]> asedeno.scripts.mit.edu Git - linux.git/blob - drivers/nvme/host/nvme.h
nvme: Remove ADMIN_ONLY state
[linux.git] / drivers / nvme / host / nvme.h
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/lightnvm.h>
15 #include <linux/sed-opal.h>
16 #include <linux/fault-inject.h>
17 #include <linux/rcupdate.h>
18
19 #include <trace/events/block.h>
20
21 extern unsigned int nvme_io_timeout;
22 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
23
24 extern unsigned int admin_timeout;
25 #define ADMIN_TIMEOUT   (admin_timeout * HZ)
26
27 #define NVME_DEFAULT_KATO       5
28 #define NVME_KATO_GRACE         10
29
30 extern struct workqueue_struct *nvme_wq;
31 extern struct workqueue_struct *nvme_reset_wq;
32 extern struct workqueue_struct *nvme_delete_wq;
33
34 enum {
35         NVME_NS_LBA             = 0,
36         NVME_NS_LIGHTNVM        = 1,
37 };
38
39 /*
40  * List of workarounds for devices that required behavior not specified in
41  * the standard.
42  */
43 enum nvme_quirks {
44         /*
45          * Prefers I/O aligned to a stripe size specified in a vendor
46          * specific Identify field.
47          */
48         NVME_QUIRK_STRIPE_SIZE                  = (1 << 0),
49
50         /*
51          * The controller doesn't handle Identify value others than 0 or 1
52          * correctly.
53          */
54         NVME_QUIRK_IDENTIFY_CNS                 = (1 << 1),
55
56         /*
57          * The controller deterministically returns O's on reads to
58          * logical blocks that deallocate was called on.
59          */
60         NVME_QUIRK_DEALLOCATE_ZEROES            = (1 << 2),
61
62         /*
63          * The controller needs a delay before starts checking the device
64          * readiness, which is done by reading the NVME_CSTS_RDY bit.
65          */
66         NVME_QUIRK_DELAY_BEFORE_CHK_RDY         = (1 << 3),
67
68         /*
69          * APST should not be used.
70          */
71         NVME_QUIRK_NO_APST                      = (1 << 4),
72
73         /*
74          * The deepest sleep state should not be used.
75          */
76         NVME_QUIRK_NO_DEEPEST_PS                = (1 << 5),
77
78         /*
79          * Supports the LighNVM command set if indicated in vs[1].
80          */
81         NVME_QUIRK_LIGHTNVM                     = (1 << 6),
82
83         /*
84          * Set MEDIUM priority on SQ creation
85          */
86         NVME_QUIRK_MEDIUM_PRIO_SQ               = (1 << 7),
87
88         /*
89          * Ignore device provided subnqn.
90          */
91         NVME_QUIRK_IGNORE_DEV_SUBNQN            = (1 << 8),
92
93         /*
94          * Broken Write Zeroes.
95          */
96         NVME_QUIRK_DISABLE_WRITE_ZEROES         = (1 << 9),
97
98         /*
99          * Force simple suspend/resume path.
100          */
101         NVME_QUIRK_SIMPLE_SUSPEND               = (1 << 10),
102
103         /*
104          * Use only one interrupt vector for all queues
105          */
106         NVME_QUIRK_SINGLE_VECTOR                = (1 << 11),
107
108         /*
109          * Use non-standard 128 bytes SQEs.
110          */
111         NVME_QUIRK_128_BYTES_SQES               = (1 << 12),
112
113         /*
114          * Prevent tag overlap between queues
115          */
116         NVME_QUIRK_SHARED_TAGS                  = (1 << 13),
117 };
118
119 /*
120  * Common request structure for NVMe passthrough.  All drivers must have
121  * this structure as the first member of their request-private data.
122  */
123 struct nvme_request {
124         struct nvme_command     *cmd;
125         union nvme_result       result;
126         u8                      retries;
127         u8                      flags;
128         u16                     status;
129         struct nvme_ctrl        *ctrl;
130 };
131
132 /*
133  * Mark a bio as coming in through the mpath node.
134  */
135 #define REQ_NVME_MPATH          REQ_DRV
136
137 enum {
138         NVME_REQ_CANCELLED              = (1 << 0),
139         NVME_REQ_USERCMD                = (1 << 1),
140 };
141
142 static inline struct nvme_request *nvme_req(struct request *req)
143 {
144         return blk_mq_rq_to_pdu(req);
145 }
146
147 static inline u16 nvme_req_qid(struct request *req)
148 {
149         if (!req->rq_disk)
150                 return 0;
151         return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
152 }
153
154 /* The below value is the specific amount of delay needed before checking
155  * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
156  * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
157  * found empirically.
