2 * Copyright (c) 2011-2014, Intel Corporation.
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms and conditions of the GNU General Public License,
6 * version 2, as published by the Free Software Foundation.
8 * This program is distributed in the hope it will be useful, but WITHOUT
9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 #include <linux/nvme.h>
18 #include <linux/cdev.h>
19 #include <linux/pci.h>
20 #include <linux/kref.h>
21 #include <linux/blk-mq.h>
22 #include <linux/lightnvm.h>
23 #include <linux/sed-opal.h>
24 #include <linux/fault-inject.h>
25 #include <linux/rcupdate.h>
27 extern unsigned int nvme_io_timeout;
28 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
30 extern unsigned int admin_timeout;
31 #define ADMIN_TIMEOUT (admin_timeout * HZ)
33 #define NVME_DEFAULT_KATO 5
34 #define NVME_KATO_GRACE 10
36 extern struct workqueue_struct *nvme_wq;
37 extern struct workqueue_struct *nvme_reset_wq;
38 extern struct workqueue_struct *nvme_delete_wq;
46 * List of workarounds for devices that required behavior not specified in
51 * Prefers I/O aligned to a stripe size specified in a vendor
52 * specific Identify field.
54 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
57 * The controller doesn't handle Identify value others than 0 or 1
60 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
63 * The controller deterministically returns O's on reads to
64 * logical blocks that deallocate was called on.
66 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
69 * The controller needs a delay before starts checking the device
70 * readiness, which is done by reading the NVME_CSTS_RDY bit.
72 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
75 * APST should not be used.
77 NVME_QUIRK_NO_APST = (1 << 4),
80 * The deepest sleep state should not be used.
82 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
85 * Supports the LighNVM command set if indicated in vs[1].
87 NVME_QUIRK_LIGHTNVM = (1 << 6),
90 * Set MEDIUM priority on SQ creation
92 NVME_QUIRK_MEDIUM_PRIO_SQ = (1 << 7),
96 * Common request structure for NVMe passthrough. All drivers must have
97 * this structure as the first member of their request-private data.
100 struct nvme_command *cmd;
101 union nvme_result result;
105 struct nvme_ctrl *ctrl;
109 * Mark a bio as coming in through the mpath node.
111 #define REQ_NVME_MPATH REQ_DRV
114 NVME_REQ_CANCELLED = (1 << 0),
115 NVME_REQ_USERCMD = (1 << 1),
118 static inline struct nvme_request *nvme_req(struct request *req)
120 return blk_mq_rq_to_pdu(req);
123 static inline u16 nvme_req_qid(struct request *req)
127 return blk_mq_unique_tag_to_hwq(blk_mq_unique_tag(req)) + 1;
130 /* The below value is the specific amount of delay needed before checking
131 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
132 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
135 #define NVME_QUIRK_DELAY_AMOUNT 2300
137 enum nvme_ctrl_state {
140 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
142 NVME_CTRL_CONNECTING,
148 enum nvme_ctrl_state state;
151 struct mutex scan_lock;
152 const struct nvme_ctrl_ops *ops;
153 struct request_queue *admin_q;
154 struct request_queue *connect_q;
157 struct blk_mq_tag_set *tagset;
158 struct blk_mq_tag_set *admin_tagset;
159 struct list_head namespaces;
160 struct rw_semaphore namespaces_rwsem;
161 struct device ctrl_device;
162 struct device *device; /* char device */
164 struct work_struct reset_work;
165 struct work_struct delete_work;
167 struct nvme_subsystem *subsys;
168 struct list_head subsys_entry;
170 struct opal_dev *opal_dev;
188 atomic_t abort_limit;
197 unsigned int shutdown_timeout;
200 unsigned long quirks;
201 struct nvme_id_power_state psd[32];
202 struct nvme_effects_log *effects;
203 struct work_struct scan_work;
204 struct work_struct async_event_work;
205 struct delayed_work ka_work;
206 struct nvme_command ka_cmd;
207 struct work_struct fw_act_work;
208 unsigned long events;
211 #ifdef CONFIG_NVME_MULTIPATH
212 /* asymmetric namespace access: */
217 struct mutex ana_lock;
218 struct nvme_ana_rsp_hdr *ana_log_buf;
220 struct timer_list anatt_timer;
221 struct work_struct ana_work;
224 /* Power saving configuration */
225 u64 ps_max_latency_us;
241 struct nvmf_ctrl_options *opts;
243 struct page *discard_page;
244 unsigned long discard_page_busy;
247 struct nvme_subsystem {
251 * Because we unregister the device on the last put we need
252 * a separate refcount.
