xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 404e077a)
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 const struct pr_ops nvme_pr_ops;
23 
24 extern unsigned int nvme_io_timeout;
25 #define NVME_IO_TIMEOUT	(nvme_io_timeout * HZ)
26 
27 extern 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 
48 extern struct workqueue_struct *nvme_wq;
49 extern struct workqueue_struct *nvme_reset_wq;
50 extern struct workqueue_struct *nvme_delete_wq;
51 
52 /*
53  * List of workarounds for devices that required behavior not specified in
54  * the standard.
55  */
56 enum 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 /*
162  * Common request structure for NVMe passthrough.  All drivers must have
163  * this structure as the first member of their request-private data.
164  */
165 struct nvme_request {
166 	struct nvme_command	*cmd;
167 	union nvme_result	result;
168 	u8			genctr;
169 	u8			retries;
170 	u8			flags;
171 	u16			status;
172 #ifdef CONFIG_NVME_MULTIPATH
173 	unsigned long		start_time;
174 #endif
175 	struct nvme_ctrl	*ctrl;
176 };
177 
178 /*
179  * Mark a bio as coming in through the mpath node.
180  */
181 #define REQ_NVME_MPATH		REQ_DRV
182 
183 enum {
184 	NVME_REQ_CANCELLED		= (1 << 0),
185 	NVME_REQ_USERCMD		= (1 << 1),
186 	NVME_MPATH_IO_STATS		= (1 << 2),
187 };
188 
189 static inline struct nvme_request *nvme_req(struct request *req)
190 {
191 	return blk_mq_rq_to_pdu(req);
192 }
193 
194 static inline u16 nvme_req_qid(struct request *req)
195 {
196 	if (!req->q->queuedata)
197 		return 0;
198 
199 	return req->mq_hctx->queue_num + 1;
200 }
201 
202 /* The below value is the specific amount of delay needed before checking
203  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
204  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
205  * found empirically.
206  */
207 #define NVME_QUIRK_DELAY_AMOUNT		2300
208 
209 /*
210  * enum nvme_ctrl_state: Controller state
211  *
212  * @NVME_CTRL_NEW:		New controller just allocated, initial state
213  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
214  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
215  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
216  *				transport
217  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
218  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
219  *				disabled/failed immediately. This state comes
220  * 				after all async event processing took place and
221  * 				before ns removal and the controller deletion
222  * 				progress
223  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
224  *				shutdown or removal. In this case we forcibly
225  *				kill all inflight I/O as they have no chance to
226  *				complete
227  */
228 enum nvme_ctrl_state {
229 	NVME_CTRL_NEW,
230 	NVME_CTRL_LIVE,
231 	NVME_CTRL_RESETTING,
232 	NVME_CTRL_CONNECTING,
233 	NVME_CTRL_DELETING,
234 	NVME_CTRL_DELETING_NOIO,
235 	NVME_CTRL_DEAD,
236 };
237 
238 struct nvme_fault_inject {
239 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
240 	struct fault_attr attr;
241 	struct dentry *parent;
242 	bool dont_retry;	/* DNR, do not retry */
243 	u16 status;		/* status code */
244 #endif
245 };
246 
247 enum nvme_ctrl_flags {
248 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
249 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
250 	NVME_CTRL_STARTED_ONCE		= 2,
251 	NVME_CTRL_STOPPED		= 3,
252 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
253 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
254 };
255 
256 struct nvme_ctrl {
257 	bool comp_seen;
258 	bool identified;
259 	enum nvme_ctrl_state state;
260 	spinlock_t lock;
261 	struct mutex scan_lock;
262 	const struct nvme_ctrl_ops *ops;
263 	struct request_queue *admin_q;
264 	struct request_queue *connect_q;
265 	struct request_queue *fabrics_q;
