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 void nvme_put_subsystem(struct nvme_subsystem
*subsys
);
93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
96 static void nvme_update_bdev_size(struct gendisk
*disk
)
98 struct block_device
*bdev
= bdget_disk(disk
, 0);
101 bd_set_nr_sectors(bdev
, get_capacity(disk
));
107 * Prepare a queue for teardown.
109 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
110 * the capacity to 0 after that to avoid blocking dispatchers that may be
111 * holding bd_butex. This will end buffered writers dirtying pages that can't
114 static void nvme_set_queue_dying(struct nvme_ns
*ns
)
116 if (test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
119 blk_set_queue_dying(ns
->queue
);
120 blk_mq_unquiesce_queue(ns
->queue
);
122 set_capacity(ns
->disk
, 0);
123 nvme_update_bdev_size(ns
->disk
);
126 static void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
129 * Only new queue scan work when admin and IO queues are both alive
131 if (ctrl
->state
== NVME_CTRL_LIVE
&& ctrl
->tagset
)
132 queue_work(nvme_wq
, &ctrl
->scan_work
);
136 * Use this function to proceed with scheduling reset_work for a controller
137 * that had previously been set to the resetting state. This is intended for
138 * code paths that can't be interrupted by other reset attempts. A hot removal
139 * may prevent this from succeeding.
141 int nvme_try_sched_reset(struct nvme_ctrl
*ctrl
)
143 if (ctrl
->state
!= NVME_CTRL_RESETTING
)
145 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
149 EXPORT_SYMBOL_GPL(nvme_try_sched_reset
);
151 int nvme_reset_ctrl(struct nvme_ctrl
*ctrl
)
153 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
155 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
159 EXPORT_SYMBOL_GPL(nvme_reset_ctrl
);
161 int nvme_reset_ctrl_sync(struct nvme_ctrl
*ctrl
)
165 ret
= nvme_reset_ctrl(ctrl
);
167 flush_work(&ctrl
->reset_work
);
168 if (ctrl
->state
!= NVME_CTRL_LIVE
)
174 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync
);
176 static void nvme_do_delete_ctrl(struct nvme_ctrl
*ctrl
)
178 dev_info(ctrl
->device
,
179 "Removing ctrl: NQN \"%s\"\n", ctrl
->opts
->subsysnqn
);
181 flush_work(&ctrl
->reset_work
);
182 nvme_stop_ctrl(ctrl
);
183 nvme_remove_namespaces(ctrl
);
184 ctrl
->ops
->delete_ctrl(ctrl
);
185 nvme_uninit_ctrl(ctrl
);
188 static void nvme_delete_ctrl_work(struct work_struct
*work
)
190 struct nvme_ctrl
*ctrl
=
191 container_of(work
, struct nvme_ctrl
, delete_work
);
193 nvme_do_delete_ctrl(ctrl
);
196 int nvme_delete_ctrl(struct nvme_ctrl
*ctrl
)
198 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
200 if (!queue_work(nvme_delete_wq
, &ctrl
->delete_work
))
204 EXPORT_SYMBOL_GPL(nvme_delete_ctrl
);
206 static void nvme_delete_ctrl_sync(struct nvme_ctrl
*ctrl
)
209 * Keep a reference until nvme_do_delete_ctrl() complete,
210 * since ->delete_ctrl can free the controller.
213 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
214 nvme_do_delete_ctrl(ctrl
);
218 static blk_status_t
nvme_error_status(u16 status
)
220 switch (status
& 0x7ff) {
221 case NVME_SC_SUCCESS
:
223 case NVME_SC_CAP_EXCEEDED
:
224 return BLK_STS_NOSPC
;
225 case NVME_SC_LBA_RANGE
:
226 case NVME_SC_CMD_INTERRUPTED
:
227 case NVME_SC_NS_NOT_READY
:
228 return BLK_STS_TARGET
;
229 case NVME_SC_BAD_ATTRIBUTES
:
230 case NVME_SC_ONCS_NOT_SUPPORTED
:
231 case NVME_SC_INVALID_OPCODE
:
232 case NVME_SC_INVALID_FIELD
:
233 case NVME_SC_INVALID_NS
:
234 return BLK_STS_NOTSUPP
;
235 case NVME_SC_WRITE_FAULT
:
236 case NVME_SC_READ_ERROR
:
237 case NVME_SC_UNWRITTEN_BLOCK
:
238 case NVME_SC_ACCESS_DENIED
:
239 case NVME_SC_READ_ONLY
:
240 case NVME_SC_COMPARE_FAILED
:
241 return BLK_STS_MEDIUM
;
242 case NVME_SC_GUARD_CHECK
:
243 case NVME_SC_APPTAG_CHECK
:
244 case NVME_SC_REFTAG_CHECK
:
245 case NVME_SC_INVALID_PI
:
246 return BLK_STS_PROTECTION
;
247 case NVME_SC_RESERVATION_CONFLICT
:
248 return BLK_STS_NEXUS
;
249 case NVME_SC_HOST_PATH_ERROR
:
250 return BLK_STS_TRANSPORT
;
252 return BLK_STS_IOERR
;
256 static void nvme_retry_req(struct request
*req
)
258 struct nvme_ns
*ns
= req
->q
->queuedata
;
259 unsigned long delay
= 0;
262 /* The mask and shift result must be <= 3 */
263 crd
= (nvme_req(req
)->status
& NVME_SC_CRD
) >> 11;
265 delay
= ns
->ctrl
->crdt
[crd
- 1] * 100;
267 nvme_req(req
)->retries
++;
268 blk_mq_requeue_request(req
, false);
269 blk_mq_delay_kick_requeue_list(req
->q
, delay
);
272 enum nvme_disposition
{
278 static inline enum nvme_disposition
nvme_decide_disposition(struct request
*req
)
280 if (likely(nvme_req(req
)->status
== 0))
283 if (blk_noretry_request(req
) ||
284 (nvme_req(req
)->status
& NVME_SC_DNR
) ||
285 nvme_req(req
)->retries
>= nvme_max_retries
)
288 if (req
->cmd_flags
& REQ_NVME_MPATH
) {
289 if (nvme_is_path_error(nvme_req(req
)->status
) ||
290 blk_queue_dying(req
->q
))
293 if (blk_queue_dying(req
->q
))
300 static inline void nvme_end_req(struct request
*req
)
302 blk_status_t status
= nvme_error_status(nvme_req(req
)->status
);
304 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED
) &&
305 req_op(req
) == REQ_OP_ZONE_APPEND
)
306 req
->__sector
= nvme_lba_to_sect(req
->q
->queuedata
,
307 le64_to_cpu(nvme_req(req
)->result
.u64
));
309 nvme_trace_bio_complete(req
, status
);
310 blk_mq_end_request(req
, status
);
313 void nvme_complete_rq(struct request
*req
)
315 trace_nvme_complete_rq(req
);
316 nvme_cleanup_cmd(req
);
318 if (nvme_req(req
)->ctrl
->kas
)
319 nvme_req(req
)->ctrl
->comp_seen
= true;
321 switch (nvme_decide_disposition(req
)) {
329 nvme_failover_req(req
);
333 EXPORT_SYMBOL_GPL(nvme_complete_rq
