1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout
= 60;
34 module_param(admin_timeout
, uint
, 0644);
35 MODULE_PARM_DESC(admin_timeout
, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout
);
38 unsigned int nvme_io_timeout
= 30;
39 module_param_named(io_timeout
, nvme_io_timeout
, uint
, 0644);
40 MODULE_PARM_DESC(io_timeout
, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout
);
43 static unsigned char shutdown_timeout
= 5;
44 module_param(shutdown_timeout
, byte
, 0644);
45 MODULE_PARM_DESC(shutdown_timeout
, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries
= 5;
48 module_param_named(max_retries
, nvme_max_retries
, byte
, 0644);
49 MODULE_PARM_DESC(max_retries
, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us
= 100000;
52 module_param(default_ps_max_latency_us
, ulong
, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us
,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst
;
57 module_param(force_apst
, bool, 0644);
58 MODULE_PARM_DESC(force_apst
, "allow APST for newly enumerated devices even if quirked off");
61 module_param(streams
, bool, 0644);
62 MODULE_PARM_DESC(streams
, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such as scan, aen handling, fw activation,
70 * keep-alive, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct
*nvme_wq
;
76 EXPORT_SYMBOL_GPL(nvme_wq
);
78 struct workqueue_struct
*nvme_reset_wq
;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq
);
81 struct workqueue_struct
*nvme_delete_wq
;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq
);
84 static LIST_HEAD(nvme_subsystems
);
85 static DEFINE_MUTEX(nvme_subsystems_lock
);
87 static DEFINE_IDA(nvme_instance_ida
);
88 static dev_t nvme_chr_devt
;
89 static struct class *nvme_class
;
90 static struct class *nvme_subsys_class
;
92 static int nvme_revalidate_disk(struct gendisk
*disk
);
93 static void nvme_put_subsystem(struct nvme_subsystem
*subsys
);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
97 static void nvme_set_queue_dying(struct nvme_ns
*ns
)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns
->disk
|| test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
105 blk_set_queue_dying(ns
->queue
);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns
->queue
);
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
111 revalidate_disk(ns
->disk
);
114 static void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
117 * Only new queue scan work when admin and IO queues are both alive
119 if (ctrl
->state
== NVME_CTRL_LIVE
&& ctrl
->tagset
)
120 queue_work(nvme_wq
, &ctrl
->scan_work
);
124 * Use this function to proceed with scheduling reset_work for a controller
125 * that had previously been set to the resetting state. This is intended for
126 * code paths that can't be interrupted by other reset attempts. A hot removal
127 * may prevent this from succeeding.
129 int nvme_try_sched_reset(struct nvme_ctrl
*ctrl
)
131 if (ctrl
->state
!= NVME_CTRL_RESETTING
)
133 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset
);
139 int nvme_reset_ctrl(struct nvme_ctrl
*ctrl
)
141 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
143 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl
);
149 int nvme_reset_ctrl_sync(struct nvme_ctrl
*ctrl
)
153 ret
= nvme_reset_ctrl(ctrl
);
155 flush_work(&ctrl
->reset_work
);
156 if (ctrl
->state
!= NVME_CTRL_LIVE
)
162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync
);
164 static void nvme_do_delete_ctrl(struct nvme_ctrl
*ctrl
)
166 dev_info(ctrl
->device
,
167 "Removing ctrl: NQN \"%s\"\n", ctrl
->opts
->subsysnqn
);
169 flush_work(&ctrl
->reset_work
);
170 nvme_stop_ctrl(ctrl
);
171 nvme_remove_namespaces(ctrl
);
172 ctrl
->ops
->delete_ctrl(ctrl
);
173 nvme_uninit_ctrl(ctrl
);
176 static void nvme_delete_ctrl_work(struct work_struct
*work
)
178 struct nvme_ctrl
*ctrl
=
179 container_of(work
, struct nvme_ctrl
, delete_work
);
181 nvme_do_delete_ctrl(ctrl
);
184 int nvme_delete_ctrl(struct nvme_ctrl
*ctrl
)
186 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
188 if (!queue_work(nvme_delete_wq
, &ctrl
->delete_work
))
192 EXPORT_SYMBOL_GPL(nvme_delete_ctrl
);
194 static void nvme_delete_ctrl_sync(struct nvme_ctrl
*ctrl
)
197 * Keep a reference until nvme_do_delete_ctrl() complete,
198 * since ->delete_ctrl can free the controller.
201 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
202 nvme_do_delete_ctrl(ctrl
);
206 static blk_status_t
nvme_error_status(u16 status
)
208 switch (status
& 0x7ff) {
209 case NVME_SC_SUCCESS
:
211 case NVME_SC_CAP_EXCEEDED
:
212 return BLK_STS_NOSPC
;
213 case NVME_SC_LBA_RANGE
:
214 case NVME_SC_CMD_INTERRUPTED
:
215 case NVME_SC_NS_NOT_READY
:
216 return BLK_STS_TARGET
;
217 case NVME_SC_BAD_ATTRIBUTES
:
218 case NVME_SC_ONCS_NOT_SUPPORTED
:
219 case NVME_SC_INVALID_OPCODE
:
220 case NVME_SC_INVALID_FIELD
:
221 case NVME_SC_INVALID_NS
:
222 return BLK_STS_NOTSUPP
;
223 case NVME_SC_WRITE_FAULT
:
224 case NVME_SC_READ_ERROR
:
225 case NVME_SC_UNWRITTEN_BLOCK
:
226 case NVME_SC_ACCESS_DENIED
:
227 case NVME_SC_READ_ONLY
:
228 case NVME_SC_COMPARE_FAILED
:
229 return BLK_STS_MEDIUM
;
230 case NVME_SC_GUARD_CHECK
:
231 case NVME_SC_APPTAG_CHECK
:
232 case NVME_SC_REFTAG_CHECK
:
233 case NVME_SC_INVALID_PI
:
234 return BLK_STS_PROTECTION
;
235 case NVME_SC_RESERVATION_CONFLICT
:
236 return BLK_STS_NEXUS
;
237 case NVME_SC_HOST_PATH_ERROR
:
238 return BLK_STS_TRANSPORT
;
240 return BLK_STS_IOERR
;
244 static inline bool nvme_req_needs_retry(struct request
*req
)
246 if (blk_noretry_request(req
))
248 if (nvme_req(req
)->status
& NVME_SC_DNR
)
250 if (nvme_req(req
)->retries
>= nvme_max_retries
)
255 static void nvme_retry_req(struct request
*req
)
257 struct nvme_ns
*ns
= req
->q
->queuedata
;
258 unsigned long delay
= 0;
261 /* The mask and shift result must be <= 3 */
262 crd
= (nvme_req(req
)->status
& NVME_SC_CRD
) >> 11;
264 delay
= ns
->ctrl
->crdt
[crd
- 1] * 100;
266 nvme_req(req
)->retries
++;
267 blk_mq_requeue_request(req
, false);
268 blk_mq_delay_kick_requeue_list(req
->q
, delay
);
271 void nvme_complete_rq(struct request
*req
)
273 blk_status_t status
= nvme_error_status(nvme_req(req
)->status
);
275 trace_nvme_complete_rq(req
);
277 nvme_cleanup_cmd(req
);
279 if (nvme_req(req
)->ctrl
->kas
)
280 nvme_req(req
)->ctrl
->comp_seen
= true;
282 if (unlikely(status
!= BLK_STS_OK
&& nvme_req_needs_retry(req
))) {
283 if ((req
->cmd_flags
& REQ_NVME_MPATH
) && nvme_failover_req(req
))
286 if (!blk_queue_dying(req
->q
)) {
292 nvme_trace_bio_complete(req
, status
);
293 blk_mq_end_request(req
, status
);
295 EXPORT_SYMBOL_GPL(nvme_complete_rq
);
297 bool nvme_cancel_request(struct request
*req
, void *data
, bool reserved
)
299 dev_dbg_ratelimited(((struct nvme_ctrl
*) data
)->device
,
300 "Cancelling I/O %d", req
->tag
);
302 /* don't abort one completed request */
303 if (blk_mq_request_completed(req
))
306 nvme_req(req
)->status
= NVME_SC_HOST_ABORTED_CMD
;
307 blk_mq_complete_request(req
);
310 EXPORT_SYMBOL_GPL(nvme_cancel_request
);
312 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
313 enum nvme_ctrl_state new_state
)
315 enum nvme_ctrl_state old_state
;
317 bool changed
= false;
319 spin_lock_irqsave(&ctrl
->lock
, flags
);
321 old_state
= ctrl
->state
;
326 case NVME_CTRL_RESETTING
:
327 case NVME_CTRL_CONNECTING
:
334 case NVME_CTRL_RESETTING
:
344 case NVME_CTRL_CONNECTING
:
347 case NVME_CTRL_RESETTING
:
354 case NVME_CTRL_DELETING
:
357 case NVME_CTRL_RESETTING
:
358 case NVME_CTRL_CONNECTING
:
367 case NVME_CTRL_DELETING
:
379 ctrl
->state
= new_state
;
380 wake_up_all(&ctrl
->state_wq
);
383 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
384 if (changed
&& ctrl
->state
== NVME_CTRL_LIVE
)
385 nvme_kick_requeue_lists(ctrl
);
388 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
391 * Returns true for sink states that can't ever transition back to live.
393 static bool nvme_state_terminal(struct nvme_ctrl
*ctrl
)
395 switch (ctrl
->state
) {
398 case NVME_CTRL_RESETTING
:
399 case NVME_CTRL_CONNECTING
:
401 case NVME_CTRL_DELETING
:
405 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl
->state
);
411 * Waits for the controller state to be resetting, or returns false if it is
412 * not possible to ever transition to that state.
414 bool nvme_wait_reset(struct nvme_ctrl
*ctrl
)
416 wait_event(ctrl
->state_wq
,
417 nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
) ||
418 nvme_state_terminal(ctrl
));
419 return ctrl
->state
== NVME_CTRL_RESETTING
;
421 EXPORT_SYMBOL_GPL(nvme_wait_reset
);
423 static void nvme_free_ns_head(struct kref
*ref
)
425 struct nvme_ns_head
*head
=
426 container_of(ref
, struct nvme_ns_head
, ref
);
428 nvme_mpath_remove_disk(head
);
429 ida_simple_remove(&head
->subsys
->ns_ida
, head
->instance
);
430 cleanup_srcu_struct(&head
->srcu
);
431 nvme_put_subsystem(head
->subsys
);
435 static void nvme_put_ns_head(struct nvme_ns_head
*head
)
437 kref_put(&head
->ref
, nvme_free_ns_head
);
440 static void nvme_free_ns(struct kref
*kref
)
442 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
445 nvme_nvm_unregister(ns
);
448 nvme_put_ns_head(ns
->head
);
449 nvme_put_ctrl(ns
->ctrl
);
453 static void nvme_put_ns(struct nvme_ns
*ns
)
455 kref_put(&ns
->kref
, nvme_free_ns
);
458 static inline void nvme_clear_nvme_request(struct request
*req
)
460 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
461 nvme_req(req
)->retries
= 0;
462 nvme_req(req
)->flags
= 0;
463 req
->rq_flags
|= RQF_DONTPREP
;
467 struct request
*nvme_alloc_request(struct request_queue
*q
,
468 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
, int qid
)
470 unsigned op
= nvme_is_write(cmd
) ? REQ_OP_DRV_OUT
: REQ_OP_DRV_IN
;
473 if (qid
== NVME_QID_ANY
) {
474 req
= blk_mq_alloc_request(q
, op
, flags
);
476 req
= blk_mq_alloc_request_hctx(q
, op
, flags
,
482 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
483 nvme_clear_nvme_request(req
);
484 nvme_req(req
)->cmd
= cmd
;
488 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
490 static int nvme_toggle_streams(struct nvme_ctrl
*ctrl
, bool enable
)
492 struct nvme_command c
;
494 memset(&c
, 0, sizeof(c
));
496 c
.directive
.opcode
= nvme_admin_directive_send
;
497 c
.directive
.nsid
= cpu_to_le32(NVME_NSID_ALL
);
498 c
.directive
.doper
= NVME_DIR_SND_ID_OP_ENABLE
;
499 c
.directive
.dtype
= NVME_DIR_IDENTIFY
;
500 c
.directive
.tdtype
= NVME_DIR_STREAMS
;
501 c
.directive
.endir
= enable
? NVME_DIR_ENDIR
: 0;
503 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, NULL
, 0);
506 static int nvme_disable_streams(struct nvme_ctrl
*ctrl
)
508 return nvme_toggle_streams(ctrl
, false);
511 static int nvme_enable_streams(struct nvme_ctrl
*ctrl
)
513 return nvme_toggle_streams(ctrl
, true);
516 static int nvme_get_stream_params(struct nvme_ctrl
*ctrl
,
517 struct streams_directive_params
*s
, u32 nsid
)
519 struct nvme_command c
;
521 memset(&c
, 0, sizeof(c
));
522 memset(s
, 0, sizeof(*s
));
524 c
.directive
.opcode
= nvme_admin_directive_recv
;
525 c
.directive
.nsid
= cpu_to_le32(nsid
);
526 c
.directive
.numd
= cpu_to_le32(nvme_bytes_to_numd(sizeof(*s
)));
527 c
.directive
.doper
= NVME_DIR_RCV_ST_OP_PARAM
;
528 c
.directive
.dtype
= NVME_DIR_STREAMS
;
530 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, s
, sizeof(*s
));
533 static int nvme_configure_directives(struct nvme_ctrl
*ctrl
)
535 struct streams_directive_params s
;
538 if (!(ctrl
->oacs
& NVME_CTRL_OACS_DIRECTIVES
))
543 ret
= nvme_enable_streams(ctrl
);
547 ret
= nvme_get_stream_params(ctrl
, &s
, NVME_NSID_ALL
);
549 goto out_disable_stream
;
551 ctrl
->nssa
= le16_to_cpu(s
.nssa
);
552 if (ctrl
->nssa
< BLK_MAX_WRITE_HINTS
- 1) {
553 dev_info(ctrl
->device
, "too few streams (%u) available\n",
555 goto out_disable_stream
;
558 ctrl
->nr_streams
= min_t(unsigned, ctrl
->nssa
, BLK_MAX_WRITE_HINTS
- 1);
559 dev_info(ctrl
->device
, "Using %u streams\n", ctrl
->nr_streams
);
563 nvme_disable_streams(ctrl
);
568 * Check if 'req' has a write hint associated with it. If it does, assign
569 * a valid namespace stream to the write.
571 static void nvme_assign_write_stream(struct nvme_ctrl
*ctrl
,
572 struct request
*req
, u16
*control
,
575 enum rw_hint streamid
= req
->write_hint
;
577 if (streamid
== WRITE_LIFE_NOT_SET
|| streamid
== WRITE_LIFE_NONE
)
581 if (WARN_ON_ONCE(streamid
> ctrl
->nr_streams
))
584 *control
|= NVME_RW_DTYPE_STREAMS
;
585 *dsmgmt
|= streamid
<< 16;
588 if (streamid
< ARRAY_SIZE(req
->q
->write_hints
))
589 req
->q
->write_hints
[streamid
] += blk_rq_bytes(req
) >> 9;
592 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
593 struct nvme_command
*cmnd
)
595 cmnd
->common
.opcode
= nvme_cmd_flush
;
596 cmnd
->common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
599 static blk_status_t
nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
600 struct nvme_command
*cmnd
)
602 unsigned short segments
= blk_rq_nr_discard_segments(req
), n
= 0;
603 struct nvme_dsm_range
*range
;
607 * Some devices do not consider the DSM 'Number of Ranges' field when
608 * determining how much data to DMA. Always allocate memory for maximum
609 * number of segments to prevent device reading beyond end of buffer.
611 static const size_t alloc_size
= sizeof(*range
) * NVME_DSM_MAX_RANGES
;
613 range
= kzalloc(alloc_size
, GFP_ATOMIC
| __GFP_NOWARN
);
616 * If we fail allocation our range, fallback to the controller
617 * discard page. If that's also busy, it's safe to return
618 * busy, as we know we can make progress once that's freed.
