projects / linux-2.6-block.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Merge tag 'perf-tools-fixes-for-v6.17-2025-09-16' of git://git.kernel.org/pub/scm...
[linux-2.6-block.git] / drivers / nvme / host / nvme.h
... / ...
CommitLineData
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#include <linux/ratelimit_types.h>
20
21#include <trace/events/block.h>
22
23extern const struct pr_ops nvme_pr_ops;
24
25extern unsigned int nvme_io_timeout;
26#define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
27
28extern unsigned int admin_timeout;
29#define NVME_ADMIN_TIMEOUT (admin_timeout * HZ)
30
31#define NVME_DEFAULT_KATO 5
32
33#ifdef CONFIG_ARCH_NO_SG_CHAIN
34#define NVME_INLINE_SG_CNT 0
35#define NVME_INLINE_METADATA_SG_CNT 0
36#else
37#define NVME_INLINE_SG_CNT 2
38#define NVME_INLINE_METADATA_SG_CNT 1
39#endif
40
41/*
42 * Default to a 4K page size, with the intention to update this
43 * path in the future to accommodate architectures with differing
44 * kernel and IO page sizes.
45 */
46#define NVME_CTRL_PAGE_SHIFT 12
47#define NVME_CTRL_PAGE_SIZE (1 << NVME_CTRL_PAGE_SHIFT)
48
49extern struct workqueue_struct *nvme_wq;
50extern struct workqueue_struct *nvme_reset_wq;
51extern struct workqueue_struct *nvme_delete_wq;
52extern struct mutex nvme_subsystems_lock;
53
54/*
55 * List of workarounds for devices that required behavior not specified in
56 * the standard.
57 */
58enum nvme_quirks {
59 /*
60 * Prefers I/O aligned to a stripe size specified in a vendor
61 * specific Identify field.
62 */
63 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
64
65 /*
66 * The controller doesn't handle Identify value others than 0 or 1
67 * correctly.
68 */
69 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
70
71 /*
72 * The controller deterministically returns 0's on reads to
73 * logical blocks that deallocate was called on.
74 */
75 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
76
77 /*
78 * The controller needs a delay before starts checking the device
79 * readiness, which is done by reading the NVME_CSTS_RDY bit.
80 */
81 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
82
83 /*
84 * APST should not be used.
85 */
86 NVME_QUIRK_NO_APST = (1 << 4),
87
88 /*
89 * The deepest sleep state should not be used.
90 */
91 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
92
93 /*
94 * Problems seen with concurrent commands
95 */
96 NVME_QUIRK_QDEPTH_ONE = (1 << 6),
97
98 /*
99 * Set MEDIUM priority on SQ creation
100 */
101 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
102
103 /*
104 * Ignore device provided subnqn.
105 */
106 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
107
108 /*
109 * Broken Write Zeroes.
110 */
111 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
112
113 /*
114 * Force simple suspend/resume path.
115 */
116 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
117
118 /*
119 * Use only one interrupt vector for all queues
120 */
121 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
122
123 /*
124 * Use non-standard 128 bytes SQEs.
125 */
126 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
127
128 /*
129 * Prevent tag overlap between queues
130 */
131 NVME_QUIRK_SHARED_TAGS = (1 << 13),
132
133 /*
134 * Don't change the value of the temperature threshold feature
135 */
136 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14),
137
138 /*
139 * The controller doesn't handle the Identify Namespace
140 * Identification Descriptor list subcommand despite claiming
141 * NVMe 1.3 compliance.
142 */
143 NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15),
144
145 /*
146 * The controller does not properly handle DMA addresses over
147 * 48 bits.
148 */
149 NVME_QUIRK_DMA_ADDRESS_BITS_48 = (1 << 16),
150
151 /*
152 * The controller requires the command_id value be limited, so skip
153 * encoding the generation sequence number.
154 */
155 NVME_QUIRK_SKIP_CID_GEN = (1 << 17),
156
157 /*
158 * Reports garbage in the namespace identifiers (eui64, nguid, uuid).
159 */
160 NVME_QUIRK_BOGUS_NID = (1 << 18),
161
162 /*
163 * No temperature thresholds for channels other than 0 (Composite).
164 */
165 NVME_QUIRK_NO_SECONDARY_TEMP_THRESH = (1 << 19),
166
167 /*
168 * Disables simple suspend/resume path.
169 */
170 NVME_QUIRK_FORCE_NO_SIMPLE_SUSPEND = (1 << 20),
171
172 /*
173 * MSI (but not MSI-X) interrupts are broken and never fire.
174 */
175 NVME_QUIRK_BROKEN_MSI = (1 << 21),
176
177 /*
178 * Align dma pool segment size to 512 bytes
179 */
180 NVME_QUIRK_DMAPOOL_ALIGN_512 = (1 << 22),
181};
182
183/*
184 * Common request structure for NVMe passthrough. All drivers must have
185 * this structure as the first member of their request-private data.