);
335 bool nvme_cancel_request(struct request
*req
, void *data
, bool reserved
)
337 dev_dbg_ratelimited(((struct nvme_ctrl
*) data
)->device
,
338 "Cancelling I/O %d", req
->tag
);
340 /* don't abort one completed request */
341 if (blk_mq_request_completed(req
))
344 nvme_req(req
)->status
= NVME_SC_HOST_ABORTED_CMD
;
345 blk_mq_complete_request(req
);
348 EXPORT_SYMBOL_GPL(nvme_cancel_request
);
350 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
351 enum nvme_ctrl_state new_state
)
353 enum nvme_ctrl_state old_state
;
355 bool changed
= false;
357 spin_lock_irqsave(&ctrl
->lock
, flags
);
359 old_state
= ctrl
->state
;
364 case NVME_CTRL_RESETTING
:
365 case NVME_CTRL_CONNECTING
:
372 case NVME_CTRL_RESETTING
:
382 case NVME_CTRL_CONNECTING
:
385 case NVME_CTRL_RESETTING
:
392 case NVME_CTRL_DELETING
:
395 case NVME_CTRL_RESETTING
:
396 case NVME_CTRL_CONNECTING
:
403 case NVME_CTRL_DELETING_NOIO
:
405 case NVME_CTRL_DELETING
:
415 case NVME_CTRL_DELETING
:
427 ctrl
->state
= new_state
;
428 wake_up_all(&ctrl
->state_wq
);
431 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
432 if (changed
&& ctrl
->state
== NVME_CTRL_LIVE
)
433 nvme_kick_requeue_lists(ctrl
);
436 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
439 * Returns true for sink states that can't ever transition back to live.
441 static bool nvme_state_terminal(struct nvme_ctrl
*ctrl
)
443 switch (ctrl
->state
) {
446 case NVME_CTRL_RESETTING
:
447 case NVME_CTRL_CONNECTING
:
449 case NVME_CTRL_DELETING
:
450 case NVME_CTRL_DELETING_NOIO
:
454 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl
->state
);
460 * Waits for the controller state to be resetting, or returns false if it is
461 * not possible to ever transition to that state.
463 bool nvme_wait_reset(struct nvme_ctrl
*ctrl
)
465 wait_event(ctrl
->state_wq
,
466 nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
) ||
467 nvme_state_terminal(ctrl
));
468 return ctrl
->state
== NVME_CTRL_RESETTING
;
470 EXPORT_SYMBOL_GPL(nvme_wait_reset
);
472 static void nvme_free_ns_head(struct kref
*ref
)
474 struct nvme_ns_head
*head
=
475 container_of(ref
, struct nvme_ns_head
, ref
);
477 nvme_mpath_remove_disk(head
);
478 ida_simple_remove(&head
->subsys
->ns_ida
, head
->instance
);
479 cleanup_srcu_struct(&head
->srcu
);
480 nvme_put_subsystem(head
->subsys
);
484 static void nvme_put_ns_head(struct nvme_ns_head
*head
)
486 kref_put(&head
->ref
, nvme_free_ns_head
);
489 static void nvme_free_ns(struct kref
*kref
)
491 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
494 nvme_nvm_unregister(ns
);
497 nvme_put_ns_head(ns
->head
);
498 nvme_put_ctrl(ns
->ctrl
);
502 void nvme_put_ns(struct nvme_ns
*ns
)
504 kref_put(&ns
->kref
, nvme_free_ns
);
506 EXPORT_SYMBOL_NS_GPL(nvme_put_ns
, NVME_TARGET_PASSTHRU
);
508 static inline void nvme_clear_nvme_request(struct request
*req
)
510 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
511 nvme_req(req
)->retries
= 0;
512 nvme_req(req
)->flags
= 0;
513 req
->rq_flags
|= RQF_DONTPREP
;
517 struct request
*nvme_alloc_request(struct request_queue
*q
,
518 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
, int qid
)
520 unsigned op
= nvme_is_write(cmd
) ? REQ_OP_DRV_OUT
: REQ_OP_DRV_IN
;
523 if (qid
== NVME_QID_ANY
) {
524 req
= blk_mq_alloc_request(q
, op
, flags
);
526 req
= blk_mq_alloc_request_hctx(q
, op
, flags
,
532 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
533 nvme_clear_nvme_request(req
);
534 nvme_req(req
)->cmd
= cmd
;
538 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
540 static int nvme_toggle_streams(struct nvme_ctrl
*ctrl
, bool enable
)
542 struct nvme_command c
;
544 memset(&c
, 0, sizeof(c
));
546 c
.directive
.opcode
= nvme_admin_directive_send
;
547 c
.directive
.nsid
= cpu_to_le32(NVME_NSID_ALL
);
548 c
.directive
.doper
= NVME_DIR_SND_ID_OP_ENABLE
;
549 c
.directive
.dtype
= NVME_DIR_IDENTIFY
;
550 c
.directive
.tdtype
= NVME_DIR_STREAMS
;
551 c
.directive
.endir
= enable
? NVME_DIR_ENDIR
: 0;
553 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, NULL
, 0);
556 static int nvme_disable_streams(struct nvme_ctrl
*ctrl
)
558 return nvme_toggle_streams(ctrl
, false);
561 static int nvme_enable_streams(struct nvme_ctrl
*ctrl
)
563 return nvme_toggle_streams(ctrl
, true);
566 static int nvme_get_stream_params(struct nvme_ctrl
*ctrl
,
567 struct streams_directive_params
*s
, u32 nsid
)
569 struct nvme_command c
;
571 memset(&c
, 0, sizeof(c
));
572 memset(s
, 0, sizeof(*s
));
574 c
.directive
.opcode
= nvme_admin_directive_recv
;
575 c
.directive
.nsid
= cpu_to_le32(nsid
);
576 c
.directive
.numd
= cpu_to_le32(nvme_bytes_to_numd(sizeof(*s
)));
577 c
.directive
.doper
= NVME_DIR_RCV_ST_OP_PARAM
;
578 c
.directive
.dtype
= NVME_DIR_STREAMS
;
580 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, s
, sizeof(*s
));
583 static int nvme_configure_directives(struct nvme_ctrl
*ctrl
)
585 struct streams_directive_params s
;
588 if (!(ctrl
->oacs
& NVME_CTRL_OACS_DIRECTIVES
))
593 ret
= nvme_enable_streams(ctrl
);
597 ret
= nvme_get_stream_params(ctrl
, &s
, NVME_NSID_ALL
);
599 goto out_disable_stream
;
601 ctrl
->nssa
= le16_to_cpu(s
.nssa
);
602 if (ctrl
->nssa
< BLK_MAX_WRITE_HINTS
- 1) {
603 dev_info(ctrl
->device
, "too few streams (%u) available\n",
605 goto out_disable_stream
;
608 ctrl
->nr_streams
= min_t(u16
, ctrl
->nssa
, BLK_MAX_WRITE_HINTS
- 1);
609 dev_info(ctrl
->device
, "Using %u streams\n", ctrl
->nr_streams
);
613 nvme_disable_streams(ctrl
);
618 * Check if 'req' has a write hint associated with it. If it does, assign
619 * a valid namespace stream to the write.