620 if (test_and_set_bit_lock(0, &ns
->ctrl
->discard_page_busy
))
621 return BLK_STS_RESOURCE
;
623 range
= page_address(ns
->ctrl
->discard_page
);
626 __rq_for_each_bio(bio
, req
) {
627 u64 slba
= nvme_sect_to_lba(ns
, bio
->bi_iter
.bi_sector
);
628 u32 nlb
= bio
->bi_iter
.bi_size
>> ns
->lba_shift
;
631 range
[n
].cattr
= cpu_to_le32(0);
632 range
[n
].nlb
= cpu_to_le32(nlb
);
633 range
[n
].slba
= cpu_to_le64(slba
);
638 if (WARN_ON_ONCE(n
!= segments
)) {
639 if (virt_to_page(range
) == ns
->ctrl
->discard_page
)
640 clear_bit_unlock(0, &ns
->ctrl
->discard_page_busy
);
643 return BLK_STS_IOERR
;
646 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
647 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
648 cmnd
->dsm
.nr
= cpu_to_le32(segments
- 1);
649 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
651 req
->special_vec
.bv_page
= virt_to_page(range
);
652 req
->special_vec
.bv_offset
= offset_in_page(range
);
653 req
->special_vec
.bv_len
= alloc_size
;
654 req
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
659 static inline blk_status_t
nvme_setup_write_zeroes(struct nvme_ns
*ns
,
660 struct request
*req
, struct nvme_command
*cmnd
)
662 if (ns
->ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
663 return nvme_setup_discard(ns
, req
, cmnd
);
665 cmnd
->write_zeroes
.opcode
= nvme_cmd_write_zeroes
;
666 cmnd
->write_zeroes
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
667 cmnd
->write_zeroes
.slba
=
668 cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
669 cmnd
->write_zeroes
.length
=
670 cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
671 cmnd
->write_zeroes
.control
= 0;
675 static inline blk_status_t
nvme_setup_rw(struct nvme_ns
*ns
,
676 struct request
*req
, struct nvme_command
*cmnd
)
678 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
682 if (req
->cmd_flags
& REQ_FUA
)
683 control
|= NVME_RW_FUA
;
684 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
685 control
|= NVME_RW_LR
;
687 if (req
->cmd_flags
& REQ_RAHEAD
)
688 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
690 cmnd
->rw
.opcode
= (rq_data_dir(req
) ? nvme_cmd_write
: nvme_cmd_read
);
691 cmnd
->rw
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
692 cmnd
->rw
.slba
= cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
693 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
695 if (req_op(req
) == REQ_OP_WRITE
&& ctrl
->nr_streams
)
696 nvme_assign_write_stream(ctrl
, req
, &control
, &dsmgmt
);
700 * If formated with metadata, the block layer always provides a
701 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
702 * we enable the PRACT bit for protection information or set the
703 * namespace capacity to zero to prevent any I/O.
705 if (!blk_integrity_rq(req
)) {
706 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns
)))
707 return BLK_STS_NOTSUPP
;
708 control
|= NVME_RW_PRINFO_PRACT
;
711 switch (ns
->pi_type
) {
712 case NVME_NS_DPS_PI_TYPE3
:
713 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
715 case NVME_NS_DPS_PI_TYPE1
:
716 case NVME_NS_DPS_PI_TYPE2
:
717 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
718 NVME_RW_PRINFO_PRCHK_REF
;
719 cmnd
->rw
.reftag
= cpu_to_le32(t10_pi_ref_tag(req
));
724 cmnd
->rw
.control
= cpu_to_le16(control
);
725 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
729 void nvme_cleanup_cmd(struct request
*req
)
731 if (req
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
732 struct nvme_ns
*ns
= req
->rq_disk
->private_data
;
733 struct page
*page
= req
->special_vec
.bv_page
;
735 if (page
== ns
->ctrl
->discard_page
)
736 clear_bit_unlock(0, &ns
->ctrl
->discard_page_busy
);
738 kfree(page_address(page
) + req
->special_vec
.bv_offset
);
741 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd
);
743 blk_status_t
nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
744 struct nvme_command
*cmd
)
746 blk_status_t ret
= BLK_STS_OK
;
748 nvme_clear_nvme_request(req
);
750 memset(cmd
, 0, sizeof(*cmd
));
751 switch (req_op(req
)) {
754 memcpy(cmd
, nvme_req(req
)->cmd
, sizeof(*cmd
));
757 nvme_setup_flush(ns
, cmd
);
759 case REQ_OP_WRITE_ZEROES
:
760 ret
= nvme_setup_write_zeroes(ns
, req
, cmd
);
763 ret
= nvme_setup_discard(ns
, req
, cmd
);
767 ret
= nvme_setup_rw(ns
, req
, cmd
);
771 return BLK_STS_IOERR
;
774 cmd
->common
.command_id
= req
->tag
;
775 trace_nvme_setup_cmd(req
, cmd
);
778 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
780 static void nvme_end_sync_rq(struct request
*rq
, blk_status_t error
)
782 struct completion
*waiting
= rq
->end_io_data
;
784 rq
->end_io_data
= NULL
;
788 static void nvme_execute_rq_polled(struct request_queue
*q
,
789 struct gendisk
*bd_disk
, struct request
*rq
, int at_head
)
791 DECLARE_COMPLETION_ONSTACK(wait
);
793 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
));
795 rq
->cmd_flags
|= REQ_HIPRI
;
796 rq
->end_io_data
= &wait
;
797 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, nvme_end_sync_rq
);
799 while (!completion_done(&wait
)) {
800 blk_poll(q
, request_to_qc_t(rq
->mq_hctx
, rq
), true);
806 * Returns 0 on success. If the result is negative, it's a Linux error code;
807 * if the result is positive, it's an NVM Express status code
809 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
810 union nvme_result
*result
, void *buffer
, unsigned bufflen
,
811 unsigned timeout
, int qid
, int at_head
,
812 blk_mq_req_flags_t flags
, bool poll
)
817 req
= nvme_alloc_request(q
, cmd
, flags
, qid
);
821 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
823 if (buffer
&& bufflen
) {
824 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
830 nvme_execute_rq_polled(req
->q
, NULL
, req
, at_head
);
832 blk_execute_rq(req
->q
, NULL
, req
, at_head
);
834 *result
= nvme_req(req
)->result
;
835 if (nvme_req(req
)->flags
& NVME_REQ_CANCELLED
)
838 ret
= nvme_req(req
)->status
;
840 blk_mq_free_request(req
);
843 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd
);
845 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
846 void *buffer
, unsigned bufflen
)
848 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0,
849 NVME_QID_ANY
, 0, 0, false);
851 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
853 static void *nvme_add_user_metadata(struct bio
*bio
, void __user
*ubuf
,
854 unsigned len
, u32 seed
, bool write
)
856 struct bio_integrity_payload
*bip
;
860 buf
= kmalloc(len
, GFP_KERNEL
);
865 if (write
&& copy_from_user(buf
, ubuf
, len
))
868 bip
= bio_integrity_alloc(bio
, GFP_KERNEL
, 1);
874 bip
->bip_iter
.bi_size
= len
;
875 bip
->bip_iter
.bi_sector
= seed
;
876 ret
= bio_integrity_add_page(bio
, virt_to_page(buf
), len
,
877 offset_in_page(buf
));
887 static int nvme_submit_user_cmd(struct request_queue
*q
,
888 struct nvme_command
*cmd
, void __user
*ubuffer
,
889 unsigned bufflen
, void __user
*meta_buffer
, unsigned meta_len
,
890 u32 meta_seed
, u64
*result
, unsigned timeout
)
892 bool write
= nvme_is_write(cmd
);
893 struct nvme_ns
*ns
= q
->queuedata
;
894 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
896 struct bio
*bio
= NULL
;
900 req
= nvme_alloc_request(q
, cmd
, 0, NVME_QID_ANY
);
904 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
905 nvme_req(req
)->flags
|= NVME_REQ_USERCMD
;
907 if (ubuffer
&& bufflen
) {
908 ret
= blk_rq_map_user(q
, req
, NULL
, ubuffer
, bufflen
,
914 if (disk
&& meta_buffer
&& meta_len
) {
915 meta
= nvme_add_user_metadata(bio
, meta_buffer
, meta_len
,
921 req
->cmd_flags
|= REQ_INTEGRITY
;
925 blk_execute_rq(req
->q
, disk
, req
, 0);
926 if (nvme_req(req
)->flags
& NVME_REQ_CANCELLED
)
929 ret
= nvme_req(req
)->status
;
931 *result
= le64_to_cpu(nvme_req(req
)->result
.u64
);
932 if (meta
&& !ret
&& !write
) {
933 if (copy_to_user(meta_buffer
, meta
, meta_len
))
939 blk_rq_unmap_user(bio
);
941 blk_mq_free_request(req
);
945 static void nvme_keep_alive_end_io(struct request
*rq
, blk_status_t status
)
947 struct nvme_ctrl
*ctrl
= rq
->end_io_data
;
949 bool startka
= false;
951 blk_mq_free_request(rq
);
954 dev_err(ctrl
->device
,
955 "failed nvme_keep_alive_end_io error=%d\n",
960 ctrl
->comp_seen
= false;
961 spin_lock_irqsave(&ctrl
->lock
, flags
);
962 if (ctrl
->state
== NVME_CTRL_LIVE
||
963 ctrl
->state
== NVME_CTRL_CONNECTING
)
965 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
967 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
970 static int nvme_keep_alive(struct nvme_ctrl
*ctrl
)
974 rq
= nvme_alloc_request(ctrl
->admin_q
, &ctrl
->ka_cmd
, BLK_MQ_REQ_RESERVED
,
979 rq
->timeout
= ctrl
->kato
* HZ
;
980 rq
->end_io_data
= ctrl
;
982 blk_execute_rq_nowait(rq
->q
, NULL
, rq
, 0, nvme_keep_alive_end_io
);
987 static void nvme_keep_alive_work(struct work_struct
*work
)
989 struct nvme_ctrl
*ctrl
= container_of(to_delayed_work(work
),
990 struct nvme_ctrl
, ka_work
);
991 bool comp_seen
= ctrl
->comp_seen
;
993 if ((ctrl
->ctratt
& NVME_CTRL_ATTR_TBKAS
) && comp_seen
) {
994 dev_dbg(ctrl
->device
,
995 "reschedule traffic based keep-alive timer\n");
996 ctrl
->comp_seen
= false;
997 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
1001 if (nvme_keep_alive(ctrl
)) {
1002 /* allocation failure, reset the controller */
1003 dev_err(ctrl
->device
, "keep-alive failed\n");
1004 nvme_reset_ctrl(ctrl
);
1009 static void nvme_start_keep_alive(struct nvme_ctrl
*ctrl
)
1011 if (unlikely(ctrl
->kato
== 0))
1014 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
1017 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
)
1019 if (unlikely(ctrl
->kato
== 0))
1022 cancel_delayed_work_sync(&ctrl
->ka_work
);
1024 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive
);
1027 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1028 * flag, thus sending any new CNS opcodes has a big chance of not working.
1029 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1030 * (but not for any later version).
1032 static bool nvme_ctrl_limited_cns(struct nvme_ctrl
*ctrl
)
1034 if (ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)
1035 return ctrl
->vs
< NVME_VS(1, 2, 0);
1036 return ctrl
->vs
< NVME_VS(1, 1, 0);
1039 static int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
1041 struct nvme_command c
= { };
1044 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1045 c
.identify
.opcode
= nvme_admin_identify
;
1046 c
.identify
.cns
= NVME_ID_CNS_CTRL
;
1048 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
1052 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
1053 sizeof(struct nvme_id_ctrl
));
1059 static int nvme_process_ns_desc(struct nvme_ctrl
*ctrl
, struct nvme_ns_ids
*ids
,
1060 struct nvme_ns_id_desc
*cur
)
1062 const char *warn_str
= "ctrl returned bogus length:";
1065 switch (cur
->nidt
) {
1066 case NVME_NIDT_EUI64
:
1067 if (cur
->nidl
!= NVME_NIDT_EUI64_LEN
) {
1068 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_EUI64\n",
1069 warn_str
, cur
->nidl
);
1072 memcpy(ids
->eui64
, data
+ sizeof(*cur
), NVME_NIDT_EUI64_LEN
);
1073 return NVME_NIDT_EUI64_LEN
;
1074 case NVME_NIDT_NGUID
:
1075 if (cur
->nidl
!= NVME_NIDT_NGUID_LEN
) {
1076 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_NGUID\n",
1077 warn_str
, cur
->nidl
);
1080 memcpy(ids
->nguid
, data
+ sizeof(*cur
), NVME_NIDT_NGUID_LEN
);
1081 return NVME_NIDT_NGUID_LEN
;
1082 case NVME_NIDT_UUID
:
1083 if (cur
->nidl
!= NVME_NIDT_UUID_LEN
) {
1084 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_UUID\n",
1085 warn_str
, cur
->nidl
);
1088 uuid_copy(&ids
->uuid
, data
+ sizeof(*cur
));
1089 return NVME_NIDT_UUID_LEN
;
1091 /* Skip unknown types */
1096 static int nvme_identify_ns_descs(struct nvme_ctrl
*ctrl
, unsigned nsid
,
1097 struct nvme_ns_ids
*ids
)
1099 struct nvme_command c
= { };
1105 c
.identify
.opcode
= nvme_admin_identify
;
1106 c
.identify
.nsid
= cpu_to_le32(nsid
);
1107 c
.identify
.cns
= NVME_ID_CNS_NS_DESC_LIST
;
1109 data
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
1113 status
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, data
,
1114 NVME_IDENTIFY_DATA_SIZE
);
1116 dev_warn(ctrl
->device
,
1117 "Identify Descriptors failed (%d)\n", status
);
1119 * Don't treat an error as fatal, as we potentially already
1120 * have a NGUID or EUI-64.
1122 if (status
> 0 && !(status
& NVME_SC_DNR
))
1127 for (pos
= 0; pos
< NVME_IDENTIFY_DATA_SIZE
; pos
+= len
) {
1128 struct nvme_ns_id_desc
*cur
= data
+ pos
;
1133 len
= nvme_process_ns_desc(ctrl
, ids
, cur
);
1137 len
+= sizeof(*cur
);
1144 static int nvme_identify_ns_list(struct nvme_ctrl
*dev
, unsigned nsid
, __le32
*ns_list
)
1146 struct nvme_command c
= { };
1148 c
.identify
.opcode
= nvme_admin_identify
;
1149 c
.identify
.cns
= NVME_ID_CNS_NS_ACTIVE_LIST
;
1150 c
.identify
.nsid
= cpu_to_le32(nsid
);
1151 return nvme_submit_sync_cmd(dev
->admin_q
, &c
, ns_list
,
1152 NVME_IDENTIFY_DATA_SIZE
);
1155 static int nvme_identify_ns(struct nvme_ctrl
*ctrl
,
1156 unsigned nsid
, struct nvme_id_ns
**id
)
1158 struct nvme_command c
= { };
1161 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1162 c
.identify
.opcode
= nvme_admin_identify
;
1163 c
.identify
.nsid
= cpu_to_le32(nsid
);
1164 c
.identify
.cns
= NVME_ID_CNS_NS
;
1166 *id
= kmalloc(sizeof(**id
), GFP_KERNEL
);
1170 error
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, *id
, sizeof(**id
));
1172 dev_warn(ctrl
->device
, "Identify namespace failed (%d)\n", error
);
1179 static int nvme_features(struct nvme_ctrl
*dev
, u8 op
, unsigned int fid
,
1180 unsigned int dword11
, void *buffer
, size_t buflen
, u32
*result
)
1182 union nvme_result res
= { 0 };
1183 struct nvme_command c
;
1186 memset(&c
, 0, sizeof(c
));
1187 c
.features
.opcode
= op
;
1188 c
.features
.fid
= cpu_to_le32(fid
);
1189 c
.features
.dword11
= cpu_to_le32(dword11
);
1191 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &res
,
1192 buffer
, buflen
, 0, NVME_QID_ANY
, 0, 0, false);
1193 if (ret
>= 0 && result
)
1194 *result
= le32_to_cpu(res
.u32
);
1198 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1199 unsigned int dword11
, void *buffer
, size_t buflen
,
1202 return nvme_features(dev
, nvme_admin_set_features
, fid
, dword11
, buffer
,
1205 EXPORT_SYMBOL_GPL(nvme_set_features
);
1207 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1208 unsigned int dword11
, void *buffer
, size_t buflen
,
1211 return nvme_features(dev
, nvme_admin_get_features
, fid
, dword11
, buffer
,
1214 EXPORT_SYMBOL_GPL(nvme_get_features
);
1216 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
1218 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
1220 int status
, nr_io_queues
;
1222 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, NULL
, 0,
1228 * Degraded controllers might return an error when setting the queue
1229 * count. We still want to be able to bring them online and offer
1230 * access to the admin queue, as that might be only way to fix them up.
