xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 6aeadf78)
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 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
251 };
252 
253 struct nvme_ctrl {
254 	bool comp_seen;
255 	bool identified;
256 	enum nvme_ctrl_state state;
257 	spinlock_t lock;
258 	struct mutex scan_lock;
259 	const struct nvme_ctrl_ops *ops;
260 	struct request_queue *admin_q;
261 	struct request_queue *connect_q;
262 	struct request_queue *fabrics_q;
263 	struct device *dev;
264 	int instance;
265 	int numa_node;
266 	struct blk_mq_tag_set *tagset;
267 	struct blk_mq_tag_set *admin_tagset;
268 	struct list_head namespaces;
269 	struct rw_semaphore namespaces_rwsem;
270 	struct device ctrl_device;
271 	struct device *device;	/* char device */
272 #ifdef CONFIG_NVME_HWMON
273 	struct device *hwmon_device;
274 #endif
275 	struct cdev cdev;
276 	struct work_struct reset_work;
277 	struct work_struct delete_work;
278 	wait_queue_head_t state_wq;
279 
280 	struct nvme_subsystem *subsys;
281 	struct list_head subsys_entry;
282 
283 	struct opal_dev *opal_dev;
284 
285 	char name[12];
286 	u16 cntlid;
287 
288 	u16 mtfa;
289 	u32 ctrl_config;
290 	u32 queue_count;
291 
292 	u64 cap;
293 	u32 max_hw_sectors;
294 	u32 max_segments;
295 	u32 max_integrity_segments;
296 	u32 max_discard_sectors;
297 	u32 max_discard_segments;
298 	u32 max_zeroes_sectors;
299 #ifdef CONFIG_BLK_DEV_ZONED
300 	u32 max_zone_append;
301 #endif
302 	u16 crdt[3];
303 	u16 oncs;
304 	u32 dmrsl;
305 	u16 oacs;
306 	u16 sqsize;
307 	u32 max_namespaces;
308 	atomic_t abort_limit;
309 	u8 vwc;
310 	u32 vs;
311 	u32 sgls;
312 	u16 kas;
313 	u8 npss;
314 	u8 apsta;
315 	u16 wctemp;
316 	u16 cctemp;
317 	u32 oaes;
318 	u32 aen_result;
319 	u32 ctratt;
320 	unsigned int shutdown_timeout;
321 	unsigned int kato;
322 	bool subsystem;
323 	unsigned long quirks;
324 	struct nvme_id_power_state psd[32];
325 	struct nvme_effects_log *effects;
326 	struct xarray cels;
327 	struct work_struct scan_work;
328 	struct work_struct async_event_work;
329 	struct delayed_work ka_work;
330 	struct delayed_work failfast_work;
331 	struct nvme_command ka_cmd;
332 	unsigned long ka_last_check_time;
333 	struct work_struct fw_act_work;
334 	unsigned long events;
335 
336 #ifdef CONFIG_NVME_MULTIPATH
337 	/* asymmetric namespace access: */
338 	u8 anacap;
339 	u8 anatt;
340 	u32 anagrpmax;
341 	u32 nanagrpid;
342 	struct mutex ana_lock;
343 	struct nvme_ana_rsp_hdr *ana_log_buf;
344 	size_t ana_log_size;
345 	struct timer_list anatt_timer;
346 	struct work_struct ana_work;
347 #endif
348 
349 #ifdef CONFIG_NVME_AUTH
350 	struct work_struct dhchap_auth_work;
351 	struct mutex dhchap_auth_mutex;
352 	struct nvme_dhchap_queue_context *dhchap_ctxs;
353 	struct nvme_dhchap_key *host_key;
354 	struct nvme_dhchap_key *ctrl_key;
355 	u16 transaction;
356 #endif
357 
358 	/* Power saving configuration */
359 	u64 ps_max_latency_us;
360 	bool apst_enabled;
361 
362 	/* PCIe only: */
363 	u16 hmmaxd;
364 	u32 hmpre;
365 	u32 hmmin;
366 	u32 hmminds;
367 
368 	/* Fabrics only */
369 	u32 ioccsz;
370 	u32 iorcsz;
371 	u16 icdoff;
372 	u16 maxcmd;
373 	int nr_reconnects;
374 	unsigned long flags;
375 	struct nvmf_ctrl_options *opts;
376 
377 	struct page *discard_page;
378 	unsigned long discard_page_busy;
379 
380 	struct nvme_fault_inject fault_inject;
381 
382 	enum nvme_ctrl_type cntrltype;
383 	enum nvme_dctype dctype;
384 };
385 
386 enum nvme_iopolicy {
387 	NVME_IOPOLICY_NUMA,
388 	NVME_IOPOLICY_RR,
389 };
390 
391 struct nvme_subsystem {
392 	int			instance;
393 	struct device		dev;
394 	/*
395 	 * Because we unregister the device on the last put we need
396 	 * a separate refcount.