266 	struct device *dev;
267 	int instance;
268 	int numa_node;
269 	struct blk_mq_tag_set *tagset;
270 	struct blk_mq_tag_set *admin_tagset;
271 	struct list_head namespaces;
272 	struct rw_semaphore namespaces_rwsem;
273 	struct device ctrl_device;
274 	struct device *device;	/* char device */
275 #ifdef CONFIG_NVME_HWMON
276 	struct device *hwmon_device;
277 #endif
278 	struct cdev cdev;
279 	struct work_struct reset_work;
280 	struct work_struct delete_work;
281 	wait_queue_head_t state_wq;
282 
283 	struct nvme_subsystem *subsys;
284 	struct list_head subsys_entry;
285 
286 	struct opal_dev *opal_dev;
287 
288 	char name[12];
289 	u16 cntlid;
290 
291 	u16 mtfa;
292 	u32 ctrl_config;
293 	u32 queue_count;
294 
295 	u64 cap;
296 	u32 max_hw_sectors;
297 	u32 max_segments;
298 	u32 max_integrity_segments;
299 	u32 max_discard_sectors;
300 	u32 max_discard_segments;
301 	u32 max_zeroes_sectors;
302 #ifdef CONFIG_BLK_DEV_ZONED
303 	u32 max_zone_append;
304 #endif
305 	u16 crdt[3];
306 	u16 oncs;
307 	u32 dmrsl;
308 	u16 oacs;
309 	u16 sqsize;
310 	u32 max_namespaces;
311 	atomic_t abort_limit;
312 	u8 vwc;
313 	u32 vs;
314 	u32 sgls;
315 	u16 kas;
316 	u8 npss;
317 	u8 apsta;
318 	u16 wctemp;
319 	u16 cctemp;
320 	u32 oaes;
321 	u32 aen_result;
322 	u32 ctratt;
323 	unsigned int shutdown_timeout;
324 	unsigned int kato;
325 	bool subsystem;
326 	unsigned long quirks;
327 	struct nvme_id_power_state psd[32];
328 	struct nvme_effects_log *effects;
329 	struct xarray cels;
330 	struct work_struct scan_work;
331 	struct work_struct async_event_work;
332 	struct delayed_work ka_work;
333 	struct delayed_work failfast_work;
334 	struct nvme_command ka_cmd;
335 	unsigned long ka_last_check_time;
336 	struct work_struct fw_act_work;
337 	unsigned long events;
338 
339 #ifdef CONFIG_NVME_MULTIPATH
340 	/* asymmetric namespace access: */
341 	u8 anacap;
342 	u8 anatt;
343 	u32 anagrpmax;
344 	u32 nanagrpid;
345 	struct mutex ana_lock;
346 	struct nvme_ana_rsp_hdr *ana_log_buf;
347 	size_t ana_log_size;
348 	struct timer_list anatt_timer;
349 	struct work_struct ana_work;
350 #endif
351 
352 #ifdef CONFIG_NVME_AUTH
353 	struct work_struct dhchap_auth_work;
354 	struct mutex dhchap_auth_mutex;
355 	struct nvme_dhchap_queue_context *dhchap_ctxs;
356 	struct nvme_dhchap_key *host_key;
357 	struct nvme_dhchap_key *ctrl_key;
358 	u16 transaction;
359 #endif
360 
361 	/* Power saving configuration */
362 	u64 ps_max_latency_us;
363 	bool apst_enabled;
364 
365 	/* PCIe only: */
366 	u16 hmmaxd;
367 	u32 hmpre;
368 	u32 hmmin;
369 	u32 hmminds;
370 
371 	/* Fabrics only */
372 	u32 ioccsz;
373 	u32 iorcsz;
374 	u16 icdoff;
375 	u16 maxcmd;
376 	int nr_reconnects;
377 	unsigned long flags;
378 	struct nvmf_ctrl_options *opts;
379 
380 	struct page *discard_page;
381 	unsigned long discard_page_busy;
382 
383 	struct nvme_fault_inject fault_inject;
384 
385 	enum nvme_ctrl_type cntrltype;
386 	enum nvme_dctype dctype;
387 };
388 
389 enum nvme_iopolicy {
390 	NVME_IOPOLICY_NUMA,
391 	NVME_IOPOLICY_RR,
392 };
393 
394 struct nvme_subsystem {
395 	int			instance;
396 	struct device		dev;
397 	/*
398 	 * Because we unregister the device on the last put we need
399 	 * a separate refcount.
400 	 */
401 	struct kref		ref;
402 	struct list_head	entry;
403 	struct mutex		lock;
404 	struct list_head	ctrls;
405 	struct list_head	nsheads;
406 	char			subnqn[NVMF_NQN_SIZE];
407 	char			serial[20];
408 	char			model[40];
409 	char			firmware_rev[8];
410 	u8			cmic;
411 	enum nvme_subsys_type	subtype;
412 	u16			vendor_id;
413 	u16			awupf;	/* 0's based awupf value. */
414 	struct ida		ns_ida;
415 #ifdef CONFIG_NVME_MULTIPATH
416 	enum nvme_iopolicy	iopolicy;
417 #endif
418 };
419 
420 /*
421  * Container structure for uniqueue namespace identifiers.