186 */
187struct nvme_request {
188 struct nvme_command *cmd;
189 union nvme_result result;
190 u8 genctr;
191 u8 retries;
192 u8 flags;
193 u16 status;
194#ifdef CONFIG_NVME_MULTIPATH
195 unsigned long start_time;
196#endif
197 struct nvme_ctrl *ctrl;
198};
199
200/*
201 * Mark a bio as coming in through the mpath node.
202 */
203#define REQ_NVME_MPATH REQ_DRV
204
205enum {
206 NVME_REQ_CANCELLED = (1 << 0),
207 NVME_REQ_USERCMD = (1 << 1),
208 NVME_MPATH_IO_STATS = (1 << 2),
209 NVME_MPATH_CNT_ACTIVE = (1 << 3),
210};
211
212static inline struct nvme_request *nvme_req(struct request *req)
213{
214 return blk_mq_rq_to_pdu(req);
215}
216
217static inline u16 nvme_req_qid(struct request *req)
218{
219 if (!req->q->queuedata)
220 return 0;
221
222 return req->mq_hctx->queue_num + 1;
223}
224
225/* The below value is the specific amount of delay needed before checking
226 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
227 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
228 * found empirically.
229 */
230#define NVME_QUIRK_DELAY_AMOUNT 2300
231
232/*
233 * enum nvme_ctrl_state: Controller state
234 *
235 * @NVME_CTRL_NEW: New controller just allocated, initial state
236 * @NVME_CTRL_LIVE: Controller is connected and I/O capable
237 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset)
238 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the
239 * transport
240 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion)
241 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not
242 * disabled/failed immediately. This state comes
243 * after all async event processing took place and
244 * before ns removal and the controller deletion
245 * progress
246 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during
247 * shutdown or removal. In this case we forcibly
248 * kill all inflight I/O as they have no chance to
249 * complete
250 */
251enum nvme_ctrl_state {
252 NVME_CTRL_NEW,
253 NVME_CTRL_LIVE,
254 NVME_CTRL_RESETTING,
255 NVME_CTRL_CONNECTING,
256 NVME_CTRL_DELETING,
257 NVME_CTRL_DELETING_NOIO,
258 NVME_CTRL_DEAD,
259};
260
261struct nvme_fault_inject {
262#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
263 struct fault_attr attr;
264 struct dentry *parent;
265 bool dont_retry; /* DNR, do not retry */
266 u16 status; /* status code */
267#endif
268};
269
270enum nvme_ctrl_flags {
271 NVME_CTRL_FAILFAST_EXPIRED = 0,
272 NVME_CTRL_ADMIN_Q_STOPPED = 1,
273 NVME_CTRL_STARTED_ONCE = 2,
274 NVME_CTRL_STOPPED = 3,
275 NVME_CTRL_SKIP_ID_CNS_CS = 4,
276 NVME_CTRL_DIRTY_CAPABILITY = 5,
277 NVME_CTRL_FROZEN = 6,
278};
279
280struct nvme_ctrl {
281 bool comp_seen;
282 bool identified;
283 bool passthru_err_log_enabled;
284 enum nvme_ctrl_state state;
285 spinlock_t lock;
286 struct mutex scan_lock;
287 const struct nvme_ctrl_ops *ops;
288 struct request_queue *admin_q;
289 struct request_queue *connect_q;
290 struct request_queue *fabrics_q;
291 struct device *dev;
292 int instance;
293 int numa_node;
294 struct blk_mq_tag_set *tagset;
295 struct blk_mq_tag_set *admin_tagset;
296 struct list_head namespaces;
297 struct mutex namespaces_lock;
298 struct srcu_struct srcu;
299 struct device ctrl_device;
300 struct device *device; /* char device */
301#ifdef CONFIG_NVME_HWMON
302 struct device *hwmon_device;
303#endif
304 struct cdev cdev;
305 struct work_struct reset_work;
306 struct work_struct delete_work;
307 wait_queue_head_t state_wq;
308
309 struct nvme_subsystem *subsys;
310 struct list_head subsys_entry;
311
312 struct opal_dev *opal_dev;
313
314 u16 cntlid;
315
316 u16 mtfa;
317 u32 ctrl_config;
