xref: /openbmc/linux/drivers/nvme/host/nvme.h (revision 5cd17f0e)
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 	 * Disables simple suspend/resume path.
162 	 */
163 	NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND	= (1 << 20),
164 
165 	/*
166 	 * MSI (but not MSI-X) interrupts are broken and never fire.
167 	 */
168 	NVME_QUIRK_BROKEN_MSI			= (1 << 21),
169 };
170 
171 /*
172  * Common request structure for NVMe passthrough.  All drivers must have
173  * this structure as the first member of their request-private data.
174  */
175 struct nvme_request {
176 	struct nvme_command	*cmd;
177 	union nvme_result	result;
178 	u8			genctr;
179 	u8			retries;
180 	u8			flags;
181 	u16			status;
182 #ifdef CONFIG_NVME_MULTIPATH
183 	unsigned long		start_time;
184 #endif
185 	struct nvme_ctrl	*ctrl;
186 };
187 
188 /*
189  * Mark a bio as coming in through the mpath node.
190  */
191 #define REQ_NVME_MPATH		REQ_DRV
192 
193 enum {
194 	NVME_REQ_CANCELLED		= (1 << 0),
195 	NVME_REQ_USERCMD		= (1 << 1),
196 	NVME_MPATH_IO_STATS		= (1 << 2),
197 };
198 
199 static inline struct nvme_request *nvme_req(struct request *req)
200 {
201 	return blk_mq_rq_to_pdu(req);
202 }
203 
204 static inline u16 nvme_req_qid(struct request *req)
205 {
206 	if (!req->q->queuedata)
207 		return 0;
208 
209 	return req->mq_hctx->queue_num + 1;
210 }
211 
212 /* The below value is the specific amount of delay needed before checking
213  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
214  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
215  * found empirically.
216  */
217 #define NVME_QUIRK_DELAY_AMOUNT		2300
218 
219 /*
220  * enum nvme_ctrl_state: Controller state
221  *
222  * @NVME_CTRL_NEW:		New controller just allocated, initial state
223  * @NVME_CTRL_LIVE:		Controller is connected and I/O capable
224  * @NVME_CTRL_RESETTING:	Controller is resetting (or scheduled reset)
225  * @NVME_CTRL_CONNECTING:	Controller is disconnected, now connecting the
226  *				transport
227  * @NVME_CTRL_DELETING:		Controller is deleting (or scheduled deletion)
228  * @NVME_CTRL_DELETING_NOIO:	Controller is deleting and I/O is not
229  *				disabled/failed immediately. This state comes
230  * 				after all async event processing took place and
231  * 				before ns removal and the controller deletion
232  * 				progress
233  * @NVME_CTRL_DEAD:		Controller is non-present/unresponsive during
234  *				shutdown or removal. In this case we forcibly
235  *				kill all inflight I/O as they have no chance to
236  *				complete
237  */
238 enum nvme_ctrl_state {
239 	NVME_CTRL_NEW,
240 	NVME_CTRL_LIVE,
241 	NVME_CTRL_RESETTING,
242 	NVME_CTRL_CONNECTING,
243 	NVME_CTRL_DELETING,
244 	NVME_CTRL_DELETING_NOIO,
245 	NVME_CTRL_DEAD,
246 };
247 
248 struct nvme_fault_inject {
249 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
250 	struct fault_attr attr;
251 	struct dentry *parent;
252 	bool dont_retry;	/* DNR, do not retry */
253 	u16 status;		/* status code */
254 #endif
255 };
256 
257 enum nvme_ctrl_flags {
258 	NVME_CTRL_FAILFAST_EXPIRED	= 0,
259 	NVME_CTRL_ADMIN_Q_STOPPED	= 1,
260 	NVME_CTRL_STARTED_ONCE		= 2,
261 	NVME_CTRL_STOPPED		= 3,
262 	NVME_CTRL_SKIP_ID_CNS_CS	= 4,
263 	NVME_CTRL_DIRTY_CAPABILITY	= 5,
264 	NVME_CTRL_FROZEN		= 6,
265 };
266 
267 struct nvme_ctrl {
268 	bool comp_seen;
269 	bool identified;
270 	enum nvme_ctrl_state state;
271 	spinlock_t lock;
272 	struct mutex scan_lock;
273 	const struct nvme_ctrl_ops *ops;
