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>
26 extern unsigned int nvme_io_timeout;
27 #define NVME_IO_TIMEOUT (nvme_io_timeout * HZ)
29 extern unsigned int admin_timeout;
30 #define ADMIN_TIMEOUT (admin_timeout * HZ)
32 #define NVME_DEFAULT_KATO 5
33 #define NVME_KATO_GRACE 10
35 extern struct workqueue_struct *nvme_wq;
36 extern struct workqueue_struct *nvme_reset_wq;
37 extern struct workqueue_struct *nvme_delete_wq;
45 * List of workarounds for devices that required behavior not specified in
50 * Prefers I/O aligned to a stripe size specified in a vendor
51 * specific Identify field.
53 NVME_QUIRK_STRIPE_SIZE = (1 << 0),
56 * The controller doesn't handle Identify value others than 0 or 1
59 NVME_QUIRK_IDENTIFY_CNS = (1 << 1),
62 * The controller deterministically returns O's on reads to
63 * logical blocks that deallocate was called on.
65 NVME_QUIRK_DEALLOCATE_ZEROES = (1 << 2),
68 * The controller needs a delay before starts checking the device
69 * readiness, which is done by reading the NVME_CSTS_RDY bit.
71 NVME_QUIRK_DELAY_BEFORE_CHK_RDY = (1 << 3),
74 * APST should not be used.
76 NVME_QUIRK_NO_APST = (1 << 4),
79 * The deepest sleep state should not be used.
81 NVME_QUIRK_NO_DEEPEST_PS = (1 << 5),
84 * Supports the LighNVM command set if indicated in vs[1].
86 NVME_QUIRK_LIGHTNVM = (1 << 6),
90 * Common request structure for NVMe passthrough. All drivers must have
91 * this structure as the first member of their request-private data.
94 struct nvme_command *cmd;
95 union nvme_result result;
102 * Mark a bio as coming in through the mpath node.
104 #define REQ_NVME_MPATH REQ_DRV
107 NVME_REQ_CANCELLED = (1 << 0),
110 static inline struct nvme_request *nvme_req(struct request *req)
112 return blk_mq_rq_to_pdu(req);
115 /* The below value is the specific amount of delay needed before checking
116 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
117 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
120 #define NVME_QUIRK_DELAY_AMOUNT 2300
122 enum nvme_ctrl_state {
125 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
127 NVME_CTRL_CONNECTING,
133 enum nvme_ctrl_state state;
136 const struct nvme_ctrl_ops *ops;
137 struct request_queue *admin_q;
138 struct request_queue *connect_q;
141 struct blk_mq_tag_set *tagset;
142 struct blk_mq_tag_set *admin_tagset;
143 struct list_head namespaces;
144 struct rw_semaphore namespaces_rwsem;
145 struct device ctrl_device;
146 struct device *device; /* char device */
148 struct work_struct reset_work;
149 struct work_struct delete_work;
151 struct nvme_subsystem *subsys;
152 struct list_head subsys_entry;
154 struct opal_dev *opal_dev;
170 atomic_t abort_limit;
178 unsigned int shutdown_timeout;
181 unsigned long quirks;
182 struct nvme_id_power_state psd[32];
183 struct nvme_effects_log *effects;
184 struct work_struct scan_work;
185 struct work_struct async_event_work;
186 struct delayed_work ka_work;
187 struct nvme_command ka_cmd;
188 struct work_struct fw_act_work;
190 /* Power saving configuration */
191 u64 ps_max_latency_us;
207 struct nvmf_ctrl_options *opts;
210 struct nvme_subsystem {
214 * Because we unregister the device on the last put we need
215 * a separate refcount.
218 struct list_head entry;
220 struct list_head ctrls;
221 struct list_head nsheads;
222 char subnqn[NVMF_NQN_SIZE];
225 char firmware_rev[8];
232 * Container structure for uniqueue namespace identifiers.
241 * Anchor structure for namespaces. There is one for each namespace in a
242 * NVMe subsystem that any of our controllers can see, and the namespace
243 * structure for each controller is chained of it. For private namespaces
244 * there is a 1:1 relation to our namespace structures, that is ->list
245 * only ever has a single entry for private namespaces.