318 u32 queue_count;
319
320 u64 cap;
321 u32 max_hw_sectors;
322 u32 max_segments;
323 u32 max_integrity_segments;
324 u32 max_zeroes_sectors;
325#ifdef CONFIG_BLK_DEV_ZONED
326 u32 max_zone_append;
327#endif
328 u16 crdt[3];
329 u16 oncs;
330 u8 dmrl;
331 u32 dmrsl;
332 u16 oacs;
333 u16 sqsize;
334 u32 max_namespaces;
335 atomic_t abort_limit;
336 u8 vwc;
337 u32 vs;
338 u32 sgls;
339 u16 kas;
340 u8 npss;
341 u8 apsta;
342 u16 wctemp;
343 u16 cctemp;
344 u32 oaes;
345 u32 aen_result;
346 u32 ctratt;
347 unsigned int shutdown_timeout;
348 unsigned int kato;
349 bool subsystem;
350 unsigned long quirks;
351 struct nvme_id_power_state psd[32];
352 struct nvme_effects_log *effects;
353 struct xarray cels;
354 struct work_struct scan_work;
355 struct work_struct async_event_work;
356 struct delayed_work ka_work;
357 struct delayed_work failfast_work;
358 struct nvme_command ka_cmd;
359 unsigned long ka_last_check_time;
360 struct work_struct fw_act_work;
361 unsigned long events;
362
363#ifdef CONFIG_NVME_MULTIPATH
364 /* asymmetric namespace access: */
365 u8 anacap;
366 u8 anatt;
367 u32 anagrpmax;
368 u32 nanagrpid;
369 struct mutex ana_lock;
370 struct nvme_ana_rsp_hdr *ana_log_buf;
371 size_t ana_log_size;
372 struct timer_list anatt_timer;
373 struct work_struct ana_work;
374 atomic_t nr_active;
375#endif
376
377#ifdef CONFIG_NVME_HOST_AUTH
378 struct work_struct dhchap_auth_work;
379 struct mutex dhchap_auth_mutex;
380 struct nvme_dhchap_queue_context *dhchap_ctxs;
381 struct nvme_dhchap_key *host_key;
382 struct nvme_dhchap_key *ctrl_key;
383 u16 transaction;
384#endif
385 key_serial_t tls_pskid;
386
387 /* Power saving configuration */
388 u64 ps_max_latency_us;
389 bool apst_enabled;
390
391 /* PCIe only: */
392 u16 hmmaxd;
393 u32 hmpre;
394 u32 hmmin;
395 u32 hmminds;
396
397 /* Fabrics only */
398 u32 ioccsz;
399 u32 iorcsz;
400 u16 icdoff;
401 u16 maxcmd;
402 int nr_reconnects;
403 unsigned long flags;
404 struct nvmf_ctrl_options *opts;
405
406 struct page *discard_page;
407 unsigned long discard_page_busy;
408
409 struct nvme_fault_inject fault_inject;
410
411 enum nvme_ctrl_type cntrltype;
412 enum nvme_dctype dctype;
413};
414
415static inline enum nvme_ctrl_state nvme_ctrl_state(struct nvme_ctrl *ctrl)
416{
417 return READ_ONCE(ctrl->state);
418}
419
420enum nvme_iopolicy {
421 NVME_IOPOLICY_NUMA,
422 NVME_IOPOLICY_RR,
423 NVME_IOPOLICY_QD,
424};
425
426struct nvme_subsystem {
427 int instance;
428 struct device dev;
429 /*
430 * Because we unregister the device on the last put we need
431 * a separate refcount.
432 */
433 struct kref ref;
434 struct list_head entry;
435 struct mutex lock;
436 struct list_head ctrls;
437 struct list_head nsheads;
438 char subnqn[NVMF_NQN_SIZE];
439 char serial[20];
440 char model[40];
441 char firmware_rev[8];
442 u8 cmic;
443 enum nvme_subsys_type subtype;
444 u16 vendor_id;
445 u16 awupf; /* 0's based value. */
446 struct ida ns_ida;
447#ifdef CONFIG_NVME_MULTIPATH
448 enum nvme_iopolicy iopolicy;
449#endif
450};
451
452/*
453 * Container structure for uniqueue namespace identifiers.
454 */
455struct nvme_ns_ids {
456 u8 eui64[8];
457 u8 nguid[16];
458 uuid_t uuid;
459 u8 csi;
460};
461
462/*
463 * Anchor structure for namespaces. There is one for each namespace in a
464 * NVMe subsystem that any of our controllers can see, and the namespace
465 * structure for each controller is chained of it. For private namespaces
466 * there is a 1:1 relation to our namespace structures, that is ->list
467 * only ever has a single entry for private namespaces.