274 	struct request_queue *admin_q;
275 	struct request_queue *connect_q;
276 	struct request_queue *fabrics_q;
277 	struct device *dev;
278 	int instance;
279 	int numa_node;
280 	struct blk_mq_tag_set *tagset;
281 	struct blk_mq_tag_set *admin_tagset;
282 	struct list_head namespaces;
283 	struct rw_semaphore namespaces_rwsem;
284 	struct device ctrl_device;
285 	struct device *device;	/* char device */
286 #ifdef CONFIG_NVME_HWMON
287 	struct device *hwmon_device;
288 #endif
289 	struct cdev cdev;
290 	struct work_struct reset_work;
291 	struct work_struct delete_work;
292 	wait_queue_head_t state_wq;
293 
294 	struct nvme_subsystem *subsys;
295 	struct list_head subsys_entry;
296 
297 	struct opal_dev *opal_dev;
298 
299 	char name[12];
300 	u16 cntlid;
301 
302 	u16 mtfa;
303 	u32 ctrl_config;
304 	u32 queue_count;
305 
306 	u64 cap;
307 	u32 max_hw_sectors;
308 	u32 max_segments;
309 	u32 max_integrity_segments;
310 	u32 max_discard_sectors;
311 	u32 max_discard_segments;
312 	u32 max_zeroes_sectors;
313 #ifdef CONFIG_BLK_DEV_ZONED
314 	u32 max_zone_append;
315 #endif
316 	u16 crdt[3];
317 	u16 oncs;
318 	u32 dmrsl;
319 	u16 oacs;
320 	u16 sqsize;
321 	u32 max_namespaces;
322 	atomic_t abort_limit;
323 	u8 vwc;
324 	u32 vs;
325 	u32 sgls;
326 	u16 kas;
327 	u8 npss;
328 	u8 apsta;
329 	u16 wctemp;
330 	u16 cctemp;
331 	u32 oaes;
332 	u32 aen_result;
333 	u32 ctratt;
334 	unsigned int shutdown_timeout;
335 	unsigned int kato;
336 	bool subsystem;
337 	unsigned long quirks;
338 	struct nvme_id_power_state psd[32];
339 	struct nvme_effects_log *effects;
340 	struct xarray cels;
341 	struct work_struct scan_work;
342 	struct work_struct async_event_work;
343 	struct delayed_work ka_work;
344 	struct delayed_work failfast_work;
345 	struct nvme_command ka_cmd;
346 	unsigned long ka_last_check_time;
347 	struct work_struct fw_act_work;
348 	unsigned long events;
349 
350 #ifdef CONFIG_NVME_MULTIPATH
351 	/* asymmetric namespace access: */
352 	u8 anacap;
353 	u8 anatt;
354 	u32 anagrpmax;
355 	u32 nanagrpid;
356 	struct mutex ana_lock;
357 	struct nvme_ana_rsp_hdr *ana_log_buf;
358 	size_t ana_log_size;
359 	struct timer_list anatt_timer;
360 	struct work_struct ana_work;
361 #endif
362 
363 #ifdef CONFIG_NVME_AUTH
364 	struct work_struct dhchap_auth_work;
365 	struct mutex dhchap_auth_mutex;
366 	struct nvme_dhchap_queue_context *dhchap_ctxs;
367 	struct nvme_dhchap_key *host_key;
368 	struct nvme_dhchap_key *ctrl_key;
369 	u16 transaction;
370 #endif
371 
372 	/* Power saving configuration */
373 	u64 ps_max_latency_us;
374 	bool apst_enabled;
375 
376 	/* PCIe only: */
377 	u16 hmmaxd;
378 	u32 hmpre;
379 	u32 hmmin;
380 	u32 hmminds;
381 
382 	/* Fabrics only */
383 	u32 ioccsz;
384 	u32 iorcsz;
385 	u16 icdoff;
386 	u16 maxcmd;
387 	int nr_reconnects;
388 	unsigned long flags;
389 	struct nvmf_ctrl_options *opts;
390 
391 	struct page *discard_page;
392 	unsigned long discard_page_busy;
393 
394 	struct nvme_fault_inject fault_inject;
395 
396 	enum nvme_ctrl_type cntrltype;
397 	enum nvme_dctype dctype;
398 };
399 
400 static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
401 {
402 	return READ_ONCE(ctrl->state);
403 }
404 
405 enum nvme_iopolicy {
406 	NVME_IOPOLICY_NUMA,
407 	NVME_IOPOLICY_RR,
408 };
409 
410 struct nvme_subsystem {
411 	int			instance;
412 	struct device		dev;
413 	/*
414 	 * Because we unregister the device on the last put we need
415 	 * a separate refcount.
