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>
19 #include <trace/events/block.h>
21 extern unsigned int nvme_io_timeout;
22 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
24 extern unsigned int admin_timeout;
25 #define ADMIN_TIMEOUT (admin_timeout * HZ)
27 #define NVME_DEFAULT_KATO 5
28 #define NVME_KATO_GRACE 10
30 extern struct workqueue_struct *nvme_wq;
31 extern struct workqueue_struct *nvme_reset_wq;
32 extern struct workqueue_struct *nvme_delete_wq;
40 * List of workarounds for devices that required behavior not specified in
45 * Prefers I/O aligned to a stripe size specified in a vendor
46 * specific Identify field.
48 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
51 * The controller doesn't handle Identify value others than 0 or 1
54 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
57 * The controller deterministically returns O's on reads to
58 * logical blocks that deallocate was called on.
60 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
63 * The controller needs a delay before starts checking the device
64 * readiness, which is done by reading the NVME_CSTS_RDY bit.
66 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
69 * APST should not be used.
71 NVME_QUIRK_NO_APST = (1 << 4),
74 * The deepest sleep state should not be used.
76 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
79 * Supports the LighNVM command set if indicated in vs[1].
81 NVME_QUIRK_LIGHTNVM = (1 << 6),
84 * Set MEDIUM priority on SQ creation
86 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
89 * Ignore device provided subnqn.
91 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
94 * Broken Write Zeroes.
96 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
99 * Force simple suspend/resume path.
101 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
104 * Use only one interrupt vector for all queues
106 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
109 * Use non-standard 128 bytes SQEs.
111 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
114 * Prevent tag overlap between queues
116 NVME_QUIRK_SHARED_TAGS = (1 << 13),
120 * Common request structure for NVMe passthrough. All drivers must have
121 * this structure as the first member of their request-private data.
123 struct nvme_request {
124 struct nvme_command *cmd;
125 union nvme_result result;
129 struct nvme_ctrl *ctrl;
133 * Mark a bio as coming in through the mpath node.
135 #define REQ_NVME_MPATH REQ_DRV
138 NVME_REQ_CANCELLED = (1 << 0),
139 NVME_REQ_USERCMD = (1 << 1),
142 static inline struct nvme_request *nvme_req(struct request *req)
144 return blk_mq_rq_to_pdu(req);
147 static inline u16 nvme_req_qid(struct request *req)
151 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
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
159 #define NVME_QUIRK_DELAY_AMOUNT 2300
161 enum nvme_ctrl_state {
164 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
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;
203 struct nvme_subsystem *subsys;
204 struct list_head subsys_entry;
206 struct opal_dev *opal_dev;
226 atomic_t abort_limit;
236 unsigned int shutdown_timeout;
239 unsigned long quirks;
240 struct nvme_id_power_state psd[32];
241 struct nvme_effects_log *effects;
242 struct work_struct scan_work;
243 struct work_struct async_event_work;
244 struct delayed_work ka_work;
245 struct nvme_command ka_cmd;
246 struct work_struct fw_act_work;
247 unsigned long events;
249 #ifdef CONFIG_NVME_MULTIPATH
250 /* asymmetric namespace access: */
255 struct mutex ana_lock;
256 struct nvme_ana_rsp_hdr *ana_log_buf;
258 struct timer_list anatt_timer;
259 struct work_struct ana_work;
262 /* Power saving configuration */
263 u64 ps_max_latency_us;
278 struct nvmf_ctrl_options *opts;
280 struct page *discard_page;
281 unsigned long discard_page_busy;
283 struct nvme_fault_inject fault_inject;
291 struct nvme_subsystem {
295 * Because we unregister the device on the last put we need
296 * a separate refcount.
299 struct list_head entry;
301 struct list_head ctrls;
302 struct list_head nsheads;
303 char subnqn[NVMF_NQN_SIZE];
306 char firmware_rev[8];
309 u16 awupf; /* 0's based awupf value. */
311 #ifdef CONFIG_NVME_MULTIPATH
312 enum nvme_iopolicy iopolicy;
317 * Container structure for uniqueue namespace identifiers.
326 * Anchor structure for namespaces. There is one for each namespace in a
327 * NVMe subsystem that any of our controllers can see, and the namespace
328 * structure for each controller is chained of it. For private namespaces
329 * there is a 1:1 relation to our namespace structures, that is ->list
330 * only ever has a single entry for private namespaces.
