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;
108 * Mark a bio as coming in through the mpath node.
110 #define REQ_NVME_MPATH REQ_DRV
113 NVME_REQ_CANCELLED = (1 << 0),
114 NVME_REQ_USERCMD = (1 << 1),
117 static inline struct nvme_request *nvme_req(struct request *req)
119 return blk_mq_rq_to_pdu(req);
122 /* The below value is the specific amount of delay needed before checking
123 * readiness in case of the PCI_DEVICE(0x1c58, 0x0003), which needs the
124 * NVME_QUIRK_DELAY_BEFORE_CHK_RDY quirk enabled. The value (in ms) was
127 #define NVME_QUIRK_DELAY_AMOUNT 2300
129 enum nvme_ctrl_state {
132 NVME_CTRL_ADMIN_ONLY, /* Only admin queue live */
134 NVME_CTRL_CONNECTING,
140 enum nvme_ctrl_state state;
143 const struct nvme_ctrl_ops *ops;
144 struct request_queue *admin_q;
145 struct request_queue *connect_q;
148 struct blk_mq_tag_set *tagset;
149 struct blk_mq_tag_set *admin_tagset;
150 struct list_head namespaces;
151 struct rw_semaphore namespaces_rwsem;
152 struct device ctrl_device;
153 struct device *device; /* char device */
155 struct work_struct reset_work;
156 struct work_struct delete_work;
158 struct nvme_subsystem *subsys;
159 struct list_head subsys_entry;
161 struct opal_dev *opal_dev;
177 atomic_t abort_limit;
186 unsigned int shutdown_timeout;
189 unsigned long quirks;
190 struct nvme_id_power_state psd[32];
191 struct nvme_effects_log *effects;
192 struct work_struct scan_work;
193 struct work_struct async_event_work;
194 struct delayed_work ka_work;
195 struct nvme_command ka_cmd;
196 struct work_struct fw_act_work;
197 unsigned long events;
199 /* Power saving configuration */
200 u64 ps_max_latency_us;
216 struct nvmf_ctrl_options *opts;
219 struct nvme_subsystem {
223 * Because we unregister the device on the last put we need
224 * a separate refcount.
227 struct list_head entry;
229 struct list_head ctrls;
230 struct list_head nsheads;
231 char subnqn[NVMF_NQN_SIZE];
234 char firmware_rev[8];
241 * Container structure for uniqueue namespace identifiers.
250 * Anchor structure for namespaces. There is one for each namespace in a
251 * NVMe subsystem that any of our controllers can see, and the namespace
252 * structure for each controller is chained of it. For private namespaces
253 * there is a 1:1 relation to our namespace structures, that is ->list
254 * only ever has a single entry for private namespaces.
256 struct nvme_ns_head {
257 #ifdef CONFIG_NVME_MULTIPATH
258 struct gendisk *disk;
259 struct nvme_ns __rcu *current_path;
260 struct bio_list requeue_list;
261 spinlock_t requeue_lock;
262 struct work_struct requeue_work;
264 struct list_head list;
265 struct srcu_struct srcu;
266 struct nvme_subsystem *subsys;
268 struct nvme_ns_ids ids;
269 struct list_head entry;
274 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
275 struct nvme_fault_inject {
276 struct fault_attr attr;
277 struct dentry *parent;
278 bool dont_retry; /* DNR, do not retry */
279 u16 status; /* status code */
284 struct list_head list;
286 struct nvme_ctrl *ctrl;
287 struct request_queue *queue;
288 struct gendisk *disk;
289 struct list_head siblings;
290 struct nvm_dev *ndev;
292 struct nvme_ns_head *head;
301 #define NVME_NS_REMOVING 0
302 #define NVME_NS_DEAD 1
305 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
306 struct nvme_fault_inject fault_inject;
311 struct nvme_ctrl_ops {
313 struct module *module;
315 #define NVME_F_FABRICS (1 << 0)
316 #define NVME_F_METADATA_SUPPORTED (1 << 1)
317 int (*reg_read32)(struct nvme_ctrl *ctrl, u32 off, u32 *val);
318 int (*reg_write32)(struct nvme_ctrl *ctrl, u32 off, u32 val);
319 int (*reg_read64)(struct nvme_ctrl *ctrl, u32 off, u64 *val);
320 void (*free_ctrl)(struct nvme_ctrl *ctrl);
321 void (*submit_async_event)(struct nvme_ctrl *ctrl);
322 void (*delete_ctrl)(struct nvme_ctrl *ctrl);
323 int (*get_address)(struct nvme_ctrl *ctrl, char *buf, int size);
324 void (*stop_ctrl)(struct nvme_ctrl *ctrl);
327 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
328 void nvme_fault_inject_init(struct nvme_ns *ns);
329 void nvme_fault_inject_fini(struct nvme_ns *ns);
330 void nvme_should_fail(struct request *req);
332 static inline void nvme_fault_inject_init(struct nvme_ns *ns) {}
333 static inline void nvme_fault_inject_fini(struct nvme_ns *ns) {}
334 static inline void nvme_should_fail(struct request *req) {}
337 static inline bool nvme_ctrl_ready(struct nvme_ctrl *ctrl)
341 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &val))
343 return val & NVME_CSTS_RDY;
346 static inline int nvme_reset_subsystem(struct nvme_ctrl *ctrl)
348 if (!ctrl->subsystem)
350 return ctrl->ops->reg_write32(ctrl, NVME_REG_NSSR, 0x4E564D65);
353 static inline u64 nvme_block_nr(struct nvme_ns *ns, sector_t sector)
355 return (sector >> (ns->lba_shift - 9));
358 static inline void nvme_cleanup_cmd(struct request *req)
360 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
361 kfree(page_address(req->special_vec.bv_page) +
362 req->special_vec.bv_offset);
366 static inline void nvme_end_request(struct request *req, __le16 status,
367 union nvme_result result)
369 struct nvme_request *rq = nvme_req(req);
371 rq->status = le16_to_cpu(status) >> 1;
373 /* inject error when permitted by fault injection framework */
374 nvme_should_fail(req);
375 blk_mq_complete_request(req);
378 static inline void nvme_get_ctrl(struct nvme_ctrl *ctrl)
380 get_device(ctrl->device);
383 static inline void nvme_put_ctrl(struct nvme_ctrl *ctrl)
385 put_device(ctrl->device);
