1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (c) 2011-2014, Intel Corporation.
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 #include <linux/wait.h>
20 #include <trace/events/block.h>
22 extern unsigned int nvme_io_timeout;
23 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
25 extern unsigned int admin_timeout;
26 #define ADMIN_TIMEOUT (admin_timeout * HZ)
28 #define NVME_DEFAULT_KATO 5
29 #define NVME_KATO_GRACE 10
31 extern struct workqueue_struct *nvme_wq;
32 extern struct workqueue_struct *nvme_reset_wq;
33 extern struct workqueue_struct *nvme_delete_wq;
41 * List of workarounds for devices that required behavior not specified in
46 * Prefers I/O aligned to a stripe size specified in a vendor
47 * specific Identify field.
49 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
52 * The controller doesn't handle Identify value others than 0 or 1
55 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
58 * The controller deterministically returns O's on reads to
59 * logical blocks that deallocate was called on.
61 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
64 * The controller needs a delay before starts checking the device
65 * readiness, which is done by reading the NVME_CSTS_RDY bit.
67 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
70 * APST should not be used.
72 NVME_QUIRK_NO_APST = (1 << 4),
75 * The deepest sleep state should not be used.
77 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
80 * Supports the LighNVM command set if indicated in vs[1].
82 NVME_QUIRK_LIGHTNVM = (1 << 6),
85 * Set MEDIUM priority on SQ creation
87 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
90 * Ignore device provided subnqn.
92 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
95 * Broken Write Zeroes.
97 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
100 * Force simple suspend/resume path.
102 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
105 * Use only one interrupt vector for all queues
107 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
110 * Use non-standard 128 bytes SQEs.
112 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
115 * Prevent tag overlap between queues
117 NVME_QUIRK_SHARED_TAGS = (1 << 13),
121 * Common request structure for NVMe passthrough. All drivers must have
122 * this structure as the first member of their request-private data.
124 struct nvme_request {
125 struct nvme_command *cmd;
126 union nvme_result result;
130 struct nvme_ctrl *ctrl;
134 * Mark a bio as coming in through the mpath node.
136 #define REQ_NVME_MPATH REQ_DRV
139 NVME_REQ_CANCELLED = (1 << 0),
140 NVME_REQ_USERCMD = (1 << 1),
143 static inline struct nvme_request *nvme_req(struct request *req)
145 return blk_mq_rq_to_pdu(req);
148 static inline u16 nvme_req_qid(struct request *req)
152 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
155 /* The below value is the specific amount of delay needed before checking
156 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
157 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
160 #define NVME_QUIRK_DELAY_AMOUNT 2300
162 enum nvme_ctrl_state {
166 NVME_CTRL_CONNECTING,
171 struct nvme_fault_inject {
172 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
173 struct fault_attr attr;
174 struct dentry *parent;
175 bool dont_retry; /* DNR, do not retry */
176 u16 status; /* status code */
182 enum nvme_ctrl_state state;
185 struct mutex scan_lock;
186 const struct nvme_ctrl_ops *ops;
187 struct request_queue *admin_q;
188 struct request_queue *connect_q;
189 struct request_queue *fabrics_q;
193 struct blk_mq_tag_set *tagset;
194 struct blk_mq_tag_set *admin_tagset;
195 struct list_head namespaces;
196 struct rw_semaphore namespaces_rwsem;
197 struct device ctrl_device;
198 struct device *device; /* char device */
200 struct work_struct reset_work;
201 struct work_struct delete_work;
202 wait_queue_head_t state_wq;
204 struct nvme_subsystem *subsys;
205 struct list_head subsys_entry;
207 struct opal_dev *opal_dev;
227 atomic_t abort_limit;
237 unsigned int shutdown_timeout;
240 unsigned long quirks;
241 struct nvme_id_power_state psd[32];
242 struct nvme_effects_log *effects;
243 struct work_struct scan_work;
244 struct work_struct async_event_work;
245 struct delayed_work ka_work;
246 struct nvme_command ka_cmd;
247 struct work_struct fw_act_work;
248 unsigned long events;
250 #ifdef CONFIG_NVME_MULTIPATH
251 /* asymmetric namespace access: */
256 struct mutex ana_lock;
257 struct nvme_ana_rsp_hdr *ana_log_buf;
259 struct timer_list anatt_timer;
260 struct work_struct ana_work;
263 /* Power saving configuration */
264 u64 ps_max_latency_us;
279 struct nvmf_ctrl_options *opts;
281 struct page *discard_page;
282 unsigned long discard_page_busy;
284 struct nvme_fault_inject fault_inject;
292 struct nvme_subsystem {
296 * Because we unregister the device on the last put we need
297 * a separate refcount.
300 struct list_head entry;
302 struct list_head ctrls;
303 struct list_head nsheads;
304 char subnqn[NVMF_NQN_SIZE];
307 char firmware_rev[8];
310 u16 awupf; /* 0's based awupf value. */
312 #ifdef CONFIG_NVME_MULTIPATH
313 enum nvme_iopolicy iopolicy;
318 * Container structure for uniqueue namespace identifiers.
327 * Anchor structure for namespaces. There is one for each namespace in a
328 * NVMe subsystem that any of our controllers can see, and the namespace
329 * structure for each controller is chained of it. For private namespaces
330 * there is a 1:1 relation to our namespace structures, that is ->list
331 * only ever has a single entry for private namespaces.