468 */
469struct nvme_ns_head {
470 struct list_head list;
471 struct srcu_struct srcu;
472 struct nvme_subsystem *subsys;
473 struct nvme_ns_ids ids;
474 u8 lba_shift;
475 u16 ms;
476 u16 pi_size;
477 u8 pi_type;
478 u8 guard_type;
479 struct list_head entry;
480 struct kref ref;
481 bool shared;
482 bool rotational;
483 bool passthru_err_log_enabled;
484 struct nvme_effects_log *effects;
485 u64 nuse;
486 unsigned ns_id;
487 int instance;
488#ifdef CONFIG_BLK_DEV_ZONED
489 u64 zsze;
490#endif
491 unsigned long features;
492
493 struct ratelimit_state rs_nuse;
494
495 struct cdev cdev;
496 struct device cdev_device;
497
498 struct gendisk *disk;
499
500 u16 nr_plids;
501 u16 *plids;
502#ifdef CONFIG_NVME_MULTIPATH
503 struct bio_list requeue_list;
504 spinlock_t requeue_lock;
505 struct work_struct requeue_work;
506 struct work_struct partition_scan_work;
507 struct mutex lock;
508 unsigned long flags;
509 struct delayed_work remove_work;
510 unsigned int delayed_removal_secs;
511#define NVME_NSHEAD_DISK_LIVE 0
512#define NVME_NSHEAD_QUEUE_IF_NO_PATH 1
513 struct nvme_ns __rcu *current_path[];
514#endif
515};
516
517static inline bool nvme_ns_head_multipath(struct nvme_ns_head *head)
518{
519 return IS_ENABLED(CONFIG_NVME_MULTIPATH) && head->disk;
520}
521
522enum nvme_ns_features {
523 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
524 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
525 NVME_NS_DEAC = 1 << 2, /* DEAC bit in Write Zeroes supported */
526};
527
528struct nvme_ns {
529 struct list_head list;
530
531 struct nvme_ctrl *ctrl;
532 struct request_queue *queue;
533 struct gendisk *disk;
534#ifdef CONFIG_NVME_MULTIPATH
535 enum nvme_ana_state ana_state;
536 u32 ana_grpid;
537#endif
538 struct list_head siblings;
539 struct kref kref;
540 struct nvme_ns_head *head;
541
542 unsigned long flags;
543#define NVME_NS_REMOVING 0
544#define NVME_NS_ANA_PENDING 2
545#define NVME_NS_FORCE_RO 3
546#define NVME_NS_READY 4
547#define NVME_NS_SYSFS_ATTR_LINK 5
548
549 struct cdev cdev;
550 struct device cdev_device;
551
552 struct nvme_fault_inject fault_inject;
553};
554
555/* NVMe ns supports metadata actions by the controller (generate/strip) */
556static inline bool nvme_ns_has_pi(struct nvme_ns_head *head)
557{
558 return head->pi_type && head->ms == head->pi_size;
559}
560
561struct nvme_ctrl_ops {
562 const char *name;
563 struct module *module;
564 unsigned int flags;
565#define NVME_F_FABRICS (1 << 0)
566#define NVME_F_METADATA_SUPPORTED (1 << 1)
567#define NVME_F_BLOCKING (1 << 2)
568
569 const struct attribute_group **dev_attr_groups;
570 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
571 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
572 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
573 void (*free_ctrl)(struct nvme_ctrl *ctrl);
574 void (*submit_async_event)(struct nvme_ctrl *ctrl);
575 int (*subsystem_reset)(struct nvme_ctrl *ctrl);
576 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
577 void (*stop_ctrl)(struct nvme_ctrl *ctrl);
578 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
579 void (*print_device_info)(struct nvme_ctrl *ctrl);
580 bool (*supports_pci_p2pdma)(struct nvme_ctrl *ctrl);
581};
582
583/*
584 * nvme command_id is constructed as such:
585 * | xxxx | xxxxxxxxxxxx |
586 * gen request tag
587 */
588#define nvme_genctr_mask(gen) (gen & 0xf)
589#define nvme_cid_install_genctr(gen) (nvme_genctr_mask(gen) << 12)
590#define nvme_genctr_from_cid(cid) ((cid & 0xf000) >> 12)
591#define nvme_tag_from_cid(cid) (cid & 0xfff)
592
593static inline u16 nvme_cid(struct request *rq)
594{
595 return nvme_cid_install_genctr(nvme_req(rq)->genctr) | rq->tag;
596}
597
598static inline struct request *nvme_find_rq(struct blk_mq_tags *tags,
599 u16 command_id)
600{
601 u8 genctr = nvme_genctr_from_cid(command_id);
602 u16 tag = nvme_tag_from_cid(command_id);
603 struct request *rq;
604
605 rq = blk_mq_tag_to_rq(tags, tag);
606 if (unlikely(!rq)) {
607 pr_err("could not locate request for tag %#x\n",
608 tag);
609 return NULL;
610 }
611 if (unlikely(nvme_genctr_mask(nvme_req(rq)->genctr) != genctr)) {
612 dev_err(nvme_req(rq)->ctrl->device,
613 "request %#x genctr mismatch (got %#x expected %#x)\n",
614 tag, genctr, nvme_genctr_mask(nvme_req(rq)->genctr));
615 return NULL;
616 }
617 return rq;
618}
619
620static inline struct request *nvme_cid_to_rq(struct blk_mq_tags *tags,
621 u16 command_id)
622{
623 return blk_mq_tag_to_rq(tags, nvme_tag_from_cid(command_id));
624}
625
626/*
627 * Return the length of the string without the space padding
628 */
629static inline int nvme_strlen(char *s, int len)
630{
631 while (s[len - 1] == ' ')
632 len--;
633 return len;
634}
635
636static inline void nvme_print_device_info(struct nvme_ctrl *ctrl)
637{
638 struct nvme_subsystem *subsys = ctrl->subsys;
639
640 if (ctrl->ops->print_device_info) {
641 ctrl->ops->print_device_info(ctrl);
642 return;
643 }
644
645 dev_err(ctrl->device,
646 "VID:%04x model:%.*s firmware:%.*s\n", subsys->vendor_id,
647 nvme_strlen(subsys->model, sizeof(subsys->model)),
648 subsys->model, nvme_strlen(subsys->firmware_rev,
649 sizeof(subsys->firmware_rev)),
650 subsys->firmware_rev);
651}
652
653#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
654void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
655 const char *dev_name);
656void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
657void nvme_should_fail(struct request *req);
658#else
659static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
660 const char *dev_name)
661{
662}
663static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
664{
665}
666static inline void nvme_should_fail(struct request *req) {}
667#endif
668
669bool nvme_wait_reset(struct nvme_ctrl *ctrl);
670int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
671
672static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
673{
674 if (!ctrl->subsystem || !ctrl->ops->subsystem_reset)
675 return -ENOTTY;
676 return ctrl->ops->subsystem_reset(ctrl);
677}
678
679/*
680 * Convert a 512B sector number to a device logical block number.