621 static void nvme_assign_write_stream(struct nvme_ctrl
*ctrl
,
622 struct request
*req
, u16
*control
,
625 enum rw_hint streamid
= req
->write_hint
;
627 if (streamid
== WRITE_LIFE_NOT_SET
|| streamid
== WRITE_LIFE_NONE
)
631 if (WARN_ON_ONCE(streamid
> ctrl
->nr_streams
))
634 *control
|= NVME_RW_DTYPE_STREAMS
;
635 *dsmgmt
|= streamid
<< 16;
638 if (streamid
< ARRAY_SIZE(req
->q
->write_hints
))
639 req
->q
->write_hints
[streamid
] += blk_rq_bytes(req
) >> 9;
642 static void nvme_setup_passthrough(struct request
*req
,
643 struct nvme_command
*cmd
)
645 memcpy(cmd
, nvme_req(req
)->cmd
, sizeof(*cmd
));
646 /* passthru commands should let the driver set the SGL flags */
647 cmd
->common
.flags
&= ~NVME_CMD_SGL_ALL
;
650 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
651 struct nvme_command
*cmnd
)
653 cmnd
->common
.opcode
= nvme_cmd_flush
;
654 cmnd
->common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
657 static blk_status_t
nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
658 struct nvme_command
*cmnd
)
660 unsigned short segments
= blk_rq_nr_discard_segments(req
), n
= 0;
661 struct nvme_dsm_range
*range
;
665 * Some devices do not consider the DSM 'Number of Ranges' field when
666 * determining how much data to DMA. Always allocate memory for maximum
667 * number of segments to prevent device reading beyond end of buffer.
669 static const size_t alloc_size
= sizeof(*range
) * NVME_DSM_MAX_RANGES
;
671 range
= kzalloc(alloc_size
, GFP_ATOMIC
| __GFP_NOWARN
);
674 * If we fail allocation our range, fallback to the controller
675 * discard page. If that's also busy, it's safe to return
676 * busy, as we know we can make progress once that's freed.
678 if (test_and_set_bit_lock(0, &ns
->ctrl
->discard_page_busy
))
679 return BLK_STS_RESOURCE
;
681 range
= page_address(ns
->ctrl
->discard_page
);
684 __rq_for_each_bio(bio
, req
) {
685 u64 slba
= nvme_sect_to_lba(ns
, bio
->bi_iter
.bi_sector
);
686 u32 nlb
= bio
->bi_iter
.bi_size
>> ns
->lba_shift
;
689 range
[n
].cattr
= cpu_to_le32(0);
690 range
[n
].nlb
= cpu_to_le32(nlb
);
691 range
[n
].slba
= cpu_to_le64(slba
);
696 if (WARN_ON_ONCE(n
!= segments
)) {
697 if (virt_to_page(range
) == ns
->ctrl
->discard_page
)
698 clear_bit_unlock(0, &ns
->ctrl
->discard_page_busy
);
701 return BLK_STS_IOERR
;
704 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
705 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
706 cmnd
->dsm
.nr
= cpu_to_le32(segments
- 1);
707 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
709 req
->special_vec
.bv_page
= virt_to_page(range
);
710 req
->special_vec
.bv_offset
= offset_in_page(range
);
711 req
->special_vec
.bv_len
= alloc_size
;
712 req
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
717 static inline blk_status_t
nvme_setup_write_zeroes(struct nvme_ns
*ns
,
718 struct request
*req
, struct nvme_command
*cmnd
)
720 if (ns
->ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
721 return nvme_setup_discard(ns
, req
, cmnd
);
723 cmnd
->write_zeroes
.opcode
= nvme_cmd_write_zeroes
;
724 cmnd
->write_zeroes
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
725 cmnd
->write_zeroes
.slba
=
726 cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
727 cmnd
->write_zeroes
.length
=
728 cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
729 cmnd
->write_zeroes
.control
= 0;
733 static inline blk_status_t
nvme_setup_rw(struct nvme_ns
*ns
,
734 struct request
*req
, struct nvme_command
*cmnd
,
737 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
741 if (req
->cmd_flags
& REQ_FUA
)
742 control
|= NVME_RW_FUA
;
743 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
744 control
|= NVME_RW_LR
;
746 if (req
->cmd_flags
& REQ_RAHEAD
)
747 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
749 cmnd
->rw
.opcode
= op
;
750 cmnd
->rw
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
751 cmnd
->rw
.slba
= cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
752 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
754 if (req_op(req
) == REQ_OP_WRITE
&& ctrl
->nr_streams
)
755 nvme_assign_write_stream(ctrl
, req
, &control
, &dsmgmt
);
759 * If formated with metadata, the block layer always provides a
760 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
761 * we enable the PRACT bit for protection information or set the
762 * namespace capacity to zero to prevent any I/O.