388 void nvme_complete_rq(struct request *req);
389 void nvme_cancel_request(struct request *req, void *data, bool reserved);
390 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
391 enum nvme_ctrl_state new_state);
392 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
393 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap);
394 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl);
395 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
396 const struct nvme_ctrl_ops *ops, unsigned long quirks);
397 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl);
398 void nvme_start_ctrl(struct nvme_ctrl *ctrl);
399 void nvme_stop_ctrl(struct nvme_ctrl *ctrl);
400 void nvme_put_ctrl(struct nvme_ctrl *ctrl);
401 int nvme_init_identify(struct nvme_ctrl *ctrl);
403 void nvme_remove_namespaces(struct nvme_ctrl *ctrl);
405 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
408 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
409 volatile union nvme_result *res);
411 void nvme_stop_queues(struct nvme_ctrl *ctrl);
412 void nvme_start_queues(struct nvme_ctrl *ctrl);
413 void nvme_kill_queues(struct nvme_ctrl *ctrl);
414 void nvme_unfreeze(struct nvme_ctrl *ctrl);
415 void nvme_wait_freeze(struct nvme_ctrl *ctrl);
416 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout);
417 void nvme_start_freeze(struct nvme_ctrl *ctrl);
419 #define NVME_QID_ANY -1
420 struct request *nvme_alloc_request(struct request_queue *q,
421 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid);
422 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
423 struct nvme_command *cmd);
424 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
425 void *buf, unsigned bufflen);
426 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
427 union nvme_result *result, void *buffer, unsigned bufflen,
428 unsigned timeout, int qid, int at_head,
429 blk_mq_req_flags_t flags);
430 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
431 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
432 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
433 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl);
434 int nvme_delete_ctrl(struct nvme_ctrl *ctrl);
435 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl);
437 int nvme_get_log_ext(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
438 u8 log_page, void *log, size_t size, u64 offset);
440 extern const struct attribute_group nvme_ns_id_attr_group;
441 extern const struct block_device_operations nvme_ns_head_ops;
443 #ifdef CONFIG_NVME_MULTIPATH
444 void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
445 struct nvme_ctrl *ctrl, int *flags);
446 void nvme_failover_req(struct request *req);
447 bool nvme_req_needs_failover(struct request *req, blk_status_t error);
448 void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl);
449 int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,struct nvme_ns_head *head);
450 void nvme_mpath_add_disk(struct nvme_ns_head *head);
451 void nvme_mpath_remove_disk(struct nvme_ns_head *head);
453 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
455 struct nvme_ns_head *head = ns->head;
457 if (head && ns == rcu_access_pointer(head->current_path))
458 rcu_assign_pointer(head->current_path, NULL);
460 struct nvme_ns *nvme_find_path(struct nvme_ns_head *head);
462 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
464 struct nvme_ns_head *head = ns->head;
466 if (head->disk && list_empty(&head->list))
467 kblockd_schedule_work(&head->requeue_work);
472 * Without the multipath code enabled, multiple controller per subsystems are
473 * visible as devices and thus we cannot use the subsystem instance.
475 static inline void nvme_set_disk_name(char *disk_name, struct nvme_ns *ns,
476 struct nvme_ctrl *ctrl, int *flags)
478 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->head->instance);
481 static inline void nvme_failover_req(struct request *req)
484 static inline bool nvme_req_needs_failover(struct request *req,
489 static inline void nvme_kick_requeue_lists(struct nvme_ctrl *ctrl)
492 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl *ctrl,
493 struct nvme_ns_head *head)
497 static inline void nvme_mpath_add_disk(struct nvme_ns_head *head)
500 static inline void nvme_mpath_remove_disk(struct nvme_ns_head *head)
503 static inline void nvme_mpath_clear_current_path(struct nvme_ns *ns)
506 static inline void nvme_mpath_check_last_path(struct nvme_ns *ns)
509 #endif /* CONFIG_NVME_MULTIPATH */
512 void nvme_nvm_update_nvm_info(struct nvme_ns *ns);
513 int nvme_nvm_register(struct nvme_ns *ns, char *disk_name, int node);
514 void nvme_nvm_unregister(struct nvme_ns *ns);
515 int nvme_nvm_register_sysfs(struct nvme_ns *ns);
516 void nvme_nvm_unregister_sysfs(struct nvme_ns *ns);
517 int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd, unsigned long arg);
519 static inline void nvme_nvm_update_nvm_info(struct nvme_ns *ns) {};
520 static inline int nvme_nvm_register(struct nvme_ns *ns, char *disk_name,
526 static inline void nvme_nvm_unregister(struct nvme_ns *ns) {};
527 static inline int nvme_nvm_register_sysfs(struct nvme_ns *ns)
531 static inline void nvme_nvm_unregister_sysfs(struct nvme_ns *ns) {};
532 static inline int nvme_nvm_ioctl(struct nvme_ns *ns, unsigned int cmd,
537 #endif /* CONFIG_NVM */
539 static inline struct nvme_ns *nvme_get_ns_from_dev(struct device *dev)
541 return dev_to_disk(dev)->private_data;
544 int __init nvme_core_init(void);
545 void nvme_core_exit(void);