332 struct nvme_ns_head {
333 struct list_head list;
334 struct srcu_struct srcu;
335 struct nvme_subsystem *subsys;
337 struct nvme_ns_ids ids;
338 struct list_head entry;
341 #ifdef CONFIG_NVME_MULTIPATH
342 struct gendisk *disk;
343 struct bio_list requeue_list;
344 spinlock_t requeue_lock;
345 struct work_struct requeue_work;
347 struct nvme_ns __rcu *current_path[];
352 struct list_head list;
354 struct nvme_ctrl *ctrl;
355 struct request_queue *queue;
356 struct gendisk *disk;
357 #ifdef CONFIG_NVME_MULTIPATH
358 enum nvme_ana_state ana_state;
361 struct list_head siblings;
362 struct nvm_dev *ndev;
364 struct nvme_ns_head *head;
373 #define NVME_NS_REMOVING 0
374 #define NVME_NS_DEAD 1
375 #define NVME_NS_ANA_PENDING 2
378 struct nvme_fault_inject fault_inject;
382 struct nvme_ctrl_ops {
384 struct module *module;
386 #define NVME_F_FABRICS (1 << 0)
387 #define NVME_F_METADATA_SUPPORTED (1 << 1)
388 #define NVME_F_PCI_P2PDMA (1 << 2)
389 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
390 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
391 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
392 void (*free_ctrl)(struct nvme_ctrl *ctrl);
393 void (*submit_async_event)(struct nvme_ctrl *ctrl);
394 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
395 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
398 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
399 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
400 const char *dev_name);
401 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
402 void nvme_should_fail(struct request *req);
404 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
405 const char *dev_name)
408 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
411 static inline void nvme_should_fail(struct request *req) {}
414 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
416 if (!ctrl->subsystem)
418 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
422 * Convert a 512B sector number to a device logical block number.
424 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
426 return sector >> (ns->lba_shift - SECTOR_SHIFT);
429 static inline void nvme_end_request(struct request *req, __le16 status,
430 union nvme_result result)
432 struct nvme_request *rq = nvme_req(req);
434 rq->status = le16_to_cpu(status) >> 1;
436 /* inject error when permitted by fault injection framework */
437 nvme_should_fail(req);
438 blk_mq_complete_request(req);
441 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
443 get_device(ctrl->device);
446 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
448 put_device(ctrl->device);
451 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
453 return !qid && command_id >= NVME_AQ_BLK_MQ_DEPTH;
456 void nvme_complete_rq(struct request *req);
457 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
458 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
459 enum nvme_ctrl_state new_state);
460 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
461 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
462 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
463 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
464 const struct nvme_ctrl_ops *ops, unsigned long quirks);
465 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
466 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
467 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
468 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
469 int nvme_init_identify(struct nvme_ctrl *ctrl);
471 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
473 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
476 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
477 volatile union nvme_result *res);
479 void nvme_stop_queues(struct nvme_ctrl *ctrl);
480 void nvme_start_queues(struct nvme_ctrl *ctrl);
481 void nvme_kill_queues(struct nvme_ctrl *ctrl);
482 void nvme_sync_queues(struct nvme_ctrl *ctrl);
483 void nvme_unfreeze(struct nvme_ctrl *ctrl);
484 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
485 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
486 void nvme_start_freeze(struct nvme_ctrl *ctrl);
488 #define NVME_QID_ANY -1
489 struct request *nvme_alloc_request(struct request_queue *q,
490 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
491 void nvme_cleanup_cmd(struct request *req);
492 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
493 struct nvme_command *cmd);
494 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
495 void *buf, unsigned bufflen);
496 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
497 union nvme_result *result, void *buffer, unsigned bufflen,
498 unsigned timeout, int qid, int at_head,
499 blk_mq_req_flags_t flags, bool poll);
500 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
501 unsigned int dword11, void *buffer, size_t buflen,
503 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
504 unsigned int dword11, void *buffer, size_t buflen,
506 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
507 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
508 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
509 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
510 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
512 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
513 void *log, size_t size, u64 offset);
515 extern const struct attribute_group *nvme_ns_id_attr_groups[];
516 extern const struct block_device_operations nvme_ns_head_ops;
518 #ifdef CONFIG_NVME_MULTIPATH
519 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
521 return ctrl->ana_log_buf != NULL;
524 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
525 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
526 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
527 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
528 struct nvme_ctrl *ctrl, int *flags);
529 void nvme_failover_req(struct request *req);
530 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
531 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
532 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
533 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
534 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
535 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
536 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
537 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
538 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
539 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
541 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
543 struct nvme_ns_head *head = ns->head;
545 if (head->disk && list_empty(&head->list))
546 kblockd_schedule_work(&head->requeue_work);
549 static inline void nvme_trace_bio_complete(struct request *req,
552 struct nvme_ns *ns = req->q->queuedata;
554 if (req->cmd_flags & REQ_NVME_MPATH)
555 trace_block_bio_complete(ns->head->disk->queue,
559 extern struct device_attribute dev_attr_ana_grpid;
560 extern struct device_attribute dev_attr_ana_state;
561 extern struct device_attribute subsys_attr_iopolicy;
564 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
569 * Without the multipath code enabled, multiple controller per subsystems are
570 * visible as devices and thus we cannot use the subsystem instance.
572 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
573 struct nvme_ctrl *ctrl, int *flags)
575 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
578 static inline void nvme_failover_req(struct request *req)
581 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
584 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
585 struct nvme_ns_head *head)
589 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
590 struct nvme_id_ns *id)
593 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
596 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
600 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
603 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
606 static inline void nvme_trace_bio_complete(struct request *req,
610 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
611 struct nvme_id_ctrl *id)
613 if (ctrl->subsys->cmic & (1 << 3))
614 dev_warn(ctrl->device,
615 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
618 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
621 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
624 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
627 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
630 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
633 #endif /* CONFIG_NVME_MULTIPATH */
636 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
637 void nvme_nvm_unregister(struct nvme_ns *ns);
638 extern const struct attribute_group nvme_nvm_attr_group;
639 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
641 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
647 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
648 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
653 #endif /* CONFIG_NVM */
655 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
657 return dev_to_disk(dev)->private_data;