422  */
423 struct nvme_ns_ids {
424 	u8	eui64[8];
425 	u8	nguid[16];
426 	uuid_t	uuid;
427 	u8	csi;
428 };
429 
430 /*
431  * Anchor structure for namespaces.  There is one for each namespace in a
432  * NVMe subsystem that any of our controllers can see, and the namespace
433  * structure for each controller is chained of it.  For private namespaces
434  * there is a 1:1 relation to our namespace structures, that is ->list
435  * only ever has a single entry for private namespaces.
436  */
437 struct nvme_ns_head {
438 	struct list_head	list;
439 	struct srcu_struct      srcu;
440 	struct nvme_subsystem	*subsys;
441 	unsigned		ns_id;
442 	struct nvme_ns_ids	ids;
443 	struct list_head	entry;
444 	struct kref		ref;
445 	bool			shared;
446 	int			instance;
447 	struct nvme_effects_log *effects;
448 
449 	struct cdev		cdev;
450 	struct device		cdev_device;
451 
452 	struct gendisk		*disk;
453 #ifdef CONFIG_NVME_MULTIPATH
454 	struct bio_list		requeue_list;
455 	spinlock_t		requeue_lock;
456 	struct work_struct	requeue_work;
457 	struct mutex		lock;
458 	unsigned long		flags;
459 #define NVME_NSHEAD_DISK_LIVE	0
460 	struct nvme_ns __rcu	*current_path[];
461 #endif
462 };
463 
464 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
465 {
466 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
467 }
468 
469 enum nvme_ns_features {
470 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
471 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
472 	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
473 };
474 
475 struct nvme_ns {
476 	struct list_head list;
477 
478 	struct nvme_ctrl *ctrl;
479 	struct request_queue *queue;
480 	struct gendisk *disk;
481 #ifdef CONFIG_NVME_MULTIPATH
482 	enum nvme_ana_state ana_state;
483 	u32 ana_grpid;
484 #endif
485 	struct list_head siblings;
486 	struct kref kref;
487 	struct nvme_ns_head *head;
488 
489 	int lba_shift;
490 	u16 ms;
491 	u16 pi_size;
492 	u16 sgs;
493 	u32 sws;
494 	u8 pi_type;
495 	u8 guard_type;
496 #ifdef CONFIG_BLK_DEV_ZONED
497 	u64 zsze;
498 #endif
499 	unsigned long features;
500 	unsigned long flags;
501 #define NVME_NS_REMOVING	0
502 #define NVME_NS_ANA_PENDING	2
503 #define NVME_NS_FORCE_RO	3
504 #define NVME_NS_READY		4
505 
506 	struct cdev		cdev;
507 	struct device		cdev_device;
508 
509 	struct nvme_fault_inject fault_inject;
510 
511 };
512 
513 /* NVMe ns supports metadata actions by the controller (generate/strip) */
514 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
515 {
516 	return ns->pi_type && ns->ms == ns->pi_size;
517 }
518 
519 struct nvme_ctrl_ops {
520 	const char *name;
521 	struct module *module;
522 	unsigned int flags;
523 #define NVME_F_FABRICS			(1 << 0)
524 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
525 #define NVME_F_BLOCKING			(1 << 2)
526 
527 	const struct attribute_group **dev_attr_groups;
528 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
529 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
530 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
531 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
532 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
533 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
534 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
535 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
536 	void (*print_device_info)(struct nvme_ctrl *ctrl);
537 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
538 };
539 
540 /*
541  * nvme command_id is constructed as such:
542  * | xxxx | xxxxxxxxxxxx |
543  *   gen    request tag
544  */
545 #define nvme_genctr_mask(gen)			(gen & 0xf)
546 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
547 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
548 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
549 
550 static inline u16 nvme_cid(struct request *rq)
551 {
552 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
553 }
554 
555 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
556 		u16 command_id)
557 {
558 	u8 genctr = nvme_genctr_from_cid(command_id);
559 	u16 tag = nvme_tag_from_cid(command_id);
560 	struct request *rq;
561 
562 	rq = blk_mq_tag_to_rq(tags, tag);
563 	if (unlikely(!rq)) {
564 		pr_err("could not locate request for tag %#x\n",
565 			tag);
566 		return NULL;
567 	}
568 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
569 		dev_err(nvme_req(rq)->ctrl->device,
570 			"request %#x genctr mismatch (got %#x expected %#x)\n",
571 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
572 		return NULL;
573 	}
574 	return rq;
575 }
576 
577 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
578                 u16 command_id)
579 {
580 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
581 }
582 
583 /*
584  * Return the length of the string without the space padding
585  */
586 static inline int nvme_strlen(char *s, int len)
587 {
588 	while (s[len - 1] == ' ')
589 		len--;
590 	return len;
591 }
592 
593 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
594 {
595 	struct nvme_subsystem *subsys = ctrl->subsys;
596 
597 	if (ctrl->ops->print_device_info) {
598 		ctrl->ops->print_device_info(ctrl);
599 		return;
600 	}
601 
602 	dev_err(ctrl->device,
603 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
604 		nvme_strlen(subsys->model, sizeof(subsys->model)),
605 		subsys->model, nvme_strlen(subsys->firmware_rev,
606 					   sizeof(subsys->firmware_rev)),
607 		subsys->firmware_rev);
608 }
609 
610 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
611 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
612 			    const char *dev_name);
613 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
614 void nvme_should_fail(struct request *req);
615 #else
616 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
617 					  const char *dev_name)
618 {
619 }