158  */
159 #define NVME_QUIRK_DELAY_AMOUNT         2300
160
161 enum nvme_ctrl_state {
162         NVME_CTRL_NEW,
163         NVME_CTRL_LIVE,
164         NVME_CTRL_RESETTING,
165         NVME_CTRL_CONNECTING,
166         NVME_CTRL_DELETING,
167         NVME_CTRL_DEAD,
168 };
169
170 struct nvme_fault_inject {
171 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
172         struct fault_attr attr;
173         struct dentry *parent;
174         bool dont_retry;        /* DNR, do not retry */
175         u16 status;             /* status code */
176 #endif
177 };
178
179 struct nvme_ctrl {
180         bool comp_seen;
181         enum nvme_ctrl_state state;
182         bool identified;
183         spinlock_t lock;
184         struct mutex scan_lock;
185         const struct nvme_ctrl_ops *ops;
186         struct request_queue *admin_q;
187         struct request_queue *connect_q;
188         struct request_queue *fabrics_q;
189         struct device *dev;
190         int instance;
191         int numa_node;
192         struct blk_mq_tag_set *tagset;
193         struct blk_mq_tag_set *admin_tagset;
194         struct list_head namespaces;
195         struct rw_semaphore namespaces_rwsem;
196         struct device ctrl_device;
197         struct device *device;  /* char device */
198         struct cdev cdev;
199         struct work_struct reset_work;
200         struct work_struct delete_work;
201
202         struct nvme_subsystem *subsys;
203         struct list_head subsys_entry;
204
205         struct opal_dev *opal_dev;
206
207         char name[12];
208         u16 cntlid;
209
210         u32 ctrl_config;
211         u16 mtfa;
212         u32 queue_count;
213
214         u64 cap;
215         u32 page_size;
216         u32 max_hw_sectors;
217         u32 max_segments;
218         u16 crdt[3];
219         u16 oncs;
220         u16 oacs;
221         u16 nssa;
222         u16 nr_streams;
223         u16 sqsize;
224         u32 max_namespaces;
225         atomic_t abort_limit;
226         u8 vwc;
227         u32 vs;
228         u32 sgls;
229         u16 kas;
230         u8 npss;
231         u8 apsta;
232         u32 oaes;
233         u32 aen_result;
234         u32 ctratt;
235         unsigned int shutdown_timeout;
236         unsigned int kato;
237         bool subsystem;
238         unsigned long quirks;
239         struct nvme_id_power_state psd[32];
240         struct nvme_effects_log *effects;
241         struct work_struct scan_work;
242         struct work_struct async_event_work;
243         struct delayed_work ka_work;
244         struct nvme_command ka_cmd;
245         struct work_struct fw_act_work;
246         unsigned long events;
247
248 #ifdef CONFIG_NVME_MULTIPATH
249         /* asymmetric namespace access: */
250         u8 anacap;
251         u8 anatt;
252         u32 anagrpmax;
253         u32 nanagrpid;
254         struct mutex ana_lock;
255         struct nvme_ana_rsp_hdr *ana_log_buf;
256         size_t ana_log_size;
257         struct timer_list anatt_timer;
258         struct work_struct ana_work;
259 #endif
260
261         /* Power saving configuration */
262         u64 ps_max_latency_us;
263         bool apst_enabled;
264
265         /* PCIe only: */
266         u32 hmpre;
267         u32 hmmin;
268         u32 hmminds;
269         u16 hmmaxd;
270
271         /* Fabrics only */
272         u32 ioccsz;
273         u32 iorcsz;
274         u16 icdoff;
275         u16 maxcmd;
276         int nr_reconnects;
277         struct nvmf_ctrl_options *opts;
278
279         struct page *discard_page;
280         unsigned long discard_page_busy;
281
282         struct nvme_fault_inject fault_inject;
283 };
284
285 enum nvme_iopolicy {
286         NVME_IOPOLICY_NUMA,
287         NVME_IOPOLICY_RR,
288 };
289
290 struct nvme_subsystem {
291         int                     instance;
292         struct device           dev;
293         /*
294          * Because we unregister the device on the last put we need
295          * a separate refcount.