764 if (!blk_integrity_rq(req
)) {
765 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns
)))
766 return BLK_STS_NOTSUPP
;
767 control
|= NVME_RW_PRINFO_PRACT
;
770 switch (ns
->pi_type
) {
771 case NVME_NS_DPS_PI_TYPE3
:
772 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
774 case NVME_NS_DPS_PI_TYPE1
:
775 case NVME_NS_DPS_PI_TYPE2
:
776 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
777 NVME_RW_PRINFO_PRCHK_REF
;
778 if (op
== nvme_cmd_zone_append
)
779 control
|= NVME_RW_APPEND_PIREMAP
;
780 cmnd
->rw
.reftag
= cpu_to_le32(t10_pi_ref_tag(req
));
785 cmnd
->rw
.control
= cpu_to_le16(control
);
786 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
790 void nvme_cleanup_cmd(struct request
*req
)
792 if (req
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
793 struct nvme_ns
*ns
= req
->rq_disk
->private_data
;
794 struct page
*page
= req
->special_vec
.bv_page
;
796 if (page
== ns
->ctrl
->discard_page
)
797 clear_bit_unlock(0, &ns
->ctrl
->discard_page_busy
);
799 kfree(page_address(page
) + req
->special_vec
.bv_offset
);
802 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd
);
804 blk_status_t
nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
805 struct nvme_command
*cmd
)
807 blk_status_t ret
= BLK_STS_OK
;
809 nvme_clear_nvme_request(req
);
811 memset(cmd
, 0, sizeof(*cmd
));
812 switch (req_op(req
)) {
815 nvme_setup_passthrough(req
, cmd
);
818 nvme_setup_flush(ns
, cmd
);
820 case REQ_OP_ZONE_RESET_ALL
:
821 case REQ_OP_ZONE_RESET
:
822 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_RESET
);
824 case REQ_OP_ZONE_OPEN
:
825 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_OPEN
);
827 case REQ_OP_ZONE_CLOSE
:
828 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_CLOSE
);
830 case REQ_OP_ZONE_FINISH
:
831 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_FINISH
);
833 case REQ_OP_WRITE_ZEROES
:
834 ret
= nvme_setup_write_zeroes(ns
, req
, cmd
);
837 ret
= nvme_setup_discard(ns
, req
, cmd
);
840 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_read
);
843 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_write
);
845 case REQ_OP_ZONE_APPEND
:
846 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_zone_append
);
850 return BLK_STS_IOERR
;
853 cmd
->common
.command_id
= req
->tag
;
854 trace_nvme_setup_cmd(req
, cmd
);
857 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
859 static void nvme_end_sync_rq(struct request
*rq
, blk_status_t error
)
861 struct completion
*waiting
= rq
->end_io_data
;
863 rq
->end_io_data
= NULL
;
867 static void nvme_execute_rq_polled(struct request_queue
*q
,
868 struct gendisk
*bd_disk
, struct request
*rq
, int at_head
)
870 DECLARE_COMPLETION_ONSTACK(wait
);
872 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
));
874 rq
->cmd_flags
|= REQ_HIPRI
;
875 rq
->end_io_data
= &wait
;
876 blk_execute_rq_nowait(q
, bd_disk
, rq
, at_head
, nvme_end_sync_rq
);
878 while (!completion_done(&wait
)) {
879 blk_poll(q
, request_to_qc_t(rq
->mq_hctx
, rq
), true);
885 * Returns 0 on success. If the result is negative, it's a Linux error code;
886 * if the result is positive, it's an NVM Express status code
888 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
889 union nvme_result
*result
, void *buffer
, unsigned bufflen
,
890 unsigned timeout
, int qid
, int at_head
,
891 blk_mq_req_flags_t flags
, bool poll
)
896 req
= nvme_alloc_request(q
, cmd
, flags
, qid
);
900 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
902 if (buffer
&& bufflen
) {
903 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
909 nvme_execute_rq_polled(req
->q
, NULL
, req
, at_head
);
911 blk_execute_rq(req
->q
, NULL
, req
, at_head
);
913 *result
= nvme_req(req
)->result
;
914 if (nvme_req(req
)->flags
& NVME_REQ_CANCELLED
)
917 ret
= nvme_req(req
)->status
;
919 blk_mq_free_request(req
);
922 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd
);
924 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
925 void *buffer
, unsigned bufflen
)
927 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0,
928 NVME_QID_ANY
, 0, 0, false);
930 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
932 static void *nvme_add_user_metadata(struct bio
*bio
, void __user
*ubuf
,
933 unsigned len
, u32 seed
, bool write
)
935 struct bio_integrity_payload
*bip
;
939 buf
= kmalloc(len
, GFP_KERNEL
);
944 if (write
&& copy_from_user(buf
, ubuf
, len
))
947 bip
= bio_integrity_alloc(bio
, GFP_KERNEL
, 1);
953 bip
->bip_iter
.bi_size
= len
;
954 bip
->bip_iter
.bi_sector
= seed
;
955 ret
= bio_integrity_add_page(bio
, virt_to_page(buf
), len
,
956 offset_in_page(buf
));
966 static u32
nvme_known_admin_effects(u8 opcode
)
969 case nvme_admin_format_nvm
:
970 return NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_NCC
|
971 NVME_CMD_EFFECTS_CSE_MASK
;
972 case nvme_admin_sanitize_nvm
:
973 return NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_CSE_MASK
;
980 u32
nvme_command_effects(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
, u8 opcode
)
985 if (ns
->head
->effects
)
986 effects
= le32_to_cpu(ns
->head
->effects
->iocs
[opcode
]);
987 if (effects
& ~(NVME_CMD_EFFECTS_CSUPP
| NVME_CMD_EFFECTS_LBCC
))
988 dev_warn(ctrl
->device
,
989 "IO command:%02x has unhandled effects:%08x\n",
995 effects
= le32_to_cpu(ctrl
->effects
->acs
[opcode
]);
996 effects
|= nvme_known_admin_effects(opcode
);
1000 EXPORT_SYMBOL_NS_GPL(nvme_command_effects
, NVME_TARGET_PASSTHRU
);
1002 static u32
nvme_passthru_start(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1005 u32 effects
= nvme_command_effects(ctrl
, ns
, opcode
);
1008 * For simplicity, IO to all namespaces is quiesced even if the command
1009 * effects say only one namespace is affected.