681 */
682static inline u64 nvme_sect_to_lba(struct nvme_ns_head *head, sector_t sector)
683{
684 return sector >> (head->lba_shift - SECTOR_SHIFT);
685}
686
687/*
688 * Convert a device logical block number to a 512B sector number.
689 */
690static inline sector_t nvme_lba_to_sect(struct nvme_ns_head *head, u64 lba)
691{
692 return lba << (head->lba_shift - SECTOR_SHIFT);
693}
694
695/*
696 * Convert byte length to nvme's 0-based num dwords
697 */
698static inline u32 nvme_bytes_to_numd(size_t len)
699{
700 return (len >> 2) - 1;
701}
702
703static inline bool nvme_is_ana_error(u16 status)
704{
705 switch (status & NVME_SCT_SC_MASK) {
706 case NVME_SC_ANA_TRANSITION:
707 case NVME_SC_ANA_INACCESSIBLE:
708 case NVME_SC_ANA_PERSISTENT_LOSS:
709 return true;
710 default:
711 return false;
712 }
713}
714
715static inline bool nvme_is_path_error(u16 status)
716{
717 /* check for a status code type of 'path related status' */
718 return (status & NVME_SCT_MASK) == NVME_SCT_PATH;
719}
720
721/*
722 * Fill in the status and result information from the CQE, and then figure out
723 * if blk-mq will need to use IPI magic to complete the request, and if yes do
724 * so. If not let the caller complete the request without an indirect function
725 * call.
726 */
727static inline bool nvme_try_complete_req(struct request *req, __le16 status,
728 union nvme_result result)
729{
730 struct nvme_request *rq = nvme_req(req);
731 struct nvme_ctrl *ctrl = rq->ctrl;
732
733 if (!(ctrl->quirks & NVME_QUIRK_SKIP_CID_GEN))
734 rq->genctr++;
735
736 rq->status = le16_to_cpu(status) >> 1;
737 rq->result = result;
738 /* inject error when permitted by fault injection framework */
739 nvme_should_fail(req);
740 if (unlikely(blk_should_fake_timeout(req->q)))
741 return true;
742 return blk_mq_complete_request_remote(req);
743}
744
745static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
746{
747 get_device(ctrl->device);
748}
749
750static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
751{
752 put_device(ctrl->device);
753}
754
755static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
756{
757 return !qid &&
758 nvme_tag_from_cid(command_id) >= NVME_AQ_BLK_MQ_DEPTH;
759}
760
761/*
762 * Returns true for sink states that can't ever transition back to live.