247 struct nvme_ns_head {
248 #ifdef CONFIG_NVME_MULTIPATH
249 struct gendisk *disk;
250 struct nvme_ns __rcu *current_path;
251 struct bio_list requeue_list;
252 spinlock_t requeue_lock;
253 struct work_struct requeue_work;
255 struct list_head list;
256 struct srcu_struct srcu;
257 struct nvme_subsystem *subsys;
259 struct nvme_ns_ids ids;
260 struct list_head entry;
265 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
266 struct nvme_fault_inject {
267 struct fault_attr attr;
268 struct dentry *parent;
269 bool dont_retry; /* DNR, do not retry */
270 u16 status; /* status code */
275 struct list_head list;
277 struct nvme_ctrl *ctrl;
278 struct request_queue *queue;
279 struct gendisk *disk;
280 struct list_head siblings;
281 struct nvm_dev *ndev;
283 struct nvme_ns_head *head;
292 #define NVME_NS_REMOVING 0
293 #define NVME_NS_DEAD 1
296 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
297 struct nvme_fault_inject fault_inject;
302 struct nvme_ctrl_ops {
304 struct module *module;
306 #define NVME_F_FABRICS (1 << 0)
307 #define NVME_F_METADATA_SUPPORTED (1 << 1)
308 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
309 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
310 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
311 void (*free_ctrl)(struct nvme_ctrl *ctrl);
312 void (*submit_async_event)(struct nvme_ctrl *ctrl);
313 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
314 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
315 int (*reinit_request)(void *data, struct request *rq);
316 void (*stop_ctrl)(struct nvme_ctrl *ctrl);
319 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
320 void nvme_fault_inject_init(struct nvme_ns *ns);
321 void nvme_fault_inject_fini(struct nvme_ns *ns);
322 void nvme_should_fail(struct request *req);
324 static inline void nvme_fault_inject_init(struct nvme_ns *ns) {}
325 static inline void nvme_fault_inject_fini(struct nvme_ns *ns) {}
326 static inline void nvme_should_fail(struct request *req) {}
329 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl)
333 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val))
335 return val & NVME_CSTS_RDY;
338 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
340 if (!ctrl->subsystem)
342 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
345 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
347 return (sector >> (ns->lba_shift - 9));
350 static inline void nvme_cleanup_cmd(struct request *req)
352 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
353 kfree(page_address(req->special_vec.bv_page) +
354 req->special_vec.bv_offset);
358 static inline void nvme_end_request(struct request *req, __le16 status,
359 union nvme_result result)
361 struct nvme_request *rq = nvme_req(req);
363 rq->status = le16_to_cpu(status) >> 1;
365 /* inject error when permitted by fault injection framework */
366 nvme_should_fail(req);
367 blk_mq_complete_request(req);
370 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
372 get_device(ctrl->device);
375 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
377 put_device(ctrl->device);
380 void nvme_complete_rq(struct request *req);
381 void nvme_cancel_request(struct request *req, void *data, bool reserved);
382 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
383 enum nvme_ctrl_state new_state);
384 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
385 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
386 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
387 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
388 const struct nvme_ctrl_ops *ops, unsigned long quirks);
389 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
390 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
391 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
392 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
393 int nvme_init_identify(struct nvme_ctrl *ctrl);
395 void nvme_queue_scan(struct nvme_ctrl *ctrl);
396 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
398 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
401 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
402 union nvme_result *res);
404 void nvme_stop_queues(struct nvme_ctrl *ctrl);
405 void nvme_start_queues(struct nvme_ctrl *ctrl);
406 void nvme_kill_queues(struct nvme_ctrl *ctrl);
407 void nvme_unfreeze(struct nvme_ctrl *ctrl);
408 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
409 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
410 void nvme_start_freeze(struct nvme_ctrl *ctrl);
411 int nvme_reinit_tagset(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set);
413 #define NVME_QID_ANY -1
414 struct request *nvme_alloc_request(struct request_queue *q,
415 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
416 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
417 struct nvme_command *cmd);
418 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
419 void *buf, unsigned bufflen);
420 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
421 union nvme_result *result, void *buffer, unsigned bufflen,
422 unsigned timeout, int qid, int at_head,
423 blk_mq_req_flags_t flags);
424 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
425 void nvme_start_keep_alive(struct nvme_ctrl *ctrl);
426 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
427 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
428 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
429 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
430 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
432 int nvme_get_log_ext(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
433 u8 log_page, void *log, size_t size, size_t offset);
435 extern const struct attribute_group nvme_ns_id_attr_group;
436 extern const struct block_device_operations nvme_ns_head_ops;
438 #ifdef CONFIG_NVME_MULTIPATH
439 void nvme_failover_req(struct request *req);
440 bool nvme_req_needs_failover(struct request *req, blk_status_t error);
441 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
442 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
443 void nvme_mpath_add_disk(struct nvme_ns_head *head);
444 void nvme_mpath_add_disk_links(struct nvme_ns *ns);
445 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
446 void nvme_mpath_remove_disk_links(struct nvme_ns *ns);
448 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
450 struct nvme_ns_head *head = ns->head;
452 if (head && ns == srcu_dereference(head->current_path, &head->srcu))
453 rcu_assign_pointer(head->current_path, NULL);
455 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
457 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
459 struct nvme_ns_head *head = ns->head;
461 if (head->disk && list_empty(&head->list))
462 kblockd_schedule_work(&head->requeue_work);
466 static inline void nvme_failover_req(struct request *req)
469 static inline bool nvme_req_needs_failover(struct request *req,
474 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
477 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
478 struct nvme_ns_head *head)
482 static inline void nvme_mpath_add_disk(struct nvme_ns_head *head)
485 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
488 static inline void nvme_mpath_add_disk_links(struct nvme_ns *ns)
491 static inline void nvme_mpath_remove_disk_links(struct nvme_ns *ns)
494 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
497 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
500 #endif /* CONFIG_NVME_MULTIPATH */
503 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
504 void nvme_nvm_unregister(struct nvme_ns *ns);
505 int nvme_nvm_register_sysfs(struct nvme_ns *ns);
506 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns);
507 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
509 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
515 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
516 static inline int nvme_nvm_register_sysfs(struct nvme_ns *ns)
520 static inline void nvme_nvm_unregister_sysfs(struct nvme_ns *ns) {};
521 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
526 #endif /* CONFIG_NVM */
528 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
530 return dev_to_disk(dev)->private_data;
533 int __init nvme_core_init(void);
534 void nvme_core_exit(void);