255 struct list_head entry;
257 struct list_head ctrls;
258 struct list_head nsheads;
259 char subnqn[NVMF_NQN_SIZE];
262 char firmware_rev[8];
269 * Container structure for uniqueue namespace identifiers.
278 * Anchor structure for namespaces. There is one for each namespace in a
279 * NVMe subsystem that any of our controllers can see, and the namespace
280 * structure for each controller is chained of it. For private namespaces
281 * there is a 1:1 relation to our namespace structures, that is ->list
282 * only ever has a single entry for private namespaces.
284 struct nvme_ns_head {
285 #ifdef CONFIG_NVME_MULTIPATH
286 struct gendisk *disk;
287 struct nvme_ns __rcu *current_path;
288 struct bio_list requeue_list;
289 spinlock_t requeue_lock;
290 struct work_struct requeue_work;
293 struct list_head list;
294 struct srcu_struct srcu;
295 struct nvme_subsystem *subsys;
297 struct nvme_ns_ids ids;
298 struct list_head entry;
303 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
304 struct nvme_fault_inject {
305 struct fault_attr attr;
306 struct dentry *parent;
307 bool dont_retry; /* DNR, do not retry */
308 u16 status; /* status code */
313 struct list_head list;
315 struct nvme_ctrl *ctrl;
316 struct request_queue *queue;
317 struct gendisk *disk;
318 #ifdef CONFIG_NVME_MULTIPATH
319 enum nvme_ana_state ana_state;
322 struct list_head siblings;
323 struct nvm_dev *ndev;
325 struct nvme_ns_head *head;
334 #define NVME_NS_REMOVING 0
335 #define NVME_NS_DEAD 1
336 #define NVME_NS_ANA_PENDING 2
339 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
340 struct nvme_fault_inject fault_inject;
345 struct nvme_ctrl_ops {
347 struct module *module;
349 #define NVME_F_FABRICS (1 << 0)
350 #define NVME_F_METADATA_SUPPORTED (1 << 1)
351 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
352 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
353 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
354 void (*free_ctrl)(struct nvme_ctrl *ctrl);
355 void (*submit_async_event)(struct nvme_ctrl *ctrl);
356 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
357 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
358 void (*stop_ctrl)(struct nvme_ctrl *ctrl);
361 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
362 void nvme_fault_inject_init(struct nvme_ns *ns);
363 void nvme_fault_inject_fini(struct nvme_ns *ns);
364 void nvme_should_fail(struct request *req);
366 static inline void nvme_fault_inject_init(struct nvme_ns *ns) {}
367 static inline void nvme_fault_inject_fini(struct nvme_ns *ns) {}
368 static inline void nvme_should_fail(struct request *req) {}
371 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl)
375 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val))
377 return val & NVME_CSTS_RDY;
380 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
382 if (!ctrl->subsystem)
384 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
387 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
389 return (sector >> (ns->lba_shift - 9));
392 static inline void nvme_end_request(struct request *req, __le16 status,
393 union nvme_result result)
395 struct nvme_request *rq = nvme_req(req);
397 rq->status = le16_to_cpu(status) >> 1;
399 /* inject error when permitted by fault injection framework */
400 nvme_should_fail(req);
401 blk_mq_complete_request(req);
404 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
406 get_device(ctrl->device);
409 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
411 put_device(ctrl->device);
414 void nvme_complete_rq(struct request *req);
415 void nvme_cancel_request(struct request *req, void *data, bool reserved);
416 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
417 enum nvme_ctrl_state new_state);
418 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
419 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
420 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
421 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
422 const struct nvme_ctrl_ops *ops, unsigned long quirks);
423 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
424 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
425 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
426 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
427 int nvme_init_identify(struct nvme_ctrl *ctrl);
429 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
431 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
434 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
435 volatile union nvme_result *res);
437 void nvme_stop_queues(struct nvme_ctrl *ctrl);
438 void nvme_start_queues(struct nvme_ctrl *ctrl);
439 void nvme_kill_queues(struct nvme_ctrl *ctrl);
440 void nvme_unfreeze(struct nvme_ctrl *ctrl);
441 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
442 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
443 void nvme_start_freeze(struct nvme_ctrl *ctrl);
445 #define NVME_QID_ANY -1
446 struct request *nvme_alloc_request(struct request_queue *q,