1233 dev_err(ctrl
->device
, "Could not set queue count (%d)\n", status
);
1236 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
1237 *count
= min(*count
, nr_io_queues
);
1242 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
1244 #define NVME_AEN_SUPPORTED \
1245 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1246 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1248 static void nvme_enable_aen(struct nvme_ctrl
*ctrl
)
1250 u32 result
, supported_aens
= ctrl
->oaes
& NVME_AEN_SUPPORTED
;
1253 if (!supported_aens
)
1256 status
= nvme_set_features(ctrl
, NVME_FEAT_ASYNC_EVENT
, supported_aens
,
1259 dev_warn(ctrl
->device
, "Failed to configure AEN (cfg %x)\n",
1262 queue_work(nvme_wq
, &ctrl
->async_event_work
);
1266 * Convert integer values from ioctl structures to user pointers, silently
1267 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1270 static void __user
*nvme_to_user_ptr(uintptr_t ptrval
)
1272 if (in_compat_syscall())
1273 ptrval
= (compat_uptr_t
)ptrval
;
1274 return (void __user
*)ptrval
;
1277 static int nvme_submit_io(struct nvme_ns
*ns
, struct nvme_user_io __user
*uio
)
1279 struct nvme_user_io io
;
1280 struct nvme_command c
;
1281 unsigned length
, meta_len
;
1282 void __user
*metadata
;
1284 if (copy_from_user(&io
, uio
, sizeof(io
)))
1289 switch (io
.opcode
) {
1290 case nvme_cmd_write
:
1292 case nvme_cmd_compare
:
1298 length
= (io
.nblocks
+ 1) << ns
->lba_shift
;
1299 meta_len
= (io
.nblocks
+ 1) * ns
->ms
;
1300 metadata
= nvme_to_user_ptr(io
.metadata
);
1302 if (ns
->features
& NVME_NS_EXT_LBAS
) {
1305 } else if (meta_len
) {
1306 if ((io
.metadata
& 3) || !io
.metadata
)
1310 memset(&c
, 0, sizeof(c
));
1311 c
.rw
.opcode
= io
.opcode
;
1312 c
.rw
.flags
= io
.flags
;
1313 c
.rw
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
1314 c
.rw
.slba
= cpu_to_le64(io
.slba
);
1315 c
.rw
.length
= cpu_to_le16(io
.nblocks
);
1316 c
.rw
.control
= cpu_to_le16(io
.control
);
1317 c
.rw
.dsmgmt
= cpu_to_le32(io
.dsmgmt
);
1318 c
.rw
.reftag
= cpu_to_le32(io
.reftag
);
1319 c
.rw
.apptag
= cpu_to_le16(io
.apptag
);
1320 c
.rw
.appmask
= cpu_to_le16(io
.appmask
);
1322 return nvme_submit_user_cmd(ns
->queue
, &c
,
1323 nvme_to_user_ptr(io
.addr
), length
,
1324 metadata
, meta_len
, lower_32_bits(io
.slba
), NULL
, 0);
1327 static u32
nvme_known_admin_effects(u8 opcode
)
1330 case nvme_admin_format_nvm
:
1331 return NVME_CMD_EFFECTS_CSUPP
| NVME_CMD_EFFECTS_LBCC
|
1332 NVME_CMD_EFFECTS_CSE_MASK
;
1333 case nvme_admin_sanitize_nvm
:
1334 return NVME_CMD_EFFECTS_CSE_MASK
;
1341 static u32
nvme_passthru_start(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1348 effects
= le32_to_cpu(ctrl
->effects
->iocs
[opcode
]);
1349 if (effects
& ~(NVME_CMD_EFFECTS_CSUPP
| NVME_CMD_EFFECTS_LBCC
))
1350 dev_warn(ctrl
->device
,
1351 "IO command:%02x has unhandled effects:%08x\n",
1357 effects
= le32_to_cpu(ctrl
->effects
->acs
[opcode
]);
1358 effects
|= nvme_known_admin_effects(opcode
);
1361 * For simplicity, IO to all namespaces is quiesced even if the command
1362 * effects say only one namespace is affected.
1364 if (effects
& (NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_CSE_MASK
)) {
1365 mutex_lock(&ctrl
->scan_lock
);
1366 mutex_lock(&ctrl
->subsys
->lock
);
1367 nvme_mpath_start_freeze(ctrl
->subsys
);
1368 nvme_mpath_wait_freeze(ctrl
->subsys
);
1369 nvme_start_freeze(ctrl
);
1370 nvme_wait_freeze(ctrl
);
1375 static void nvme_update_formats(struct nvme_ctrl
*ctrl
)
1379 down_read(&ctrl
->namespaces_rwsem
);
1380 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
1381 if (ns
->disk
&& nvme_revalidate_disk(ns
->disk
))
1382 nvme_set_queue_dying(ns
);
1383 up_read(&ctrl
->namespaces_rwsem
);
1386 static void nvme_passthru_end(struct nvme_ctrl
*ctrl
, u32 effects
)
1389 * Revalidate LBA changes prior to unfreezing. This is necessary to
1390 * prevent memory corruption if a logical block size was changed by
1393 if (effects
& NVME_CMD_EFFECTS_LBCC
)
1394 nvme_update_formats(ctrl
);
1395 if (effects
& (NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_CSE_MASK
)) {
1396 nvme_unfreeze(ctrl
);
1397 nvme_mpath_unfreeze(ctrl
->subsys
);
1398 mutex_unlock(&ctrl
->subsys
->lock
);
1399 nvme_remove_invalid_namespaces(ctrl
, NVME_NSID_ALL
);
1400 mutex_unlock(&ctrl
->scan_lock
);
1402 if (effects
& NVME_CMD_EFFECTS_CCC
)
1403 nvme_init_identify(ctrl
);
1404 if (effects
& (NVME_CMD_EFFECTS_NIC
| NVME_CMD_EFFECTS_NCC
)) {
1405 nvme_queue_scan(ctrl
);
1406 flush_work(&ctrl
->scan_work
);
1410 static int nvme_user_cmd(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1411 struct nvme_passthru_cmd __user
*ucmd
)
1413 struct nvme_passthru_cmd cmd
;
1414 struct nvme_command c
;
1415 unsigned timeout
= 0;
1420 if (!capable(CAP_SYS_ADMIN
))
1422 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
1427 memset(&c
, 0, sizeof(c
));
1428 c
.common
.opcode
= cmd
.opcode
;
1429 c
.common
.flags
= cmd
.flags
;
1430 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
1431 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
1432 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
1433 c
.common
.cdw10
= cpu_to_le32(cmd
.cdw10
);
1434 c
.common
.cdw11
= cpu_to_le32(cmd
.cdw11
);
1435 c
.common
.cdw12
= cpu_to_le32(cmd
.cdw12
);
1436 c
.common
.cdw13
= cpu_to_le32(cmd
.cdw13
);
1437 c
.common
.cdw14
= cpu_to_le32(cmd
.cdw14
);
1438 c
.common
.cdw15
= cpu_to_le32(cmd
.cdw15
);
1441 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
1443 effects
= nvme_passthru_start(ctrl
, ns
, cmd
.opcode
);
1444 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
1445 nvme_to_user_ptr(cmd
.addr
), cmd
.data_len
,
1446 nvme_to_user_ptr(cmd
.metadata
), cmd
.metadata_len
,
1447 0, &result
, timeout
);
1448 nvme_passthru_end(ctrl
, effects
);
1451 if (put_user(result
, &ucmd
->result
))
1458 static int nvme_user_cmd64(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1459 struct nvme_passthru_cmd64 __user
*ucmd
)
1461 struct nvme_passthru_cmd64 cmd
;
1462 struct nvme_command c
;
1463 unsigned timeout
= 0;
1467 if (!capable(CAP_SYS_ADMIN
))
1469 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
1474 memset(&c
, 0, sizeof(c
));
1475 c
.common
.opcode
= cmd
.opcode
;
1476 c
.common
.flags
= cmd
.flags
;
1477 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
1478 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
1479 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
1480 c
.common
.cdw10
= cpu_to_le32(cmd
.cdw10
);
1481 c
.common
.cdw11
= cpu_to_le32(cmd
.cdw11
);
1482 c
.common
.cdw12
= cpu_to_le32(cmd
.cdw12
);
1483 c
.common
.cdw13
= cpu_to_le32(cmd
.cdw13
);
1484 c
.common
.cdw14
= cpu_to_le32(cmd
.cdw14
);
1485 c
.common
.cdw15
= cpu_to_le32(cmd
.cdw15
);
1488 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
1490 effects
= nvme_passthru_start(ctrl
, ns
, cmd
.opcode
);
1491 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
1492 nvme_to_user_ptr(cmd
.addr
), cmd
.data_len
,
1493 nvme_to_user_ptr(cmd
.metadata
), cmd
.metadata_len
,
1494 0, &cmd
.result
, timeout
);
1495 nvme_passthru_end(ctrl
, effects
);
1498 if (put_user(cmd
.result
, &ucmd
->result
))
1506 * Issue ioctl requests on the first available path. Note that unlike normal
1507 * block layer requests we will not retry failed request on another controller.
1509 static struct nvme_ns
*nvme_get_ns_from_disk(struct gendisk
*disk
,
1510 struct nvme_ns_head
**head
, int *srcu_idx
)
1512 #ifdef CONFIG_NVME_MULTIPATH
1513 if (disk
->fops
== &nvme_ns_head_ops
) {
1516 *head
= disk
->private_data
;
1517 *srcu_idx
= srcu_read_lock(&(*head
)->srcu
);
1518 ns
= nvme_find_path(*head
);
1520 srcu_read_unlock(&(*head
)->srcu
, *srcu_idx
);
1526 return disk
->private_data
;
1529 static void nvme_put_ns_from_disk(struct nvme_ns_head
*head
, int idx
)
1532 srcu_read_unlock(&head
->srcu
, idx
);
1535 static bool is_ctrl_ioctl(unsigned int cmd
)
1537 if (cmd
== NVME_IOCTL_ADMIN_CMD
|| cmd
== NVME_IOCTL_ADMIN64_CMD
)
1539 if (is_sed_ioctl(cmd
))
1544 static int nvme_handle_ctrl_ioctl(struct nvme_ns
*ns
, unsigned int cmd
,
1546 struct nvme_ns_head
*head
,
1549 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1552 nvme_get_ctrl(ns
->ctrl
);
1553 nvme_put_ns_from_disk(head
, srcu_idx
);
1556 case NVME_IOCTL_ADMIN_CMD
:
1557 ret
= nvme_user_cmd(ctrl
, NULL
, argp
);
1559 case NVME_IOCTL_ADMIN64_CMD
:
1560 ret
= nvme_user_cmd64(ctrl
, NULL
, argp
);
1563 ret
= sed_ioctl(ctrl
->opal_dev
, cmd
, argp
);
1566 nvme_put_ctrl(ctrl
);
1570 static int nvme_ioctl(struct block_device
*bdev
, fmode_t mode
,
1571 unsigned int cmd
, unsigned long arg
)
1573 struct nvme_ns_head
*head
= NULL
;
1574 void __user
*argp
= (void __user
*)arg
;
1578 ns
= nvme_get_ns_from_disk(bdev
->bd_disk
, &head
, &srcu_idx
);
1580 return -EWOULDBLOCK
;
1583 * Handle ioctls that apply to the controller instead of the namespace
1584 * seperately and drop the ns SRCU reference early. This avoids a
1585 * deadlock when deleting namespaces using the passthrough interface.
1587 if (is_ctrl_ioctl(cmd
))
1588 return nvme_handle_ctrl_ioctl(ns
, cmd
, argp
, head
, srcu_idx
);
1592 force_successful_syscall_return();
1593 ret
= ns
->head
->ns_id
;
1595 case NVME_IOCTL_IO_CMD
:
1596 ret
= nvme_user_cmd(ns
->ctrl
, ns
, argp
);
1598 case NVME_IOCTL_SUBMIT_IO
:
1599 ret
= nvme_submit_io(ns
, argp
);
1601 case NVME_IOCTL_IO64_CMD
:
1602 ret
= nvme_user_cmd64(ns
->ctrl
, ns
, argp
);
1606 ret
= nvme_nvm_ioctl(ns
, cmd
, arg
);
1611 nvme_put_ns_from_disk(head
, srcu_idx
);
1615 #ifdef CONFIG_COMPAT
1616 struct nvme_user_io32
{
1629 } __attribute__((__packed__
));
1631 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1633 static int nvme_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1634 unsigned int cmd
, unsigned long arg
)
1637 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1638 * between 32 bit programs and 64 bit kernel.
1639 * The cause is that the results of sizeof(struct nvme_user_io),
1640 * which is used to define NVME_IOCTL_SUBMIT_IO,
1641 * are not same between 32 bit compiler and 64 bit compiler.
1642 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1643 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1644 * Other IOCTL numbers are same between 32 bit and 64 bit.
1645 * So there is nothing to do regarding to other IOCTL numbers.
1647 if (cmd
== NVME_IOCTL_SUBMIT_IO32
)
1648 return nvme_ioctl(bdev
, mode
, NVME_IOCTL_SUBMIT_IO
, arg
);
1650 return nvme_ioctl(bdev
, mode
, cmd
, arg
);
1653 #define nvme_compat_ioctl NULL
1654 #endif /* CONFIG_COMPAT */
1656 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
1658 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
1660 #ifdef CONFIG_NVME_MULTIPATH
1661 /* should never be called due to GENHD_FL_HIDDEN */
1662 if (WARN_ON_ONCE(ns
->head
->disk
))
1665 if (!kref_get_unless_zero(&ns
->kref
))
1667 if (!try_module_get(ns
->ctrl
->ops
->module
))
1678 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
1680 struct nvme_ns
*ns
= disk
->private_data
;
1682 module_put(ns
->ctrl
->ops
->module
);
1686 static int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1688 /* some standard values */
1689 geo
->heads
= 1 << 6;
1690 geo
->sectors
= 1 << 5;
1691 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
1695 #ifdef CONFIG_BLK_DEV_INTEGRITY
1696 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1697 u32 max_integrity_segments
)
1699 struct blk_integrity integrity
;
1701 memset(&integrity
, 0, sizeof(integrity
));
1703 case NVME_NS_DPS_PI_TYPE3
:
1704 integrity
.profile
= &t10_pi_type3_crc
;
1705 integrity
.tag_size
= sizeof(u16
) + sizeof(u32
);
1706 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1708 case NVME_NS_DPS_PI_TYPE1
:
1709 case NVME_NS_DPS_PI_TYPE2
:
1710 integrity
.profile
= &t10_pi_type1_crc
;
1711 integrity
.tag_size
= sizeof(u16
);
1712 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1715 integrity
.profile
= NULL
;
1718 integrity
.tuple_size
= ms
;
1719 blk_integrity_register(disk
, &integrity
);
1720 blk_queue_max_integrity_segments(disk
->queue
, max_integrity_segments
);
1723 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1724 u32 max_integrity_segments
)
1727 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1729 static void nvme_config_discard(struct gendisk
*disk
, struct nvme_ns
*ns
)
1731 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1732 struct request_queue
*queue
= disk
->queue
;
1733 u32 size
= queue_logical_block_size(queue
);
1735 if (!(ctrl
->oncs
& NVME_CTRL_ONCS_DSM
)) {
1736 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, queue
);
1740 if (ctrl
->nr_streams
&& ns
->sws
&& ns
->sgs
)
1741 size
*= ns
->sws
* ns
->sgs
;
1743 BUILD_BUG_ON(PAGE_SIZE
/ sizeof(struct nvme_dsm_range
) <
1744 NVME_DSM_MAX_RANGES
);
1746 queue
->limits
.discard_alignment
= 0;
1747 queue
->limits
.discard_granularity
= size
;
1749 /* If discard is already enabled, don't reset queue limits */
1750 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD
, queue
))
1753 blk_queue_max_discard_sectors(queue
, UINT_MAX
);
1754 blk_queue_max_discard_segments(queue
, NVME_DSM_MAX_RANGES
);
1756 if (ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
1757 blk_queue_max_write_zeroes_sectors(queue
, UINT_MAX
);
1760 static void nvme_config_write_zeroes(struct gendisk
*disk
, struct nvme_ns
*ns
)
1764 if (!(ns
->ctrl
->oncs
& NVME_CTRL_ONCS_WRITE_ZEROES
) ||
1765 (ns
->ctrl
->quirks
& NVME_QUIRK_DISABLE_WRITE_ZEROES
))
1768 * Even though NVMe spec explicitly states that MDTS is not
1769 * applicable to the write-zeroes:- "The restriction does not apply to
1770 * commands that do not transfer data between the host and the
1771 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1772 * In order to be more cautious use controller's max_hw_sectors value
1773 * to configure the maximum sectors for the write-zeroes which is
1774 * configured based on the controller's MDTS field in the
1775 * nvme_init_identify() if available.