763 */
764static inline bool nvme_state_terminal(struct nvme_ctrl *ctrl)
765{
766 switch (nvme_ctrl_state(ctrl)) {
767 case NVME_CTRL_NEW:
768 case NVME_CTRL_LIVE:
769 case NVME_CTRL_RESETTING:
770 case NVME_CTRL_CONNECTING:
771 return false;
772 case NVME_CTRL_DELETING:
773 case NVME_CTRL_DELETING_NOIO:
774 case NVME_CTRL_DEAD:
775 return true;
776 default:
777 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
778 return true;
779 }
780}
781
782void nvme_end_req(struct request *req);
783void nvme_complete_rq(struct request *req);
784void nvme_complete_batch_req(struct request *req);
785
786static __always_inline void nvme_complete_batch(struct io_comp_batch *iob,
787 void (*fn)(struct request *rq))
788{
789 struct request *req;
790
791 rq_list_for_each(&iob->req_list, req) {
792 fn(req);
793 nvme_complete_batch_req(req);
794 }
795 blk_mq_end_request_batch(iob);
796}
797
798blk_status_t nvme_host_path_error(struct request *req);
799bool nvme_cancel_request(struct request *req, void *data);
800void nvme_cancel_tagset(struct nvme_ctrl *ctrl);
801void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl);
802bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
803 enum nvme_ctrl_state new_state);
804int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown);
805int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
806int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
807 const struct nvme_ctrl_ops *ops, unsigned long quirks);
808int nvme_add_ctrl(struct nvme_ctrl *ctrl);
809void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
810void nvme_start_ctrl(struct nvme_ctrl *ctrl);
811void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
812int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended);
813int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
814 const struct blk_mq_ops *ops, unsigned int cmd_size);
815void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl);
816int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
817 const struct blk_mq_ops *ops, unsigned int nr_maps,
818 unsigned int cmd_size);
819void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl);
820
821void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
822
823void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
824 volatile union nvme_result *res);
825
826void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl);
827void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl);
828void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl);
829void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl);
830void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl);
831void nvme_sync_queues(struct nvme_ctrl *ctrl);
832void nvme_sync_io_queues(struct nvme_ctrl *ctrl);
833void nvme_unfreeze(struct nvme_ctrl *ctrl);
834void nvme_wait_freeze(struct nvme_ctrl *ctrl);
835int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
836void nvme_start_freeze(struct nvme_ctrl *ctrl);
837
838static inline enum req_op nvme_req_op(struct nvme_command *cmd)
839{
840 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
841}
842
843#define NVME_QID_ANY -1
844void nvme_init_request(struct request *req, struct nvme_command *cmd);
845void nvme_cleanup_cmd(struct request *req);
846blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req);
847blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
848 struct request *req);
849bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
850 bool queue_live, enum nvme_ctrl_state state);
851
852static inline bool nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
853 bool queue_live)
854{
855 enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
856
857 if (likely(state == NVME_CTRL_LIVE))
858 return true;
859 if (ctrl->ops->flags & NVME_F_FABRICS && state == NVME_CTRL_DELETING)
860 return queue_live;
861 return __nvme_check_ready(ctrl, rq, queue_live, state);
862}
863
864/*
865 * NSID shall be unique for all shared namespaces, or if at least one of the
866 * following conditions is met:
867 * 1. Namespace Management is supported by the controller
868 * 2. ANA is supported by the controller
869 * 3. NVM Set are supported by the controller
870 *
871 * In other case, private namespace are not required to report a unique NSID.
872 */
873static inline bool nvme_is_unique_nsid(struct nvme_ctrl *ctrl,
874 struct nvme_ns_head *head)
875{
876 return head->shared ||
877 (ctrl->oacs & NVME_CTRL_OACS_NS_MNGT_SUPP) ||
878 (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA) ||
879 (ctrl->ctratt & NVME_CTRL_CTRATT_NVM_SETS);
880}
881
882/*
883 * Flags for __nvme_submit_sync_cmd()
884 */
885typedef __u32 __bitwise nvme_submit_flags_t;
886
887enum {
888 /* Insert request at the head of the queue */
889 NVME_SUBMIT_AT_HEAD = (__force nvme_submit_flags_t)(1 << 0),
890 /* Set BLK_MQ_REQ_NOWAIT when allocating request */
891 NVME_SUBMIT_NOWAIT = (__force nvme_submit_flags_t)(1 << 1),
892 /* Set BLK_MQ_REQ_RESERVED when allocating request */
893 NVME_SUBMIT_RESERVED = (__force nvme_submit_flags_t)(1 << 2),
894 /* Retry command when NVME_STATUS_DNR is not set in the result */
895 NVME_SUBMIT_RETRY = (__force nvme_submit_flags_t)(1 << 3),
896};
897
898int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
899 void *buf, unsigned bufflen);
900int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
901 union nvme_result *result, void *buffer, unsigned bufflen,
902 int qid, nvme_submit_flags_t flags);
903int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
904 unsigned int dword11, void *buffer, size_t buflen,
905 void *result);
906int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
