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 extern unsigned int nvme_io_timeout;
20 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
22 extern unsigned int admin_timeout;
23 #define ADMIN_TIMEOUT (admin_timeout * HZ)
25 #define NVME_DEFAULT_KATO 5
26 #define NVME_KATO_GRACE 10
28 extern struct workqueue_struct *nvme_wq;
29 extern struct workqueue_struct *nvme_reset_wq;
30 extern struct workqueue_struct *nvme_delete_wq;
38 * List of workarounds for devices that required behavior not specified in
43 * Prefers I/O aligned to a stripe size specified in a vendor
44 * specific Identify field.
46 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
49 * The controller doesn't handle Identify value others than 0 or 1
52 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
55 * The controller deterministically returns O's on reads to
56 * logical blocks that deallocate was called on.
58 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
61 * The controller needs a delay before starts checking the device
62 * readiness, which is done by reading the NVME_CSTS_RDY bit.
64 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
67 * APST should not be used.
69 NVME_QUIRK_NO_APST = (1 << 4),
72 * The deepest sleep state should not be used.
74 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
77 * Supports the LighNVM command set if indicated in vs[1].
79 NVME_QUIRK_LIGHTNVM = (1 << 6),
82 * Set MEDIUM priority on SQ creation
84 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
87 * Ignore device provided subnqn.
89 NVME_QUIRK_IGNORE_DEV_SUBNQN = (1 << 8),
92 * Broken Write Zeroes.
94 NVME_QUIRK_DISABLE_WRITE_ZEROES = (1 << 9),
97 * Force simple suspend/resume path.
99 NVME_QUIRK_SIMPLE_SUSPEND = (1 << 10),
103 * Common request structure for NVMe passthrough. All drivers must have
104 * this structure as the first member of their request-private data.
106 struct nvme_request {
107 struct nvme_command *cmd;
108 union nvme_result result;
112 struct nvme_ctrl *ctrl;
116 * Mark a bio as coming in through the mpath node.
118 #define REQ_NVME_MPATH REQ_DRV
121 NVME_REQ_CANCELLED = (1 << 0),
122 NVME_REQ_USERCMD = (1 << 1),
125 static inline struct nvme_request *nvme_req(struct request *req)
127 return blk_mq_rq_to_pdu(req);
130 static inline u16 nvme_req_qid(struct request *req)
134 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
137 /* The below value is the specific amount of delay needed before checking
138 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
139 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
142 #define NVME_QUIRK_DELAY_AMOUNT 2300
144 enum nvme_ctrl_state {
147 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
149 NVME_CTRL_CONNECTING,
154 struct nvme_fault_inject {
155 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
156 struct fault_attr attr;
157 struct dentry *parent;
158 bool dont_retry; /* DNR, do not retry */
159 u16 status; /* status code */
165 enum nvme_ctrl_state state;
168 struct mutex scan_lock;
169 const struct nvme_ctrl_ops *ops;
170 struct request_queue *admin_q;
171 struct request_queue *connect_q;
175 struct blk_mq_tag_set *tagset;
176 struct blk_mq_tag_set *admin_tagset;
177 struct list_head namespaces;
178 struct rw_semaphore namespaces_rwsem;
179 struct device ctrl_device;
180 struct device *device; /* char device */
182 struct work_struct reset_work;
183 struct work_struct delete_work;
185 struct nvme_subsystem *subsys;
186 struct list_head subsys_entry;
188 struct opal_dev *opal_dev;
207 atomic_t abort_limit;
217 unsigned int shutdown_timeout;
220 unsigned long quirks;
221 struct nvme_id_power_state psd[32];
222 struct nvme_effects_log *effects;
223 struct work_struct scan_work;
224 struct work_struct async_event_work;
225 struct delayed_work ka_work;
226 struct nvme_command ka_cmd;
227 struct work_struct fw_act_work;
228 unsigned long events;
230 #ifdef CONFIG_NVME_MULTIPATH
231 /* asymmetric namespace access: */
236 struct mutex ana_lock;
237 struct nvme_ana_rsp_hdr *ana_log_buf;
239 struct timer_list anatt_timer;
240 struct work_struct ana_work;
243 /* Power saving configuration */
244 u64 ps_max_latency_us;
260 struct nvmf_ctrl_options *opts;
262 struct page *discard_page;
263 unsigned long discard_page_busy;
265 struct nvme_fault_inject fault_inject;
273 struct nvme_subsystem {
277 * Because we unregister the device on the last put we need
278 * a separate refcount.
281 struct list_head entry;
283 struct list_head ctrls;
284 struct list_head nsheads;
285 char subnqn[NVMF_NQN_SIZE];
288 char firmware_rev[8];
291 u16 awupf; /* 0's based awupf value. */
293 #ifdef CONFIG_NVME_MULTIPATH
294 enum nvme_iopolicy iopolicy;
299 * Container structure for uniqueue namespace identifiers.
308 * Anchor structure for namespaces. There is one for each namespace in a
309 * NVMe subsystem that any of our controllers can see, and the namespace
310 * structure for each controller is chained of it. For private namespaces
311 * there is a 1:1 relation to our namespace structures, that is ->list
312 * only ever has a single entry for private namespaces.