1777 if (ns
->ctrl
->max_hw_sectors
== UINT_MAX
)
1778 max_blocks
= (u64
)USHRT_MAX
+ 1;
1780 max_blocks
= ns
->ctrl
->max_hw_sectors
+ 1;
1782 blk_queue_max_write_zeroes_sectors(disk
->queue
,
1783 nvme_lba_to_sect(ns
, max_blocks
));
1786 static int nvme_report_ns_ids(struct nvme_ctrl
*ctrl
, unsigned int nsid
,
1787 struct nvme_id_ns
*id
, struct nvme_ns_ids
*ids
)
1789 memset(ids
, 0, sizeof(*ids
));
1791 if (ctrl
->vs
>= NVME_VS(1, 1, 0))
1792 memcpy(ids
->eui64
, id
->eui64
, sizeof(id
->eui64
));
1793 if (ctrl
->vs
>= NVME_VS(1, 2, 0))
1794 memcpy(ids
->nguid
, id
->nguid
, sizeof(id
->nguid
));
1795 if (ctrl
->vs
>= NVME_VS(1, 3, 0))
1796 return nvme_identify_ns_descs(ctrl
, nsid
, ids
);
1800 static bool nvme_ns_ids_valid(struct nvme_ns_ids
*ids
)
1802 return !uuid_is_null(&ids
->uuid
) ||
1803 memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)) ||
1804 memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
));
1807 static bool nvme_ns_ids_equal(struct nvme_ns_ids
*a
, struct nvme_ns_ids
*b
)
1809 return uuid_equal(&a
->uuid
, &b
->uuid
) &&
1810 memcmp(&a
->nguid
, &b
->nguid
, sizeof(a
->nguid
)) == 0 &&
1811 memcmp(&a
->eui64
, &b
->eui64
, sizeof(a
->eui64
)) == 0;
1814 static int nvme_setup_streams_ns(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1815 u32
*phys_bs
, u32
*io_opt
)
1817 struct streams_directive_params s
;
1820 if (!ctrl
->nr_streams
)
1823 ret
= nvme_get_stream_params(ctrl
, &s
, ns
->head
->ns_id
);
1827 ns
->sws
= le32_to_cpu(s
.sws
);
1828 ns
->sgs
= le16_to_cpu(s
.sgs
);
1831 *phys_bs
= ns
->sws
* (1 << ns
->lba_shift
);
1833 *io_opt
= *phys_bs
* ns
->sgs
;
1839 static void nvme_update_disk_info(struct gendisk
*disk
,
1840 struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1842 sector_t capacity
= nvme_lba_to_sect(ns
, le64_to_cpu(id
->nsze
));
1843 unsigned short bs
= 1 << ns
->lba_shift
;
1844 u32 atomic_bs
, phys_bs
, io_opt
= 0;
1846 if (ns
->lba_shift
> PAGE_SHIFT
) {
1847 /* unsupported block size, set capacity to 0 later */
1850 blk_mq_freeze_queue(disk
->queue
);
1851 blk_integrity_unregister(disk
);
1853 atomic_bs
= phys_bs
= bs
;
1854 nvme_setup_streams_ns(ns
->ctrl
, ns
, &phys_bs
, &io_opt
);
1855 if (id
->nabo
== 0) {
1857 * Bit 1 indicates whether NAWUPF is defined for this namespace
1858 * and whether it should be used instead of AWUPF. If NAWUPF ==
1859 * 0 then AWUPF must be used instead.
1861 if (id
->nsfeat
& NVME_NS_FEAT_ATOMICS
&& id
->nawupf
)
1862 atomic_bs
= (1 + le16_to_cpu(id
->nawupf
)) * bs
;
1864 atomic_bs
= (1 + ns
->ctrl
->subsys
->awupf
) * bs
;
1867 if (id
->nsfeat
& NVME_NS_FEAT_IO_OPT
) {
1868 /* NPWG = Namespace Preferred Write Granularity */
1869 phys_bs
= bs
* (1 + le16_to_cpu(id
->npwg
));
1870 /* NOWS = Namespace Optimal Write Size */
1871 io_opt
= bs
* (1 + le16_to_cpu(id
->nows
));
1874 blk_queue_logical_block_size(disk
->queue
, bs
);
1876 * Linux filesystems assume writing a single physical block is
1877 * an atomic operation. Hence limit the physical block size to the
1878 * value of the Atomic Write Unit Power Fail parameter.
1880 blk_queue_physical_block_size(disk
->queue
, min(phys_bs
, atomic_bs
));
1881 blk_queue_io_min(disk
->queue
, phys_bs
);
1882 blk_queue_io_opt(disk
->queue
, io_opt
);
1885 * The block layer can't support LBA sizes larger than the page size
1886 * yet, so catch this early and don't allow block I/O.
1888 if (ns
->lba_shift
> PAGE_SHIFT
)
1892 * Register a metadata profile for PI, or the plain non-integrity NVMe
1893 * metadata masquerading as Type 0 if supported, otherwise reject block
1894 * I/O to namespaces with metadata except when the namespace supports
1895 * PI, as it can strip/insert in that case.
1898 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY
) &&
1899 (ns
->features
& NVME_NS_METADATA_SUPPORTED
))
1900 nvme_init_integrity(disk
, ns
->ms
, ns
->pi_type
,
1901 ns
->ctrl
->max_integrity_segments
);
1902 else if (!nvme_ns_has_pi(ns
))
1906 set_capacity_revalidate_and_notify(disk
, capacity
, false);
1908 nvme_config_discard(disk
, ns
);
1909 nvme_config_write_zeroes(disk
, ns
);
1911 if (id
->nsattr
& NVME_NS_ATTR_RO
)
1912 set_disk_ro(disk
, true);
1914 set_disk_ro(disk
, false);
1916 blk_mq_unfreeze_queue(disk
->queue
);
1919 static void __nvme_revalidate_disk(struct gendisk
*disk
, struct nvme_id_ns
*id
)
1921 struct nvme_ns
*ns
= disk
->private_data
;
1925 * If identify namespace failed, use default 512 byte block size so
1926 * block layer can use before failing read/write for 0 capacity.
1928 ns
->lba_shift
= id
->lbaf
[id
->flbas
& NVME_NS_FLBAS_LBA_MASK
].ds
;
1929 if (ns
->lba_shift
== 0)
1932 if ((ns
->ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) &&
1933 is_power_of_2(ns
->ctrl
->max_hw_sectors
))
1934 iob
= ns
->ctrl
->max_hw_sectors
;
1936 iob
= nvme_lba_to_sect(ns
, le16_to_cpu(id
->noiob
));
1939 ns
->ms
= le16_to_cpu(id
->lbaf
[id
->flbas
& NVME_NS_FLBAS_LBA_MASK
].ms
);
1940 /* the PI implementation requires metadata equal t10 pi tuple size */
1941 if (ns
->ms
== sizeof(struct t10_pi_tuple
))
1942 ns
->pi_type
= id
->dps
& NVME_NS_DPS_PI_MASK
;
1947 if (id
->flbas
& NVME_NS_FLBAS_META_EXT
)
1948 ns
->features
|= NVME_NS_EXT_LBAS
;
1951 * For PCI, Extended logical block will be generated by the
1952 * controller. Non-extended format can be generated by the
1955 if (ns
->ctrl
->ops
->flags
& NVME_F_METADATA_SUPPORTED
) {
1956 if (!(ns
->features
& NVME_NS_EXT_LBAS
))
1957 ns
->features
|= NVME_NS_METADATA_SUPPORTED
;
1962 blk_queue_chunk_sectors(ns
->queue
, rounddown_pow_of_two(iob
));
1963 nvme_update_disk_info(disk
, ns
, id
);
1964 #ifdef CONFIG_NVME_MULTIPATH
1965 if (ns
->head
->disk
) {
1966 nvme_update_disk_info(ns
->head
->disk
, ns
, id
);
1967 blk_queue_stack_limits(ns
->head
->disk
->queue
, ns
->queue
);
1968 revalidate_disk(ns
->head
->disk
);
1973 static int nvme_revalidate_disk(struct gendisk
*disk
)
1975 struct nvme_ns
*ns
= disk
->private_data
;
1976 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1977 struct nvme_id_ns
*id
;
1978 struct nvme_ns_ids ids
;
1981 if (test_bit(NVME_NS_DEAD
, &ns
->flags
)) {
1982 set_capacity(disk
, 0);
1986 ret
= nvme_identify_ns(ctrl
, ns
->head
->ns_id
, &id
);
1990 if (id
->ncap
== 0) {
1995 ret
= nvme_report_ns_ids(ctrl
, ns
->head
->ns_id
, id
, &ids
);
1999 if (!nvme_ns_ids_equal(&ns
->head
->ids
, &ids
)) {
2000 dev_err(ctrl
->device
,
2001 "identifiers changed for nsid %d\n", ns
->head
->ns_id
);
2006 __nvme_revalidate_disk(disk
, id
);
2011 * Only fail the function if we got a fatal error back from the
2012 * device, otherwise ignore the error and just move on.
2014 if (ret
== -ENOMEM
|| (ret
> 0 && !(ret
& NVME_SC_DNR
)))
2017 ret
= blk_status_to_errno(nvme_error_status(ret
));
2021 static char nvme_pr_type(enum pr_type type
)
2024 case PR_WRITE_EXCLUSIVE
:
2026 case PR_EXCLUSIVE_ACCESS
:
2028 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
2030 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
2032 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
2034 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
2041 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
2042 u64 key
, u64 sa_key
, u8 op
)
2044 struct nvme_ns_head
*head
= NULL
;
2046 struct nvme_command c
;
2048 u8 data
[16] = { 0, };
2050 ns
= nvme_get_ns_from_disk(bdev
->bd_disk
, &head
, &srcu_idx
);
2052 return -EWOULDBLOCK
;
2054 put_unaligned_le64(key
, &data
[0]);
2055 put_unaligned_le64(sa_key
, &data
[8]);
2057 memset(&c
, 0, sizeof(c
));
2058 c
.common
.opcode
= op
;
2059 c
.common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
2060 c
.common
.cdw10
= cpu_to_le32(cdw10
);
2062 ret
= nvme_submit_sync_cmd(ns
->queue
, &c
, data
, 16);
2063 nvme_put_ns_from_disk(head
, srcu_idx
);
2067 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
2068 u64
new, unsigned flags
)
2072 if (flags
& ~PR_FL_IGNORE_KEY
)
2075 cdw10
= old
? 2 : 0;
2076 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
2077 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
2078 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
2081 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
2082 enum pr_type type
, unsigned flags
)
2086 if (flags
& ~PR_FL_IGNORE_KEY
)
2089 cdw10
= nvme_pr_type(type
) << 8;
2090 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
2091 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
2094 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
2095 enum pr_type type
, bool abort
)
2097 u32 cdw10
= nvme_pr_type(type
) << 8 | (abort
? 2 : 1);
2098 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
2101 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
2103 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
2104 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
2107 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2109 u32 cdw10
= nvme_pr_type(type
) << 8 | (key
? 1 << 3 : 0);
2110 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
2113 static const struct pr_ops nvme_pr_ops
= {
2114 .pr_register
= nvme_pr_register
,
2115 .pr_reserve
= nvme_pr_reserve
,
2116 .pr_release
= nvme_pr_release
,
2117 .pr_preempt
= nvme_pr_preempt
,
2118 .pr_clear
= nvme_pr_clear
,
2121 #ifdef CONFIG_BLK_SED_OPAL
2122 int nvme_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
, size_t len
,
2125 struct nvme_ctrl
*ctrl
= data
;
2126 struct nvme_command cmd
;
2128 memset(&cmd
, 0, sizeof(cmd
));
2130 cmd
.common
.opcode
= nvme_admin_security_send
;
2132 cmd
.common
.opcode
= nvme_admin_security_recv
;
2133 cmd
.common
.nsid
= 0;
2134 cmd
.common
.cdw10
= cpu_to_le32(((u32
)secp
) << 24 | ((u32
)spsp
) << 8);
2135 cmd
.common
.cdw11
= cpu_to_le32(len
);
2137 return __nvme_submit_sync_cmd(ctrl
->admin_q
, &cmd
, NULL
, buffer
, len
,
2138 ADMIN_TIMEOUT
, NVME_QID_ANY
, 1, 0, false);
2140 EXPORT_SYMBOL_GPL(nvme_sec_submit
);
2141 #endif /* CONFIG_BLK_SED_OPAL */
2143 static const struct block_device_operations nvme_fops
= {
2144 .owner
= THIS_MODULE
,
2145 .ioctl
= nvme_ioctl
,
2146 .compat_ioctl
= nvme_compat_ioctl
,
2148 .release
= nvme_release
,
2149 .getgeo
= nvme_getgeo
,
2150 .revalidate_disk
= nvme_revalidate_disk
,
2151 .pr_ops
= &nvme_pr_ops
,
2154 #ifdef CONFIG_NVME_MULTIPATH
2155 static int nvme_ns_head_open(struct block_device
*bdev
, fmode_t mode
)
2157 struct nvme_ns_head
*head
= bdev
->bd_disk
->private_data
;
2159 if (!kref_get_unless_zero(&head
->ref
))
2164 static void nvme_ns_head_release(struct gendisk
*disk
, fmode_t mode
)
2166 nvme_put_ns_head(disk
->private_data
);
2169 const struct block_device_operations nvme_ns_head_ops
= {
2170 .owner
= THIS_MODULE
,
2171 .open
= nvme_ns_head_open
,
2172 .release
= nvme_ns_head_release
,
2173 .ioctl
= nvme_ioctl
,
2174 .compat_ioctl
= nvme_compat_ioctl
,
2175 .getgeo
= nvme_getgeo
,
2176 .pr_ops
= &nvme_pr_ops
,
2178 #endif /* CONFIG_NVME_MULTIPATH */
2180 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
2182 unsigned long timeout
=
2183 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
2184 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
2187 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2190 if ((csts
& NVME_CSTS_RDY
) == bit
)
2193 usleep_range(1000, 2000);
2194 if (fatal_signal_pending(current
))
2196 if (time_after(jiffies
, timeout
)) {
2197 dev_err(ctrl
->device
,
2198 "Device not ready; aborting %s, CSTS=0x%x\n",
2199 enabled
? "initialisation" : "reset", csts
);
2208 * If the device has been passed off to us in an enabled state, just clear
2209 * the enabled bit. The spec says we should set the 'shutdown notification
2210 * bits', but doing so may cause the device to complete commands to the
2211 * admin queue ... and we don't know what memory that might be pointing at!
2213 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
)
2217 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2218 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
2220 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2224 if (ctrl
->quirks
& NVME_QUIRK_DELAY_BEFORE_CHK_RDY
)
2225 msleep(NVME_QUIRK_DELAY_AMOUNT
);
2227 return nvme_wait_ready(ctrl
, ctrl
->cap
, false);
2229 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
2231 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
)
2234 * Default to a 4K page size, with the intention to update this
2235 * path in the future to accomodate architectures with differing
2236 * kernel and IO page sizes.