907 unsigned int dword11, void *buffer, size_t buflen,
908 void *result);
909int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
910void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
911int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
912int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
913int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
914void nvme_queue_scan(struct nvme_ctrl *ctrl);
915int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
916 void *log, size_t size, u64 offset);
917bool nvme_tryget_ns_head(struct nvme_ns_head *head);
918void nvme_put_ns_head(struct nvme_ns_head *head);
919int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
920 const struct file_operations *fops, struct module *owner);
921void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device);
922int nvme_ioctl(struct block_device *bdev, blk_mode_t mode,
923 unsigned int cmd, unsigned long arg);
924long nvme_ns_chr_ioctl(struct file *file, unsigned int cmd, unsigned long arg);
925int nvme_ns_head_ioctl(struct block_device *bdev, blk_mode_t mode,
926 unsigned int cmd, unsigned long arg);
927long nvme_ns_head_chr_ioctl(struct file *file, unsigned int cmd,
928 unsigned long arg);
929long nvme_dev_ioctl(struct file *file, unsigned int cmd,
930 unsigned long arg);
931int nvme_ns_chr_uring_cmd_iopoll(struct io_uring_cmd *ioucmd,
932 struct io_comp_batch *iob, unsigned int poll_flags);
933int nvme_ns_chr_uring_cmd(struct io_uring_cmd *ioucmd,
934 unsigned int issue_flags);
935int nvme_ns_head_chr_uring_cmd(struct io_uring_cmd *ioucmd,
936 unsigned int issue_flags);
937int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
938 struct nvme_id_ns **id);
939int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo);
940int nvme_dev_uring_cmd(struct io_uring_cmd *ioucmd, unsigned int issue_flags);
941
942extern const struct attribute_group *nvme_ns_attr_groups[];
943extern const struct attribute_group nvme_ns_mpath_attr_group;
944extern const struct pr_ops nvme_pr_ops;
945extern const struct block_device_operations nvme_ns_head_ops;
946extern const struct attribute_group nvme_dev_attrs_group;
947extern const struct attribute_group *nvme_subsys_attrs_groups[];
948extern const struct attribute_group *nvme_dev_attr_groups[];
949extern const struct block_device_operations nvme_bdev_ops;
950
951void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
952struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
953#ifdef CONFIG_NVME_MULTIPATH
954static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
955{
956 return ctrl->ana_log_buf != NULL;
957}
958
959void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
960void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
961void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
962void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys);
963void nvme_failover_req(struct request *req);
964void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
965int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
966void nvme_mpath_add_sysfs_link(struct nvme_ns_head *ns);
967void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns);
968void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid);
969void nvme_mpath_put_disk(struct nvme_ns_head *head);
970int nvme_mpath_init_identify(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
971void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl);
972void nvme_mpath_update(struct nvme_ctrl *ctrl);
973void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
974void nvme_mpath_stop(struct nvme_ctrl *ctrl);
975bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
976void nvme_mpath_revalidate_paths(struct nvme_ns *ns);
977void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
978void nvme_mpath_remove_disk(struct nvme_ns_head *head);
979void nvme_mpath_start_request(struct request *rq);
980void nvme_mpath_end_request(struct request *rq);
981
982static inline void nvme_trace_bio_complete(struct request *req)
983{
984 struct nvme_ns *ns = req->q->queuedata;
985
986 if ((req->cmd_flags & REQ_NVME_MPATH) && req->bio)
987 trace_block_bio_complete(ns->head->disk->queue, req->bio);
988}
989
990extern bool multipath;
991extern struct device_attribute dev_attr_ana_grpid;
992extern struct device_attribute dev_attr_ana_state;
993extern struct device_attribute dev_attr_queue_depth;
994extern struct device_attribute dev_attr_numa_nodes;
995extern struct device_attribute dev_attr_delayed_removal_secs;
996extern struct device_attribute subsys_attr_iopolicy;
997
998static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
999{
1000 return disk->fops == &nvme_ns_head_ops;
1001}
1002static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1003{
1004 if (test_bit(NVME_NSHEAD_QUEUE_IF_NO_PATH, &head->flags))
1005 return true;
1006 return false;
1007}
1008#else
1009#define multipath false
1010static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
1011{
1012 return false;
1013}
1014static inline void nvme_failover_req(struct request *req)
1015{
1016}
1017static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
1018{
1019}
1020static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
1021 struct nvme_ns_head *head)
1022{
1023 return 0;
1024}
1025static inline void nvme_mpath_add_disk(struct nvme_ns *ns, __le32 anagrpid)
1026{
1027}
1028static inline void nvme_mpath_put_disk(struct nvme_ns_head *head)
1029{
1030}
1031static inline void nvme_mpath_add_sysfs_link(struct nvme_ns *ns)
1032{
1033}
1034static inline void nvme_mpath_remove_sysfs_link(struct nvme_ns *ns)
1035{
1036}