1011 if (effects
& NVME_CMD_EFFECTS_CSE_MASK
) {
1012 mutex_lock(&ctrl
->scan_lock
);
1013 mutex_lock(&ctrl
->subsys
->lock
);
1014 nvme_mpath_start_freeze(ctrl
->subsys
);
1015 nvme_mpath_wait_freeze(ctrl
->subsys
);
1016 nvme_start_freeze(ctrl
);
1017 nvme_wait_freeze(ctrl
);
1022 static void nvme_passthru_end(struct nvme_ctrl
*ctrl
, u32 effects
)
1024 if (effects
& NVME_CMD_EFFECTS_CSE_MASK
) {
1025 nvme_unfreeze(ctrl
);
1026 nvme_mpath_unfreeze(ctrl
->subsys
);
1027 mutex_unlock(&ctrl
->subsys
->lock
);
1028 nvme_remove_invalid_namespaces(ctrl
, NVME_NSID_ALL
);
1029 mutex_unlock(&ctrl
->scan_lock
);
1031 if (effects
& NVME_CMD_EFFECTS_CCC
)
1032 nvme_init_identify(ctrl
);
1033 if (effects
& (NVME_CMD_EFFECTS_NIC
| NVME_CMD_EFFECTS_NCC
)) {
1034 nvme_queue_scan(ctrl
);
1035 flush_work(&ctrl
->scan_work
);
1039 void nvme_execute_passthru_rq(struct request
*rq
)
1041 struct nvme_command
*cmd
= nvme_req(rq
)->cmd
;
1042 struct nvme_ctrl
*ctrl
= nvme_req(rq
)->ctrl
;
1043 struct nvme_ns
*ns
= rq
->q
->queuedata
;
1044 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
1047 effects
= nvme_passthru_start(ctrl
, ns
, cmd
->common
.opcode
);
1048 blk_execute_rq(rq
->q
, disk
, rq
, 0);
1049 nvme_passthru_end(ctrl
, effects
);
1051 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq
, NVME_TARGET_PASSTHRU
);
1053 static int nvme_submit_user_cmd(struct request_queue
*q
,
1054 struct nvme_command
*cmd
, void __user
*ubuffer
,
1055 unsigned bufflen
, void __user
*meta_buffer
, unsigned meta_len
,
1056 u32 meta_seed
, u64
*result
, unsigned timeout
)
1058 bool write
= nvme_is_write(cmd
);
1059 struct nvme_ns
*ns
= q
->queuedata
;
1060 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
1061 struct request
*req
;
1062 struct bio
*bio
= NULL
;
1066 req
= nvme_alloc_request(q
, cmd
, 0, NVME_QID_ANY
);
1068 return PTR_ERR(req
);
1070 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
1071 nvme_req(req
)->flags
|= NVME_REQ_USERCMD
;
1073 if (ubuffer
&& bufflen
) {
1074 ret
= blk_rq_map_user(q
, req
, NULL
, ubuffer
, bufflen
,
1079 bio
->bi_disk
= disk
;
1080 if (disk
&& meta_buffer
&& meta_len
) {
1081 meta
= nvme_add_user_metadata(bio
, meta_buffer
, meta_len
,
1084 ret
= PTR_ERR(meta
);
1087 req
->cmd_flags
|= REQ_INTEGRITY
;
1091 nvme_execute_passthru_rq(req
);
1092 if (nvme_req(req
)->flags
& NVME_REQ_CANCELLED
)
1095 ret
= nvme_req(req
)->status
;
1097 *result
= le64_to_cpu(nvme_req(req
)->result
.u64
);
1098 if (meta
&& !ret
&& !write
) {
1099 if (copy_to_user(meta_buffer
, meta
, meta_len
))
1105 blk_rq_unmap_user(bio
);
1107 blk_mq_free_request(req
);
1111 static void nvme_keep_alive_end_io(struct request
*rq
, blk_status_t status
)
1113 struct nvme_ctrl
*ctrl
= rq
->end_io_data
;
1114 unsigned long flags
;
1115 bool startka
= false;
1117 blk_mq_free_request(rq
);
1120 dev_err(ctrl
->device
,
1121 "failed nvme_keep_alive_end_io error=%d\n",
1126 ctrl
->comp_seen
= false;
1127 spin_lock_irqsave(&ctrl
->lock
, flags
);
1128 if (ctrl
->state
== NVME_CTRL_LIVE
||
1129 ctrl
->state
== NVME_CTRL_CONNECTING
)
1131 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
1133 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
1136 static int nvme_keep_alive(struct nvme_ctrl
*ctrl
)
1140 rq
= nvme_alloc_request(ctrl
->admin_q
, &ctrl
->ka_cmd
, BLK_MQ_REQ_RESERVED
,
1145 rq
->timeout
= ctrl
->kato
* HZ
;
1146 rq
->end_io_data
= ctrl
;
1148 blk_execute_rq_nowait(rq
->q
, NULL
, rq
, 0, nvme_keep_alive_end_io
);
1153 static void nvme_keep_alive_work(struct work_struct
*work
)
1155 struct nvme_ctrl
*ctrl
= container_of(to_delayed_work(work
),
1156 struct nvme_ctrl
, ka_work
);
1157 bool comp_seen
= ctrl
->comp_seen
;
1159 if ((ctrl
->ctratt
& NVME_CTRL_ATTR_TBKAS
) && comp_seen
) {
1160 dev_dbg(ctrl
->device
,
1161 "reschedule traffic based keep-alive timer\n");
1162 ctrl
->comp_seen
= false;
1163 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
1167 if (nvme_keep_alive(ctrl
)) {
1168 /* allocation failure, reset the controller */
1169 dev_err(ctrl
->device
, "keep-alive failed\n");
1170 nvme_reset_ctrl(ctrl
);
1175 static void nvme_start_keep_alive(struct nvme_ctrl
*ctrl
)
1177 if (unlikely(ctrl
->kato
== 0))
1180 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
);
1183 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
)
1185 if (unlikely(ctrl
->kato
== 0))
1188 cancel_delayed_work_sync(&ctrl
->ka_work
);
1190 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive
);
1193 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1194 * flag, thus sending any new CNS opcodes has a big chance of not working.
1195 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1196 * (but not for any later version).