314 struct nvme_ns_head {
315 struct list_head list;
316 struct srcu_struct srcu;
317 struct nvme_subsystem *subsys;
319 struct nvme_ns_ids ids;
320 struct list_head entry;
323 #ifdef CONFIG_NVME_MULTIPATH
324 struct gendisk *disk;
325 struct bio_list requeue_list;
326 spinlock_t requeue_lock;
327 struct work_struct requeue_work;
329 struct nvme_ns __rcu *current_path[];
334 struct list_head list;
336 struct nvme_ctrl *ctrl;
337 struct request_queue *queue;
338 struct gendisk *disk;
339 #ifdef CONFIG_NVME_MULTIPATH
340 enum nvme_ana_state ana_state;
343 struct list_head siblings;
344 struct nvm_dev *ndev;
346 struct nvme_ns_head *head;
355 #define NVME_NS_REMOVING 0
356 #define NVME_NS_DEAD 1
357 #define NVME_NS_ANA_PENDING 2
360 struct nvme_fault_inject fault_inject;
364 struct nvme_ctrl_ops {
366 struct module *module;
368 #define NVME_F_FABRICS (1 << 0)
369 #define NVME_F_METADATA_SUPPORTED (1 << 1)
370 #define NVME_F_PCI_P2PDMA (1 << 2)
371 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
372 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
373 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
374 void (*free_ctrl)(struct nvme_ctrl *ctrl);
375 void (*submit_async_event)(struct nvme_ctrl *ctrl);
376 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
377 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
380 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
381 void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
382 const char *dev_name);
383 void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inject);
384 void nvme_should_fail(struct request *req);
386 static inline void nvme_fault_inject_init(struct nvme_fault_inject *fault_inj,
387 const char *dev_name)
390 static inline void nvme_fault_inject_fini(struct nvme_fault_inject *fault_inj)
393 static inline void nvme_should_fail(struct request *req) {}
396 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
398 if (!ctrl->subsystem)
400 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
403 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
405 return (sector >> (ns->lba_shift - 9));
408 static inline void nvme_end_request(struct request *req, __le16 status,
409 union nvme_result result)
411 struct nvme_request *rq = nvme_req(req);
413 rq->status = le16_to_cpu(status) >> 1;
415 /* inject error when permitted by fault injection framework */
416 nvme_should_fail(req);
417 blk_mq_complete_request(req);
420 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
422 get_device(ctrl->device);
425 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
427 put_device(ctrl->device);
430 void nvme_complete_rq(struct request *req);
431 bool nvme_cancel_request(struct request *req, void *data, bool reserved);
432 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
433 enum nvme_ctrl_state new_state);
434 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
435 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
436 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
437 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
438 const struct nvme_ctrl_ops *ops, unsigned long quirks);
439 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
440 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
441 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
442 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
443 int nvme_init_identify(struct nvme_ctrl *ctrl);
445 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
447 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
450 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
451 volatile union nvme_result *res);
453 void nvme_stop_queues(struct nvme_ctrl *ctrl);
454 void nvme_start_queues(struct nvme_ctrl *ctrl);
455 void nvme_kill_queues(struct nvme_ctrl *ctrl);
456 void nvme_sync_queues(struct nvme_ctrl *ctrl);
457 void nvme_unfreeze(struct nvme_ctrl *ctrl);
458 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
459 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
460 void nvme_start_freeze(struct nvme_ctrl *ctrl);
462 #define NVME_QID_ANY -1
463 struct request *nvme_alloc_request(struct request_queue *q,
464 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
465 void nvme_cleanup_cmd(struct request *req);
466 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
467 struct nvme_command *cmd);
468 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
469 void *buf, unsigned bufflen);
470 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
471 union nvme_result *result, void *buffer, unsigned bufflen,
472 unsigned timeout, int qid, int at_head,
473 blk_mq_req_flags_t flags, bool poll);
474 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
475 unsigned int dword11, void *buffer, size_t buflen,
477 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
478 unsigned int dword11, void *buffer, size_t buflen,
480 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
481 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
482 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
483 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
484 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
486 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
487 void *log, size_t size, u64 offset);
489 extern const struct attribute_group *nvme_ns_id_attr_groups[];
490 extern const struct block_device_operations nvme_ns_head_ops;
492 #ifdef CONFIG_NVME_MULTIPATH
493 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
495 return ctrl->ana_log_buf != NULL;
498 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
499 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
500 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
501 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
502 struct nvme_ctrl *ctrl, int *flags);
503 void nvme_failover_req(struct request *req);
504 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
505 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
506 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
507 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
508 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
509 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
510 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
511 bool nvme_mpath_clear_current_path(struct nvme_ns *ns);
512 void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl);
513 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
515 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
517 struct nvme_ns_head *head = ns->head;
519 if (head->disk && list_empty(&head->list))
520 kblockd_schedule_work(&head->requeue_work);
523 extern struct device_attribute dev_attr_ana_grpid;
524 extern struct device_attribute dev_attr_ana_state;
525 extern struct device_attribute subsys_attr_iopolicy;
528 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
533 * Without the multipath code enabled, multiple controller per subsystems are
534 * visible as devices and thus we cannot use the subsystem instance.
536 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
537 struct nvme_ctrl *ctrl, int *flags)
539 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
542 static inline void nvme_failover_req(struct request *req)
545 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
548 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
549 struct nvme_ns_head *head)
553 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
554 struct nvme_id_ns *id)
557 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
560 static inline bool nvme_mpath_clear_current_path(struct nvme_ns *ns)
564 static inline void nvme_mpath_clear_ctrl_paths(struct nvme_ctrl *ctrl)
567 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
570 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
571 struct nvme_id_ctrl *id)
573 if (ctrl->subsys->cmic & (1 << 3))
574 dev_warn(ctrl->device,
575 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
578 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
581 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
584 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
587 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
590 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
593 #endif /* CONFIG_NVME_MULTIPATH */
596 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
597 void nvme_nvm_unregister(struct nvme_ns *ns);
598 extern const struct attribute_group nvme_nvm_attr_group;
599 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
601 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
607 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
608 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
613 #endif /* CONFIG_NVM */
615 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
617 return dev_to_disk(dev)->private_data;