447 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
448 void nvme_cleanup_cmd(struct request *req);
449 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
450 struct nvme_command *cmd);
451 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
452 void *buf, unsigned bufflen);
453 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
454 union nvme_result *result, void *buffer, unsigned bufflen,
455 unsigned timeout, int qid, int at_head,
456 blk_mq_req_flags_t flags);
457 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
458 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
459 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
460 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
461 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
462 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
464 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
465 void *log, size_t size, u64 offset);
467 extern const struct attribute_group *nvme_ns_id_attr_groups[];
468 extern const struct block_device_operations nvme_ns_head_ops;
470 #ifdef CONFIG_NVME_MULTIPATH
471 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
473 return ctrl->ana_log_buf != NULL;
476 void nvme_mpath_unfreeze(struct nvme_subsystem *subsys);
477 void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys);
478 void nvme_mpath_start_freeze(struct nvme_subsystem *subsys);
479 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
480 struct nvme_ctrl *ctrl, int *flags);
481 bool nvme_failover_req(struct request *req);
482 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
483 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
484 void nvme_mpath_add_disk(struct nvme_ns *ns, struct nvme_id_ns *id);
485 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
486 int nvme_mpath_init(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id);
487 void nvme_mpath_uninit(struct nvme_ctrl *ctrl);
488 void nvme_mpath_stop(struct nvme_ctrl *ctrl);
490 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
492 struct nvme_ns_head *head = ns->head;
494 if (head && ns == rcu_access_pointer(head->current_path))
495 rcu_assign_pointer(head->current_path, NULL);
497 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
499 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
501 struct nvme_ns_head *head = ns->head;
503 if (head->disk && list_empty(&head->list))
504 kblockd_schedule_work(&head->requeue_work);
507 static inline void nvme_mpath_update_disk_size(struct gendisk *disk)
509 struct block_device *bdev = bdget_disk(disk, 0);
512 bd_set_size(bdev, get_capacity(disk) << SECTOR_SHIFT);
517 extern struct device_attribute dev_attr_ana_grpid;
518 extern struct device_attribute dev_attr_ana_state;
521 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl *ctrl)
526 * Without the multipath code enabled, multiple controller per subsystems are
527 * visible as devices and thus we cannot use the subsystem instance.
529 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
530 struct nvme_ctrl *ctrl, int *flags)
532 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
535 static inline bool nvme_failover_req(struct request *req)
539 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
542 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
543 struct nvme_ns_head *head)
547 static inline void nvme_mpath_add_disk(struct nvme_ns *ns,
548 struct nvme_id_ns *id)
551 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
554 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
557 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
560 static inline int nvme_mpath_init(struct nvme_ctrl *ctrl,
561 struct nvme_id_ctrl *id)
563 if (ctrl->subsys->cmic & (1 << 3))
564 dev_warn(ctrl->device,
565 "Please enable CONFIG_NVME_MULTIPATH for full support of multi-port devices.\n");
568 static inline void nvme_mpath_uninit(struct nvme_ctrl *ctrl)
571 static inline void nvme_mpath_stop(struct nvme_ctrl *ctrl)
574 static inline void nvme_mpath_unfreeze(struct nvme_subsystem *subsys)
577 static inline void nvme_mpath_wait_freeze(struct nvme_subsystem *subsys)
580 static inline void nvme_mpath_start_freeze(struct nvme_subsystem *subsys)
583 static inline void nvme_mpath_update_disk_size(struct gendisk *disk)
586 #endif /* CONFIG_NVME_MULTIPATH */
589 void nvme_nvm_update_nvm_info(struct nvme_ns *ns);
590 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
591 void nvme_nvm_unregister(struct nvme_ns *ns);
592 extern const struct attribute_group nvme_nvm_attr_group;
593 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
595 static inline void nvme_nvm_update_nvm_info(struct nvme_ns *ns) {};
596 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
602 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
603 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
608 #endif /* CONFIG_NVM */
610 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
612 return dev_to_disk(dev)->private_data;
615 int __init nvme_core_init(void);
616 void nvme_core_exit(void);