1198 static bool nvme_ctrl_limited_cns(struct nvme_ctrl
*ctrl
)
1200 if (ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)
1201 return ctrl
->vs
< NVME_VS(1, 2, 0);
1202 return ctrl
->vs
< NVME_VS(1, 1, 0);
1205 static int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
1207 struct nvme_command c
= { };
1210 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1211 c
.identify
.opcode
= nvme_admin_identify
;
1212 c
.identify
.cns
= NVME_ID_CNS_CTRL
;
1214 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
1218 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
1219 sizeof(struct nvme_id_ctrl
));
1225 static bool nvme_multi_css(struct nvme_ctrl
*ctrl
)
1227 return (ctrl
->ctrl_config
& NVME_CC_CSS_MASK
) == NVME_CC_CSS_CSI
;
1230 static int nvme_process_ns_desc(struct nvme_ctrl
*ctrl
, struct nvme_ns_ids
*ids
,
1231 struct nvme_ns_id_desc
*cur
, bool *csi_seen
)
1233 const char *warn_str
= "ctrl returned bogus length:";
1236 switch (cur
->nidt
) {
1237 case NVME_NIDT_EUI64
:
1238 if (cur
->nidl
!= NVME_NIDT_EUI64_LEN
) {
1239 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_EUI64\n",
1240 warn_str
, cur
->nidl
);
1243 memcpy(ids
->eui64
, data
+ sizeof(*cur
), NVME_NIDT_EUI64_LEN
);
1244 return NVME_NIDT_EUI64_LEN
;
1245 case NVME_NIDT_NGUID
:
1246 if (cur
->nidl
!= NVME_NIDT_NGUID_LEN
) {
1247 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_NGUID\n",
1248 warn_str
, cur
->nidl
);
1251 memcpy(ids
->nguid
, data
+ sizeof(*cur
), NVME_NIDT_NGUID_LEN
);
1252 return NVME_NIDT_NGUID_LEN
;
1253 case NVME_NIDT_UUID
:
1254 if (cur
->nidl
!= NVME_NIDT_UUID_LEN
) {
1255 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_UUID\n",
1256 warn_str
, cur
->nidl
);
1259 uuid_copy(&ids
->uuid
, data
+ sizeof(*cur
));
1260 return NVME_NIDT_UUID_LEN
;
1262 if (cur
->nidl
!= NVME_NIDT_CSI_LEN
) {
1263 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_CSI\n",
1264 warn_str
, cur
->nidl
);
1267 memcpy(&ids
->csi
, data
+ sizeof(*cur
), NVME_NIDT_CSI_LEN
);
1269 return NVME_NIDT_CSI_LEN
;
1271 /* Skip unknown types */
1276 static int nvme_identify_ns_descs(struct nvme_ctrl
*ctrl
, unsigned nsid
,
1277 struct nvme_ns_ids
*ids
)
1279 struct nvme_command c
= { };
1280 bool csi_seen
= false;
1281 int status
, pos
, len
;
1284 if (ctrl
->vs
< NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl
))
1286 if (ctrl
->quirks
& NVME_QUIRK_NO_NS_DESC_LIST
)
1289 c
.identify
.opcode
= nvme_admin_identify
;
1290 c
.identify
.nsid
= cpu_to_le32(nsid
);
1291 c
.identify
.cns
= NVME_ID_CNS_NS_DESC_LIST
;
1293 data
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
1297 status
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, data
,
1298 NVME_IDENTIFY_DATA_SIZE
);
1300 dev_warn(ctrl
->device
,
1301 "Identify Descriptors failed (%d)\n", status
);
1305 for (pos
= 0; pos
< NVME_IDENTIFY_DATA_SIZE
; pos
+= len
) {
1306 struct nvme_ns_id_desc
*cur
= data
+ pos
;
1311 len
= nvme_process_ns_desc(ctrl
, ids
, cur
, &csi_seen
);
1315 len
+= sizeof(*cur
);
1318 if (nvme_multi_css(ctrl
) && !csi_seen
) {
1319 dev_warn(ctrl
->device
, "Command set not reported for nsid:%d\n",
1329 static int nvme_identify_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
,
1330 struct nvme_ns_ids
*ids
, struct nvme_id_ns
**id
)
1332 struct nvme_command c
= { };
1335 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1336 c
.identify
.opcode
= nvme_admin_identify
;
1337 c
.identify
.nsid
= cpu_to_le32(nsid
);
1338 c
.identify
.cns
= NVME_ID_CNS_NS
;
1340 *id
= kmalloc(sizeof(**id
), GFP_KERNEL
);
1344 error
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, *id
, sizeof(**id
));
1346 dev_warn(ctrl
->device
, "Identify namespace failed (%d)\n", error
);
1351 if ((*id
)->ncap
== 0) /* namespace not allocated or attached */
1354 if (ctrl
->vs
>= NVME_VS(1, 1, 0) &&
1355 !memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
1356 memcpy(ids
->eui64
, (*id
)->eui64
, sizeof(ids
->eui64
));
1357 if (ctrl
->vs
>= NVME_VS(1, 2, 0) &&
1358 !memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
1359 memcpy(ids
->nguid
, (*id
)->nguid
, sizeof(ids
->nguid
));
1368 static int nvme_features(struct nvme_ctrl
*dev
, u8 op
, unsigned int fid
,
1369 unsigned int dword11
, void *buffer
, size_t buflen
, u32
*result
)
1371 union nvme_result res
= { 0 };
1372 struct nvme_command c
;
1375 memset(&c
, 0, sizeof(c
));
1376 c
.features
.opcode
= op
;
1377 c
.features
.fid
= cpu_to_le32(fid
);
1378 c
.features
.dword11
= cpu_to_le32(dword11
);
1380 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &res
,
1381 buffer
, buflen
, 0, NVME_QID_ANY
, 0, 0, false);
1382 if (ret
>= 0 && result
)
1383 *result
= le32_to_cpu(res
.u32
);
1387 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1388 unsigned int dword11
, void *buffer
, size_t buflen
,
1391 return nvme_features(dev
, nvme_admin_set_features
, fid
, dword11
, buffer
,
1394 EXPORT_SYMBOL_GPL(nvme_set_features
);
1396 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1397 unsigned int dword11
, void *buffer
, size_t buflen
,
1400 return nvme_features(dev
, nvme_admin_get_features
, fid
, dword11
, buffer
,
1403 EXPORT_SYMBOL_GPL(nvme_get_features
);
1405 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
1407 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
1409 int status
, nr_io_queues
;
1411 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, NULL
, 0,
1417 * Degraded controllers might return an error when setting the queue
1418 * count. We still want to be able to bring them online and offer
1419 * access to the admin queue, as that might be only way to fix them up.