397 	 */
398 	struct kref		ref;
399 	struct list_head	entry;
400 	struct mutex		lock;
401 	struct list_head	ctrls;
402 	struct list_head	nsheads;
403 	char			subnqn[NVMF_NQN_SIZE];
404 	char			serial[20];
405 	char			model[40];
406 	char			firmware_rev[8];
407 	u8			cmic;
408 	enum nvme_subsys_type	subtype;
409 	u16			vendor_id;
410 	u16			awupf;	/* 0's based awupf value. */
411 	struct ida		ns_ida;
412 #ifdef CONFIG_NVME_MULTIPATH
413 	enum nvme_iopolicy	iopolicy;
414 #endif
415 };
416 
417 /*
418  * Container structure for uniqueue namespace identifiers.
419  */
420 struct nvme_ns_ids {
421 	u8	eui64[8];
422 	u8	nguid[16];
423 	uuid_t	uuid;
424 	u8	csi;
425 };
426 
427 /*
428  * Anchor structure for namespaces.  There is one for each namespace in a
429  * NVMe subsystem that any of our controllers can see, and the namespace
430  * structure for each controller is chained of it.  For private namespaces
431  * there is a 1:1 relation to our namespace structures, that is ->list
432  * only ever has a single entry for private namespaces.
433  */
434 struct nvme_ns_head {
435 	struct list_head	list;
436 	struct srcu_struct      srcu;
437 	struct nvme_subsystem	*subsys;
438 	unsigned		ns_id;
439 	struct nvme_ns_ids	ids;
440 	struct list_head	entry;
441 	struct kref		ref;
442 	bool			shared;
443 	int			instance;
444 	struct nvme_effects_log *effects;
445 
446 	struct cdev		cdev;
447 	struct device		cdev_device;
448 
449 	struct gendisk		*disk;
450 #ifdef CONFIG_NVME_MULTIPATH
451 	struct bio_list		requeue_list;
452 	spinlock_t		requeue_lock;
453 	struct work_struct	requeue_work;
454 	struct mutex		lock;
455 	unsigned long		flags;
456 #define NVME_NSHEAD_DISK_LIVE	0
457 	struct nvme_ns __rcu	*current_path[];
458 #endif
459 };
460 
461 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
462 {
463 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
464 }
465 
466 enum nvme_ns_features {
467 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
468 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
469 	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
470 };
471 
472 struct nvme_ns {
473 	struct list_head list;
474 
475 	struct nvme_ctrl *ctrl;
476 	struct request_queue *queue;
477 	struct gendisk *disk;
478 #ifdef CONFIG_NVME_MULTIPATH
479 	enum nvme_ana_state ana_state;
480 	u32 ana_grpid;
481 #endif
482 	struct list_head siblings;
483 	struct kref kref;
484 	struct nvme_ns_head *head;
485 
486 	int lba_shift;
487 	u16 ms;
488 	u16 pi_size;
489 	u16 sgs;
490 	u32 sws;
491 	u8 pi_type;
492 	u8 guard_type;
493 #ifdef CONFIG_BLK_DEV_ZONED
494 	u64 zsze;
495 #endif
496 	unsigned long features;
497 	unsigned long flags;
498 #define NVME_NS_REMOVING	0
499 #define NVME_NS_ANA_PENDING	2
500 #define NVME_NS_FORCE_RO	3
501 #define NVME_NS_READY		4
502 
503 	struct cdev		cdev;
504 	struct device		cdev_device;
505 
506 	struct nvme_fault_inject fault_inject;
507 
508 };
509 
510 /* NVMe ns supports metadata actions by the controller (generate/strip) */
511 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
512 {
513 	return ns->pi_type && ns->ms == ns->pi_size;
514 }
515 
516 struct nvme_ctrl_ops {
517 	const char *name;
518 	struct module *module;
519 	unsigned int flags;
520 #define NVME_F_FABRICS			(1 << 0)
521 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
522 #define NVME_F_BLOCKING			(1 << 2)
523 
524 	const struct attribute_group **dev_attr_groups;
525 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
526 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