2238 unsigned dev_page_min
, page_shift
= 12;
2241 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &ctrl
->cap
);
2243 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
2246 dev_page_min
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12;
2248 if (page_shift
< dev_page_min
) {
2249 dev_err(ctrl
->device
,
2250 "Minimum device page size %u too large for host (%u)\n",
2251 1 << dev_page_min
, 1 << page_shift
);
2255 ctrl
->page_size
= 1 << page_shift
;
2257 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
2258 ctrl
->ctrl_config
|= (page_shift
- 12) << NVME_CC_MPS_SHIFT
;
2259 ctrl
->ctrl_config
|= NVME_CC_AMS_RR
| NVME_CC_SHN_NONE
;
2260 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
2261 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
2263 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2266 return nvme_wait_ready(ctrl
, ctrl
->cap
, true);
2268 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
2270 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
2272 unsigned long timeout
= jiffies
+ (ctrl
->shutdown_timeout
* HZ
);
2276 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2277 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
2279 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2283 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2284 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
2288 if (fatal_signal_pending(current
))
2290 if (time_after(jiffies
, timeout
)) {
2291 dev_err(ctrl
->device
,
2292 "Device shutdown incomplete; abort shutdown\n");
2299 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
2301 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
2302 struct request_queue
*q
)
2306 if (ctrl
->max_hw_sectors
) {
2308 (ctrl
->max_hw_sectors
/ (ctrl
->page_size
>> 9)) + 1;
2310 max_segments
= min_not_zero(max_segments
, ctrl
->max_segments
);
2311 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
2312 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
2314 blk_queue_virt_boundary(q
, ctrl
->page_size
- 1);
2315 if (ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
)
2317 blk_queue_write_cache(q
, vwc
, vwc
);
2320 static int nvme_configure_timestamp(struct nvme_ctrl
*ctrl
)
2325 if (!(ctrl
->oncs
& NVME_CTRL_ONCS_TIMESTAMP
))
2328 ts
= cpu_to_le64(ktime_to_ms(ktime_get_real()));
2329 ret
= nvme_set_features(ctrl
, NVME_FEAT_TIMESTAMP
, 0, &ts
, sizeof(ts
),
2332 dev_warn_once(ctrl
->device
,
2333 "could not set timestamp (%d)\n", ret
);
2337 static int nvme_configure_acre(struct nvme_ctrl
*ctrl
)
2339 struct nvme_feat_host_behavior
*host
;
2342 /* Don't bother enabling the feature if retry delay is not reported */
2346 host
= kzalloc(sizeof(*host
), GFP_KERNEL
);
2350 host
->acre
= NVME_ENABLE_ACRE
;
2351 ret
= nvme_set_features(ctrl
, NVME_FEAT_HOST_BEHAVIOR
, 0,
2352 host
, sizeof(*host
), NULL
);
2357 static int nvme_configure_apst(struct nvme_ctrl
*ctrl
)
2360 * APST (Autonomous Power State Transition) lets us program a
2361 * table of power state transitions that the controller will
2362 * perform automatically. We configure it with a simple
2363 * heuristic: we are willing to spend at most 2% of the time
2364 * transitioning between power states. Therefore, when running
2365 * in any given state, we will enter the next lower-power
2366 * non-operational state after waiting 50 * (enlat + exlat)
2367 * microseconds, as long as that state's exit latency is under
2368 * the requested maximum latency.
2370 * We will not autonomously enter any non-operational state for
2371 * which the total latency exceeds ps_max_latency_us. Users
2372 * can set ps_max_latency_us to zero to turn off APST.
2376 struct nvme_feat_auto_pst
*table
;
2382 * If APST isn't supported or if we haven't been initialized yet,
2383 * then don't do anything.
2388 if (ctrl
->npss
> 31) {
2389 dev_warn(ctrl
->device
, "NPSS is invalid; not using APST\n");
2393 table
= kzalloc(sizeof(*table
), GFP_KERNEL
);
2397 if (!ctrl
->apst_enabled
|| ctrl
->ps_max_latency_us
== 0) {
2398 /* Turn off APST. */
2400 dev_dbg(ctrl
->device
, "APST disabled\n");
2402 __le64 target
= cpu_to_le64(0);
2406 * Walk through all states from lowest- to highest-power.
2407 * According to the spec, lower-numbered states use more
2408 * power. NPSS, despite the name, is the index of the
2409 * lowest-power state, not the number of states.
2411 for (state
= (int)ctrl
->npss
; state
>= 0; state
--) {
2412 u64 total_latency_us
, exit_latency_us
, transition_ms
;
2415 table
->entries
[state
] = target
;
2418 * Don't allow transitions to the deepest state
2419 * if it's quirked off.
2421 if (state
== ctrl
->npss
&&
2422 (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
))
2426 * Is this state a useful non-operational state for
2427 * higher-power states to autonomously transition to?
2429 if (!(ctrl
->psd
[state
].flags
&
2430 NVME_PS_FLAGS_NON_OP_STATE
))
2434 (u64
)le32_to_cpu(ctrl
->psd
[state
].exit_lat
);
2435 if (exit_latency_us
> ctrl
->ps_max_latency_us
)
2440 le32_to_cpu(ctrl
->psd
[state
].entry_lat
);
2443 * This state is good. Use it as the APST idle
2444 * target for higher power states.
2446 transition_ms
= total_latency_us
+ 19;
2447 do_div(transition_ms
, 20);
2448 if (transition_ms
> (1 << 24) - 1)
2449 transition_ms
= (1 << 24) - 1;
2451 target
= cpu_to_le64((state
<< 3) |
2452 (transition_ms
<< 8));
2457 if (total_latency_us
> max_lat_us
)
2458 max_lat_us
= total_latency_us
;
2464 dev_dbg(ctrl
->device
, "APST enabled but no non-operational states are available\n");
2466 dev_dbg(ctrl
->device
, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2467 max_ps
, max_lat_us
, (int)sizeof(*table
), table
);
2471 ret
= nvme_set_features(ctrl
, NVME_FEAT_AUTO_PST
, apste
,
2472 table
, sizeof(*table
), NULL
);
2474 dev_err(ctrl
->device
, "failed to set APST feature (%d)\n", ret
);
2480 static void nvme_set_latency_tolerance(struct device
*dev
, s32 val
)
2482 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
2486 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT
:
2487 case PM_QOS_LATENCY_ANY
:
2495 if (ctrl
->ps_max_latency_us
!= latency
) {
2496 ctrl
->ps_max_latency_us
= latency
;
2497 nvme_configure_apst(ctrl
);
2501 struct nvme_core_quirk_entry
{
2503 * NVMe model and firmware strings are padded with spaces. For
2504 * simplicity, strings in the quirk table are padded with NULLs
2510 unsigned long quirks
;
2513 static const struct nvme_core_quirk_entry core_quirks
[] = {
2516 * This Toshiba device seems to die using any APST states. See:
2517 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2520 .mn
= "THNSF5256GPUK TOSHIBA",
2521 .quirks
= NVME_QUIRK_NO_APST
,
2525 * This LiteON CL1-3D*-Q11 firmware version has a race
2526 * condition associated with actions related to suspend to idle
2527 * LiteON has resolved the problem in future firmware
2531 .quirks
= NVME_QUIRK_SIMPLE_SUSPEND
,
2535 /* match is null-terminated but idstr is space-padded. */
2536 static bool string_matches(const char *idstr
, const char *match
, size_t len
)
2543 matchlen
= strlen(match
);
2544 WARN_ON_ONCE(matchlen
> len
);
2546 if (memcmp(idstr
, match
, matchlen
))
2549 for (; matchlen
< len
; matchlen
++)
2550 if (idstr
[matchlen
] != ' ')
2556 static bool quirk_matches(const struct nvme_id_ctrl
*id
,
2557 const struct nvme_core_quirk_entry
*q
)
2559 return q
->vid
== le16_to_cpu(id
->vid
) &&
2560 string_matches(id
->mn
, q
->mn
, sizeof(id
->mn
)) &&
2561 string_matches(id
->fr
, q
->fr
, sizeof(id
->fr
));
2564 static void nvme_init_subnqn(struct nvme_subsystem
*subsys
, struct nvme_ctrl
*ctrl
,
2565 struct nvme_id_ctrl
*id
)
2570 if(!(ctrl
->quirks
& NVME_QUIRK_IGNORE_DEV_SUBNQN
)) {
2571 nqnlen
= strnlen(id
->subnqn
, NVMF_NQN_SIZE
);
2572 if (nqnlen
> 0 && nqnlen
< NVMF_NQN_SIZE
) {
2573 strlcpy(subsys
->subnqn
, id
->subnqn
, NVMF_NQN_SIZE
);
2577 if (ctrl
->vs
>= NVME_VS(1, 2, 1))
2578 dev_warn(ctrl
->device
, "missing or invalid SUBNQN field.\n");
2581 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2582 off
= snprintf(subsys
->subnqn
, NVMF_NQN_SIZE
,
2583 "nqn.2014.08.org.nvmexpress:%04x%04x",
2584 le16_to_cpu(id
->vid
), le16_to_cpu(id
->ssvid
));
2585 memcpy(subsys
->subnqn
+ off
, id
->sn
, sizeof(id
->sn
));
2586 off
+= sizeof(id
->sn
);
2587 memcpy(subsys
->subnqn
+ off
, id
->mn
, sizeof(id
->mn
));
2588 off
+= sizeof(id
->mn
);
2589 memset(subsys
->subnqn
+ off
, 0, sizeof(subsys
->subnqn
) - off
);
2592 static void nvme_release_subsystem(struct device
*dev
)
2594 struct nvme_subsystem
*subsys
=
2595 container_of(dev
, struct nvme_subsystem
, dev
);
2597 if (subsys
->instance
>= 0)
2598 ida_simple_remove(&nvme_instance_ida
, subsys
->instance
);
2602 static void nvme_destroy_subsystem(struct kref
*ref
)
2604 struct nvme_subsystem
*subsys
=
2605 container_of(ref
, struct nvme_subsystem
, ref
);
2607 mutex_lock(&nvme_subsystems_lock
);
2608 list_del(&subsys
->entry
);
2609 mutex_unlock(&nvme_subsystems_lock
);
2611 ida_destroy(&subsys
->ns_ida
);
2612 device_del(&subsys
->dev
);
2613 put_device(&subsys
->dev
);
2616 static void nvme_put_subsystem(struct nvme_subsystem
*subsys
)
2618 kref_put(&subsys
->ref
, nvme_destroy_subsystem
);
2621 static struct nvme_subsystem
*__nvme_find_get_subsystem(const char *subsysnqn
)
2623 struct nvme_subsystem
*subsys
;
2625 lockdep_assert_held(&nvme_subsystems_lock
);
2628 * Fail matches for discovery subsystems. This results
2629 * in each discovery controller bound to a unique subsystem.
2630 * This avoids issues with validating controller values
2631 * that can only be true when there is a single unique subsystem.
2632 * There may be multiple and completely independent entities
2633 * that provide discovery controllers.
2635 if (!strcmp(subsysnqn
, NVME_DISC_SUBSYS_NAME
))
2638 list_for_each_entry(subsys
, &nvme_subsystems
, entry
) {
2639 if (strcmp(subsys
->subnqn
, subsysnqn
))
2641 if (!kref_get_unless_zero(&subsys
->ref
))
2649 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2650 struct device_attribute subsys_attr_##_name = \
2651 __ATTR(_name, _mode, _show, NULL)
2653 static ssize_t
nvme_subsys_show_nqn(struct device
*dev
,
2654 struct device_attribute
*attr
,
2657 struct nvme_subsystem
*subsys
=
2658 container_of(dev
, struct nvme_subsystem
, dev
);
2660 return snprintf(buf
, PAGE_SIZE
, "%s\n", subsys
->subnqn
);
2662 static SUBSYS_ATTR_RO(subsysnqn
, S_IRUGO
, nvme_subsys_show_nqn
);
2664 #define nvme_subsys_show_str_function(field) \
2665 static ssize_t subsys_##field##_show(struct device *dev, \
2666 struct device_attribute *attr, char *buf) \
2668 struct nvme_subsystem *subsys = \
2669 container_of(dev, struct nvme_subsystem, dev); \
2670 return sprintf(buf, "%.*s\n", \
2671 (int)sizeof(subsys->field), subsys->field); \
2673 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2675 nvme_subsys_show_str_function(model
);
2676 nvme_subsys_show_str_function(serial
);
2677 nvme_subsys_show_str_function(firmware_rev
);
2679 static struct attribute
*nvme_subsys_attrs
[] = {
2680 &subsys_attr_model
.attr
,
2681 &subsys_attr_serial
.attr
,
2682 &subsys_attr_firmware_rev
.attr
,
2683 &subsys_attr_subsysnqn
.attr
,
2684 #ifdef CONFIG_NVME_MULTIPATH
2685 &subsys_attr_iopolicy
.attr
,
2690 static struct attribute_group nvme_subsys_attrs_group
= {
2691 .attrs
= nvme_subsys_attrs
,
2694 static const struct attribute_group
*nvme_subsys_attrs_groups
[] = {
2695 &nvme_subsys_attrs_group
,
2699 static bool nvme_validate_cntlid(struct nvme_subsystem
*subsys
,
2700 struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2702 struct nvme_ctrl
*tmp
;
2704 lockdep_assert_held(&nvme_subsystems_lock
);
2706 list_for_each_entry(tmp
, &subsys
->ctrls
, subsys_entry
) {
2707 if (nvme_state_terminal(tmp
))
2710 if (tmp
->cntlid
== ctrl
->cntlid
) {
2711 dev_err(ctrl
->device
,
2712 "Duplicate cntlid %u with %s, rejecting\n",
2713 ctrl
->cntlid
, dev_name(tmp
->device
));
2717 if ((id
->cmic
& NVME_CTRL_CMIC_MULTI_CTRL
) ||
2718 (ctrl
->opts
&& ctrl
->opts
->discovery_nqn
))
2721 dev_err(ctrl
->device
,
2722 "Subsystem does not support multiple controllers\n");
2729 static int nvme_init_subsystem(struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2731 struct nvme_subsystem
*subsys
, *found
;
2734 subsys
= kzalloc(sizeof(*subsys
), GFP_KERNEL
);
2738 subsys
->instance
= -1;
2739 mutex_init(&subsys
->lock
);
2740 kref_init(&subsys
->ref
);
2741 INIT_LIST_HEAD(&subsys
->ctrls
);
2742 INIT_LIST_HEAD(&subsys
->nsheads
);
2743 nvme_init_subnqn(subsys
, ctrl
, id
);
2744 memcpy(subsys
->serial
, id
->sn
, sizeof(subsys
->serial
));
2745 memcpy(subsys
->model
, id
->mn
, sizeof(subsys
->model
));
2746 memcpy(subsys
->firmware_rev
, id
->fr
, sizeof(subsys
->firmware_rev
));
2747 subsys
->vendor_id
= le16_to_cpu(id
->vid
);
2748 subsys
->cmic
= id
->cmic
;
2749 subsys
->awupf
= le16_to_cpu(id
->awupf
);
2750 #ifdef CONFIG_NVME_MULTIPATH
2751 subsys
->iopolicy
= NVME_IOPOLICY_NUMA
;
2754 subsys
->dev
.class = nvme_subsys_class
;
2755 subsys
->dev
.release
= nvme_release_subsystem
;
2756 subsys
->dev
.groups
= nvme_subsys_attrs_groups
;
2757 dev_set_name(&subsys
->dev
, "nvme-subsys%d", ctrl
->instance
);
2758 device_initialize(&subsys
->dev
);
2760 mutex_lock(&nvme_subsystems_lock
);
2761 found
= __nvme_find_get_subsystem(subsys
->subnqn
);
2763 put_device(&subsys
->dev
);
2766 if (!nvme_validate_cntlid(subsys
, ctrl
, id
)) {
2768 goto out_put_subsystem
;
2771 ret
= device_add(&subsys
->dev
);
2773 dev_err(ctrl
->device
,
2774 "failed to register subsystem device.\n");
2775 put_device(&subsys
->dev
);
2778 ida_init(&subsys
->ns_ida
);
2779 list_add_tail(&subsys
->entry
, &nvme_subsystems
);
2782 ret
= sysfs_create_link(&subsys
->dev
.kobj
, &ctrl
->device
->kobj
,
2783 dev_name(ctrl
->device
));
2785 dev_err(ctrl
->device
,
2786 "failed to create sysfs link from subsystem.\n");
2787 goto out_put_subsystem
;
2791 subsys
->instance
= ctrl
->instance
;
2792 ctrl
->subsys
= subsys
;
2793 list_add_tail(&ctrl
->subsys_entry
, &subsys
->ctrls
);
2794 mutex_unlock(&nvme_subsystems_lock
);
2798 nvme_put_subsystem(subsys
);
2800 mutex_unlock(&nvme_subsystems_lock
);
2804 int nvme_get_log(struct nvme_ctrl
*ctrl
, u32 nsid
, u8 log_page
, u8 lsp
,
2805 void *log
, size_t size
, u64 offset
)
2807 struct nvme_command c
= { };
2808 u32 dwlen
= nvme_bytes_to_numd(size
);
2810 c
.get_log_page
.opcode
= nvme_admin_get_log_page
;
2811 c
.get_log_page
.nsid
= cpu_to_le32(nsid
);
2812 c
.get_log_page
.lid
= log_page
;
2813 c
.get_log_page
.lsp
= lsp
;
2814 c
.get_log_page
.numdl
= cpu_to_le16(dwlen
& ((1 << 16) - 1));
2815 c
.get_log_page
.numdu
= cpu_to_le16(dwlen
>> 16);
2816 c
.get_log_page
.lpol
= cpu_to_le32(lower_32_bits(offset
));
2817 c
.get_log_page
.lpou
= cpu_to_le32(upper_32_bits(offset
));
2819 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, log
, size
);
2822 static int nvme_get_effects_log(struct nvme_ctrl
*ctrl
)
2827 ctrl
->effects
= kzalloc(sizeof(*ctrl
->effects
), GFP_KERNEL
);
2832 ret
= nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_CMD_EFFECTS
, 0,
2833 ctrl
->effects
, sizeof(*ctrl
->effects
), 0);
2835 kfree(ctrl
->effects
);
2836 ctrl
->effects
= NULL
;
2842 * Initialize the cached copies of the Identify data and various controller
2843 * register in our nvme_ctrl structure. This should be called as soon as
2844 * the admin queue is fully up and running.