416 	 */
417 	struct kref		ref;
418 	struct list_head	entry;
419 	struct mutex		lock;
420 	struct list_head	ctrls;
421 	struct list_head	nsheads;
422 	char			subnqn[NVMF_NQN_SIZE];
423 	char			serial[20];
424 	char			model[40];
425 	char			firmware_rev[8];
426 	u8			cmic;
427 	enum nvme_subsys_type	subtype;
428 	u16			vendor_id;
429 	u16			awupf;	/* 0's based awupf value. */
430 	struct ida		ns_ida;
431 #ifdef CONFIG_NVME_MULTIPATH
432 	enum nvme_iopolicy	iopolicy;
433 #endif
434 };
435 
436 /*
437  * Container structure for uniqueue namespace identifiers.
438  */
439 struct nvme_ns_ids {
440 	u8	eui64[8];
441 	u8	nguid[16];
442 	uuid_t	uuid;
443 	u8	csi;
444 };
445 
446 /*
447  * Anchor structure for namespaces.  There is one for each namespace in a
448  * NVMe subsystem that any of our controllers can see, and the namespace
449  * structure for each controller is chained of it.  For private namespaces
450  * there is a 1:1 relation to our namespace structures, that is ->list
451  * only ever has a single entry for private namespaces.
452  */
453 struct nvme_ns_head {
454 	struct list_head	list;
455 	struct srcu_struct      srcu;
456 	struct nvme_subsystem	*subsys;
457 	unsigned		ns_id;
458 	struct nvme_ns_ids	ids;
459 	struct list_head	entry;
460 	struct kref		ref;
461 	bool			shared;
462 	int			instance;
463 	struct nvme_effects_log *effects;
464 
465 	struct cdev		cdev;
466 	struct device		cdev_device;
467 
468 	struct gendisk		*disk;
469 #ifdef CONFIG_NVME_MULTIPATH
470 	struct bio_list		requeue_list;
471 	spinlock_t		requeue_lock;
472 	struct work_struct	requeue_work;
473 	struct mutex		lock;
474 	unsigned long		flags;
475 #define NVME_NSHEAD_DISK_LIVE	0
476 	struct nvme_ns __rcu	*current_path[];
477 #endif
478 };
479 
480 static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
481 {
482 	return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
483 }
484 
485 enum nvme_ns_features {
486 	NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
487 	NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
488 	NVME_NS_DEAC,		/* DEAC bit in Write Zeores supported */
489 };
490 
491 struct nvme_ns {
492 	struct list_head list;
493 
494 	struct nvme_ctrl *ctrl;
495 	struct request_queue *queue;
496 	struct gendisk *disk;
497 #ifdef CONFIG_NVME_MULTIPATH
498 	enum nvme_ana_state ana_state;
499 	u32 ana_grpid;
500 #endif
501 	struct list_head siblings;
502 	struct kref kref;
503 	struct nvme_ns_head *head;
504 
505 	int lba_shift;
506 	u16 ms;
507 	u16 pi_size;
508 	u16 sgs;
509 	u32 sws;
510 	u8 pi_type;
511 	u8 guard_type;
512 #ifdef CONFIG_BLK_DEV_ZONED
513 	u64 zsze;
514 #endif
515 	unsigned long features;
516 	unsigned long flags;
517 #define NVME_NS_REMOVING	0
518 #define NVME_NS_ANA_PENDING	2
519 #define NVME_NS_FORCE_RO	3
520 #define NVME_NS_READY		4
521 
522 	struct cdev		cdev;
523 	struct device		cdev_device;
524 
525 	struct nvme_fault_inject fault_inject;
526 
527 };
528 
529 /* NVMe ns supports metadata actions by the controller (generate/strip) */
530 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
531 {
532 	return ns->pi_type && ns->ms == ns->pi_size;
533 }
534 
535 struct nvme_ctrl_ops {
536 	const char *name;
537 	struct module *module;
538 	unsigned int flags;
539 #define NVME_F_FABRICS			(1 << 0)
540 #define NVME_F_METADATA_SUPPORTED	(1 << 1)
541 #define NVME_F_BLOCKING			(1 << 2)
542 
543 	const struct attribute_group **dev_attr_groups;
544 	int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
545 	int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
546 	int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
547 	void (*free_ctrl)(struct nvme_ctrl *ctrl);
