this structure as the first member of their request-private data.
146  */
147 struct nvme_request {
148 	struct nvme_command	*cmd;
149 	union nvme_result	result;
150 	u8			genctr;
151 	u8			retries;
152 	u8			flags;
153 	u16			status;
154 	struct nvme_ctrl	*ctrl;
155 };
156 
157 /*
158  * Mark a bio as coming in through the mpath node.
159  */
160 #define REQ_NVME_MPATH		REQ_DRV
161 
162 enum {
163 	NVME_REQ_CANCELLED		= (1 << 0),
164 	NVME_REQ_USERCMD		= (1 << 1),
165 };
166 
167 static inline struct nvme_request *nvme_req(struct request *req)
168 {
169 	return blk_mq_rq_to_pdu(req);
170 }
171 
172 static inline u16 nvme_req_qid(struct request *req)
173 {
174 	if (!req->q->queuedata)
175 		return 0;
176 
177 	return req->mq_hctx->queue_num + 1;
178 }
179 
180 /* The below value is the specific amount of delay needed before checking
181  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
182  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
183  * found empirically.
184  */
185 #define NVME_QUIRK_DELAY_AMOUNT		2300
186 
187 /*
188  * enum nvme_ctrl_state: Controller state
189  *
190  * @NVME_CTRL_NEW:		New controller just allocated, initial state
191  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
192  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
193  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
194  *				transport
195  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
196  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
197  *				disabled/failed immediately. This state comes
198  * 				after all async event processing took place and
199  * 				before ns removal and the controller deletion
200  * 				progress
201  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
202  *				shutdown or removal. In this case we forcibly
203  *				kill all inflight I/O as they have no chance to
204  *				complete
205  */
206 enum nvme_ctrl_state {
207 	NVME_CTRL_NEW,
208 	NVME_CTRL_LIVE,
209 	NVME_CTRL_RESETTING,
210 	NVME_CTRL_CONNECTING,
211 	NVME_CTRL_DELETING,
212 	NVME_CTRL_DELETING_NOIO,
213 	NVME_CTRL_DEAD,
214 };
215 
216 struct nvme_fault_inject {
217 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
218 	struct fault_attr attr;
219 	struct dentry *parent;
220 	bool dont_retry;	/* DNR, do not retry */
221 	u16 status;		/* status code */
222 #endif
223 };
224 
225 struct nvme_ctrl {
226 	bool comp_seen;
227 	enum nvme_ctrl_state state;
228 	bool identified;
229 	spinlock_t lock;
230 	struct mutex scan_lock;
231 	const struct nvme_ctrl_ops *ops;
232 	struct request_queue *admin_q;
233 	struct request_queue *connect_q;
234 	struct request_queue *fabrics_q;
235 	struct device *dev;
236 	int instance;
237 	int numa_node;
238 	struct blk_mq_tag_set *tagset;
239 	struct blk_mq_tag_set *admin_tagset;
240 	struct list_head namespaces;
241 	struct rw_semaphore namespaces_rwsem;
242 	struct device ctrl_device;
243 	struct device *device;	/* char device */
244 #ifdef CONFIG_NVME_HWMON
245 	struct device *hwmon_device;
246 #endif
247 	struct cdev cdev;
248 	struct work_struct reset_work;
249 	struct work_struct delete_work;
250 	wait_queue_head_t state_wq;
251 
252 	struct nvme_subsystem *subsys;
253 	struct list_head subsys_entry;
254 
255 	struct opal_dev *opal_dev;
256 
257 	char name[12];