2846 int nvme_init_identify(struct nvme_ctrl
*ctrl
)
2848 struct nvme_id_ctrl
*id
;
2849 int ret
, page_shift
;
2851 bool prev_apst_enabled
;
2853 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
2855 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
2858 page_shift
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12;
2859 ctrl
->sqsize
= min_t(int, NVME_CAP_MQES(ctrl
->cap
), ctrl
->sqsize
);
2861 if (ctrl
->vs
>= NVME_VS(1, 1, 0))
2862 ctrl
->subsystem
= NVME_CAP_NSSRC(ctrl
->cap
);
2864 ret
= nvme_identify_ctrl(ctrl
, &id
);
2866 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
2870 if (id
->lpa
& NVME_CTRL_LPA_CMD_EFFECTS_LOG
) {
2871 ret
= nvme_get_effects_log(ctrl
);
2876 if (!(ctrl
->ops
->flags
& NVME_F_FABRICS
))
2877 ctrl
->cntlid
= le16_to_cpu(id
->cntlid
);
2879 if (!ctrl
->identified
) {
2882 ret
= nvme_init_subsystem(ctrl
, id
);
2887 * Check for quirks. Quirk can depend on firmware version,
2888 * so, in principle, the set of quirks present can change
2889 * across a reset. As a possible future enhancement, we
2890 * could re-scan for quirks every time we reinitialize
2891 * the device, but we'd have to make sure that the driver
2892 * behaves intelligently if the quirks change.
2894 for (i
= 0; i
< ARRAY_SIZE(core_quirks
); i
++) {
2895 if (quirk_matches(id
, &core_quirks
[i
]))
2896 ctrl
->quirks
|= core_quirks
[i
].quirks
;
2900 if (force_apst
&& (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
)) {
2901 dev_warn(ctrl
->device
, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2902 ctrl
->quirks
&= ~NVME_QUIRK_NO_DEEPEST_PS
;
2905 ctrl
->crdt
[0] = le16_to_cpu(id
->crdt1
);
2906 ctrl
->crdt
[1] = le16_to_cpu(id
->crdt2
);
2907 ctrl
->crdt
[2] = le16_to_cpu(id
->crdt3
);
2909 ctrl
->oacs
= le16_to_cpu(id
->oacs
);
2910 ctrl
->oncs
= le16_to_cpu(id
->oncs
);
2911 ctrl
->mtfa
= le16_to_cpu(id
->mtfa
);
2912 ctrl
->oaes
= le32_to_cpu(id
->oaes
);
2913 ctrl
->wctemp
= le16_to_cpu(id
->wctemp
);
2914 ctrl
->cctemp
= le16_to_cpu(id
->cctemp
);
2916 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
2917 ctrl
->vwc
= id
->vwc
;
2919 max_hw_sectors
= 1 << (id
->mdts
+ page_shift
- 9);
2921 max_hw_sectors
= UINT_MAX
;
2922 ctrl
->max_hw_sectors
=
2923 min_not_zero(ctrl
->max_hw_sectors
, max_hw_sectors
);
2925 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
2926 ctrl
->sgls
= le32_to_cpu(id
->sgls
);
2927 ctrl
->kas
= le16_to_cpu(id
->kas
);
2928 ctrl
->max_namespaces
= le32_to_cpu(id
->mnan
);
2929 ctrl
->ctratt
= le32_to_cpu(id
->ctratt
);
2933 u32 transition_time
= le32_to_cpu(id
->rtd3e
) / 1000000;
2935 ctrl
->shutdown_timeout
= clamp_t(unsigned int, transition_time
,
2936 shutdown_timeout
, 60);
2938 if (ctrl
->shutdown_timeout
!= shutdown_timeout
)
2939 dev_info(ctrl
->device
,
2940 "Shutdown timeout set to %u seconds\n",
2941 ctrl
->shutdown_timeout
);
2943 ctrl
->shutdown_timeout
= shutdown_timeout
;
2945 ctrl
->npss
= id
->npss
;
2946 ctrl
->apsta
= id
->apsta
;
2947 prev_apst_enabled
= ctrl
->apst_enabled
;
2948 if (ctrl
->quirks
& NVME_QUIRK_NO_APST
) {
2949 if (force_apst
&& id
->apsta
) {
2950 dev_warn(ctrl
->device
, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2951 ctrl
->apst_enabled
= true;
2953 ctrl
->apst_enabled
= false;
2956 ctrl
->apst_enabled
= id
->apsta
;
2958 memcpy(ctrl
->psd
, id
->psd
, sizeof(ctrl
->psd
));
2960 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
2961 ctrl
->icdoff
= le16_to_cpu(id
->icdoff
);
2962 ctrl
->ioccsz
= le32_to_cpu(id
->ioccsz
);
2963 ctrl
->iorcsz
= le32_to_cpu(id
->iorcsz
);
2964 ctrl
->maxcmd
= le16_to_cpu(id
->maxcmd
);
2967 * In fabrics we need to verify the cntlid matches the
2970 if (ctrl
->cntlid
!= le16_to_cpu(id
->cntlid
)) {
2971 dev_err(ctrl
->device
,
2972 "Mismatching cntlid: Connect %u vs Identify "
2974 ctrl
->cntlid
, le16_to_cpu(id
->cntlid
));
2979 if (!ctrl
->opts
->discovery_nqn
&& !ctrl
->kas
) {
2980 dev_err(ctrl
->device
,
2981 "keep-alive support is mandatory for fabrics\n");
2986 ctrl
->hmpre
= le32_to_cpu(id
->hmpre
);
2987 ctrl
->hmmin
= le32_to_cpu(id
->hmmin
);
2988 ctrl
->hmminds
= le32_to_cpu(id
->hmminds
);
2989 ctrl
->hmmaxd
= le16_to_cpu(id
->hmmaxd
);
2992 ret
= nvme_mpath_init(ctrl
, id
);
2998 if (ctrl
->apst_enabled
&& !prev_apst_enabled
)
2999 dev_pm_qos_expose_latency_tolerance(ctrl
->device
);
3000 else if (!ctrl
->apst_enabled
&& prev_apst_enabled
)
3001 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
3003 ret
= nvme_configure_apst(ctrl
);
3007 ret
= nvme_configure_timestamp(ctrl
);
3011 ret
= nvme_configure_directives(ctrl
);
3015 ret
= nvme_configure_acre(ctrl
);
3019 if (!ctrl
->identified
)
3020 nvme_hwmon_init(ctrl
);
3022 ctrl
->identified
= true;
3030 EXPORT_SYMBOL_GPL(nvme_init_identify
);
3032 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
3034 struct nvme_ctrl
*ctrl
=
3035 container_of(inode
->i_cdev
, struct nvme_ctrl
, cdev
);
3037 switch (ctrl
->state
) {
3038 case NVME_CTRL_LIVE
:
3041 return -EWOULDBLOCK
;
3044 file
->private_data
= ctrl
;
3048 static int nvme_dev_user_cmd(struct nvme_ctrl
*ctrl
, void __user
*argp
)
3053 down_read(&ctrl
->namespaces_rwsem
);
3054 if (list_empty(&ctrl
->namespaces
)) {
3059 ns
= list_first_entry(&ctrl
->namespaces
, struct nvme_ns
, list
);
3060 if (ns
!= list_last_entry(&ctrl
->namespaces
, struct nvme_ns
, list
)) {
3061 dev_warn(ctrl
->device
,
3062 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3067 dev_warn(ctrl
->device
,
3068 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3069 kref_get(&ns
->kref
);
3070 up_read(&ctrl
->namespaces_rwsem
);
3072 ret
= nvme_user_cmd(ctrl
, ns
, argp
);
3077 up_read(&ctrl
->namespaces_rwsem
);
3081 static long nvme_dev_ioctl(struct file
*file
, unsigned int cmd
,
3084 struct nvme_ctrl
*ctrl
= file
->private_data
;
3085 void __user
*argp
= (void __user
*)arg
;
3088 case NVME_IOCTL_ADMIN_CMD
:
3089 return nvme_user_cmd(ctrl
, NULL
, argp
);
3090 case NVME_IOCTL_ADMIN64_CMD
:
3091 return nvme_user_cmd64(ctrl
, NULL
, argp
);
3092 case NVME_IOCTL_IO_CMD
:
3093 return nvme_dev_user_cmd(ctrl
, argp
);
3094 case NVME_IOCTL_RESET
:
3095 dev_warn(ctrl
->device
, "resetting controller\n");
3096 return nvme_reset_ctrl_sync(ctrl
);
3097 case NVME_IOCTL_SUBSYS_RESET
:
3098 return nvme_reset_subsystem(ctrl
);
3099 case NVME_IOCTL_RESCAN
:
3100 nvme_queue_scan(ctrl
);
3107 static const struct file_operations nvme_dev_fops
= {
3108 .owner
= THIS_MODULE
,
3109 .open
= nvme_dev_open
,
3110 .unlocked_ioctl
= nvme_dev_ioctl
,
3111 .compat_ioctl
= compat_ptr_ioctl
,
3114 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
3115 struct device_attribute
*attr
, const char *buf
,
3118 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3121 ret
= nvme_reset_ctrl_sync(ctrl
);
3126 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
3128 static ssize_t
nvme_sysfs_rescan(struct device
*dev
,
3129 struct device_attribute
*attr
, const char *buf
,
3132 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3134 nvme_queue_scan(ctrl
);
3137 static DEVICE_ATTR(rescan_controller
, S_IWUSR
, NULL
, nvme_sysfs_rescan
);
3139 static inline struct nvme_ns_head
*dev_to_ns_head(struct device
*dev
)
3141 struct gendisk
*disk
= dev_to_disk(dev
);
3143 if (disk
->fops
== &nvme_fops
)
3144 return nvme_get_ns_from_dev(dev
)->head
;
3146 return disk
->private_data
;
3149 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
3152 struct nvme_ns_head
*head
= dev_to_ns_head(dev
);
3153 struct nvme_ns_ids
*ids
= &head
->ids
;
3154 struct nvme_subsystem
*subsys
= head
->subsys
;
3155 int serial_len
= sizeof(subsys
->serial
);
3156 int model_len
= sizeof(subsys
->model
);
3158 if (!uuid_is_null(&ids
->uuid
))
3159 return sprintf(buf
, "uuid.%pU\n", &ids
->uuid
);
3161 if (memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3162 return sprintf(buf
, "eui.%16phN\n", ids
->nguid
);
3164 if (memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3165 return sprintf(buf
, "eui.%8phN\n", ids
->eui64
);
3167 while (serial_len
> 0 && (subsys
->serial
[serial_len
- 1] == ' ' ||
3168 subsys
->serial
[serial_len
- 1] == '\0'))
3170 while (model_len
> 0 && (subsys
->model
[model_len
- 1] == ' ' ||
3171 subsys
->model
[model_len
- 1] == '\0'))
3174 return sprintf(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", subsys
->vendor_id
,
3175 serial_len
, subsys
->serial
, model_len
, subsys
->model
,
3178 static DEVICE_ATTR_RO(wwid
);
3180 static ssize_t
nguid_show(struct device
*dev
, struct device_attribute
*attr
,
3183 return sprintf(buf
, "%pU\n", dev_to_ns_head(dev
)->ids
.nguid
);
3185 static DEVICE_ATTR_RO(nguid
);
3187 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
3190 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3192 /* For backward compatibility expose the NGUID to userspace if
3193 * we have no UUID set
3195 if (uuid_is_null(&ids
->uuid
)) {
3196 printk_ratelimited(KERN_WARNING
3197 "No UUID available providing old NGUID\n");
3198 return sprintf(buf
, "%pU\n", ids
->nguid
);
3200 return sprintf(buf
, "%pU\n", &ids
->uuid
);
3202 static DEVICE_ATTR_RO(uuid
);
3204 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
3207 return sprintf(buf
, "%8ph\n", dev_to_ns_head(dev
)->ids
.eui64
);
3209 static DEVICE_ATTR_RO(eui
);
3211 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
3214 return sprintf(buf
, "%d\n", dev_to_ns_head(dev
)->ns_id
);
3216 static DEVICE_ATTR_RO(nsid
);
3218 static struct attribute
*nvme_ns_id_attrs
[] = {
3219 &dev_attr_wwid
.attr
,
3220 &dev_attr_uuid
.attr
,
3221 &dev_attr_nguid
.attr
,
3223 &dev_attr_nsid
.attr
,
3224 #ifdef CONFIG_NVME_MULTIPATH
3225 &dev_attr_ana_grpid
.attr
,
3226 &dev_attr_ana_state
.attr
,
3231 static umode_t
nvme_ns_id_attrs_are_visible(struct kobject
*kobj
,
3232 struct attribute
*a
, int n
)
3234 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3235 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3237 if (a
== &dev_attr_uuid
.attr
) {
3238 if (uuid_is_null(&ids
->uuid
) &&
3239 !memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3242 if (a
== &dev_attr_nguid
.attr
) {
3243 if (!memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3246 if (a
== &dev_attr_eui
.attr
) {
3247 if (!memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3250 #ifdef CONFIG_NVME_MULTIPATH
3251 if (a
== &dev_attr_ana_grpid
.attr
|| a
== &dev_attr_ana_state
.attr
) {
3252 if (dev_to_disk(dev
)->fops
!= &nvme_fops
) /* per-path attr */
3254 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev
)->ctrl
))
3261 static const struct attribute_group nvme_ns_id_attr_group
= {
3262 .attrs
= nvme_ns_id_attrs
,
3263 .is_visible
= nvme_ns_id_attrs_are_visible
,
3266 const struct attribute_group
*nvme_ns_id_attr_groups
[] = {
3267 &nvme_ns_id_attr_group
,
3269 &nvme_nvm_attr_group
,
3274 #define nvme_show_str_function(field) \
3275 static ssize_t field##_show(struct device *dev, \
3276 struct device_attribute *attr, char *buf) \
3278 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3279 return sprintf(buf, "%.*s\n", \
3280 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3282 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3284 nvme_show_str_function(model
);
3285 nvme_show_str_function(serial
);
3286 nvme_show_str_function(firmware_rev
);
3288 #define nvme_show_int_function(field) \
3289 static ssize_t field##_show(struct device *dev, \
3290 struct device_attribute *attr, char *buf) \
3292 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3293 return sprintf(buf, "%d\n", ctrl->field); \
3295 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3297 nvme_show_int_function(cntlid
);
3298 nvme_show_int_function(numa_node
);
3299 nvme_show_int_function(queue_count
);
3300 nvme_show_int_function(sqsize
);
3302 static ssize_t
nvme_sysfs_delete(struct device
*dev
,
3303 struct device_attribute
*attr
, const char *buf
,
3306 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3308 /* Can't delete non-created controllers */
3312 if (device_remove_file_self(dev
, attr
))
3313 nvme_delete_ctrl_sync(ctrl
);
3316 static DEVICE_ATTR(delete_controller
, S_IWUSR
, NULL
, nvme_sysfs_delete
);
3318 static ssize_t
nvme_sysfs_show_transport(struct device
*dev
,
3319 struct device_attribute
*attr
,
3322 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3324 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->ops
->name
);
3326 static DEVICE_ATTR(transport
, S_IRUGO
, nvme_sysfs_show_transport
, NULL
);
3328 static ssize_t
nvme_sysfs_show_state(struct device
*dev
,
3329 struct device_attribute
*attr
,
3332 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3333 static const char *const state_name
[] = {
3334 [NVME_CTRL_NEW
] = "new",
3335 [NVME_CTRL_LIVE
] = "live",
3336 [NVME_CTRL_RESETTING
] = "resetting",
3337 [NVME_CTRL_CONNECTING
] = "connecting",
3338 [NVME_CTRL_DELETING
] = "deleting",
3339 [NVME_CTRL_DEAD
] = "dead",
3342 if ((unsigned)ctrl
->state
< ARRAY_SIZE(state_name
) &&
3343 state_name
[ctrl
->state
])
3344 return sprintf(buf
, "%s\n", state_name
[ctrl
->state
]);
3346 return sprintf(buf
, "unknown state\n");
3349 static DEVICE_ATTR(state
, S_IRUGO
, nvme_sysfs_show_state