548 	void (*submit_async_event)(struct nvme_ctrl *ctrl);
549 	void (*delete_ctrl)(struct nvme_ctrl *ctrl);
550 	void (*stop_ctrl)(struct nvme_ctrl *ctrl);
551 	int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
552 	void (*print_device_info)(struct nvme_ctrl *ctrl);
553 	bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
554 };
555 
556 /*
557  * nvme command_id is constructed as such:
558  * | xxxx | xxxxxxxxxxxx |
559  *   gen    request tag
560  */
561 #define nvme_genctr_mask(gen)			(gen & 0xf)
562 #define nvme_cid_install_genctr(gen)		(nvme_genctr_mask(gen) << 12)
563 #define nvme_genctr_from_cid(cid)		((cid & 0xf000) >> 12)
564 #define nvme_tag_from_cid(cid)			(cid & 0xfff)
565 
566 static inline u16 nvme_cid(struct request *rq)
567 {
568 	return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
569 }
570 
571 static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
572 		u16 command_id)
573 {
574 	u8 genctr = nvme_genctr_from_cid(command_id);
575 	u16 tag = nvme_tag_from_cid(command_id);
576 	struct request *rq;
577 
578 	rq = blk_mq_tag_to_rq(tags, tag);
579 	if (unlikely(!rq)) {
580 		pr_err("could not locate request for tag %#x\n",
581 			tag);
582 		return NULL;
583 	}
584 	if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
585 		dev_err(nvme_req(rq)->ctrl->device,
586 			"request %#x genctr mismatch (got %#x expected %#x)\n",
587 			tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
588 		return NULL;
589 	}
590 	return rq;
591 }
592 
593 static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
594                 u16 command_id)
595 {
596 	return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
597 }
598 
599 /*
600  * Return the length of the string without the space padding
601  */
602 static inline int nvme_strlen(char *s, int len)
603 {
604 	while (s[len - 1] == ' ')
605 		len--;
606 	return len;
607 }
608 
609 static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
610 {
611 	struct nvme_subsystem *subsys = ctrl->subsys;
612 
613 	if (ctrl->ops->print_device_info) {
614 		ctrl->ops->print_device_info(ctrl);
615 		return;
616 	}
617 
618 	dev_err(ctrl->device,
619 		"VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
620 		nvme_strlen(subsys->model, sizeof(subsys->model)),
621 		subsys->model, nvme_strlen(subsys->firmware_rev,
622 					   sizeof(subsys->firmware_rev)),
623 		subsys->firmware_rev);
624 }
625 
626 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
627 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
628 			    const char *dev_name);
629 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
630 void nvme_should_fail(struct request *req);
631 #else
632 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
633 					  const char *dev_name)
634 {
635 }
636 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
637 {
638 }
639 static inline void nvme_should_fail(struct request *req) {}
640 #endif
641 
642 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
643 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
644 
645 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
646 {
647 	int ret;
648 
649 	if (!ctrl->subsystem)
650 		return -ENOTTY;
651 	if (!nvme_wait_reset(ctrl))
652 		return -EBUSY;
653 
654 	ret = ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
655 	if (ret)
656 		return ret;
657 
658 	return nvme_try_sched_reset(ctrl);
659 }
660 
661 /*
662  * Convert a 512B sector number to a device logical block number.
663  */
664 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
665 {
666 	return sector >> (ns->lba_shift - SECTOR_SHIFT);
667 }
668 
669 /*
670  * Convert a device logical block number to a 512B sector number.
