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 * Use only one interrupt vector for all queues
101 NVME_QUIRK_SINGLE_VECTOR = (1 << 10),
104 * Use non-standard 128 bytes SQEs.
106 NVME_QUIRK_128_BYTES_SQES = (1 << 11),
109 * Prevent tag overlap between queues
111 NVME_QUIRK_SHARED_TAGS = (1 << 12),
115 * Common request structure for NVMe passthrough. All drivers must have
116 * this structure as the first member of their request-private data.
118 struct nvme_request {
119 struct nvme_command *cmd;
120 union nvme_result result;
124 struct nvme_ctrl *ctrl;
128 * Mark a bio as coming in through the mpath node.
130 #define REQ_NVME_MPATH REQ_DRV
133 NVME_REQ_CANCELLED = (1 << 0),
134 NVME_REQ_USERCMD = (1 << 1),
137 static inline struct nvme_request *nvme_req(struct request *req)
139 return blk_mq_rq_to_pdu(req);
142 static inline u16 nvme_req_qid(struct request *req)
146 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
149 /* The below value is the specific amount of delay needed before checking
150 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
151 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
154 #define NVME_QUIRK_DELAY_AMOUNT 2300
156 enum nvme_ctrl_state {
159 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
161 NVME_CTRL_CONNECTING,
166 struct nvme_fault_inject {
167 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
168 struct fault_attr attr;
169 struct dentry *parent;
170 bool dont_retry; /* DNR, do not retry */
171 u16 status; /* status code */
177 enum nvme_ctrl_state state;
180 struct mutex scan_lock;
181 const struct nvme_ctrl_ops *ops;
182 struct request_queue *admin_q;
183 struct request_queue *connect_q;
187 struct blk_mq_tag_set *tagset;
188 struct blk_mq_tag_set *admin_tagset;
189 struct list_head namespaces;
190 struct rw_semaphore namespaces_rwsem;
191 struct device ctrl_device;
192 struct device *device; /* char device */
194 struct work_struct reset_work;
195 struct work_struct delete_work;
197 struct nvme_subsystem *subsys;
198 struct list_head subsys_entry;
200 struct opal_dev *opal_dev;
219 atomic_t abort_limit;
229 unsigned int shutdown_timeout;
232 unsigned long quirks;
233 struct nvme_id_power_state psd[32];
234 struct nvme_effects_log *effects;
235 struct work_struct scan_work;
236 struct work_struct async_event_work;
237 struct delayed_work ka_work;
238 struct nvme_command ka_cmd;
239 struct work_struct fw_act_work;
240 unsigned long events;
242 #ifdef CONFIG_NVME_MULTIPATH
243 /* asymmetric namespace access: */
248 struct mutex ana_lock;
249 struct nvme_ana_rsp_hdr *ana_log_buf;
251 struct timer_list anatt_timer;
252 struct work_struct ana_work;
255 /* Power saving configuration */
256 u64 ps_max_latency_us;
272 struct nvmf_ctrl_options *opts;
274 struct page *discard_page;
275 unsigned long discard_page_busy;
277 struct nvme_fault_inject fault_inject;
285 struct nvme_subsystem {
289 * Because we unregister the device on the last put we need
290 * a separate refcount.
293 struct list_head entry;
295 struct list_head ctrls;
296 struct list_head nsheads;
297 char subnqn[NVMF_NQN_SIZE];
300 char firmware_rev[8];
303 u16 awupf; /* 0's based awupf value. */
305 #ifdef CONFIG_NVME_MULTIPATH
306 enum nvme_iopolicy iopolicy;
311 * Container structure for uniqueue namespace identifiers.
320 * Anchor structure for namespaces. There is one for each namespace in a
321 * NVMe subsystem that any of our controllers can see, and the namespace
322 * structure for each controller is chained of it. For private namespaces
323 * there is a 1:1 relation to our namespace structures, that is ->list
324 * only ever has a single entry for private namespaces.