527 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
528 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
529 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
530 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
531 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
532 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
533 	void (*print_device_info)(struct nvme_ctrl *ctrl);
534 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
535 };
536 
537 /*
538  * nvme command_id is constructed as such:
539  * | xxxx | xxxxxxxxxxxx |
540  *   gen    request tag
541  */
542 #define nvme_genctr_mask(gen)			(gen & 0xf)
543 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
544 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
545 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
546 
547 static inline u16 nvme_cid(struct request *rq)
548 {
549 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
550 }
551 
552 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
553 		u16 command_id)
554 {
555 	u8 genctr = nvme_genctr_from_cid(command_id);
556 	u16 tag = nvme_tag_from_cid(command_id);
557 	struct request *rq;
558 
559 	rq = blk_mq_tag_to_rq(tags, tag);
560 	if (unlikely(!rq)) {
561 		pr_err("could not locate request for tag %#x\n",
562 			tag);
563 		return NULL;
564 	}
565 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
566 		dev_err(nvme_req(rq)->ctrl->device,
567 			"request %#x genctr mismatch (got %#x expected %#x)\n",
568 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
569 		return NULL;
570 	}
571 	return rq;
572 }
573 
574 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
575                 u16 command_id)
576 {
577 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
578 }
579 
580 /*
581  * Return the length of the string without the space padding
582  */
583 static inline int nvme_strlen(char *s, int len)
584 {
585 	while (s[len - 1] == ' ')
586 		len--;
587 	return len;
588 }
589 
590 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
591 {
592 	struct nvme_subsystem *subsys = ctrl->subsys;
593 
594 	if (ctrl->ops->print_device_info) {
595 		ctrl->ops->print_device_info(ctrl);
596 		return;
597 	}
598 
599 	dev_err(ctrl->device,
600 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
601 		nvme_strlen(subsys->model, sizeof(subsys->model)),
602 		subsys->model, nvme_strlen(subsys->firmware_rev,
603 					   sizeof(subsys->firmware_rev)),
604 		subsys->firmware_rev);
605 }
606 
607 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
608 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
609 			    const char *dev_name);
610 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
611 void nvme_should_fail(struct request *req);
612 #else
613 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
614 					  const char *dev_name)
615 {
616 }
617 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
618 {
619 }
620 static inline void nvme_should_fail(struct request *req) {}
621 #endif
622 
623 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
624 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
625 
626 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
627 {
628 	int ret;
629 
630 	if (!ctrl->subsystem)
631 		return -ENOTTY;
632 	if (!nvme_wait_reset(ctrl))
633 		return -EBUSY;
634 
635 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
636 	if (ret)
637 		return ret;
638 
639 	return nvme_try_sched_reset(ctrl);
640 }
641 
642 /*
643  * Convert a 512B sector number to a device logical block number.