671  */
672 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
673 {
674 	return lba << (ns->lba_shift - SECTOR_SHIFT);
675 }
676 
677 /*
678  * Convert byte length to nvme's 0-based num dwords
679  */
680 static inline u32 nvme_bytes_to_numd(size_t len)
681 {
682 	return (len >> 2) - 1;
683 }
684 
685 static inline bool nvme_is_ana_error(u16 status)
686 {
687 	switch (status & 0x7ff) {
688 	case NVME_SC_ANA_TRANSITION:
689 	case NVME_SC_ANA_INACCESSIBLE:
690 	case NVME_SC_ANA_PERSISTENT_LOSS:
691 		return true;
692 	default:
693 		return false;
694 	}
695 }
696 
697 static inline bool nvme_is_path_error(u16 status)
698 {
699 	/* check for a status code type of 'path related status' */
700 	return (status & 0x700) == 0x300;
701 }
702 
703 /*
704  * Fill in the status and result information from the CQE, and then figure out
705  * if blk-mq will need to use IPI magic to complete the request, and if yes do
706  * so.  If not let the caller complete the request without an indirect function
707  * call.
708  */
709 static inline bool nvme_try_complete_req(struct request *req, __le16 status,
710 		union nvme_result result)
711 {
712 	struct nvme_request *rq = nvme_req(req);
713 	struct nvme_ctrl *ctrl = rq->ctrl;
714 
715 	if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
716 		rq->genctr++;
717 
718 	rq->status = le16_to_cpu(status) >> 1;
719 	rq->result = result;
720 	/* inject error when permitted by fault injection framework */
721 	nvme_should_fail(req);
722 	if (unlikely(blk_should_fake_timeout(req->q)))
723 		return true;
724 	return blk_mq_complete_request_remote(req);
725 }
726 
727 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
728 {
729 	get_device(ctrl->device);
730 }
731 
732 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
733 {
734 	put_device(ctrl->device);
735 }
736 
737 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
738 {
739 	return !qid &&
740 		nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
741 }
742 
743 void nvme_complete_rq(struct request *req);
744 void nvme_complete_batch_req(struct request *req);
745 
746 static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
747 						void (*fn)(struct request *rq))
748 {
749 	struct request *req;
750 
751 	rq_list_for_each(&iob->req_list, req) {
752 		fn(req);
753 		nvme_complete_batch_req(req);
754 	}
755 	blk_mq_end_request_batch(iob);
756 }
757 
758 blk_status_t nvme_host_path_error(struct request *req);
759 bool nvme_cancel_request(struct request *req, void *data);
760 void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
761 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
762 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
763 		enum nvme_ctrl_state new_state);
764 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
765 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
766 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
767 		const struct nvme_ctrl_ops *ops, unsigned long quirks);
768 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
769 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
770 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
771 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
772 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
773 		const struct blk_mq_ops *ops, unsigned int cmd_size);
774 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
775 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
776 		const struct blk_mq_ops *ops, unsigned int nr_maps,
777 		unsigned int cmd_size);
778 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
779 
780 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
781 
782 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
783 		volatile union nvme_result *res);
784 
785 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
786 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
787 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
788 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
789 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
790 void nvme_sync_queues(struct nvme_ctrl *ctrl);
791 void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
792 void nvme_unfreeze(struct nvme_ctrl *ctrl);
793 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
794 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
795 void nvme_start_freeze(struct nvme_ctrl *ctrl);
796 
797 static inline enum req_op nvme_req_op(struct nvme_command *cmd)
798 {
799 	return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
800 }
801 
802 #define NVME_QID_ANY -1
803 void nvme_init_request(struct request *req, struct nvme_command *cmd);
804 void nvme_cleanup_cmd(struct request *req);
805 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
806 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
807 		struct request *req);
808 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
809 		bool queue_live);
810 
811 static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
812 		bool queue_live)
813 {
814 	if (likely(ctrl->state == NVME_CTRL_LIVE))
815 		return true;
816 	if (ctrl->ops->flags & NVME_F_FABRICS &&
817 	    ctrl->state == NVME_CTRL_DELETING)
818 		return queue_live;
819 	return __nvme_check_ready(ctrl, rq, queue_live);
820 }
821 
822 /*
823  * NSID shall be unique for all shared namespaces, or if at least one of the
824  * following conditions is met:
825  *   1. Namespace Management is supported by the controller
826  *   2. ANA is supported by the controller
827  *   3. NVM Set are supported by the controller
828  *
829  * In other case, private namespace are not required to report a unique NSID.