1422 dev_err(ctrl
->device
, "Could not set queue count (%d)\n", status
);
1425 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
1426 *count
= min(*count
, nr_io_queues
);
1431 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
1433 #define NVME_AEN_SUPPORTED \
1434 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1435 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1437 static void nvme_enable_aen(struct nvme_ctrl
*ctrl
)
1439 u32 result
, supported_aens
= ctrl
->oaes
& NVME_AEN_SUPPORTED
;
1442 if (!supported_aens
)
1445 status
= nvme_set_features(ctrl
, NVME_FEAT_ASYNC_EVENT
, supported_aens
,
1448 dev_warn(ctrl
->device
, "Failed to configure AEN (cfg %x)\n",
1451 queue_work(nvme_wq
, &ctrl
->async_event_work
);
1455 * Convert integer values from ioctl structures to user pointers, silently
1456 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1459 static void __user
*nvme_to_user_ptr(uintptr_t ptrval
)
1461 if (in_compat_syscall())
1462 ptrval
= (compat_uptr_t
)ptrval
;
1463 return (void __user
*)ptrval
;
1466 static int nvme_submit_io(struct nvme_ns
*ns
, struct nvme_user_io __user
*uio
)
1468 struct nvme_user_io io
;
1469 struct nvme_command c
;
1470 unsigned length
, meta_len
;
1471 void __user
*metadata
;
1473 if (copy_from_user(&io
, uio
, sizeof(io
)))
1478 switch (io
.opcode
) {
1479 case nvme_cmd_write
:
1481 case nvme_cmd_compare
:
1487 length
= (io
.nblocks
+ 1) << ns
->lba_shift
;
1488 meta_len
= (io
.nblocks
+ 1) * ns
->ms
;
1489 metadata
= nvme_to_user_ptr(io
.metadata
);
1491 if (ns
->features
& NVME_NS_EXT_LBAS
) {
1494 } else if (meta_len
) {
1495 if ((io
.metadata
& 3) || !io
.metadata
)
1499 memset(&c
, 0, sizeof(c
));
1500 c
.rw
.opcode
= io
.opcode
;
1501 c
.rw
.flags
= io
.flags
;
1502 c
.rw
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
1503 c
.rw
.slba
= cpu_to_le64(io
.slba
);
1504 c
.rw
.length
= cpu_to_le16(io
.nblocks
);
1505 c
.rw
.control
= cpu_to_le16(io
.control
);
1506 c
.rw
.dsmgmt
= cpu_to_le32(io
.dsmgmt
);
1507 c
.rw
.reftag
= cpu_to_le32(io
.reftag
);
1508 c
.rw
.apptag
= cpu_to_le16(io
.apptag
);
1509 c
.rw
.appmask
= cpu_to_le16(io
.appmask
);
1511 return nvme_submit_user_cmd(ns
->queue
, &c
,
1512 nvme_to_user_ptr(io
.addr
), length
,
1513 metadata
, meta_len
, lower_32_bits(io
.slba
), NULL
, 0);
1516 static int nvme_user_cmd(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1517 struct nvme_passthru_cmd __user
*ucmd
)
1519 struct nvme_passthru_cmd cmd
;
1520 struct nvme_command c
;
1521 unsigned timeout
= 0;
1525 if (!capable(CAP_SYS_ADMIN
))
1527 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
1532 memset(&c
, 0, sizeof(c
));
1533 c
.common
.opcode
= cmd
.opcode
;
1534 c
.common
.flags
= cmd
.flags
;
1535 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
1536 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
1537 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
1538 c
.common
.cdw10
= cpu_to_le32(cmd
.cdw10
);
1539 c
.common
.cdw11
= cpu_to_le32(cmd
.cdw11
);
1540 c
.common
.cdw12
= cpu_to_le32(cmd
.cdw12
);
1541 c
.common
.cdw13
= cpu_to_le32(cmd
.cdw13
);
1542 c
.common
.cdw14
= cpu_to_le32(cmd
.cdw14
);
1543 c
.common
.cdw15
= cpu_to_le32(cmd
.cdw15
);
1546 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
1548 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
1549 nvme_to_user_ptr(cmd
.addr
), cmd
.data_len
,
1550 nvme_to_user_ptr(cmd
.metadata
), cmd
.metadata_len
,
1551 0, &result
, timeout
);
1554 if (put_user(result
, &ucmd
->result
))
1561 static int nvme_user_cmd64(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1562 struct nvme_passthru_cmd64 __user
*ucmd
)
1564 struct nvme_passthru_cmd64 cmd
;
1565 struct nvme_command c
;
1566 unsigned timeout
= 0;
1569 if (!capable(CAP_SYS_ADMIN
))
1571 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
1576 memset(&c
, 0, sizeof(c
));
1577 c
.common
.opcode
= cmd
.opcode
;
1578 c
.common
.flags
= cmd
.flags
;
1579 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
1580 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
1581 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
1582 c
.common
.cdw10
= cpu_to_le32(cmd
.cdw10
);
1583 c
.common
.cdw11
= cpu_to_le32(cmd
.cdw11
);
1584 c
.common
.cdw12
= cpu_to_le32(cmd
.cdw12
);
1585 c
.common
.cdw13
= cpu_to_le32(cmd
.cdw13
);
1586 c
.common
.cdw14
= cpu_to_le32(cmd
.cdw14
);
1587 c
.common
.cdw15
= cpu_to_le32(cmd
.cdw15
);
1590 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
1592 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
1593 nvme_to_user_ptr(cmd
.addr
), cmd
.data_len
,
1594 nvme_to_user_ptr(cmd
.metadata
), cmd
.metadata_len
,
1595 0, &cmd
.result
, timeout
);
1598 if (put_user(cmd
.result
, &ucmd
->result
))
1606 * Issue ioctl requests on the first available path. Note that unlike normal
1607 * block layer requests we will not retry failed request on another controller.
1609 struct nvme_ns
*nvme_get_ns_from_disk(struct gendisk
*disk
,
1610 struct nvme_ns_head
**head
, int *srcu_idx
)
1612 #ifdef CONFIG_NVME_MULTIPATH
1613 if (disk
->fops
== &nvme_ns_head_ops
) {
1616 *head
= disk
->private_data
;
1617 *srcu_idx
= srcu_read_lock(&(*head
)->srcu
);
1618 ns
= nvme_find_path(*head
);
1620 srcu_read_unlock(&(*head
)->srcu
, *srcu_idx
);
1626 return disk
->private_data
;
1629 void nvme_put_ns_from_disk(struct nvme_ns_head
*head
, int idx
)
1632 srcu_read_unlock(&head
->srcu
, idx
);
1635 static bool is_ctrl_ioctl(unsigned int cmd
)
1637 if (cmd
== NVME_IOCTL_ADMIN_CMD
|| cmd
== NVME_IOCTL_ADMIN64_CMD
)
1639 if (is_sed_ioctl(cmd
))
1644 static int nvme_handle_ctrl_ioctl(struct nvme_ns
*ns
, unsigned int cmd
,
1646 struct nvme_ns_head
*head
,
1649 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1652 nvme_get_ctrl(ns
->ctrl
);
1653 nvme_put_ns_from_disk(head
, srcu_idx
);
1656 case NVME_IOCTL_ADMIN_CMD
:
1657 ret
= nvme_user_cmd(ctrl
, NULL
, argp
);
1659 case NVME_IOCTL_ADMIN64_CMD
:
1660 ret
= nvme_user_cmd64(ctrl
, NULL
, argp
);
1663 ret
= sed_ioctl(ctrl
->opal_dev
, cmd
, argp
);
1666 nvme_put_ctrl(ctrl
);
1670 static int nvme_ioctl(struct block_device
*bdev
, fmode_t mode
,
1671 unsigned int cmd
, unsigned long arg
)
1673 struct nvme_ns_head
*head
= NULL
;
1674 void __user
*argp
= (void __user
*)arg
;
1678 ns
= nvme_get_ns_from_disk(bdev
->bd_disk
, &head
, &srcu_idx
);
1680 return -EWOULDBLOCK
;
1683 * Handle ioctls that apply to the controller instead of the namespace
1684 * seperately and drop the ns SRCU reference early. This avoids a
1685 * deadlock when deleting namespaces using the passthrough interface.