258 	u16 cntlid;
259 
260 	u32 ctrl_config;
261 	u16 mtfa;
262 	u32 queue_count;
263 
264 	u64 cap;
265 	u32 max_hw_sectors;
266 	u32 max_segments;
267 	u32 max_integrity_segments;
268 	u32 max_discard_sectors;
269 	u32 max_discard_segments;
270 	u32 max_zeroes_sectors;
271 #ifdef CONFIG_BLK_DEV_ZONED
272 	u32 max_zone_append;
273 #endif
274 	u16 crdt[3];
275 	u16 oncs;
276 	u16 oacs;
277 	u16 nssa;
278 	u16 nr_streams;
279 	u16 sqsize;
280 	u32 max_namespaces;
281 	atomic_t abort_limit;
282 	u8 vwc;
283 	u32 vs;
284 	u32 sgls;
285 	u16 kas;
286 	u8 npss;
287 	u8 apsta;
288 	u16 wctemp;
289 	u16 cctemp;
290 	u32 oaes;
291 	u32 aen_result;
292 	u32 ctratt;
293 	unsigned int shutdown_timeout;
294 	unsigned int kato;
295 	bool subsystem;
296 	unsigned long quirks;
297 	struct nvme_id_power_state psd[32];
298 	struct nvme_effects_log *effects;
299 	struct xarray cels;
300 	struct work_struct scan_work;
301 	struct work_struct async_event_work;
302 	struct delayed_work ka_work;
303 	struct delayed_work failfast_work;
304 	struct nvme_command ka_cmd;
305 	struct work_struct fw_act_work;
306 	unsigned long events;
307 
308 #ifdef CONFIG_NVME_MULTIPATH
309 	/* asymmetric namespace access: */
310 	u8 anacap;
311 	u8 anatt;
312 	u32 anagrpmax;
313 	u32 nanagrpid;
314 	struct mutex ana_lock;
315 	struct nvme_ana_rsp_hdr *ana_log_buf;
316 	size_t ana_log_size;
317 	struct timer_list anatt_timer;
318 	struct work_struct ana_work;
319 #endif
320 
321 	/* Power saving configuration */
322 	u64 ps_max_latency_us;
323 	bool apst_enabled;
324 
325 	/* PCIe only: */
326 	u32 hmpre;
327 	u32 hmmin;
328 	u32 hmminds;
329 	u16 hmmaxd;
330 
331 	/* Fabrics only */
332 	u32 ioccsz;
333 	u32 iorcsz;
334 	u16 icdoff;
335 	u16 maxcmd;
336 	int nr_reconnects;
337 	unsigned long flags;
338 #define NVME_CTRL_FAILFAST_EXPIRED	0
339 	struct nvmf_ctrl_options *opts;
340 
341 	struct page *discard_page;
342 	unsigned long discard_page_busy;
343 
344 	struct nvme_fault_inject fault_inject;
345 };
346 
347 enum nvme_iopolicy {
348 	NVME_IOPOLICY_NUMA,
349 	NVME_IOPOLICY_RR,
350 };
351 
352 struct nvme_subsystem {
353 	int			instance;
354 	struct device		dev;
355 	/*
356 	 * Because we unregister the device on the last put we need
357 	 * a separate refcount.
358 	 */
359 	struct kref		ref;
360 	struct list_head	entry;
361 	struct mutex		lock;
362 	struct list_head	ctrls;
363 	struct list_head	nsheads;
364 	char			subnqn[NVMF_NQN_SIZE];
365 	char			serial[20];
366 	char			model[40];
367 	char			firmware_rev[8];
368 	u8			cmic;
369 	u16			vendor_id;
370 	u16			awupf;	/* 0's based awupf value. */
371 	struct ida		ns_ida;
372 #ifdef CONFIG_NVME_MULTIPATH
373 	enum nvme_iopolicy	iopolicy;
374 #endif
375 };
376 
377 /*
378  * Container structure for uniqueue namespace identifiers.
379  */
380 struct nvme_ns_ids {
381 	u8	eui64[8];
382 	u8	nguid[16];
383 	uuid_t	uuid;
384 	u8	csi;
385 };
386 
387 /*
388  * Anchor structure for namespaces.  There is one for each namespace in a
389  * NVMe subsystem that any of our controllers can see, and the namespace
390  * structure for each controller is chained of it.  For private namespaces
391  * there is a 1:1 relation to our namespace structures, that is ->list
392  * only ever has a single entry for private namespaces.
