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 const struct nvme_ctrl_ops
*ops
;
152 struct request_queue
*admin_q
;
153 struct request_queue
*connect_q
;
156 struct blk_mq_tag_set
*tagset
;
157 struct blk_mq_tag_set
*admin_tagset
;
158 struct list_head namespaces
;
159 struct rw_semaphore namespaces_rwsem
;
160 struct device ctrl_device
;
161 struct device
*device
; /* char device */
163 struct work_struct reset_work
;
164 struct work_struct delete_work
;
166 struct nvme_subsystem
*subsys
;
167 struct list_head subsys_entry
;
169 struct opal_dev
*opal_dev
;
187 atomic_t abort_limit
;
196 unsigned int shutdown_timeout
;
199 unsigned long quirks
;
200 struct nvme_id_power_state psd
[32];
201 struct nvme_effects_log
*effects
;
202 struct work_struct scan_work
;
203 struct work_struct async_event_work
;
204 struct delayed_work ka_work
;
205 struct nvme_command ka_cmd
;
206 struct work_struct fw_act_work
;
207 unsigned long events
;
209 #ifdef CONFIG_NVME_MULTIPATH
210 /* asymmetric namespace access: */
215 struct mutex ana_lock
;
216 struct nvme_ana_rsp_hdr
*ana_log_buf
;
218 struct timer_list anatt_timer
;
219 struct work_struct ana_work
;
222 /* Power saving configuration */
223 u64 ps_max_latency_us
;
239 struct nvmf_ctrl_options
*opts
;
242 struct nvme_subsystem
{
246 * Because we unregister the device on the last put we need
247 * a separate refcount.
250 struct list_head entry
;
252 struct list_head ctrls
;
253 struct list_head nsheads
;
254 char subnqn
[NVMF_NQN_SIZE
];
257 char firmware_rev
[8];
264 * Container structure for uniqueue namespace identifiers.
273 * Anchor structure for namespaces. There is one for each namespace in a
274 * NVMe subsystem that any of our controllers can see, and the namespace
275 * structure for each controller is chained of it. For private namespaces
276 * there is a 1:1 relation to our namespace structures, that is ->list
277 * only ever has a single entry for private namespaces.
279 struct nvme_ns_head
{
280 struct list_head list
;
281 struct srcu_struct srcu
;
282 struct nvme_subsystem
*subsys
;
284 struct nvme_ns_ids ids
;
285 struct list_head entry
;
288 #ifdef CONFIG_NVME_MULTIPATH
289 struct gendisk
*disk
;
290 struct bio_list requeue_list
;
291 spinlock_t requeue_lock
;
292 struct work_struct requeue_work
;
294 struct nvme_ns __rcu
*current_path
[];
298 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
299 struct nvme_fault_inject
{
300 struct fault_attr attr
;
301 struct dentry
*parent
;
302 bool dont_retry
; /* DNR, do not retry */
303 u16 status
; /* status code */
308 struct list_head list
;
310 struct nvme_ctrl
*ctrl
;
311 struct request_queue
*queue
;
312 struct gendisk
*disk
;
313 #ifdef CONFIG_NVME_MULTIPATH
314 enum nvme_ana_state ana_state
;
317 struct list_head siblings
;
318 struct nvm_dev
*ndev
;
320 struct nvme_ns_head
*head
;
329 #define NVME_NS_REMOVING 0
330 #define NVME_NS_DEAD 1
331 #define NVME_NS_ANA_PENDING 2
334 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
335 struct nvme_fault_inject fault_inject
;
340 struct nvme_ctrl_ops
{
342 struct module
*module
;
344 #define NVME_F_FABRICS (1 << 0)
345 #define NVME_F_METADATA_SUPPORTED (1 << 1)
346 #define NVME_F_PCI_P2PDMA (1 << 2)
347 int (*reg_read32
)(struct nvme_ctrl
*ctrl
, u32 off
, u32
*val
);
348 int (*reg_write32
)(struct nvme_ctrl
*ctrl
, u32 off
, u32 val
);
349 int (*reg_read64
)(struct nvme_ctrl
*ctrl
, u32 off
, u64
*val
);
350 void (*free_ctrl
)(struct nvme_ctrl
*ctrl
);
351 void (*submit_async_event
)(struct nvme_ctrl
*ctrl
);
352 void (*delete_ctrl
)(struct nvme_ctrl
*ctrl
);
353 int (*get_address
)(struct nvme_ctrl
*ctrl
, char *buf
, int size
);
