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>
19 #include <linux/t10-pi.h>
21 #include <trace/events/block.h>
23 extern unsigned int nvme_io_timeout;
24 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
26 extern unsigned int admin_timeout;
27 #define ADMIN_TIMEOUT (admin_timeout * HZ)
29 #define NVME_DEFAULT_KATO 5
30 #define NVME_KATO_GRACE 10
32 #ifdef CONFIG_ARCH_NO_SG_CHAIN
33 #define NVME_INLINE_SG_CNT 0
34 #define NVME_INLINE_METADATA_SG_CNT 0
36 #define NVME_INLINE_SG_CNT 2
37 #define NVME_INLINE_METADATA_SG_CNT 1
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.
45 #define NVME_CTRL_PAGE_SHIFT 12
46 #define NVME_CTRL_PAGE_SIZE (1 << NVME_CTRL_PAGE_SHIFT)
48 extern struct workqueue_struct *nvme_wq;
49 extern struct workqueue_struct *nvme_reset_wq;
50 extern struct workqueue_struct *nvme_delete_wq;
58 * List of workarounds for devices that required behavior not specified in
63 * Prefers I/O aligned to a stripe size specified in a vendor
64 * specific Identify field.
66 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
69 * The controller doesn't handle Identify value others than 0 or 1
72 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
75 * The controller deterministically returns O's on reads to
76 * logical blocks that deallocate was called on.
78 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
81 * The controller needs a delay before starts checking the device
82 * readiness, which is done by reading the NVME_CSTS_RDY bit.
84 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
87 * APST should not be used.
89 NVME_QUIRK_NO_APST = (1 << 4),
92 * The deepest sleep state should not be used.
94 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
97 * Supports the LighNVM command set if indicated in vs[1].
99 NVME_QUIRK_LIGHTNVM = (1 << 6),
102 * Set MEDIUM priority on SQ creation
104 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
107 * Ignore device provided subnqn.
109 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
112 * Broken Write Zeroes.
114 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
117 * Force simple suspend/resume path.
119 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
122 * Use only one interrupt vector for all queues
124 NVME_QUIRK_SINGLE_VECTOR = (1 << 11),
127 * Use non-standard 128 bytes SQEs.
129 NVME_QUIRK_128_BYTES_SQES = (1 << 12),
132 * Prevent tag overlap between queues
134 NVME_QUIRK_SHARED_TAGS = (1 << 13),
137 * Don't change the value of the temperature threshold feature
139 NVME_QUIRK_NO_TEMP_THRESH_CHANGE = (1 << 14),
142 * The controller doesn't handle the Identify Namespace
143 * Identification Descriptor list subcommand despite claiming
144 * NVMe 1.3 compliance.
146 NVME_QUIRK_NO_NS_DESC_LIST = (1 << 15),
150 * Common request structure for NVMe passthrough. All drivers must have
151 * this structure as the first member of their request-private data.
153 struct nvme_request {
154 struct nvme_command *cmd;
155 union nvme_result result;
159 struct nvme_ctrl *ctrl;
163 * Mark a bio as coming in through the mpath node.