644  */
645 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
646 {
647 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
648 }
649 
650 /*
651  * Convert a device logical block number to a 512B sector number.
652  */
653 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
654 {
655 	return lba << (ns->lba_shift - SECTOR_SHIFT);
656 }
657 
658 /*
659  * Convert byte length to nvme's 0-based num dwords
660  */
661 static inline u32 nvme_bytes_to_numd(size_t len)
662 {
663 	return (len >> 2) - 1;
664 }
665 
666 static inline bool nvme_is_ana_error(u16 status)
667 {
668 	switch (status & 0x7ff) {
669 	case NVME_SC_ANA_TRANSITION:
670 	case NVME_SC_ANA_INACCESSIBLE:
671 	case NVME_SC_ANA_PERSISTENT_LOSS:
672 		return true;
673 	default:
674 		return false;
675 	}
676 }
677 
678 static inline bool nvme_is_path_error(u16 status)
679 {
680 	/* check for a status code type of 'path related status' */
681 	return (status & 0x700) == 0x300;
682 }
683 
684 /*
685  * Fill in the status and result information from the CQE, and then figure out
686  * if blk-mq will need to use IPI magic to complete the request, and if yes do
687  * so.  If not let the caller complete the request without an indirect function
688  * call.
689  */
690 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
691 		union nvme_result result)
692 {
693 	struct nvme_request *rq = nvme_req(req);
694 	struct nvme_ctrl *ctrl = rq->ctrl;
695 
696 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
697 		rq->genctr++;
698 
699 	rq->status = le16_to_cpu(status) >> 1;
700 	rq->result = result;
701 	/* inject error when permitted by fault injection framework */
702 	nvme_should_fail(req);
703 	if (unlikely(blk_should_fake_timeout(req->q)))
704 		return true;
705 	return blk_mq_complete_request_remote(req);
706 }
707 
708 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
709 {
710 	get_device(ctrl->device);
711 }
712 
713 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
714 {
715 	put_device(ctrl->device);
716 }
717 
718 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
719 {
720 	return !qid &&
721 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
722 }
723 
724 void nvme_complete_rq(struct request *req);
725 void nvme_complete_batch_req(struct request *req);
726 
727 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
728 						void (*fn)(struct request *rq))
729 {
730 	struct request *req;
731 
732 	rq_list_for_each(&iob->req_list, req) {
733 		fn(req);
734 		nvme_complete_batch_req(req);
735 	}
736 	blk_mq_end_request_batch(iob);
737 }
738 
739 blk_status_t nvme_host_path_error(struct request *req);
740 bool nvme_cancel_request(struct request *req, void *data);
741 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
742 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
743 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
744 		enum nvme_ctrl_state new_state);
745 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
746 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
747 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
748 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
749 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
750 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
751 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
752 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
753 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
754 		const struct blk_mq_ops *ops, unsigned int cmd_size);
755 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
756 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
757 		const struct blk_mq_ops *ops, unsigned int nr_maps,
758 		unsigned int cmd_size);
759 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
760 
761 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
762 
763 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
764 		volatile union nvme_result *res);
765 
766 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
767 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
768 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
769 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
770 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
771 void nvme_sync_queues(struct nvme_ctrl *ctrl);
772 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
773 void nvme_unfreeze(struct nvme_ctrl *ctrl);
774 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
775 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
776 void nvme_start_freeze(struct nvme_ctrl *ctrl);
777 
778 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
779 {
780 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
781 }
782 
783 #define NVME_QID_ANY -1
784 void nvme_init_request(struct request *req, struct nvme_command *cmd);
785 void nvme_cleanup_cmd(struct request *req);
786 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
787 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
788 		struct request *req);
789 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
790 		bool queue_live);
791 
792 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
793 		bool queue_live)
794 {
795 	if (likely(ctrl->state == NVME_CTRL_LIVE))
796 		return true;
797 	if (ctrl->ops->flags & NVME_F_FABRICS &&
798 	    ctrl->state == NVME_CTRL_DELETING)
799 		return queue_live;
800 	return __nvme_check_ready(ctrl, rq, queue_live);
801 }
802 
803 /*
804  * NSID shall be unique for all shared namespaces, or if at least one of the
805  * following conditions is met:
806  *   1. Namespace Management is supported by the controller
807  *   2. ANA is supported by the controller
808  *   3. NVM Set are supported by the controller
809  *
810  * In other case, private namespace are not required to report a unique NSID.