354 void (*stop_ctrl
)(struct nvme_ctrl
*ctrl
);
357 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
358 void nvme_fault_inject_init(struct nvme_ns
*ns
);
359 void nvme_fault_inject_fini(struct nvme_ns
*ns
);
360 void nvme_should_fail(struct request
*req
);
362 static inline void nvme_fault_inject_init(struct nvme_ns
*ns
) {}
363 static inline void nvme_fault_inject_fini(struct nvme_ns
*ns
) {}
364 static inline void nvme_should_fail(struct request
*req
) {}
367 static inline bool nvme_ctrl_ready(struct nvme_ctrl
*ctrl
)
371 if (ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &val
))
373 return val
& NVME_CSTS_RDY
;
376 static inline int nvme_reset_subsystem(struct nvme_ctrl
*ctrl
)
378 if (!ctrl
->subsystem
)
380 return ctrl
->ops
->reg_write32(ctrl
, NVME_REG_NSSR
, 0x4E564D65);
383 static inline u64
nvme_block_nr(struct nvme_ns
*ns
, sector_t sector
)
385 return (sector
>> (ns
->lba_shift
- 9));
388 static inline void nvme_end_request(struct request
*req
, __le16 status
,
389 union nvme_result result
)
391 struct nvme_request
*rq
= nvme_req(req
);
393 rq
->status
= le16_to_cpu(status
) >> 1;
395 /* inject error when permitted by fault injection framework */
396 nvme_should_fail(req
);
397 blk_mq_complete_request(req
);
400 static inline void nvme_get_ctrl(struct nvme_ctrl
*ctrl
)
402 get_device(ctrl
->device
);
405 static inline void nvme_put_ctrl(struct nvme_ctrl
*ctrl
)
407 put_device(ctrl
->device
);
410 void nvme_complete_rq(struct request
*req
);
411 void nvme_cancel_request(struct request
*req
, void *data
, bool reserved
);
412 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
413 enum nvme_ctrl_state new_state
);
414 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
);
415 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
);
416 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
);
417 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
418 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
);
419 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
);
420 void nvme_start_ctrl(struct nvme_ctrl
*ctrl
);
421 void nvme_stop_ctrl(struct nvme_ctrl
*ctrl
);
422 void nvme_put_ctrl(struct nvme_ctrl
*ctrl
);
423 int nvme_init_identify(struct nvme_ctrl
*ctrl
);
425 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
);
427 int nvme_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
, size_t len
,
430 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
, __le16 status
,
431 volatile union nvme_result
*res
);
433 void nvme_stop_queues(struct nvme_ctrl
*ctrl
);
434 void nvme_start_queues(struct nvme_ctrl
*ctrl
);
435 void nvme_kill_queues(struct nvme_ctrl
*ctrl
);
436 void nvme_unfreeze(struct nvme_ctrl
*ctrl
);
437 void nvme_wait_freeze(struct nvme_ctrl
*ctrl
);
438 void nvme_wait_freeze_timeout(struct nvme_ctrl
*ctrl
, long timeout
);
439 void nvme_start_freeze(struct nvme_ctrl
*ctrl
);
441 #define NVME_QID_ANY -1
442 struct request
*nvme_alloc_request(struct request_queue
*q
,
443 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
, int qid
);
444 void nvme_cleanup_cmd(struct request
*req
);
445 blk_status_t
nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
446 struct nvme_command
*cmd
);
447 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
448 void *buf
, unsigned bufflen
);
449 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
450 union nvme_result
*result
, void *buffer
, unsigned bufflen
,
451 unsigned timeout
, int qid
, int at_head
,
452 blk_mq_req_flags_t flags
);
453 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
);
454 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
);
455 int nvme_reset_ctrl(struct nvme_ctrl
*ctrl
);