830  */
831 static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
832 		struct nvme_ns_head *head)
833 {
834 	return head->shared ||
835 		(ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
836 		(ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
837 		(ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
838 }
839 
840 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
841 		void *buf, unsigned bufflen);
842 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
843 		union nvme_result *result, void *buffer, unsigned bufflen,
844 		int qid, int at_head,
845 		blk_mq_req_flags_t flags);
846 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
847 		      unsigned int dword11, void *buffer, size_t buflen,
848 		      u32 *result);
849 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
850 		      unsigned int dword11, void *buffer, size_t buflen,
851 		      u32 *result);
852 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
853 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
854 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
855 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
856 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
857 void nvme_queue_scan(struct nvme_ctrl *ctrl);
858 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
859 		void *log, size_t size, u64 offset);
860 bool nvme_tryget_ns_head(struct nvme_ns_head *head);
861 void nvme_put_ns_head(struct nvme_ns_head *head);
862 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
863 		const struct file_operations *fops, struct module *owner);
864 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
865 int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
866 		unsigned int cmd, unsigned long arg);
867 long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
868 int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
869 		unsigned int cmd, unsigned long arg);
870 long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
871 		unsigned long arg);
872 long nvme_dev_ioctl(struct file *file, unsigned int cmd,
873 		unsigned long arg);
874 int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
875 		struct io_comp_batch *iob, unsigned int poll_flags);
876 int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
877 		unsigned int issue_flags);
878 int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
879 		unsigned int issue_flags);
880 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
881 int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
882 
883 extern const struct attribute_group *nvme_ns_id_attr_groups[];
884 extern const struct pr_ops nvme_pr_ops;
885 extern const struct block_device_operations nvme_ns_head_ops;
886 extern const struct attribute_group nvme_dev_attrs_group;
887 extern const struct attribute_group *nvme_subsys_attrs_groups[];
888 extern const struct attribute_group *nvme_dev_attr_groups[];
889 extern const struct block_device_operations nvme_bdev_ops;
890 
891 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
892 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
893 #ifdef CONFIG_NVME_MULTIPATH
894 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
895 {
896 	return ctrl->ana_log_buf != NULL;
897 }
898 
899 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
900 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
901 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
902 void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
903 void nvme_failover_req(struct request *req);
904 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
905 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
906 void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
907 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
908 int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
909 void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
910 void nvme_mpath_update(struct nvme_ctrl *ctrl);
911 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
912 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
913 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
914 void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
915 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
916 void nvme_mpath_shutdown_disk(struct nvme_ns_head *head);
917 void nvme_mpath_start_request(struct request *rq);
918 void nvme_mpath_end_request(struct request *rq);
919 
920 static inline void nvme_trace_bio_complete(struct request *req)
921 {
922 	struct nvme_ns *ns = req->q->queuedata;
923 
924 	if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
925 		trace_block_bio_complete(ns->head->disk->queue, req->bio);
926 }
927 
928 extern bool multipath;
929 extern struct device_attribute dev_attr_ana_grpid;
930 extern struct device_attribute dev_attr_ana_state;
931 extern struct device_attribute subsys_attr_iopolicy;
932 
933 #else
934 #define multipath false
935 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
936 {
937 	return false;
938 }
939 static inline void nvme_failover_req(struct request *req)
940 {
941 }
942 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
943 {
944 }
945 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
946 		struct nvme_ns_head *head)
947 {
948 	return 0;
949 }
950 static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
951 {
952 }
953 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
954 {
955 }
956 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
957 {
958 	return false;
959 }
960 static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
961 {
962 }
963 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
964 {
965 }
966 static inline void nvme_mpath_shutdown_disk(struct nvme_ns_head *head)