165 #define REQ_NVME_MPATH REQ_DRV
168 NVME_REQ_CANCELLED = (1 << 0),
169 NVME_REQ_USERCMD = (1 << 1),
172 static inline struct nvme_request *nvme_req(struct request *req)
174 return blk_mq_rq_to_pdu(req);
177 static inline u16 nvme_req_qid(struct request *req)
181 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
184 /* The below value is the specific amount of delay needed before checking
185 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
186 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
189 #define NVME_QUIRK_DELAY_AMOUNT 2300
192 * enum nvme_ctrl_state: Controller state
194 * @NVME_CTRL_NEW: New controller just allocated, initial state
195 * @NVME_CTRL_LIVE: Controller is connected and I/O capable
196 * @NVME_CTRL_RESETTING: Controller is resetting (or scheduled reset)
197 * @NVME_CTRL_CONNECTING: Controller is disconnected, now connecting the
199 * @NVME_CTRL_DELETING: Controller is deleting (or scheduled deletion)
200 * @NVME_CTRL_DELETING_NOIO: Controller is deleting and I/O is not
201 * disabled/failed immediately. This state comes
202 * after all async event processing took place and
203 * before ns removal and the controller deletion
205 * @NVME_CTRL_DEAD: Controller is non-present/unresponsive during
206 * shutdown or removal. In this case we forcibly
207 * kill all inflight I/O as they have no chance to
210 enum nvme_ctrl_state {
214 NVME_CTRL_CONNECTING,
216 NVME_CTRL_DELETING_NOIO,
220 struct nvme_fault_inject {
221 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
222 struct fault_attr attr;
223 struct dentry *parent;
224 bool dont_retry; /* DNR, do not retry */
225 u16 status; /* status code */
230 struct list_head entry;
231 struct nvme_effects_log log;
237 enum nvme_ctrl_state state;
240 struct mutex scan_lock;
241 const struct nvme_ctrl_ops *ops;
242 struct request_queue *admin_q;
243 struct request_queue *connect_q;
244 struct request_queue *fabrics_q;
248 struct blk_mq_tag_set *tagset;
249 struct blk_mq_tag_set *admin_tagset;
250 struct list_head namespaces;
251 struct rw_semaphore namespaces_rwsem;
252 struct device ctrl_device;
253 struct device *device; /* char device */
255 struct work_struct reset_work;
256 struct work_struct delete_work;
257 wait_queue_head_t state_wq;
259 struct nvme_subsystem *subsys;
260 struct list_head subsys_entry;
262 struct opal_dev *opal_dev;
274 u32 max_integrity_segments;
275 #ifdef CONFIG_BLK_DEV_ZONED
285 atomic_t abort_limit;
297 unsigned int shutdown_timeout;
300 unsigned long quirks;
301 struct nvme_id_power_state psd[32];
302 struct nvme_effects_log *effects;
303 struct list_head cels;
304 struct work_struct scan_work;
305 struct work_struct async_event_work;
306 struct delayed_work ka_work;
307 struct nvme_command ka_cmd;
308 struct work_struct fw_act_work;
309 unsigned long events;
312 #ifdef CONFIG_NVME_MULTIPATH
313 /* asymmetric namespace access: */
318 struct mutex ana_lock;
319 struct nvme_ana_rsp_hdr *ana_log_buf;
321 struct timer_list anatt_timer;
322 struct work_struct ana_work;
325 /* Power saving configuration */
326 u64 ps_max_latency_us;
341 struct nvmf_ctrl_options *opts;
343 struct page *discard_page;
344 unsigned long discard_page_busy;
346 struct nvme_fault_inject fault_inject;
354 struct nvme_subsystem {
358 * Because we unregister the device on the last put we need
359 * a separate refcount.
362 struct list_head entry;
364 struct list_head ctrls;
365 struct list_head nsheads;
366 char subnqn[NVMF_NQN_SIZE];
369 char firmware_rev[8];
372 u16 awupf; /* 0's based awupf value. */
374 #ifdef CONFIG_NVME_MULTIPATH
375 enum nvme_iopolicy iopolicy;
380 * Container structure for uniqueue namespace identifiers.
390 * Anchor structure for namespaces. There is one for each namespace in a
391 * NVMe subsystem that any of our controllers can see, and the namespace
392 * structure for each controller is chained of it. For private namespaces
393 * there is a 1:1 relation to our namespace structures, that is ->list
394 * only ever has a single entry for private namespaces.