456 int nvme_reset_ctrl_sync(struct nvme_ctrl
*ctrl
);
457 int nvme_delete_ctrl(struct nvme_ctrl
*ctrl
);
458 int nvme_delete_ctrl_sync(struct nvme_ctrl
*ctrl
);
460 int nvme_get_log(struct nvme_ctrl
*ctrl
, u32 nsid
, u8 log_page
, u8 lsp
,
461 void *log
, size_t size
, u64 offset
);
463 extern const struct attribute_group
*nvme_ns_id_attr_groups
[];
464 extern const struct block_device_operations nvme_ns_head_ops
;
466 #ifdef CONFIG_NVME_MULTIPATH
467 bool nvme_ctrl_use_ana(struct nvme_ctrl
*ctrl
);
468 void nvme_set_disk_name(char *disk_name
, struct nvme_ns
*ns
,
469 struct nvme_ctrl
*ctrl
, int *flags
);
470 void nvme_failover_req(struct request
*req
);
471 void nvme_kick_requeue_lists(struct nvme_ctrl
*ctrl
);
472 int nvme_mpath_alloc_disk(struct nvme_ctrl
*ctrl
,struct nvme_ns_head
*head
);
473 void nvme_mpath_add_disk(struct nvme_ns
*ns
, struct nvme_id_ns
*id
);
474 void nvme_mpath_remove_disk(struct nvme_ns_head
*head
);
475 int nvme_mpath_init(struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
);
476 void nvme_mpath_uninit(struct nvme_ctrl
*ctrl
);
477 void nvme_mpath_stop(struct nvme_ctrl
*ctrl
);
478 void nvme_mpath_clear_current_path(struct nvme_ns
*ns
);
479 struct nvme_ns
*nvme_find_path(struct nvme_ns_head
*head
);
481 static inline void nvme_mpath_check_last_path(struct nvme_ns
*ns
)
483 struct nvme_ns_head
*head
= ns
->head
;
485 if (head
->disk
&& list_empty(&head
->list
))
486 kblockd_schedule_work(&head
->requeue_work
);
489 extern struct device_attribute dev_attr_ana_grpid
;
490 extern struct device_attribute dev_attr_ana_state
;
493 static inline bool nvme_ctrl_use_ana(struct nvme_ctrl
*ctrl
)
498 * Without the multipath code enabled, multiple controller per subsystems are
499 * visible as devices and thus we cannot use the subsystem instance.
501 static inline void nvme_set_disk_name(char *disk_name
, struct nvme_ns
*ns
,
502 struct nvme_ctrl
*ctrl
, int *flags
)
504 sprintf(disk_name
, "nvme%dn%d", ctrl
->instance
, ns
->head
->instance
);
507 static inline void nvme_failover_req(struct request
*req
)
510 static inline void nvme_kick_requeue_lists(struct nvme_ctrl
*ctrl
)
513 static inline int nvme_mpath_alloc_disk(struct nvme_ctrl
*ctrl
,
514 struct nvme_ns_head
*head
)
518 static inline void nvme_mpath_add_disk(struct nvme_ns
*ns
,
519 struct nvme_id_ns
*id
)
522 static inline void nvme_mpath_remove_disk(struct nvme_ns_head
*head
)
525 static inline void nvme_mpath_clear_current_path(struct nvme_ns
*ns
)
528 static inline void nvme_mpath_check_last_path(struct nvme_ns
*ns
)
531 static inline int nvme_mpath_init(struct nvme_ctrl
*ctrl
,
532 struct nvme_id_ctrl
*id
)
536 static inline void nvme_mpath_uninit(struct nvme_ctrl
*ctrl
)
539 static inline void nvme_mpath_stop(struct nvme_ctrl
*ctrl
)
542 #endif /* CONFIG_NVME_MULTIPATH */
545 void nvme_nvm_update_nvm_info(struct nvme_ns
*ns
);
546 int nvme_nvm_register(struct nvme_ns
*ns
, char *disk_name
, int node
);
547 void nvme_nvm_unregister(struct nvme_ns
*ns
);
548 extern const struct attribute_group nvme_nvm_attr_group
;
549 int nvme_nvm_ioctl(struct nvme_ns
*ns
, unsigned int cmd
, unsigned long arg
);
551 static inline void nvme_nvm_update_nvm_info(struct nvme_ns
*ns
) {};
552 static inline int nvme_nvm_register(struct nvme_ns
*ns
, char *disk_name
,
558 static inline void nvme_nvm_unregister(struct nvme_ns
*ns
) {};
559 static inline int nvme_nvm_ioctl(struct nvme_ns
*ns
, unsigned int cmd
,
564 #endif /* CONFIG_NVM */
566 static inline struct nvme_ns
*nvme_get_ns_from_dev(struct device
*dev
)
568 return dev_to_disk(dev
)->private_data
;
571 int __init
nvme_core_init(void);
572 void nvme_core_exit(void);