1037static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
1038{
1039 return false;
1040}
1041static inline void nvme_mpath_revalidate_paths(struct nvme_ns *ns)
1042{
1043}
1044static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
1045{
1046}
1047static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
1048{
1049}
1050static inline void nvme_trace_bio_complete(struct request *req)
1051{
1052}
1053static inline void nvme_mpath_init_ctrl(struct nvme_ctrl *ctrl)
1054{
1055}
1056static inline int nvme_mpath_init_identify(struct nvme_ctrl *ctrl,
1057 struct nvme_id_ctrl *id)
1058{
1059 if (ctrl->subsys->cmic & NVME_CTRL_CMIC_ANA)
1060 dev_warn(ctrl->device,
1061"Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
1062 return 0;
1063}
1064static inline void nvme_mpath_update(struct nvme_ctrl *ctrl)
1065{
1066}
1067static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
1068{
1069}
1070static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
1071{
1072}
1073static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
1074{
1075}
1076static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
1077{
1078}
1079static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
1080{
1081}
1082static inline void nvme_mpath_default_iopolicy(struct nvme_subsystem *subsys)
1083{
1084}
1085static inline void nvme_mpath_start_request(struct request *rq)
1086{
1087}
1088static inline void nvme_mpath_end_request(struct request *rq)
1089{
1090}
1091static inline bool nvme_disk_is_ns_head(struct gendisk *disk)
1092{
1093 return false;
1094}
1095static inline bool nvme_mpath_queue_if_no_path(struct nvme_ns_head *head)
1096{
1097 return false;
1098}
1099#endif /* CONFIG_NVME_MULTIPATH */
1100
1101int nvme_ns_get_unique_id(struct nvme_ns *ns, u8 id[16],
1102 enum blk_unique_id type);
1103
1104struct nvme_zone_info {
1105 u64 zone_size;
1106 unsigned int max_open_zones;
1107 unsigned int max_active_zones;
1108};
1109
1110int nvme_ns_report_zones(struct nvme_ns *ns, sector_t sector,
1111 unsigned int nr_zones, report_zones_cb cb, void *data);
1112int nvme_query_zone_info(struct nvme_ns *ns, unsigned lbaf,
1113 struct nvme_zone_info *zi);
1114void nvme_update_zone_info(struct nvme_ns *ns, struct queue_limits *lim,
1115 struct nvme_zone_info *zi);
1116#ifdef CONFIG_BLK_DEV_ZONED
1117blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
1118 struct nvme_command *cmnd,
1119 enum nvme_zone_mgmt_action action);
1120#else
1121static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
1122 struct request *req, struct nvme_command *cmnd,
1123 enum nvme_zone_mgmt_action action)
1124{
1125 return BLK_STS_NOTSUPP;
1126}
1127#endif
1128
1129static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
1130{
1131 struct gendisk *disk = dev_to_disk(dev);
1132
1133 WARN_ON(nvme_disk_is_ns_head(disk));
1134 return disk->private_data;
1135}
1136
1137#ifdef CONFIG_NVME_HWMON
1138int nvme_hwmon_init(struct nvme_ctrl *ctrl);
1139void nvme_hwmon_exit(struct nvme_ctrl *ctrl);
1140#else
1141static inline int nvme_hwmon_init(struct nvme_ctrl *ctrl)
1142{
1143 return 0;
1144}
1145
1146static inline void nvme_hwmon_exit(struct nvme_ctrl *ctrl)
1147{
1148}
1149#endif
1150
1151static inline void nvme_start_request(struct request *rq)
1152{
1153 if (rq->cmd_flags & REQ_NVME_MPATH)
1154 nvme_mpath_start_request(rq);
1155 blk_mq_start_request(rq);
1156}
1157
1158static inline bool nvme_ctrl_sgl_supported(struct nvme_ctrl *ctrl)
1159{
1160 return ctrl->sgls & (NVME_CTRL_SGLS_BYTE_ALIGNED |
1161 NVME_CTRL_SGLS_DWORD_ALIGNED);
1162}
1163
1164static inline bool nvme_ctrl_meta_sgl_supported(struct nvme_ctrl *ctrl)
1165{
1166 if (ctrl->ops->flags & NVME_F_FABRICS)
1167 return true;
1168 return ctrl->sgls & NVME_CTRL_SGLS_MSDS;
1169}
1170
1171#ifdef CONFIG_NVME_HOST_AUTH
1172int __init nvme_init_auth(void);
1173void __exit nvme_exit_auth(void);
1174int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl);
1175void nvme_auth_stop(struct nvme_ctrl *ctrl);
1176int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid);
1177int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid);
1178void nvme_auth_free(struct nvme_ctrl *ctrl);
1179void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl);
1180#else
1181static inline int nvme_auth_init_ctrl(struct nvme_ctrl *ctrl)
1182{
1183 return 0;
1184}
1185static inline int __init nvme_init_auth(void)
1186{
1187 return 0;
1188}
1189static inline void __exit nvme_exit_auth(void)
1190{
1191}
1192static inline void nvme_auth_stop(struct nvme_ctrl *ctrl) {};
1193static inline int nvme_auth_negotiate(struct nvme_ctrl *ctrl, int qid)
1194{
1195 return -EPROTONOSUPPORT;
1196}
1197static inline int nvme_auth_wait(struct nvme_ctrl *ctrl, int qid)
1198{
1199 return -EPROTONOSUPPORT;
1200}
1201static inline void nvme_auth_free(struct nvme_ctrl *ctrl) {};
1202static inline void nvme_auth_revoke_tls_key(struct nvme_ctrl *ctrl) {};
1203#endif
1204
1205u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1206 u8 opcode);
1207u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode);
1208int nvme_execute_rq(struct request *rq, bool at_head);
1209void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1210 struct nvme_command *cmd, int status);
1211struct nvme_ctrl *nvme_ctrl_from_file(struct file *file);
1212struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
1213bool nvme_get_ns(struct nvme_ns *ns);
1214void nvme_put_ns(struct nvme_ns *ns);
1215
1216static inline bool nvme_multi_css(struct nvme_ctrl *ctrl)
1217{
1218 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1219}
1220
1221#endif /* _NVME_H */
Linux 6.x block layer and io_uring tree(s)