393  */
394 struct nvme_ns_head {
395 	struct list_head	list;
396 	struct srcu_struct      srcu;
397 	struct nvme_subsystem	*subsys;
398 	unsigned		ns_id;
399 	struct nvme_ns_ids	ids;
400 	struct list_head	entry;
401 	struct kref		ref;
402 	bool			shared;
403 	int			instance;
404 	struct nvme_effects_log *effects;
405 
406 	struct cdev		cdev;
407 	struct device		cdev_device;
408 
409 	struct gendisk		*disk;
410 #ifdef CONFIG_NVME_MULTIPATH
411 	struct bio_list		requeue_list;
412 	spinlock_t		requeue_lock;
413 	struct work_struct	requeue_work;
414 	struct mutex		lock;
415 	unsigned long		flags;
416 #define NVME_NSHEAD_DISK_LIVE	0
417 	struct nvme_ns __rcu	*current_path[];
418 #endif
419 };
420 
421 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
422 {
423 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
424 }
425 
426 enum nvme_ns_features {
427 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
428 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
429 };
430 
431 struct nvme_ns {
432 	struct list_head list;
433 
434 	struct nvme_ctrl *ctrl;
435 	struct request_queue *queue;
436 	struct gendisk *disk;
437 #ifdef CONFIG_NVME_MULTIPATH
438 	enum nvme_ana_state ana_state;
439 	u32 ana_grpid;
440 #endif
441 	struct list_head siblings;
442 	struct kref kref;
443 	struct nvme_ns_head *head;
444 
445 	int lba_shift;
446 	u16 ms;
447 	u16 sgs;
448 	u32 sws;
449 	u8 pi_type;
450 #ifdef CONFIG_BLK_DEV_ZONED
451 	u64 zsze;
452 #endif
453 	unsigned long features;
454 	unsigned long flags;
455 #define NVME_NS_REMOVING	0
456 #define NVME_NS_DEAD     	1
457 #define NVME_NS_ANA_PENDING	2
458 #define NVME_NS_FORCE_RO	3
459 #define NVME_NS_READY		4
460 
461 	struct cdev		cdev;
462 	struct device		cdev_device;
463 
464 	struct nvme_fault_inject fault_inject;
465 
466 };
467 
468 /* NVMe ns supports metadata actions by the controller (generate/strip) */
469 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
470 {
471 	return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
472 }
473 
474 struct nvme_ctrl_ops {
475 	const char *name;
476 	struct module *module;
477 	unsigned int flags;
478 #define NVME_F_FABRICS			(1 << 0)
479 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
480 #define NVME_F_PCI_P2PDMA		(1 << 2)
481 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
482 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
483 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
484 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
485 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
486 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
487 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
488 };
489 
490 /*
491  * nvme command_id is constructed as such:
492  * | xxxx | xxxxxxxxxxxx |
493  *   gen    request tag
494  */
495 #define nvme_genctr_mask(gen)			(gen & 0xf)
496 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
497 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
498 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
499 
500 static inline u16 nvme_cid(struct request *rq)
501 {
502 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
503 }
504 
505 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
506 		u16 command_id)
507 {
508 	u8 genctr = nvme_genctr_from_cid(command_id);
509 	u16 tag = nvme_tag_from_cid(command_id);
510 	struct request *rq;
511 
512 	rq = blk_mq_tag_to_rq(tags, tag);
513 	if (unlikely(!rq)) {
514 		pr_err("could not locate request for tag %#x\n",
515 			tag);
516 		return NULL;
517 	}
518 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
519 		dev_err(nvme_req(rq)->ctrl->device,
520 			"request %#x genctr mismatch (got %#x expected %#x)\n",
521 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
522 		return NULL;
523 	}
524 	return rq;
525 }
526 
527 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
528                 u16 command_id)
529 {
530 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
531 }
532 
533 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
534 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
535 			    const char *dev_name);
536 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
537 void nvme_should_fail(struct request *req);
538 #else
539 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
540 					  const char *dev_name)
541 {
542 }
543 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
544 {
545 }
546 static inline void nvme_should_fail(struct request *req) {}
547 #endif
548 
549 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
550 {
551 	if (!ctrl->subsystem)
552 		return -ENOTTY;
553 	return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
554 }
555 
556 /*
557  * Convert a 512B sector number to a device logical block number.
558  */
559 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
560 {
561 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
562 }
563 
564 /*
565  * Convert a device logical block number to a 512B sector number.