, NULL
);
3351 static ssize_t
nvme_sysfs_show_subsysnqn(struct device
*dev
,
3352 struct device_attribute
*attr
,
3355 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3357 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->subsys
->subnqn
);
3359 static DEVICE_ATTR(subsysnqn
, S_IRUGO
, nvme_sysfs_show_subsysnqn
, NULL
);
3361 static ssize_t
nvme_sysfs_show_hostnqn(struct device
*dev
,
3362 struct device_attribute
*attr
,
3365 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3367 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->opts
->host
->nqn
);
3369 static DEVICE_ATTR(hostnqn
, S_IRUGO
, nvme_sysfs_show_hostnqn
, NULL
);
3371 static ssize_t
nvme_sysfs_show_hostid(struct device
*dev
,
3372 struct device_attribute
*attr
,
3375 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3377 return snprintf(buf
, PAGE_SIZE
, "%pU\n", &ctrl
->opts
->host
->id
);
3379 static DEVICE_ATTR(hostid
, S_IRUGO
, nvme_sysfs_show_hostid
, NULL
);
3381 static ssize_t
nvme_sysfs_show_address(struct device
*dev
,
3382 struct device_attribute
*attr
,
3385 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3387 return ctrl
->ops
->get_address(ctrl
, buf
, PAGE_SIZE
);
3389 static DEVICE_ATTR(address
, S_IRUGO
, nvme_sysfs_show_address
, NULL
);
3391 static struct attribute
*nvme_dev_attrs
[] = {
3392 &dev_attr_reset_controller
.attr
,
3393 &dev_attr_rescan_controller
.attr
,
3394 &dev_attr_model
.attr
,
3395 &dev_attr_serial
.attr
,
3396 &dev_attr_firmware_rev
.attr
,
3397 &dev_attr_cntlid
.attr
,
3398 &dev_attr_delete_controller
.attr
,
3399 &dev_attr_transport
.attr
,
3400 &dev_attr_subsysnqn
.attr
,
3401 &dev_attr_address
.attr
,
3402 &dev_attr_state
.attr
,
3403 &dev_attr_numa_node
.attr
,
3404 &dev_attr_queue_count
.attr
,
3405 &dev_attr_sqsize
.attr
,
3406 &dev_attr_hostnqn
.attr
,
3407 &dev_attr_hostid
.attr
,
3411 static umode_t
nvme_dev_attrs_are_visible(struct kobject
*kobj
,
3412 struct attribute
*a
, int n
)
3414 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3415 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3417 if (a
== &dev_attr_delete_controller
.attr
&& !ctrl
->ops
->delete_ctrl
)
3419 if (a
== &dev_attr_address
.attr
&& !ctrl
->ops
->get_address
)
3421 if (a
== &dev_attr_hostnqn
.attr
&& !ctrl
->opts
)
3423 if (a
== &dev_attr_hostid
.attr
&& !ctrl
->opts
)
3429 static struct attribute_group nvme_dev_attrs_group
= {
3430 .attrs
= nvme_dev_attrs
,
3431 .is_visible
= nvme_dev_attrs_are_visible
,
3434 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
3435 &nvme_dev_attrs_group
,
3439 static struct nvme_ns_head
*nvme_find_ns_head(struct nvme_subsystem
*subsys
,
3442 struct nvme_ns_head
*h
;
3444 lockdep_assert_held(&subsys
->lock
);
3446 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3447 if (h
->ns_id
== nsid
&& kref_get_unless_zero(&h
->ref
))
3454 static int __nvme_check_ids(struct nvme_subsystem
*subsys
,
3455 struct nvme_ns_head
*new)
3457 struct nvme_ns_head
*h
;
3459 lockdep_assert_held(&subsys
->lock
);
3461 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3462 if (nvme_ns_ids_valid(&new->ids
) &&
3463 nvme_ns_ids_equal(&new->ids
, &h
->ids
))
3470 static struct nvme_ns_head
*nvme_alloc_ns_head(struct nvme_ctrl
*ctrl
,
3471 unsigned nsid
, struct nvme_ns_ids
*ids
)
3473 struct nvme_ns_head
*head
;
3474 size_t size
= sizeof(*head
);
3477 #ifdef CONFIG_NVME_MULTIPATH
3478 size
+= num_possible_nodes() * sizeof(struct nvme_ns
*);
3481 head
= kzalloc(size
, GFP_KERNEL
);
3484 ret
= ida_simple_get(&ctrl
->subsys
->ns_ida
, 1, 0, GFP_KERNEL
);
3487 head
->instance
= ret
;
3488 INIT_LIST_HEAD(&head
->list
);
3489 ret
= init_srcu_struct(&head
->srcu
);
3491 goto out_ida_remove
;
3492 head
->subsys
= ctrl
->subsys
;
3495 kref_init(&head
->ref
);
3497 ret
= __nvme_check_ids(ctrl
->subsys
, head
);
3499 dev_err(ctrl
->device
,
3500 "duplicate IDs for nsid %d\n", nsid
);
3501 goto out_cleanup_srcu
;
3504 ret
= nvme_mpath_alloc_disk(ctrl
, head
);
3506 goto out_cleanup_srcu
;
3508 list_add_tail(&head
->entry
, &ctrl
->subsys
->nsheads
);
3510 kref_get(&ctrl
->subsys
->ref
);
3514 cleanup_srcu_struct(&head
->srcu
);
3516 ida_simple_remove(&ctrl
->subsys
->ns_ida
, head
->instance
);
3521 ret
= blk_status_to_errno(nvme_error_status(ret
));
3522 return ERR_PTR(ret
);
3525 static int nvme_init_ns_head(struct nvme_ns
*ns
, unsigned nsid
,
3526 struct nvme_id_ns
*id
)
3528 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
3529 bool is_shared
= id
->nmic
& NVME_NS_NMIC_SHARED
;
3530 struct nvme_ns_head
*head
= NULL
;
3531 struct nvme_ns_ids ids
;
3534 ret
= nvme_report_ns_ids(ctrl
, nsid
, id
, &ids
);
3538 return blk_status_to_errno(nvme_error_status(ret
));
3541 mutex_lock(&ctrl
->subsys
->lock
);
3542 head
= nvme_find_ns_head(ctrl
->subsys
, nsid
);
3544 head
= nvme_alloc_ns_head(ctrl
, nsid
, &ids
);
3546 ret
= PTR_ERR(head
);
3549 head
->shared
= is_shared
;
3552 if (!is_shared
|| !head
->shared
) {
3553 dev_err(ctrl
->device
,
3554 "Duplicate unshared namespace %d\n", nsid
);
3555 goto out_put_ns_head
;
3557 if (!nvme_ns_ids_equal(&head
->ids
, &ids
)) {
3558 dev_err(ctrl
->device
,
3559 "IDs don't match for shared namespace %d\n",
3561 goto out_put_ns_head
;
3565 list_add_tail(&ns
->siblings
, &head
->list
);
3567 mutex_unlock(&ctrl
->subsys
->lock
);
3571 nvme_put_ns_head(head
);
3573 mutex_unlock(&ctrl
->subsys
->lock
);
3577 static int ns_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3579 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
3580 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
3582 return nsa
->head
->ns_id
- nsb
->head
->ns_id
;
3585 static struct nvme_ns
*nvme_find_get_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3587 struct nvme_ns
*ns
, *ret
= NULL
;
3589 down_read(&ctrl
->namespaces_rwsem
);
3590 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
3591 if (ns
->head
->ns_id
== nsid
) {
3592 if (!kref_get_unless_zero(&ns
->kref
))
3597 if (ns
->head
->ns_id
> nsid
)
3600 up_read(&ctrl
->namespaces_rwsem
);
3604 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3607 struct gendisk
*disk
;
3608 struct nvme_id_ns
*id
;
3609 char disk_name
[DISK_NAME_LEN
];
3610 int node
= ctrl
->numa_node
, flags
= GENHD_FL_EXT_DEVT
, ret
;
3612 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
3616 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
3617 if (IS_ERR(ns
->queue
))
3620 if (ctrl
->opts
&& ctrl
->opts
->data_digest
)
3621 ns
->queue
->backing_dev_info
->capabilities
3622 |= BDI_CAP_STABLE_WRITES
;
3624 blk_queue_flag_set(QUEUE_FLAG_NONROT
, ns
->queue
);
3625 if (ctrl
->ops
->flags
& NVME_F_PCI_P2PDMA
)
3626 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA
, ns
->queue
);
3628 ns
->queue
->queuedata
= ns
;
3631 kref_init(&ns
->kref
);
3632 ns
->lba_shift
= 9; /* set to a default value for 512 until disk is validated */
3634 blk_queue_logical_block_size(ns
->queue
, 1 << ns
->lba_shift
);
3635 nvme_set_queue_limits(ctrl
, ns
->queue
);
3637 ret
= nvme_identify_ns(ctrl
, nsid
, &id
);
3639 goto out_free_queue
;
3641 if (id
->ncap
== 0) /* no namespace (legacy quirk) */
3644 ret
= nvme_init_ns_head(ns
, nsid
, id
);
3647 nvme_set_disk_name(disk_name
, ns
, ctrl
, &flags
);
3649 disk
= alloc_disk_node(0, node
);
3653 disk
->fops
= &nvme_fops
;
3654 disk
->private_data
= ns
;
3655 disk
->queue
= ns
->queue
;
3656 disk
->flags
= flags
;
3657 memcpy(disk
->disk_name
, disk_name
, DISK_NAME_LEN
);
3660 __nvme_revalidate_disk(disk
, id
);
3662 if ((ctrl
->quirks
& NVME_QUIRK_LIGHTNVM
) && id
->vs
[0] == 0x1) {
3663 ret
= nvme_nvm_register(ns
, disk_name
, node
);
3665 dev_warn(ctrl
->device
, "LightNVM init failure\n");
3670 down_write(&ctrl
->namespaces_rwsem
);
3671 list_add_tail(&ns
->list
, &ctrl
->namespaces
);
3672 up_write(&ctrl
->namespaces_rwsem
);
3674 nvme_get_ctrl(ctrl
);
3676 device_add_disk(ctrl
->device
, ns
->disk
, nvme_ns_id_attr_groups
);
3678 nvme_mpath_add_disk(ns
, id
);
3679 nvme_fault_inject_init(&ns
->fault_inject
, ns
->disk
->disk_name
);
3684 /* prevent double queue cleanup */
3685 ns
->disk
->queue
= NULL
;
3688 mutex_lock(&ctrl
->subsys
->lock
);
3689 list_del_rcu(&ns
->siblings
);
3690 if (list_empty(&ns
->head
->list
))
3691 list_del_init(&ns
->head
->entry
);
3692 mutex_unlock(&ctrl
->subsys
->lock
);
3693 nvme_put_ns_head(ns
->head
);
3697 blk_cleanup_queue(ns
->queue
);
3702 static void nvme_ns_remove(struct nvme_ns
*ns
)
3704 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
3707 nvme_fault_inject_fini(&ns
->fault_inject
);
3709 mutex_lock(&ns
->ctrl
->subsys
->lock
);
3710 list_del_rcu(&ns
->siblings
);
3711 if (list_empty(&ns
->head
->list
))
3712 list_del_init(&ns
->head
->entry
);
3713 mutex_unlock(&ns
->ctrl
->subsys
->lock
);
3715 synchronize_rcu(); /* guarantee not available in head->list */
3716 nvme_mpath_clear_current_path(ns
);
3717 synchronize_srcu(&ns
->head
->srcu
); /* wait for concurrent submissions */
3719 if (ns
->disk
&& ns
->disk
->flags
& GENHD_FL_UP
) {
3720 del_gendisk(ns
->disk
);
3721 blk_cleanup_queue(ns
->queue
);
3722 if (blk_get_integrity(ns
->disk
))
3723 blk_integrity_unregister(ns
->disk
);
3726 down_write(&ns
->ctrl
->namespaces_rwsem
);
3727 list_del_init(&ns
->list
);
3728 up_write(&ns
->ctrl
->namespaces_rwsem
);
3730 nvme_mpath_check_last_path(ns
);
3734 static void nvme_ns_remove_by_nsid(struct nvme_ctrl
*ctrl
, u32 nsid
)
3736 struct nvme_ns
*ns
= nvme_find_get_ns(ctrl
, nsid
);
3744 static void nvme_validate_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3748 ns
= nvme_find_get_ns(ctrl
, nsid
);
3750 if (ns
->disk
&& revalidate_disk(ns
->disk
))
3754 nvme_alloc_ns(ctrl
, nsid
);
3757 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
3760 struct nvme_ns
*ns
, *next
;
3763 down_write(&ctrl
->namespaces_rwsem
);
3764 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
3765 if (ns
->head
->ns_id
> nsid
|| test_bit(NVME_NS_DEAD
, &ns
->flags
))
3766 list_move_tail(&ns
->list
, &rm_list
);
3768 up_write(&ctrl
->namespaces_rwsem
);
3770 list_for_each_entry_safe(ns
, next
, &rm_list
, list
)
3775 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
)
3777 const int nr_entries
= NVME_IDENTIFY_DATA_SIZE
/ sizeof(__le32
);
3782 if (nvme_ctrl_limited_cns(ctrl
))
3785 ns_list
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
3790 ret
= nvme_identify_ns_list(ctrl
, prev
, ns_list
);
3794 for (i
= 0; i
< nr_entries
; i
++) {
3795 u32 nsid
= le32_to_cpu(ns_list
[i
]);
3797 if (!nsid
) /* end of the list? */
3799 nvme_validate_ns(ctrl
, nsid
);
3800 while (++prev
< nsid
)
3801 nvme_ns_remove_by_nsid(ctrl
, prev
);
3805 nvme_remove_invalid_namespaces(ctrl
, prev
);
3811 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
)
3813 struct nvme_id_ctrl
*id
;
3816 if (nvme_identify_ctrl(ctrl
, &id
))
3818 nn
= le32_to_cpu(id
->nn
);
3821 for (i
= 1; i
<= nn
; i
++)
3822 nvme_validate_ns(ctrl
, i
);
3824 nvme_remove_invalid_namespaces(ctrl
, nn
);
3827 static void nvme_clear_changed_ns_log(struct nvme_ctrl
*ctrl
)
3829 size_t log_size
= NVME_MAX_CHANGED_NAMESPACES
* sizeof(__le32
);
3833 log
= kzalloc(log_size
, GFP_KERNEL
);
3838 * We need to read the log to clear the AEN, but we don't want to rely
3839 * on it for the changed namespace information as userspace could have
3840 * raced with us in reading the log page, which could cause us to miss
3843 error
= nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_CHANGED_NS
, 0, log
,
3846 dev_warn(ctrl
->device
,
3847 "reading changed ns log failed: %d\n", error
);
3852 static void nvme_scan_work(struct work_struct
*work
)
3854 struct nvme_ctrl
*ctrl
=
3855 container_of(work
, struct nvme_ctrl
, scan_work
);
3857 /* No tagset on a live ctrl means IO queues could not created */
3858 if (ctrl
->state
!= NVME_CTRL_LIVE
|| !ctrl
->tagset
)
3861 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
)) {
3862 dev_info(ctrl
->device
, "rescanning namespaces.\n");
3863 nvme_clear_changed_ns_log(ctrl
);
3866 mutex_lock(&ctrl
->scan_lock
);
3867 if (nvme_scan_ns_list(ctrl
) != 0)
3868 nvme_scan_ns_sequential(ctrl
);
3869 mutex_unlock(&ctrl
->scan_lock
);
3871 down_write(&ctrl
->namespaces_rwsem
);
3872 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
3873 up_write(&ctrl
->namespaces_rwsem
);
3877 * This function iterates the namespace list unlocked to allow recovery from
3878 * controller failure. It is up to the caller to ensure the namespace list is
3879 * not modified by scan work while this function is executing.
3881 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
3883 struct nvme_ns
*ns
, *next
;
3887 * make sure to requeue I/O to all namespaces as these
3888 * might result from the scan itself and must complete
3889 * for the scan_work to make progress
3891 nvme_mpath_clear_ctrl_paths(ctrl
);
3893 /* prevent racing with ns scanning */
3894 flush_work(&ctrl
->scan_work
);
3897 * The dead states indicates the controller was not gracefully
3898 * disconnected. In that case, we won't be able to flush any data while
3899 * removing the namespaces' disks; fail all the queues now to avoid
3900 * potentially having to clean up the failed sync later.