333 struct nvme_ns_head {
334 struct list_head list;
335 struct srcu_struct srcu;
336 struct nvme_subsystem *subsys;
338 struct nvme_ns_ids ids;
339 struct list_head entry;
342 #ifdef CONFIG_NVME_MULTIPATH
343 struct gendisk *disk;
344 struct bio_list requeue_list;
345 spinlock_t requeue_lock;
346 struct work_struct requeue_work;
348 struct nvme_ns __rcu *current_path[];
353 struct list_head list;
355 struct nvme_ctrl *ctrl;
356 struct request_queue *queue;
357 struct gendisk *disk;
358 #ifdef CONFIG_NVME_MULTIPATH
359 enum nvme_ana_state ana_state;
362 struct list_head siblings;
363 struct nvm_dev *ndev;
365 struct nvme_ns_head *head;
374 #define NVME_NS_REMOVING 0
375 #define NVME_NS_DEAD 1
376 #define NVME_NS_ANA_PENDING 2
379 struct nvme_fault_inject fault_inject;
383 struct nvme_ctrl_ops {
385 struct module *module;
387 #define NVME_F_FABRICS (1 << 0)
388 #define NVME_F_METADATA_SUPPORTED (1 << 1)
389 #define NVME_F_PCI_P2PDMA (1 << 2)
390 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
391 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
392 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
393 void (*free_ctrl)(struct nvme_ctrl *ctrl);
394 void (*submit_async_event)(struct nvme_ctrl *ctrl);
395 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
396 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
399 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
400 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
401 const char *dev_name);
402 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
403 void nvme_should_fail(struct request *req);
405 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
406 const char *dev_name)
409 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
412 static inline void nvme_should_fail(struct request *req) {}
415 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
417 if (!ctrl->subsystem)
419 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
422 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
424 return (sector >> (ns->lba_shift - 9));
427 static inline void nvme_end_request(struct request *req, __le16 status,
428 union nvme_result result)
430 struct nvme_request *rq = nvme_req(req);
432 rq->status = le16_to_cpu(status) >> 1;
434 /* inject error when permitted by fault injection framework */
435 nvme_should_fail(req);
436 blk_mq_complete_request(req);
439 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
441 get_device(ctrl->device);
444 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
446 put_device(ctrl->device);
449 void nvme_complete_rq(struct request *req);
450 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
451 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
452 enum nvme_ctrl_state new_state);
453 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
454 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
455 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
456 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
457 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
458 const struct nvme_ctrl_ops *ops, unsigned long quirks);
459 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
460 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
461 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
462 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
463 int nvme_init_identify(struct nvme_ctrl *ctrl);
465 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
467 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
470 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
471 volatile union nvme_result *res);
473 void nvme_stop_queues(struct nvme_ctrl *ctrl);
474 void nvme_start_queues(struct nvme_ctrl *ctrl);
475 void nvme_kill_queues(struct nvme_ctrl *ctrl);
476 void nvme_sync_queues(struct nvme_ctrl *ctrl);
477 void nvme_unfreeze(struct nvme_ctrl *ctrl);
478 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
479 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
480 void nvme_start_freeze(struct nvme_ctrl *ctrl);
482 #define NVME_QID_ANY -1
483 struct request *nvme_alloc_request(struct request_queue *q,
484 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
485 void nvme_cleanup_cmd(struct request *req);
486 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
487 struct nvme_command *cmd);
488 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
489 void *buf, unsigned bufflen);
490 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
491 union nvme_result *result, void *buffer, unsigned bufflen,
492 unsigned timeout, int qid, int at_head,
493 blk_mq_req_flags_t flags, bool poll);
494 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
495 unsigned int dword11, void *buffer, size_t buflen,
497 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
498 unsigned int dword11, void *buffer, size_t buflen,
500 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
501 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
502 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
503 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
504 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
505 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
507 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
508 void *log, size_t size, u64 offset);
510 extern const struct attribute_group *nvme_ns_id_attr_groups[];
511 extern const struct block_device_operations nvme_ns_head_ops;
513 #ifdef CONFIG_NVME_MULTIPATH
514 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
516 return ctrl->ana_log_buf != NULL;
519 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
520 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
521 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
522 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
523 struct nvme_ctrl *ctrl, int *flags);
524 void nvme_failover_req(struct request *req);
525 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
526 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
527 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
528 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
529 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
530 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
531 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
532 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
533 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
534 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
536 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
538 struct nvme_ns_head *head = ns->head;
540 if (head->disk && list_empty(&head->list))
541 kblockd_schedule_work(&head->requeue_work);
544 static inline void nvme_trace_bio_complete(struct request *req,
547 struct nvme_ns *ns = req->q->queuedata;
549 if (req->cmd_flags & REQ_NVME_MPATH)
550 trace_block_bio_complete(ns->head->disk->queue,
554 extern struct device_attribute dev_attr_ana_grpid;
555 extern struct device_attribute dev_attr_ana_state;
556 extern struct device_attribute subsys_attr_iopolicy;
559 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
564 * Without the multipath code enabled, multiple controller per subsystems are
565 * visible as devices and thus we cannot use the subsystem instance.
567 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
568 struct nvme_ctrl *ctrl, int *flags)
570 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
573 static inline void nvme_failover_req(struct request *req)
576 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
579 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
580 struct nvme_ns_head *head)
584 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
585 struct nvme_id_ns *id)
588 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
591 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
595 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
598 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
601 static inline void nvme_trace_bio_complete(struct request *req,
605 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
606 struct nvme_id_ctrl *id)
608 if (ctrl->subsys->cmic & (1 << 3))
609 dev_warn(ctrl->device,
610 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
613 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
616 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
619 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
622 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
625 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
628 #endif /* CONFIG_NVME_MULTIPATH */
631 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
632 void nvme_nvm_unregister(struct nvme_ns *ns);
633 extern const struct attribute_group nvme_nvm_attr_group;
634 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
636 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
642 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
643 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
648 #endif /* CONFIG_NVM */
650 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
652 return dev_to_disk(dev)->private_data;