326 struct nvme_ns_head {
327 struct list_head list;
328 struct srcu_struct srcu;
329 struct nvme_subsystem *subsys;
331 struct nvme_ns_ids ids;
332 struct list_head entry;
335 #ifdef CONFIG_NVME_MULTIPATH
336 struct gendisk *disk;
337 struct bio_list requeue_list;
338 spinlock_t requeue_lock;
339 struct work_struct requeue_work;
341 struct nvme_ns __rcu *current_path[];
346 struct list_head list;
348 struct nvme_ctrl *ctrl;
349 struct request_queue *queue;
350 struct gendisk *disk;
351 #ifdef CONFIG_NVME_MULTIPATH
352 enum nvme_ana_state ana_state;
355 struct list_head siblings;
356 struct nvm_dev *ndev;
358 struct nvme_ns_head *head;
367 #define NVME_NS_REMOVING 0
368 #define NVME_NS_DEAD 1
369 #define NVME_NS_ANA_PENDING 2
372 struct nvme_fault_inject fault_inject;
376 struct nvme_ctrl_ops {
378 struct module *module;
380 #define NVME_F_FABRICS (1 << 0)
381 #define NVME_F_METADATA_SUPPORTED (1 << 1)
382 #define NVME_F_PCI_P2PDMA (1 << 2)
383 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
384 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
385 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
386 void (*free_ctrl)(struct nvme_ctrl *ctrl);
387 void (*submit_async_event)(struct nvme_ctrl *ctrl);
388 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
389 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
392 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
393 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
394 const char *dev_name);
395 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
396 void nvme_should_fail(struct request *req);
398 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
399 const char *dev_name)
402 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
405 static inline void nvme_should_fail(struct request *req) {}
408 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
410 if (!ctrl->subsystem)
412 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
415 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
417 return (sector >> (ns->lba_shift - 9));
420 static inline void nvme_end_request(struct request *req, __le16 status,
421 union nvme_result result)
423 struct nvme_request *rq = nvme_req(req);
425 rq->status = le16_to_cpu(status) >> 1;
427 /* inject error when permitted by fault injection framework */
428 nvme_should_fail(req);
429 blk_mq_complete_request(req);
432 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
434 get_device(ctrl->device);
437 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
439 put_device(ctrl->device);
442 void nvme_complete_rq(struct request *req);
443 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
444 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
445 enum nvme_ctrl_state new_state);
446 int nvme_disable_ctrl(struct nvme_ctrl *ctrl);
447 int nvme_enable_ctrl(struct nvme_ctrl *ctrl);
448 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
449 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
450 const struct nvme_ctrl_ops *ops, unsigned long quirks);
451 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
452 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
453 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
454 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
455 int nvme_init_identify(struct nvme_ctrl *ctrl);
457 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
459 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
462 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
463 volatile union nvme_result *res);
465 void nvme_stop_queues(struct nvme_ctrl *ctrl);
466 void nvme_start_queues(struct nvme_ctrl *ctrl);
467 void nvme_kill_queues(struct nvme_ctrl *ctrl);
468 void nvme_sync_queues(struct nvme_ctrl *ctrl);
469 void nvme_unfreeze(struct nvme_ctrl *ctrl);
470 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
471 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
472 void nvme_start_freeze(struct nvme_ctrl *ctrl);
474 #define NVME_QID_ANY -1
475 struct request *nvme_alloc_request(struct request_queue *q,
476 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
477 void nvme_cleanup_cmd(struct request *req);
478 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
479 struct nvme_command *cmd);
480 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
481 void *buf, unsigned bufflen);
482 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
483 union nvme_result *result, void *buffer, unsigned bufflen,
484 unsigned timeout, int qid, int at_head,
485 blk_mq_req_flags_t flags, bool poll);
486 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
487 unsigned int dword11, void *buffer, size_t buflen,
489 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
490 unsigned int dword11, void *buffer, size_t buflen,
492 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
493 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
494 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
495 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
496 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
498 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
499 void *log, size_t size, u64 offset);
501 extern const struct attribute_group *nvme_ns_id_attr_groups[];
502 extern const struct block_device_operations nvme_ns_head_ops;
504 #ifdef CONFIG_NVME_MULTIPATH
505 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
507 return ctrl->ana_log_buf != NULL;
510 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
511 struct nvme_ctrl *ctrl, int *flags);
512 void nvme_failover_req(struct request *req);
513 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
514 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
515 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
516 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
517 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
518 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
519 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
520 void nvme_mpath_clear_current_path(struct nvme_ns *ns);
521 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
523 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
525 struct nvme_ns_head *head = ns->head;
527 if (head->disk && list_empty(&head->list))
528 kblockd_schedule_work(&head->requeue_work);
531 static inline void nvme_trace_bio_complete(struct request *req,
534 struct nvme_ns *ns = req->q->queuedata;
536 if (req->cmd_flags & REQ_NVME_MPATH)
537 trace_block_bio_complete(ns->head->disk->queue,
541 extern struct device_attribute dev_attr_ana_grpid;
542 extern struct device_attribute dev_attr_ana_state;
543 extern struct device_attribute subsys_attr_iopolicy;
546 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
551 * Without the multipath code enabled, multiple controller per subsystems are
552 * visible as devices and thus we cannot use the subsystem instance.
554 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
555 struct nvme_ctrl *ctrl, int *flags)
557 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
560 static inline void nvme_failover_req(struct request *req)
563 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
566 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
567 struct nvme_ns_head *head)
571 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
572 struct nvme_id_ns *id)
575 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
578 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
581 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
584 static inline void nvme_trace_bio_complete(struct request *req,
588 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
589 struct nvme_id_ctrl *id)
591 if (ctrl->subsys->cmic & (1 << 3))
592 dev_warn(ctrl->device,
593 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
596 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
599 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
602 #endif /* CONFIG_NVME_MULTIPATH */
605 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
606 void nvme_nvm_unregister(struct nvme_ns *ns);
607 extern const struct attribute_group nvme_nvm_attr_group;
608 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
610 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
616 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
617 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
622 #endif /* CONFIG_NVM */
624 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
626 return dev_to_disk(dev)->private_data;