296          */
297         struct kref             ref;
298         struct list_head        entry;
299         struct mutex            lock;
300         struct list_head        ctrls;
301         struct list_head        nsheads;
302         char                    subnqn[NVMF_NQN_SIZE];
303         char                    serial[20];
304         char                    model[40];
305         char                    firmware_rev[8];
306         u8                      cmic;
307         u16                     vendor_id;
308         u16                     awupf;  /* 0's based awupf value. */
309         struct ida              ns_ida;
310 #ifdef CONFIG_NVME_MULTIPATH
311         enum nvme_iopolicy      iopolicy;
312 #endif
313 };
314
315 /*
316  * Container structure for uniqueue namespace identifiers.
317  */
318 struct nvme_ns_ids {
319         u8      eui64[8];
320         u8      nguid[16];
321         uuid_t  uuid;
322 };
323
324 /*
325  * Anchor structure for namespaces.  There is one for each namespace in a
326  * NVMe subsystem that any of our controllers can see, and the namespace
327  * structure for each controller is chained of it.  For private namespaces
328  * there is a 1:1 relation to our namespace structures, that is ->list
329  * only ever has a single entry for private namespaces.
330  */
331 struct nvme_ns_head {
332         struct list_head        list;
333         struct srcu_struct      srcu;
334         struct nvme_subsystem   *subsys;
335         unsigned                ns_id;
336         struct nvme_ns_ids      ids;
337         struct list_head        entry;
338         struct kref             ref;
339         int                     instance;
340 #ifdef CONFIG_NVME_MULTIPATH
341         struct gendisk          *disk;
342         struct bio_list         requeue_list;
343         spinlock_t              requeue_lock;
344         struct work_struct      requeue_work;
345         struct mutex            lock;
346         struct nvme_ns __rcu    *current_path[];
347 #endif
348 };
349
350 struct nvme_ns {
351         struct list_head list;
352
353         struct nvme_ctrl *ctrl;
354         struct request_queue *queue;
355         struct gendisk *disk;
356 #ifdef CONFIG_NVME_MULTIPATH
357         enum nvme_ana_state ana_state;
358         u32 ana_grpid;
359 #endif
360         struct list_head siblings;
361         struct nvm_dev *ndev;
362         struct kref kref;
363         struct nvme_ns_head *head;
364
365         int lba_shift;
366         u16 ms;
367         u16 sgs;
368         u32 sws;
369         bool ext;
370         u8 pi_type;
371         unsigned long flags;
372 #define NVME_NS_REMOVING        0
373 #define NVME_NS_DEAD            1
374 #define NVME_NS_ANA_PENDING     2
375         u16 noiob;
376
377         struct nvme_fault_inject fault_inject;
378
379 };
380
381 struct nvme_ctrl_ops {
382         const char *name;
383         struct module *module;
384         unsigned int flags;
385 #define NVME_F_FABRICS                  (1 << 0)
386 #define NVME_F_METADATA_SUPPORTED       (1 << 1)
387 #define NVME_F_PCI_P2PDMA               (1 << 2)
388         int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
389         int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
390         int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
391         void (*free_ctrl)(struct nvme_ctrl *ctrl);
392         void (*submit_async_event)(struct nvme_ctrl *ctrl);
393         void (*delete_ctrl)(struct nvme_ctrl *ctrl);
394         int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
395 };
396
397 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
398 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
399                             const char *dev_name);
400 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
401 void nvme_should_fail(struct request *req);
402 #else
403 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
404                                           const char *dev_name)
405 {
406 }
407 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
408 {
409 }
410 static inline void nvme_should_fail(struct request *req) {}
411 #endif
412
413 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
414 {
415         if (!ctrl->subsystem)
416                 return -ENOTTY;
417         return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
418 }
419
420 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
421 {
422         return (sector >> (ns->lba_shift - 9));
423 }
424
425 static inline void nvme_end_request(struct request *req, __le16 status,
426                 union nvme_result result)
427 {
428         struct nvme_request *rq = nvme_req(req);
429
430         rq->status = le16_to_cpu(status) >> 1;
431         rq->result = result;
432         /* inject error when permitted by fault injection framework */
433         nvme_should_fail(req);
434         blk_mq_complete_request(req);
435 }
436
437 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
438 {
439         get_device(ctrl->device);
440 }
441
442 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
443 {
444         put_device(ctrl->device);
445 }
446
447 void nvme_complete_rq(struct request *req);
448 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
449 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
450                 enum nvme_ctrl_state new_state);
451 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
452 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
453 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
454 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
455                 const struct nvme_ctrl_ops *ops, unsigned long quirks);
456 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
457 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
458 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
459 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
460 int nvme_init_identify(struct nvme_ctrl *ctrl);
461
462 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
463
464 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
465                 bool send);
466
467 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
468                 volatile union nvme_result *res);
469
470 void nvme_stop_queues(struct nvme_ctrl *ctrl);
471 void nvme_start_queues(struct nvme_ctrl *ctrl);
472 void nvme_kill_queues(struct nvme_ctrl *ctrl);
473 void nvme_sync_queues(struct nvme_ctrl *ctrl);