3902 if (ctrl
->state
== NVME_CTRL_DEAD
)
3903 nvme_kill_queues(ctrl
);
3905 down_write(&ctrl
->namespaces_rwsem
);
3906 list_splice_init(&ctrl
->namespaces
, &ns_list
);
3907 up_write(&ctrl
->namespaces_rwsem
);
3909 list_for_each_entry_safe(ns
, next
, &ns_list
, list
)
3912 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
3914 static int nvme_class_uevent(struct device
*dev
, struct kobj_uevent_env
*env
)
3916 struct nvme_ctrl
*ctrl
=
3917 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
3918 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3921 ret
= add_uevent_var(env
, "NVME_TRTYPE=%s", ctrl
->ops
->name
);
3926 ret
= add_uevent_var(env
, "NVME_TRADDR=%s", opts
->traddr
);
3930 ret
= add_uevent_var(env
, "NVME_TRSVCID=%s",
3931 opts
->trsvcid
?: "none");
3935 ret
= add_uevent_var(env
, "NVME_HOST_TRADDR=%s",
3936 opts
->host_traddr
?: "none");
3941 static void nvme_aen_uevent(struct nvme_ctrl
*ctrl
)
3943 char *envp
[2] = { NULL
, NULL
};
3944 u32 aen_result
= ctrl
->aen_result
;
3946 ctrl
->aen_result
= 0;
3950 envp
[0] = kasprintf(GFP_KERNEL
, "NVME_AEN=%#08x", aen_result
);
3953 kobject_uevent_env(&ctrl
->device
->kobj
, KOBJ_CHANGE
, envp
);
3957 static void nvme_async_event_work(struct work_struct
*work
)
3959 struct nvme_ctrl
*ctrl
=
3960 container_of(work
, struct nvme_ctrl
, async_event_work
);
3962 nvme_aen_uevent(ctrl
);
3963 ctrl
->ops
->submit_async_event(ctrl
);
3966 static bool nvme_ctrl_pp_status(struct nvme_ctrl
*ctrl
)
3971 if (ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
))
3977 return ((ctrl
->ctrl_config
& NVME_CC_ENABLE
) && (csts
& NVME_CSTS_PP
));
3980 static void nvme_get_fw_slot_info(struct nvme_ctrl
*ctrl
)
3982 struct nvme_fw_slot_info_log
*log
;
3984 log
= kmalloc(sizeof(*log
), GFP_KERNEL
);
3988 if (nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_FW_SLOT
, 0, log
,
3990 dev_warn(ctrl
->device
, "Get FW SLOT INFO log error\n");
3994 static void nvme_fw_act_work(struct work_struct
*work
)
3996 struct nvme_ctrl
*ctrl
= container_of(work
,
3997 struct nvme_ctrl
, fw_act_work
);
3998 unsigned long fw_act_timeout
;
4001 fw_act_timeout
= jiffies
+
4002 msecs_to_jiffies(ctrl
->mtfa
* 100);
4004 fw_act_timeout
= jiffies
+
4005 msecs_to_jiffies(admin_timeout
* 1000);
4007 nvme_stop_queues(ctrl
);
4008 while (nvme_ctrl_pp_status(ctrl
)) {
4009 if (time_after(jiffies
, fw_act_timeout
)) {
4010 dev_warn(ctrl
->device
,
4011 "Fw activation timeout, reset controller\n");
4012 nvme_try_sched_reset(ctrl
);
4018 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_LIVE
))
4021 nvme_start_queues(ctrl
);
4022 /* read FW slot information to clear the AER */
4023 nvme_get_fw_slot_info(ctrl
);
4026 static void nvme_handle_aen_notice(struct nvme_ctrl
*ctrl
, u32 result
)
4028 u32 aer_notice_type
= (result
& 0xff00) >> 8;
4030 trace_nvme_async_event(ctrl
, aer_notice_type
);
4032 switch (aer_notice_type
) {
4033 case NVME_AER_NOTICE_NS_CHANGED
:
4034 set_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
);
4035 nvme_queue_scan(ctrl
);
4037 case NVME_AER_NOTICE_FW_ACT_STARTING
:
4039 * We are (ab)using the RESETTING state to prevent subsequent
4040 * recovery actions from interfering with the controller's
4041 * firmware activation.
4043 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
4044 queue_work(nvme_wq
, &ctrl
->fw_act_work
);
4046 #ifdef CONFIG_NVME_MULTIPATH
4047 case NVME_AER_NOTICE_ANA
:
4048 if (!ctrl
->ana_log_buf
)
4050 queue_work(nvme_wq
, &ctrl
->ana_work
);
4053 case NVME_AER_NOTICE_DISC_CHANGED
:
4054 ctrl
->aen_result
= result
;
4057 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
4061 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
, __le16 status
,
4062 volatile union nvme_result
*res
)
4064 u32 result
= le32_to_cpu(res
->u32
);
4065 u32 aer_type
= result
& 0x07;
4067 if (le16_to_cpu(status
) >> 1 != NVME_SC_SUCCESS
)
4071 case NVME_AER_NOTICE
:
4072 nvme_handle_aen_notice(ctrl
, result
);
4074 case NVME_AER_ERROR
:
4075 case NVME_AER_SMART
:
4078 trace_nvme_async_event(ctrl
, aer_type
);
4079 ctrl
->aen_result
= result
;
4084 queue_work(nvme_wq
, &ctrl
->async_event_work
);
4086 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
4088 void nvme_stop_ctrl(struct nvme_ctrl
*ctrl
)
4090 nvme_mpath_stop(ctrl
);
4091 nvme_stop_keep_alive(ctrl
);
4092 flush_work(&ctrl
->async_event_work
);
4093 cancel_work_sync(&ctrl
->fw_act_work
);
4095 EXPORT_SYMBOL_GPL(nvme_stop_ctrl
);
4097 void nvme_start_ctrl(struct nvme_ctrl
*ctrl
)
4100 nvme_start_keep_alive(ctrl
);
4102 nvme_enable_aen(ctrl
);
4104 if (ctrl
->queue_count
> 1) {
4105 nvme_queue_scan(ctrl
);
4106 nvme_start_queues(ctrl
);
4108 ctrl
->created
= true;
4110 EXPORT_SYMBOL_GPL(nvme_start_ctrl
);
4112 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
4114 nvme_fault_inject_fini(&ctrl
->fault_inject
);
4115 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
4116 cdev_device_del(&ctrl
->cdev
, ctrl
->device
);
4117 nvme_put_ctrl(ctrl
);
4119 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
4121 static void nvme_free_ctrl(struct device
*dev
)
4123 struct nvme_ctrl
*ctrl
=
4124 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
4125 struct nvme_subsystem
*subsys
= ctrl
->subsys
;
4127 if (subsys
&& ctrl
->instance
!= subsys
->instance
)
4128 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4130 kfree(ctrl
->effects
);
4131 nvme_mpath_uninit(ctrl
);
4132 __free_page(ctrl
->discard_page
);
4135 mutex_lock(&nvme_subsystems_lock
);
4136 list_del(&ctrl
->subsys_entry
);
4137 sysfs_remove_link(&subsys
->dev
.kobj
, dev_name(ctrl
->device
));
4138 mutex_unlock(&nvme_subsystems_lock
);
4141 ctrl
->ops
->free_ctrl(ctrl
);
4144 nvme_put_subsystem(subsys
);
4148 * Initialize a NVMe controller structures. This needs to be called during
4149 * earliest initialization so that we have the initialized structured around
4152 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
4153 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
4157 ctrl
->state
= NVME_CTRL_NEW
;
4158 spin_lock_init(&ctrl
->lock
);
4159 mutex_init(&ctrl
->scan_lock
);
4160 INIT_LIST_HEAD(&ctrl
->namespaces
);
4161 init_rwsem(&ctrl
->namespaces_rwsem
);
4164 ctrl
->quirks
= quirks
;
4165 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
4166 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
4167 INIT_WORK(&ctrl
->fw_act_work
, nvme_fw_act_work
);
4168 INIT_WORK(&ctrl
->delete_work
, nvme_delete_ctrl_work
);
4169 init_waitqueue_head(&ctrl
->state_wq
);
4171 INIT_DELAYED_WORK(&ctrl
->ka_work
, nvme_keep_alive_work
);
4172 memset(&ctrl
->ka_cmd
, 0, sizeof(ctrl
->ka_cmd
));
4173 ctrl
->ka_cmd
.common
.opcode
= nvme_admin_keep_alive
;
4175 BUILD_BUG_ON(NVME_DSM_MAX_RANGES
* sizeof(struct nvme_dsm_range
) >
4177 ctrl
->discard_page
= alloc_page(GFP_KERNEL
);
4178 if (!ctrl
->discard_page
) {
4183 ret
= ida_simple_get(&nvme_instance_ida
, 0, 0, GFP_KERNEL
);
4186 ctrl
->instance
= ret
;
4188 device_initialize(&ctrl
->ctrl_device
);
4189 ctrl
->device
= &ctrl
->ctrl_device
;
4190 ctrl
->device
->devt
= MKDEV(MAJOR(nvme_chr_devt
), ctrl
->instance
);
4191 ctrl
->device
->class = nvme_class
;
4192 ctrl
->device
->parent
= ctrl
->dev
;
4193 ctrl
->device
->groups
= nvme_dev_attr_groups
;
4194 ctrl
->device
->release
= nvme_free_ctrl
;
4195 dev_set_drvdata(ctrl
->device
, ctrl
);
4196 ret
= dev_set_name(ctrl
->device
, "nvme%d", ctrl
->instance
);
4198 goto out_release_instance
;
4200 nvme_get_ctrl(ctrl
);
4201 cdev_init(&ctrl
->cdev
, &nvme_dev_fops
);
4202 ctrl
->cdev
.owner
= ops
->module
;
4203 ret
= cdev_device_add(&ctrl
->cdev
, ctrl
->device
);
4208 * Initialize latency tolerance controls. The sysfs files won't
4209 * be visible to userspace unless the device actually supports APST.
4211 ctrl
->device
->power
.set_latency_tolerance
= nvme_set_latency_tolerance
;
4212 dev_pm_qos_update_user_latency_tolerance(ctrl
->device
,
4213 min(default_ps_max_latency_us
, (unsigned long)S32_MAX
));
4215 nvme_fault_inject_init(&ctrl
->fault_inject
, dev_name(ctrl
->device
));
4219 nvme_put_ctrl(ctrl
);
4220 kfree_const(ctrl
->device
->kobj
.name
);
4221 out_release_instance
:
4222 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4224 if (ctrl
->discard_page
)
4225 __free_page(ctrl
->discard_page
);
4228 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
4231 * nvme_kill_queues(): Ends all namespace queues
4232 * @ctrl: the dead controller that needs to end
4234 * Call this function when the driver determines it is unable to get the
4235 * controller in a state capable of servicing IO.
4237 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
4241 down_read(&ctrl
->namespaces_rwsem
);
4243 /* Forcibly unquiesce queues to avoid blocking dispatch */
4244 if (ctrl
->admin_q
&& !blk_queue_dying(ctrl
->admin_q
))
4245 blk_mq_unquiesce_queue(ctrl
->admin_q
);
4247 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4248 nvme_set_queue_dying(ns
);
4250 up_read(&ctrl
->namespaces_rwsem
);
4252 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
4254 void nvme_unfreeze(struct nvme_ctrl
*ctrl
)
4258 down_read(&ctrl
->namespaces_rwsem
);
4259 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4260 blk_mq_unfreeze_queue(ns
->queue
);
4261 up_read(&ctrl
->namespaces_rwsem
);
4263 EXPORT_SYMBOL_GPL(nvme_unfreeze
);
4265 void nvme_wait_freeze_timeout(struct nvme_ctrl
*ctrl
, long timeout
)
4269 down_read(&ctrl
->namespaces_rwsem
);
4270 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
4271 timeout
= blk_mq_freeze_queue_wait_timeout(ns
->queue
, timeout
);
4275 up_read(&ctrl
->namespaces_rwsem
);
4277 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout
);
4279 void nvme_wait_freeze(struct nvme_ctrl
*ctrl
)
4283 down_read(&ctrl
->namespaces_rwsem
);
4284 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4285 blk_mq_freeze_queue_wait(ns
->queue
);
4286 up_read(&ctrl
->namespaces_rwsem
);
4288 EXPORT_SYMBOL_GPL(nvme_wait_freeze
);
4290 void nvme_start_freeze(struct nvme_ctrl
*ctrl
)
4294 down_read(&ctrl
->namespaces_rwsem
);
4295 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4296 blk_freeze_queue_start(ns
->queue
);
4297 up_read(&ctrl
->namespaces_rwsem
);
4299 EXPORT_SYMBOL_GPL(nvme_start_freeze
);
4301 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
4305 down_read(&ctrl
->namespaces_rwsem
);
4306 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4307 blk_mq_quiesce_queue(ns
->queue
);
4308 up_read(&ctrl
->namespaces_rwsem
);
4310 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
4312 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
4316 down_read(&ctrl
->namespaces_rwsem
);
4317 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4318 blk_mq_unquiesce_queue(ns
->queue
);
4319 up_read(&ctrl
->namespaces_rwsem
);
4321 EXPORT_SYMBOL_GPL(nvme_start_queues
);
4324 void nvme_sync_queues(struct nvme_ctrl
*ctrl
)
4328 down_read(&ctrl
->namespaces_rwsem
);
4329 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4330 blk_sync_queue(ns
->queue
);
4331 up_read(&ctrl
->namespaces_rwsem
);
4334 blk_sync_queue(ctrl
->admin_q
);
4336 EXPORT_SYMBOL_GPL(nvme_sync_queues
);
4339 * Check we didn't inadvertently grow the command structure sizes:
4341 static inline void _nvme_check_size(void)
4343 BUILD_BUG_ON(sizeof(struct nvme_common_command
) != 64);
4344 BUILD_BUG_ON(sizeof(struct nvme_rw_command
) != 64);
4345 BUILD_BUG_ON(sizeof(struct nvme_identify
) != 64);
4346 BUILD_BUG_ON(sizeof(struct nvme_features
) != 64);
4347 BUILD_BUG_ON(sizeof(struct nvme_download_firmware
) != 64);
4348 BUILD_BUG_ON(sizeof(struct nvme_format_cmd
) != 64);
4349 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd
) != 64);
4350 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd
) != 64);
4351 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd
) != 64);
4352 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command
) != 64);
4353 BUILD_BUG_ON(sizeof(struct nvme_command
) != 64);
4354 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl
) != NVME_IDENTIFY_DATA_SIZE
);
4355 BUILD_BUG_ON(sizeof(struct nvme_id_ns
) != NVME_IDENTIFY_DATA_SIZE
);
4356 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type
) != 64);
4357 BUILD_BUG_ON(sizeof(struct nvme_smart_log
) != 512);
4358 BUILD_BUG_ON(sizeof(struct nvme_dbbuf
) != 64);
4359 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd
) != 64);
4363 static int __init
nvme_core_init(void)
4365 int result
= -ENOMEM
;
4369 nvme_wq
= alloc_workqueue("nvme-wq",
4370 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4374 nvme_reset_wq
= alloc_workqueue("nvme-reset-wq",
4375 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4379 nvme_delete_wq
= alloc_workqueue("nvme-delete-wq",
4380 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4381 if (!nvme_delete_wq
)
4382 goto destroy_reset_wq
;
4384 result
= alloc_chrdev_region(&nvme_chr_devt
, 0, NVME_MINORS
, "nvme");
4386 goto destroy_delete_wq
;
4388 nvme_class
= class_create(THIS_MODULE
, "nvme");
4389 if (IS_ERR(nvme_class
)) {
4390 result
= PTR_ERR(nvme_class
);
4391 goto unregister_chrdev
;
4393 nvme_class
->dev_uevent
= nvme_class_uevent
;
4395 nvme_subsys_class
= class_create(THIS_MODULE
, "nvme-subsystem");
4396 if (IS_ERR(nvme_subsys_class
)) {
4397 result
= PTR_ERR(nvme_subsys_class
);
4403 class_destroy(nvme_class
);
4405 unregister_chrdev_region(nvme_chr_devt
, NVME_MINORS
);
4407 destroy_workqueue(nvme_delete_wq
);
4409 destroy_workqueue(nvme_reset_wq
);
4411 destroy_workqueue(nvme_wq
);
4416 static void __exit
nvme_core_exit(void)
4418 class_destroy(nvme_subsys_class
);
4419 class_destroy(nvme_class
);
4420 unregister_chrdev_region(nvme_chr_devt
, NVME_MINORS
);
4421 destroy_workqueue(nvme_delete_wq
);
4422 destroy_workqueue(nvme_reset_wq
);
4423 destroy_workqueue(nvme_wq
);
4424 ida_destroy(&nvme_instance_ida
);
4427 MODULE_LICENSE("GPL");
4428 MODULE_VERSION("1.0");
4429 module_init(nvme_core_init
);
4430 module_exit(nvme_core_exit
);