396 struct nvme_ns_head {
397 struct list_head list;
398 struct srcu_struct srcu;
399 struct nvme_subsystem *subsys;
401 struct nvme_ns_ids ids;
402 struct list_head entry;
406 struct nvme_effects_log *effects;
407 #ifdef CONFIG_NVME_MULTIPATH
408 struct gendisk *disk;
409 struct bio_list requeue_list;
410 spinlock_t requeue_lock;
411 struct work_struct requeue_work;
414 #define NVME_NSHEAD_DISK_LIVE 0
415 struct nvme_ns __rcu *current_path[];
419 enum nvme_ns_features {
420 NVME_NS_EXT_LBAS = 1 << 0, /* support extended LBA format */
421 NVME_NS_METADATA_SUPPORTED = 1 << 1, /* support getting generated md */
425 struct list_head list;
427 struct nvme_ctrl *ctrl;
428 struct request_queue *queue;
429 struct gendisk *disk;
430 #ifdef CONFIG_NVME_MULTIPATH
431 enum nvme_ana_state ana_state;
434 struct list_head siblings;
435 struct nvm_dev *ndev;
437 struct nvme_ns_head *head;
444 #ifdef CONFIG_BLK_DEV_ZONED
447 unsigned long features;
449 #define NVME_NS_REMOVING 0
450 #define NVME_NS_DEAD 1
451 #define NVME_NS_ANA_PENDING 2
453 struct nvme_fault_inject fault_inject;
457 /* NVMe ns supports metadata actions by the controller (generate/strip) */
458 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
460 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
463 struct nvme_ctrl_ops {
465 struct module *module;
467 #define NVME_F_FABRICS (1 << 0)
468 #define NVME_F_METADATA_SUPPORTED (1 << 1)
469 #define NVME_F_PCI_P2PDMA (1 << 2)
470 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
471 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
472 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
473 void (*free_ctrl)(struct nvme_ctrl *ctrl);
474 void (*submit_async_event)(struct nvme_ctrl *ctrl);
475 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
476 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
479 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
480 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
481 const char *dev_name);
482 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
483 void nvme_should_fail(struct request *req);
485 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
486 const char *dev_name)
489 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
492 static inline void nvme_should_fail(struct request *req) {}
495 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
497 if (!ctrl->subsystem)
499 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
503 * Convert a 512B sector number to a device logical block number.
505 static inline u64 nvme_sect_to_lba(struct nvme_ns *ns, sector_t sector)
507 return sector >> (ns->lba_shift - SECTOR_SHIFT);
511 * Convert a device logical block number to a 512B sector number.
513 static inline sector_t nvme_lba_to_sect(struct nvme_ns *ns, u64 lba)
515 return lba << (ns->lba_shift - SECTOR_SHIFT);
519 * Convert byte length to nvme's 0-based num dwords
521 static inline u32 nvme_bytes_to_numd(size_t len)
523 return (len >> 2) - 1;
526 static inline bool nvme_end_request(struct request *req, __le16 status,
527 union nvme_result result)
529 struct nvme_request *rq = nvme_req(req);
531 rq->status = le16_to_cpu(status) >> 1;
533 /* inject error when permitted by fault injection framework */
534 nvme_should_fail(req);
535 if (unlikely(blk_should_fake_timeout(req->q)))
537 return blk_mq_complete_request_remote(req);
540 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
542 get_device(ctrl->device);
545 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
547 put_device(ctrl->device);
550 static inline bool nvme_is_aen_req(u16 qid, __u16 command_id)
552 return !qid && command_id >= NVME_AQ_BLK_MQ_DEPTH;
555 void nvme_complete_rq(struct request *req);
556 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
557 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
558 enum nvme_ctrl_state new_state);
559 bool nvme_wait_reset(struct nvme_ctrl *ctrl);
560 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
561 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
562 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
563 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
564 const struct nvme_ctrl_ops *ops, unsigned long quirks);
565 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
566 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
567 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
568 int nvme_init_identify(struct nvme_ctrl *ctrl);
570 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
572 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
575 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
576 volatile union nvme_result *res);
578 void nvme_stop_queues(struct nvme_ctrl *ctrl);
579 void nvme_start_queues(struct nvme_ctrl *ctrl);
580 void nvme_kill_queues(struct nvme_ctrl *ctrl);
581 void nvme_sync_queues(struct nvme_ctrl *ctrl);
582 void nvme_unfreeze(struct nvme_ctrl *ctrl);
583 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
584 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
585 void nvme_start_freeze(struct nvme_ctrl *ctrl);
587 #define NVME_QID_ANY -1
588 struct request *nvme_alloc_request(struct request_queue *q,
589 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
590 void nvme_cleanup_cmd(struct request *req);
591 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
592 struct nvme_command *cmd);
593 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
594 void *buf, unsigned bufflen);
595 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
596 union nvme_result *result, void *buffer, unsigned bufflen,
597 unsigned timeout, int qid, int at_head,
598 blk_mq_req_flags_t flags, bool poll);
599 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
600 unsigned int dword11, void *buffer, size_t buflen,
602 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
603 unsigned int dword11, void *buffer, size_t buflen,
605 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