566  */
567 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
568 {
569 	return lba << (ns->lba_shift - SECTOR_SHIFT);
570 }
571 
572 /*
573  * Convert byte length to nvme's 0-based num dwords
574  */
575 static inline u32 nvme_bytes_to_numd(size_t len)
576 {
577 	return (len >> 2) - 1;
578 }
579 
580 static inline bool nvme_is_ana_error(u16 status)
581 {
582 	switch (status & 0x7ff) {
583 	case NVME_SC_ANA_TRANSITION:
584 	case NVME_SC_ANA_INACCESSIBLE:
585 	case NVME_SC_ANA_PERSISTENT_LOSS:
586 		return true;
587 	default:
588 		return false;
589 	}
590 }
591 
592 static inline bool nvme_is_path_error(u16 status)
593 {
594 	/* check for a status code type of 'path related status' */
595 	return (status & 0x700) == 0x300;
596 }
597 
598 /*
599  * Fill in the status and result information from the CQE, and then figure out
600  * if blk-mq will need to use IPI magic to complete the request, and if yes do
601  * so.  If not let the caller complete the request without an indirect function
602  * call.
603  */
604 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
605 		union nvme_result result)
606 {
607 	struct nvme_request *rq = nvme_req(req);
608 
609 	rq->status = le16_to_cpu(status) >> 1;
610 	rq->result = result;
611 	/* inject error when permitted by fault injection framework */
612 	nvme_should_fail(req);
613 	if (unlikely(blk_should_fake_timeout(req->q)))
614 		return true;
615 	return blk_mq_complete_request_remote(req);
616 }
617 
618 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
619 {
620 	get_device(ctrl->device);
621 }
622 
623 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
624 {
625 	put_device(ctrl->device);
626 }
627 
628 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
629 {
630 	return !qid &&
631 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
632 }
633 
634 void nvme_complete_rq(struct request *req);
635 blk_status_t nvme_host_path_error(struct request *req);
636 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
637 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
638 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
639 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
640 		enum nvme_ctrl_state new_state);
641 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
642 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
643 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
644 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
645 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
646 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
647 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
648 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
649 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
650 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl);
651 
652 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
653 
654 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
655 		bool send);
656 
657 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
658 		volatile union nvme_result *res);
659 
660 void nvme_stop_queues(struct nvme_ctrl *ctrl);
661 void nvme_start_queues(struct nvme_ctrl *ctrl);
662 void nvme_kill_queues(struct nvme_ctrl *ctrl);
663 void nvme_sync_queues(struct nvme_ctrl *ctrl);
664 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
665 void nvme_unfreeze(struct nvme_ctrl *ctrl);
666 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
667 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
668 void nvme_start_freeze(struct nvme_ctrl *ctrl);
669 
670 #define NVME_QID_ANY -1
671 struct request *nvme_alloc_request(struct request_queue *q,
672 		struct nvme_command *cmd, blk_mq_req_flags_t flags);
673 void nvme_cleanup_cmd(struct request *req);
674 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
675 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
676 		struct request *req);
677 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
678 		bool queue_live);
679 
680 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
681 		bool queue_live)
682 {
683 	if (likely(ctrl->state == NVME_CTRL_LIVE))
684 		return true;
685 	if (ctrl->ops->flags & NVME_F_FABRICS &&
686 	    ctrl->state == NVME_CTRL_DELETING)
687 		return true;
688 	return __nvme_check_ready(ctrl, rq, queue_live);
689 }
690 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
691 		void *buf, unsigned bufflen);
692 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
693 		union nvme_result *result, void *buffer, unsigned bufflen,
694 		unsigned timeout, int qid, int at_head,
695 		blk_mq_req_flags_t flags);
696 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
697 		      unsigned int dword11, void *buffer, size_t buflen,
698 		      u32 *result);
699 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
700 		      unsigned int dword11, void *buffer, size_t buflen,
701 		      u32 *result);
702 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
703 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
704 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
705 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
706 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
707 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
708 void nvme_queue_scan(struct nvme_ctrl *ctrl);
709 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
710 		void *log, size_t size, u64 offset);
711 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
712 void nvme_put_ns_head(struct nvme_ns_head *head);
713 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
714 		const struct file_operations *fops, struct module *owner);
715 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
716 int nvme_ioctl(struct block_device *bdev, fmode_t mode,
717 		unsigned int cmd, unsigned long arg);