474 void nvme_unfreeze(struct nvme_ctrl *ctrl);
475 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
476 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
477 void nvme_start_freeze(struct nvme_ctrl *ctrl);
478
479 #define NVME_QID_ANY -1
480 struct request *nvme_alloc_request(struct request_queue *q,
481                 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
482 void nvme_cleanup_cmd(struct request *req);
483 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
484                 struct nvme_command *cmd);
485 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
486                 void *buf, unsigned bufflen);
487 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
488                 union nvme_result *result, void *buffer, unsigned bufflen,
489                 unsigned timeout, int qid, int at_head,
490                 blk_mq_req_flags_t flags, bool poll);
491 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
492                       unsigned int dword11, void *buffer, size_t buflen,
493                       u32 *result);
494 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
495                       unsigned int dword11, void *buffer, size_t buflen,
496                       u32 *result);
497 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
498 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
499 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
500 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
501 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
502
503 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
504                 void *log, size_t size, u64 offset);
505
506 extern const struct attribute_group *nvme_ns_id_attr_groups[];
507 extern const struct block_device_operations nvme_ns_head_ops;
508
509 #ifdef CONFIG_NVME_MULTIPATH
510 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
511 {
512         return ctrl->ana_log_buf != NULL;
513 }
514
515 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
516 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
517 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
518 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
519                         struct nvme_ctrl *ctrl, int *flags);
520 void nvme_failover_req(struct request *req);
521 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
522 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
523 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
524 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
525 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
526 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
527 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
528 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
529 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
530 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
531
532 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
533 {
534         struct nvme_ns_head *head = ns->head;
535
536         if (head->disk && list_empty(&head->list))
537                 kblockd_schedule_work(&head->requeue_work);
538 }
539
540 static inline void nvme_trace_bio_complete(struct request *req,
541         blk_status_t status)
542 {
543         struct nvme_ns *ns = req->q->queuedata;
544
545         if (req->cmd_flags & REQ_NVME_MPATH)
546                 trace_block_bio_complete(ns->head->disk->queue,
547                                          req->bio, status);
548 }
549
550 extern struct device_attribute dev_attr_ana_grpid;
551 extern struct device_attribute dev_attr_ana_state;
552 extern struct device_attribute subsys_attr_iopolicy;
553
554 #else
555 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
556 {
557         return false;
558 }
559 /*
560  * Without the multipath code enabled, multiple controller per subsystems are
561  * visible as devices and thus we cannot use the subsystem instance.
562  */
563 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
564                                       struct nvme_ctrl *ctrl, int *flags)
565 {
566         sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
567 }
568
569 static inline void nvme_failover_req(struct request *req)
570 {
571 }
572 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
573 {
574 }
575 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
576                 struct nvme_ns_head *head)
577 {
578         return 0;
579 }
580 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
581                 struct nvme_id_ns *id)
582 {
583 }
584 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
585 {
586 }
587 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
588 {
589         return false;
590 }
591 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
592 {
593 }
594 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
595 {
596 }
597 static inline void nvme_trace_bio_complete(struct request *req,
598         blk_status_t status)
599 {
600 }
601 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
602                 struct nvme_id_ctrl *id)
603 {
604         if (ctrl->subsys->cmic & (1 << 3))
605                 dev_warn(ctrl->device,
606 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
607         return 0;
608 }
609 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
610 {
611 }
612 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
613 {
614 }
615 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
616 {
617 }
618 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
619 {
620 }
621 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
622 {
623 }
624 #endif /* CONFIG_NVME_MULTIPATH */
625
626 #ifdef CONFIG_NVM
627 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
628 void nvme_nvm_unregister(struct nvme_ns *ns);
629 extern const struct attribute_group nvme_nvm_attr_group;
630 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
631 #else
632 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
633                                     int node)
634 {
635         return 0;
636 }
637
638 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
639 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
640                                                         unsigned long arg)
641 {
642         return -ENOTTY;
643 }
644 #endif /* CONFIG_NVM */
645
646 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
647 {
648         return dev_to_disk(dev)->private_data;
649 }
650
651 #endif /* _NVME_H */