811  */
812 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
813 		struct nvme_ns_head *head)
814 {
815 	return head->shared ||
816 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
817 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
818 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
819 }
820 
821 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
822 		void *buf, unsigned bufflen);
823 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
824 		union nvme_result *result, void *buffer, unsigned bufflen,
825 		int qid, int at_head,
826 		blk_mq_req_flags_t flags);
827 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
828 		      unsigned int dword11, void *buffer, size_t buflen,
829 		      u32 *result);
830 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
831 		      unsigned int dword11, void *buffer, size_t buflen,
832 		      u32 *result);
833 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
834 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
835 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
836 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
837 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
838 void nvme_queue_scan(struct nvme_ctrl *ctrl);
839 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
840 		void *log, size_t size, u64 offset);
841 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
842 void nvme_put_ns_head(struct nvme_ns_head *head);
843 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
844 		const struct file_operations *fops, struct module *owner);
845 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
846 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
847 		unsigned int cmd, unsigned long arg);
848 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
849 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
850 		unsigned int cmd, unsigned long arg);
851 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
852 		unsigned long arg);
853 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
854 		unsigned long arg);
855 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
856 		struct io_comp_batch *iob, unsigned int poll_flags);
857 int nvme_ns_head_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
858 		struct io_comp_batch *iob, unsigned int poll_flags);
859 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
860 		unsigned int issue_flags);
861 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
862 		unsigned int issue_flags);
863 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
864 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
865 
866 extern const struct attribute_group *nvme_ns_id_attr_groups[];
867 extern const struct pr_ops nvme_pr_ops;
868 extern const struct block_device_operations nvme_ns_head_ops;
869 extern const struct attribute_group nvme_dev_attrs_group;
870 extern const struct attribute_group *nvme_subsys_attrs_groups[];
871 extern const struct attribute_group *nvme_dev_attr_groups[];
872 extern const struct block_device_operations nvme_bdev_ops;
873 
874 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
875 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
876 #ifdef CONFIG_NVME_MULTIPATH
877 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
878 {
879 	return ctrl->ana_log_buf != NULL;
880 }
881 
882 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
883 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
884 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
885 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
886 void nvme_failover_req(struct request *req);
887 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
888 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
889 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
890 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
891 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
892 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
893 void nvme_mpath_update(struct nvme_ctrl *ctrl);
894 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
895 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
896 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
897 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
898 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
899 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
900 void nvme_mpath_start_request(struct request *rq);
901 void nvme_mpath_end_request(struct request *rq);
902 
903 static inline void nvme_trace_bio_complete(struct request *req)
904 {
905 	struct nvme_ns *ns = req->q->queuedata;
906 
907 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
908 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
909 }
910 
911 extern bool multipath;
912 extern struct device_attribute dev_attr_ana_grpid;
913 extern struct device_attribute dev_attr_ana_state;
914 extern struct device_attribute subsys_attr_iopolicy;
915 
916 #else
917 #define multipath false
918 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
919 {
920 	return false;
921 }
922 static inline void nvme_failover_req(struct request *req)
923 {
924 }
925 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
926 {
927 }
928 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
929 		struct nvme_ns_head *head)
930 {
931 	return 0;
932 }
933 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
934 {
935 }
936 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
937 {
938 }
939 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
940 {
941 	return false;
942 }
943 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