620 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
621 {
622 }
623 static inline void nvme_should_fail(struct request *req) {}
624 #endif
625 
626 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
627 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
628 
629 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
630 {
631 	int ret;
632 
633 	if (!ctrl->subsystem)
634 		return -ENOTTY;
635 	if (!nvme_wait_reset(ctrl))
636 		return -EBUSY;
637 
638 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
639 	if (ret)
640 		return ret;
641 
642 	return nvme_try_sched_reset(ctrl);
643 }
644 
645 /*
646  * Convert a 512B sector number to a device logical block number.
647  */
648 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
649 {
650 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
651 }
652 
653 /*
654  * Convert a device logical block number to a 512B sector number.
655  */
656 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
657 {
658 	return lba << (ns->lba_shift - SECTOR_SHIFT);
659 }
660 
661 /*
662  * Convert byte length to nvme's 0-based num dwords
663  */
664 static inline u32 nvme_bytes_to_numd(size_t len)
665 {
666 	return (len >> 2) - 1;
667 }
668 
669 static inline bool nvme_is_ana_error(u16 status)
670 {
671 	switch (status & 0x7ff) {
672 	case NVME_SC_ANA_TRANSITION:
673 	case NVME_SC_ANA_INACCESSIBLE:
674 	case NVME_SC_ANA_PERSISTENT_LOSS:
675 		return true;
676 	default:
677 		return false;
678 	}
679 }
680 
681 static inline bool nvme_is_path_error(u16 status)
682 {
683 	/* check for a status code type of 'path related status' */
684 	return (status & 0x700) == 0x300;
685 }
686 
687 /*
688  * Fill in the status and result information from the CQE, and then figure out
689  * if blk-mq will need to use IPI magic to complete the request, and if yes do
690  * so.  If not let the caller complete the request without an indirect function
691  * call.
692  */
693 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
694 		union nvme_result result)
695 {
696 	struct nvme_request *rq = nvme_req(req);
697 	struct nvme_ctrl *ctrl = rq->ctrl;
698 
699 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
700 		rq->genctr++;
701 
702 	rq->status = le16_to_cpu(status) >> 1;
703 	rq->result = result;
704 	/* inject error when permitted by fault injection framework */
705 	nvme_should_fail(req);
706 	if (unlikely(blk_should_fake_timeout(req->q)))
707 		return true;
708 	return blk_mq_complete_request_remote(req);
709 }
710 
711 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
712 {
713 	get_device(ctrl->device);
714 }
715 
716 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
717 {
718 	put_device(ctrl->device);
719 }
720 
721 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
722 {
723 	return !qid &&
724 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
725 }
726 
727 void nvme_complete_rq(struct request *req);
728 void nvme_complete_batch_req(struct request *req);
729 
730 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
731 						void (*fn)(struct request *rq))
732 {
733 	struct request *req;
734 
735 	rq_list_for_each(&iob->req_list, req) {
736 		fn(req);
737 		nvme_complete_batch_req(req);
738 	}
739 	blk_mq_end_request_batch(iob);
740 }
741 
742 blk_status_t nvme_host_path_error(struct request *req);
743 bool nvme_cancel_request(struct request *req, void *data);
744 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
745 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
746 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
747 		enum nvme_ctrl_state new_state);
748 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
749 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
750 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
751 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
752 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
753 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
754 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
755 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
756 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
757 		const struct blk_mq_ops *ops, unsigned int cmd_size);
758 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
759 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
760 		const struct blk_mq_ops *ops, unsigned int nr_maps,
761 		unsigned int cmd_size);
762 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
763 
764 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
765 
766 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
767 		volatile union nvme_result *res);
768 
769 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
770 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
771 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
772 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
773 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
774 void nvme_sync_queues(struct nvme_ctrl *ctrl);
775 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
776 void nvme_unfreeze(struct nvme_ctrl *ctrl);
777 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
778 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
779 void nvme_start_freeze(struct nvme_ctrl *ctrl);
780 
781 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
782 {
783 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
784 }
785 
786 #define NVME_QID_ANY -1
787 void nvme_init_request(struct request *req, struct nvme_command *cmd);
788 void nvme_cleanup_cmd(struct request *req);
789 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
790 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
791 		struct request *req);
792 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
793 		bool queue_live);
794 
795 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
796 		bool queue_live)
797 {