606 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
607 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
608 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
609 int nvme_try_sched_reset(struct nvme_ctrl *ctrl);
610 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
612 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
613 void *log, size_t size, u64 offset);
614 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
615 struct nvme_ns_head **head, int *srcu_idx);
616 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx);
618 extern const struct attribute_group *nvme_ns_id_attr_groups[];
619 extern const struct block_device_operations nvme_ns_head_ops;
621 #ifdef CONFIG_NVME_MULTIPATH
622 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
624 return ctrl->ana_log_buf != NULL;
627 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
628 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
629 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
630 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
631 struct nvme_ctrl *ctrl, int *flags);
632 bool nvme_failover_req(struct request *req);
633 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
634 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
635 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
636 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
637 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
638 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
639 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
640 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
641 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
642 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
643 blk_qc_t nvme_ns_head_submit_bio(struct bio *bio);
645 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
647 struct nvme_ns_head *head = ns->head;
649 if (head->disk && list_empty(&head->list))
650 kblockd_schedule_work(&head->requeue_work);
653 static inline void nvme_trace_bio_complete(struct request *req,
656 struct nvme_ns *ns = req->q->queuedata;
658 if (req->cmd_flags & REQ_NVME_MPATH)
659 trace_block_bio_complete(ns->head->disk->queue, req->bio);
662 static inline void nvme_mpath_update_disk_size(struct gendisk *disk)
664 struct block_device *bdev = bdget_disk(disk, 0);
667 bd_set_size(bdev, get_capacity(disk) << SECTOR_SHIFT);
672 extern struct device_attribute dev_attr_ana_grpid;
673 extern struct device_attribute dev_attr_ana_state;
674 extern struct device_attribute subsys_attr_iopolicy;
677 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
682 * Without the multipath code enabled, multiple controller per subsystems are
683 * visible as devices and thus we cannot use the subsystem instance.
685 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
686 struct nvme_ctrl *ctrl, int *flags)
688 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
691 static inline bool nvme_failover_req(struct request *req)
695 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
698 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
699 struct nvme_ns_head *head)
703 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
704 struct nvme_id_ns *id)
707 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
710 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
714 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
717 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
720 static inline void nvme_trace_bio_complete(struct request *req,
724 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
725 struct nvme_id_ctrl *id)
727 if (ctrl->subsys->cmic & (1 << 3))
728 dev_warn(ctrl->device,
729 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
732 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
735 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
738 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
741 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
744 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
747 static inline void nvme_mpath_update_disk_size(struct gendisk *disk)
750 #endif /* CONFIG_NVME_MULTIPATH */
752 #ifdef CONFIG_BLK_DEV_ZONED
753 int nvme_update_zone_info(struct gendisk *disk, struct nvme_ns *ns,
756 int nvme_report_zones(struct gendisk *disk, sector_t sector,
757 unsigned int nr_zones, report_zones_cb cb, void *data);
759 blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns, struct request *req,
760 struct nvme_command *cmnd,
761 enum nvme_zone_mgmt_action action);
763 #define nvme_report_zones NULL
765 static inline blk_status_t nvme_setup_zone_mgmt_send(struct nvme_ns *ns,
766 struct request *req, struct nvme_command *cmnd,
767 enum nvme_zone_mgmt_action action)
769 return BLK_STS_NOTSUPP;
772 static inline int nvme_update_zone_info(struct gendisk *disk,
776 dev_warn(ns->ctrl->device,
777 "Please enable CONFIG_BLK_DEV_ZONED to support ZNS devices\n");
778 return -EPROTONOSUPPORT;
783 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
784 void nvme_nvm_unregister(struct nvme_ns *ns);
785 extern const struct attribute_group nvme_nvm_attr_group;
786 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
788 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
794 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
795 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
800 #endif /* CONFIG_NVM */
802 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
804 return dev_to_disk(dev)->private_data;
807 #ifdef CONFIG_NVME_HWMON
808 void nvme_hwmon_init(struct nvme_ctrl *ctrl);
810 static inline void nvme_hwmon_init(struct nvme_ctrl *ctrl) { }
813 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
815 void nvme_execute_passthru_rq(struct request *rq);
816 struct nvme_ctrl *nvme_ctrl_get_by_path(const char *path);
817 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid);
818 void nvme_put_ns(struct nvme_ns *ns);