944 {
945 }
946 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
947 {
948 }
949 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
950 {
951 }
952 static inline void nvme_trace_bio_complete(struct request *req)
953 {
954 }
955 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
956 {
957 }
958 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
959 		struct nvme_id_ctrl *id)
960 {
961 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
962 		dev_warn(ctrl->device,
963 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
964 	return 0;
965 }
966 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
967 {
968 }
969 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
970 {
971 }
972 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
973 {
974 }
975 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
976 {
977 }
978 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
979 {
980 }
981 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
982 {
983 }
984 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
985 {
986 }
987 static inline void nvme_mpath_start_request(struct request *rq)
988 {
989 }
990 static inline void nvme_mpath_end_request(struct request *rq)
991 {
992 }
993 #endif /* CONFIG_NVME_MULTIPATH */
994 
995 int nvme_revalidate_zones(struct nvme_ns *ns);
996 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
997 		unsigned int nr_zones, report_zones_cb cb, void *data);
998 #ifdef CONFIG_BLK_DEV_ZONED
999 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1000 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1001 				       struct nvme_command *cmnd,
1002 				       enum nvme_zone_mgmt_action action);
1003 #else
1004 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1005 		struct request *req, struct nvme_command *cmnd,
1006 		enum nvme_zone_mgmt_action action)
1007 {
1008 	return BLK_STS_NOTSUPP;
1009 }
1010 
1011 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1012 {
1013 	dev_warn(ns->ctrl->device,
1014 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1015 	return -EPROTONOSUPPORT;
1016 }
1017 #endif
1018 
1019 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1020 {
1021 	return dev_to_disk(dev)->private_data;
1022 }
1023 
1024 #ifdef CONFIG_NVME_HWMON
1025 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1026 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1027 #else
1028 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1029 {
1030 	return 0;
1031 }
1032 
1033 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1034 {
1035 }
1036 #endif
1037 
1038 static inline void nvme_start_request(struct request *rq)
1039 {
1040 	if (rq->cmd_flags & REQ_NVME_MPATH)
1041 		nvme_mpath_start_request(rq);
1042 	blk_mq_start_request(rq);
1043 }
1044 
1045 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1046 {
1047 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1048 }
1049 
1050 #ifdef CONFIG_NVME_AUTH
1051 int __init nvme_init_auth(void);
1052 void __exit nvme_exit_auth(void);
1053 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1054 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1055 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1056 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1057 void nvme_auth_free(struct nvme_ctrl *ctrl);
1058 #else
1059 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1060 {
1061 	return 0;
1062 }
1063 static inline int __init nvme_init_auth(void)
1064 {
1065 	return 0;
1066 }
1067 static inline void __exit nvme_exit_auth(void)
1068 {
1069 }
1070 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1071 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1072 {
1073 	return -EPROTONOSUPPORT;
1074 }
1075 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1076 {
1077 	return NVME_SC_AUTH_REQUIRED;
1078 }
1079 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1080 #endif
1081 
1082 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1083 			 u8 opcode);
1084 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1085 int nvme_execute_rq(struct request *rq, bool at_head);
1086 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1087 		       struct nvme_command *cmd, int status);
1088 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1089 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1090 void nvme_put_ns(struct nvme_ns *ns);
1091 
1092 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1093 {
1094 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1095 }
1096 
1097 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1098 const unsigned char *nvme_get_error_status_str(u16 status);
1099 const unsigned char *nvme_get_opcode_str(u8 opcode);
1100 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1101 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1102 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1103 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1104 {
1105 	return "I/O Error";
1106 }
1107 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1108 {
1109 	return "I/O Cmd";
1110 }
1111 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1112 {
1113 	return "Admin Cmd";
1114 }
1115 
1116 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1117 {
1118 	return "Fabrics Cmd";
1119 }
1120 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1121 
1122 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1123 {
1124 	if (opcode == nvme_fabrics_command)
1125 		return nvme_get_fabrics_opcode_str(fctype);
1126 	return qid ? nvme_get_opcode_str(opcode) :
1127 		nvme_get_admin_opcode_str(opcode);
1128 }
1129 #endif /* _NVME_H */
1130