798 	if (likely(ctrl->state == NVME_CTRL_LIVE))
799 		return true;
800 	if (ctrl->ops->flags & NVME_F_FABRICS &&
801 	    ctrl->state == NVME_CTRL_DELETING)
802 		return queue_live;
803 	return __nvme_check_ready(ctrl, rq, queue_live);
804 }
805 
806 /*
807  * NSID shall be unique for all shared namespaces, or if at least one of the
808  * following conditions is met:
809  *   1. Namespace Management is supported by the controller
810  *   2. ANA is supported by the controller
811  *   3. NVM Set are supported by the controller
812  *
813  * In other case, private namespace are not required to report a unique NSID.
814  */
815 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
816 		struct nvme_ns_head *head)
817 {
818 	return head->shared ||
819 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
820 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
821 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
822 }
823 
824 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
825 		void *buf, unsigned bufflen);
826 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
827 		union nvme_result *result, void *buffer, unsigned bufflen,
828 		int qid, int at_head,
829 		blk_mq_req_flags_t flags);
830 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
831 		      unsigned int dword11, void *buffer, size_t buflen,
832 		      u32 *result);
833 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
834 		      unsigned int dword11, void *buffer, size_t buflen,
835 		      u32 *result);
836 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
837 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
838 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
839 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
840 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
841 void nvme_queue_scan(struct nvme_ctrl *ctrl);
842 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
843 		void *log, size_t size, u64 offset);
844 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
845 void nvme_put_ns_head(struct nvme_ns_head *head);
846 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
847 		const struct file_operations *fops, struct module *owner);
848 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
849 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
850 		unsigned int cmd, unsigned long arg);
851 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
852 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
853 		unsigned int cmd, unsigned long arg);
854 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
855 		unsigned long arg);
856 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
857 		unsigned long arg);
858 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
859 		struct io_comp_batch *iob, unsigned int poll_flags);
860 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
861 		unsigned int issue_flags);
862 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
863 		unsigned int issue_flags);
864 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
865 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
866 
867 extern const struct attribute_group *nvme_ns_id_attr_groups[];
868 extern const struct pr_ops nvme_pr_ops;
869 extern const struct block_device_operations nvme_ns_head_ops;
870 extern const struct attribute_group nvme_dev_attrs_group;
871 extern const struct attribute_group *nvme_subsys_attrs_groups[];
872 extern const struct attribute_group *nvme_dev_attr_groups[];
873 extern const struct block_device_operations nvme_bdev_ops;
874 
875 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
876 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
877 #ifdef CONFIG_NVME_MULTIPATH
878 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
879 {
880 	return ctrl->ana_log_buf != NULL;
881 }
882 
883 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
884 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
885 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
886 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
887 void nvme_failover_req(struct request *req);
888 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
889 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
890 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
891 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
892 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
893 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
894 void nvme_mpath_update(struct nvme_ctrl *ctrl);
895 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
896 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
897 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
898 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
899 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
900 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
901 void nvme_mpath_start_request(struct request *rq);
902 void nvme_mpath_end_request(struct request *rq);
903 
904 static inline void nvme_trace_bio_complete(struct request *req)
905 {
906 	struct nvme_ns *ns = req->q->queuedata;
907 
908 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
909 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
910 }
911 
912 extern bool multipath;
913 extern struct device_attribute dev_attr_ana_grpid;
914 extern struct device_attribute dev_attr_ana_state;
915 extern struct device_attribute subsys_attr_iopolicy;
916 
917 #else
918 #define multipath false
919 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
920 {
921 	return false;
922 }
923 static inline void nvme_failover_req(struct request *req)
924 {
925 }
926 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
927 {
928 }
929 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
930 		struct nvme_ns_head *head)
931 {
932 	return 0;
933 }
934 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