967 {
968 }
969 static inline void nvme_trace_bio_complete(struct request *req)
970 {
971 }
972 static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
973 {
974 }
975 static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
976 		struct nvme_id_ctrl *id)
977 {
978 	if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
979 		dev_warn(ctrl->device,
980 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
981 	return 0;
982 }
983 static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
984 {
985 }
986 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
987 {
988 }
989 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
990 {
991 }
992 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
993 {
994 }
995 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
996 {
997 }
998 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
999 {
1000 }
1001 static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1002 {
1003 }
1004 static inline void nvme_mpath_start_request(struct request *rq)
1005 {
1006 }
1007 static inline void nvme_mpath_end_request(struct request *rq)
1008 {
1009 }
1010 #endif /* CONFIG_NVME_MULTIPATH */
1011 
1012 int nvme_revalidate_zones(struct nvme_ns *ns);
1013 int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1014 		unsigned int nr_zones, report_zones_cb cb, void *data);
1015 #ifdef CONFIG_BLK_DEV_ZONED
1016 int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf);
1017 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1018 				       struct nvme_command *cmnd,
1019 				       enum nvme_zone_mgmt_action action);
1020 #else
1021 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1022 		struct request *req, struct nvme_command *cmnd,
1023 		enum nvme_zone_mgmt_action action)
1024 {
1025 	return BLK_STS_NOTSUPP;
1026 }
1027 
1028 static inline int nvme_update_zone_info(struct nvme_ns *ns, unsigned lbaf)
1029 {
1030 	dev_warn(ns->ctrl->device,
1031 		 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
1032 	return -EPROTONOSUPPORT;
1033 }
1034 #endif
1035 
1036 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1037 {
1038 	return dev_to_disk(dev)->private_data;
1039 }
1040 
1041 #ifdef CONFIG_NVME_HWMON
1042 int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1043 void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1044 #else
1045 static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1046 {
1047 	return 0;
1048 }
1049 
1050 static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1051 {
1052 }
1053 #endif
1054 
1055 static inline void nvme_start_request(struct request *rq)
1056 {
1057 	if (rq->cmd_flags & REQ_NVME_MPATH)
1058 		nvme_mpath_start_request(rq);
1059 	blk_mq_start_request(rq);
1060 }
1061 
1062 static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1063 {
1064 	return ctrl->sgls & ((1 << 0) | (1 << 1));
1065 }
1066 
1067 #ifdef CONFIG_NVME_AUTH
1068 int __init nvme_init_auth(void);
1069 void __exit nvme_exit_auth(void);
1070 int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1071 void nvme_auth_stop(struct nvme_ctrl *ctrl);
1072 int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1073 int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1074 void nvme_auth_free(struct nvme_ctrl *ctrl);
1075 #else
1076 static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1077 {
1078 	return 0;
1079 }
1080 static inline int __init nvme_init_auth(void)
1081 {
1082 	return 0;
1083 }
1084 static inline void __exit nvme_exit_auth(void)
1085 {
1086 }
1087 static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1088 static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1089 {
1090 	return -EPROTONOSUPPORT;
1091 }
1092 static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1093 {
1094 	return NVME_SC_AUTH_REQUIRED;
1095 }
1096 static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1097 #endif
1098 
1099 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1100 			 u8 opcode);
1101 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1102 int nvme_execute_rq(struct request *rq, bool at_head);
1103 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1104 		       struct nvme_command *cmd, int status);
1105 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1106 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1107 void nvme_put_ns(struct nvme_ns *ns);
1108 
1109 static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1110 {
1111 	return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1112 }
1113 
1114 #ifdef CONFIG_NVME_VERBOSE_ERRORS
1115 const unsigned char *nvme_get_error_status_str(u16 status);
1116 const unsigned char *nvme_get_opcode_str(u8 opcode);
1117 const unsigned char *nvme_get_admin_opcode_str(u8 opcode);
1118 const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode);
1119 #else /* CONFIG_NVME_VERBOSE_ERRORS */
1120 static inline const unsigned char *nvme_get_error_status_str(u16 status)
1121 {
1122 	return "I/O Error";
1123 }
1124 static inline const unsigned char *nvme_get_opcode_str(u8 opcode)
1125 {
1126 	return "I/O Cmd";
1127 }
1128 static inline const unsigned char *nvme_get_admin_opcode_str(u8 opcode)
1129 {
1130 	return "Admin Cmd";
1131 }
1132 
1133 static inline const unsigned char *nvme_get_fabrics_opcode_str(u8 opcode)
1134 {
1135 	return "Fabrics Cmd";
1136 }
1137 #endif /* CONFIG_NVME_VERBOSE_ERRORS */
1138 
1139 static inline const unsigned char *nvme_opcode_str(int qid, u8 opcode, u8 fctype)
1140 {
1141 	if (opcode == nvme_fabrics_command)
1142 		return nvme_get_fabrics_opcode_str(fctype);
1143 	return qid ? nvme_get_opcode_str(opcode) :
1144 		nvme_get_admin_opcode_str(opcode);
1145 }
1146 #endif /* _NVME_H */
1147