718 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
719 int nvme_ns_head_ioctl(struct block_device *bdev, fmode_t mode,
720 		unsigned int cmd, unsigned long arg);
721 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
722 		unsigned long arg);
723 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
724 		unsigned long arg);
725 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
726 
727 extern const struct attribute_group *nvme_ns_id_attr_groups[];
728 extern const struct pr_ops nvme_pr_ops;
729 extern const struct block_device_operations nvme_ns_head_ops;
730 
731 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
732 #ifdef CONFIG_NVME_MULTIPATH
733 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
734 {
735 	return ctrl->ana_log_buf != NULL;
736 }
737 
738 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
739 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
740 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
741 bool nvme_mpath_set_disk_name(struct nvme_ns *ns, char *disk_name, int *flags);
742 void nvme_failover_req(struct request *req);
743 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
744 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
745 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
746 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
747 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
748 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
749 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
750 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
751 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
752 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
753 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
754 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
755 
756 static inline void nvme_trace_bio_complete(struct request *req)
757 {
758 	struct nvme_ns *ns = req->q->queuedata;
759 
760 	if (req->cmd_flags & REQ_NVME_MPATH)
761 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
762 }
763 
764 extern struct device_attribute dev_attr_ana_grpid;
765 extern struct device_attribute dev_attr_ana_state;
766 extern struct device_attribute subsys_attr_iopolicy;
767 
768 #else
769 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
770 {
771 	return false;
772 }
773 static inline bool nvme_mpath_set_disk_name(struct nvme_ns *ns, char *disk_name,
774 		int *flags)
775 {
776 	return false;
777 }
778 static inline void nvme_failover_req(struct request *req)
779 {
780 }
781 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
782 {
783 }
784 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
785 		struct nvme_ns_head *head)
786 {
787 	return 0;
788 }
789 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
790 		struct nvme_id_ns *id)
791 {
792 }
793 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
794 {
795 }
796 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
797 {
798 	return false;
799 }
800 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
801 {
802 }
803 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
804 {
805 }
806 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
807 {
808 }
809 static inline void nvme_trace_bio_complete(struct request *req)
810 {
811 }
812 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
813 {
814 }
815 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
816 		struct nvme_id_ctrl *id)
817 {
818 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
819 		dev_warn(ctrl->device,
820 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
821 	return 0;
822 }
823 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
824 {
825 }
826 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
827 {
828 }
829 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
830 {
831 }
832 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
833 {
834 }
835 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
836 {
837 }
838 #endif /* CONFIG_NVME_MULTIPATH */
839 
840 int nvme_revalidate_zones(struct nvme_ns *ns);
841 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
842 		unsigned int nr_zones, report_zones_cb cb, void *data);
843 #ifdef CONFIG_BLK_DEV_ZONED
844 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
845 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
846 				       struct nvme_command *cmnd,
847 				       enum nvme_zone_mgmt_action action);
848 #else
849 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
850 		struct request *req, struct nvme_command *cmnd,
851 		enum nvme_zone_mgmt_action action)
852 {
853 	return BLK_STS_NOTSUPP;
854 }
855 
856 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
857 {
858 	dev_warn(ns->ctrl->device,
859 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
860 	return -EPROTONOSUPPORT;
861 }
862 #endif
863 
864 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
865 {
866 	return dev_to_disk(dev)->private_data;
867 }
868 
869 #ifdef CONFIG_NVME_HWMON
870 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
871 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
872 #else
873 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
874 {
875 	return 0;
876 }
877 
878 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
879 {
880 }
881 #endif
882 
883 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
884 {
885 	return ctrl->sgls & ((1 << 0) | (1 << 1));
886 }
887 
888 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
889 			 u8 opcode);
890 int nvme_execute_passthru_rq(struct request *rq);
891 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
892 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
893 void nvme_put_ns(struct nvme_ns *ns);
894 
895 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
896 {
897 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
898 }
899 
900 #endif /* _NVME_H */
901