935 {
936 }
937 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
938 {
939 }
940 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
941 {
942 	return false;
943 }
944 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
945 {
946 }
947 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
948 {
949 }
950 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
951 {
952 }
953 static inline void nvme_trace_bio_complete(struct request *req)
954 {
955 }
956 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
957 {
958 }
959 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
960 		struct nvme_id_ctrl *id)
961 {
962 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
963 		dev_warn(ctrl->device,
964 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
965 	return 0;
966 }
967 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
968 {
969 }
970 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
971 {
972 }
973 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
974 {
975 }
976 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
977 {
978 }
979 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
980 {
981 }
982 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
983 {
984 }
985 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
986 {
987 }
988 static inline void nvme_mpath_start_request(struct request *rq)
989 {
990 }
991 static inline void nvme_mpath_end_request(struct request *rq)
992 {
993 }
994 #endif /* CONFIG_NVME_MULTIPATH */
995 
996 int nvme_revalidate_zones(struct nvme_ns *ns);
997 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
998 		unsigned int nr_zones, report_zones_cb cb, void *data);
999 #ifdef CONFIG_BLK_DEV_ZONED
1000 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1001 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1002 				       struct nvme_command *cmnd,
1003 				       enum nvme_zone_mgmt_action action);
1004 #else
1005 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1006 		struct request *req, struct nvme_command *cmnd,
1007 		enum nvme_zone_mgmt_action action)
1008 {
1009 	return BLK_STS_NOTSUPP;
1010 }
1011 
1012 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1013 {
1014 	dev_warn(ns->ctrl->device,
1015 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1016 	return -EPROTONOSUPPORT;
1017 }
1018 #endif
1019 
1020 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1021 {
1022 	return dev_to_disk(dev)->private_data;
1023 }
1024 
1025 #ifdef CONFIG_NVME_HWMON
1026 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1027 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1028 #else
1029 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1030 {
1031 	return 0;
1032 }
1033 
1034 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1035 {
1036 }
1037 #endif
1038 
1039 static inline void nvme_start_request(struct request *rq)
1040 {
1041 	if (rq->cmd_flags & REQ_NVME_MPATH)
1042 		nvme_mpath_start_request(rq);
1043 	blk_mq_start_request(rq);
1044 }
1045 
1046 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1047 {
1048 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1049 }
1050 
1051 #ifdef CONFIG_NVME_AUTH
1052 int __init nvme_init_auth(void);
1053 void __exit nvme_exit_auth(void);
1054 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1055 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1056 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1057 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1058 void nvme_auth_free(struct nvme_ctrl *ctrl);
1059 #else
1060 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1061 {
1062 	return 0;
1063 }
1064 static inline int __init nvme_init_auth(void)
1065 {
1066 	return 0;
1067 }
1068 static inline void __exit nvme_exit_auth(void)
1069 {
1070 }
1071 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1072 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1073 {
1074 	return -EPROTONOSUPPORT;
1075 }
1076 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1077 {
1078 	return NVME_SC_AUTH_REQUIRED;
1079 }
1080 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1081 #endif
1082 
1083 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1084 			 u8 opcode);
1085 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1086 int nvme_execute_rq(struct request *rq, bool at_head);
1087 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1088 		       struct nvme_command *cmd, int status);
1089 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1090 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1091 void nvme_put_ns(struct nvme_ns *ns);
1092 
1093 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1094 {
1095 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1096 }
1097 
1098 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1099 const unsigned char *nvme_get_error_status_str(u16 status);
1100 const unsigned char *nvme_get_opcode_str(u8 opcode);
1101 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1102 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1103 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1104 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1105 {
1106 	return "I/O Error";
1107 }
1108 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1109 {
1110 	return "I/O Cmd";
1111 }
1112 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1113 {
1114 	return "Admin Cmd";
1115 }
1116 
1117 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1118 {
1119 	return "Fabrics Cmd";
1120 }
1121 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1122 
1123 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1124 {
1125 	if (opcode == nvme_fabrics_command)
1126 		return nvme_get_fabrics_opcode_str(fctype);
1127 	return qid ? nvme_get_opcode_str(opcode) :
1128 		nvme_get_admin_opcode_str(opcode);
1129 }
1130 #endif /* _NVME_H */
1131