1687 if (is_ctrl_ioctl(cmd
))
1688 return nvme_handle_ctrl_ioctl(ns
, cmd
, argp
, head
, srcu_idx
);
1692 force_successful_syscall_return();
1693 ret
= ns
->head
->ns_id
;
1695 case NVME_IOCTL_IO_CMD
:
1696 ret
= nvme_user_cmd(ns
->ctrl
, ns
, argp
);
1698 case NVME_IOCTL_SUBMIT_IO
:
1699 ret
= nvme_submit_io(ns
, argp
);
1701 case NVME_IOCTL_IO64_CMD
:
1702 ret
= nvme_user_cmd64(ns
->ctrl
, ns
, argp
);
1706 ret
= nvme_nvm_ioctl(ns
, cmd
, arg
);
1711 nvme_put_ns_from_disk(head
, srcu_idx
);
1715 #ifdef CONFIG_COMPAT
1716 struct nvme_user_io32
{
1729 } __attribute__((__packed__
));
1731 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1733 static int nvme_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
1734 unsigned int cmd
, unsigned long arg
)
1737 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1738 * between 32 bit programs and 64 bit kernel.
1739 * The cause is that the results of sizeof(struct nvme_user_io),
1740 * which is used to define NVME_IOCTL_SUBMIT_IO,
1741 * are not same between 32 bit compiler and 64 bit compiler.
1742 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1743 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1744 * Other IOCTL numbers are same between 32 bit and 64 bit.
1745 * So there is nothing to do regarding to other IOCTL numbers.
1747 if (cmd
== NVME_IOCTL_SUBMIT_IO32
)
1748 return nvme_ioctl(bdev
, mode
, NVME_IOCTL_SUBMIT_IO
, arg
);
1750 return nvme_ioctl(bdev
, mode
, cmd
, arg
);
1753 #define nvme_compat_ioctl NULL
1754 #endif /* CONFIG_COMPAT */
1756 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
1758 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
1760 #ifdef CONFIG_NVME_MULTIPATH
1761 /* should never be called due to GENHD_FL_HIDDEN */
1762 if (WARN_ON_ONCE(ns
->head
->disk
))
1765 if (!kref_get_unless_zero(&ns
->kref
))
1767 if (!try_module_get(ns
->ctrl
->ops
->module
))
1778 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
1780 struct nvme_ns
*ns
= disk
->private_data
;
1782 module_put(ns
->ctrl
->ops
->module
);
1786 static int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1788 /* some standard values */
1789 geo
->heads
= 1 << 6;
1790 geo
->sectors
= 1 << 5;
1791 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
1795 #ifdef CONFIG_BLK_DEV_INTEGRITY
1796 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1797 u32 max_integrity_segments
)
1799 struct blk_integrity integrity
;
1801 memset(&integrity
, 0, sizeof(integrity
));
1803 case NVME_NS_DPS_PI_TYPE3
:
1804 integrity
.profile
= &t10_pi_type3_crc
;
1805 integrity
.tag_size
= sizeof(u16
) + sizeof(u32
);
1806 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1808 case NVME_NS_DPS_PI_TYPE1
:
1809 case NVME_NS_DPS_PI_TYPE2
:
1810 integrity
.profile
= &t10_pi_type1_crc
;
1811 integrity
.tag_size
= sizeof(u16
);
1812 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1815 integrity
.profile
= NULL
;
1818 integrity
.tuple_size
= ms
;
1819 blk_integrity_register(disk
, &integrity
);
1820 blk_queue_max_integrity_segments(disk
->queue
, max_integrity_segments
);
1823 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1824 u32 max_integrity_segments
)
1827 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1829 static void nvme_config_discard(struct gendisk
*disk
, struct nvme_ns
*ns
)
1831 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1832 struct request_queue
*queue
= disk
->queue
;
1833 u32 size
= queue_logical_block_size(queue
);
1835 if (!(ctrl
->oncs
& NVME_CTRL_ONCS_DSM
)) {
1836 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, queue
);
1840 if (ctrl
->nr_streams
&& ns
->sws
&& ns
->sgs
)
1841 size
*= ns
->sws
* ns
->sgs
;
1843 BUILD_BUG_ON(PAGE_SIZE
/ sizeof(struct nvme_dsm_range
) <
1844 NVME_DSM_MAX_RANGES
);
1846 queue
->limits
.discard_alignment
= 0;
1847 queue
->limits
.discard_granularity
= size
;
1849 /* If discard is already enabled, don't reset queue limits */
1850 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD
, queue
))
1853 blk_queue_max_discard_sectors(queue
, UINT_MAX
);
1854 blk_queue_max_discard_segments(queue
, NVME_DSM_MAX_RANGES
);
1856 if (ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
1857 blk_queue_max_write_zeroes_sectors(queue
, UINT_MAX
);
1860 static void nvme_config_write_zeroes(struct gendisk
*disk
, struct nvme_ns
*ns
)
1864 if (!(ns
->ctrl
->oncs
& NVME_CTRL_ONCS_WRITE_ZEROES
) ||
1865 (ns
->ctrl
->quirks
& NVME_QUIRK_DISABLE_WRITE_ZEROES
))
1868 * Even though NVMe spec explicitly states that MDTS is not
1869 * applicable to the write-zeroes:- "The restriction does not apply to
1870 * commands that do not transfer data between the host and the
1871 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1872 * In order to be more cautious use controller's max_hw_sectors value
1873 * to configure the maximum sectors for the write-zeroes which is
1874 * configured based on the controller's MDTS field in the
1875 * nvme_init_identify() if available.
1877 if (ns
->ctrl
->max_hw_sectors
== UINT_MAX
)
1878 max_blocks
= (u64
)USHRT_MAX
+ 1;
1880 max_blocks
= ns
->ctrl
->max_hw_sectors
+ 1;
1882 blk_queue_max_write_zeroes_sectors(disk
->queue
,
1883 nvme_lba_to_sect(ns
, max_blocks
));
1886 static bool nvme_ns_ids_valid(struct nvme_ns_ids
*ids
)
1888 return !uuid_is_null(&ids
->uuid
) ||
1889 memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)) ||
1890 memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
));
1893 static bool nvme_ns_ids_equal(struct nvme_ns_ids
*a
, struct nvme_ns_ids
*b
)
1895 return uuid_equal(&a
->uuid
, &b
->uuid
) &&
1896 memcmp(&a
->nguid
, &b
->nguid
, sizeof(a
->nguid
)) == 0 &&
1897 memcmp(&a
->eui64
, &b
->eui64
, sizeof(a
->eui64
)) == 0 &&
1901 static int nvme_setup_streams_ns(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1902 u32
*phys_bs
, u32
*io_opt
)
1904 struct streams_directive_params s
;
1907 if (!ctrl
->nr_streams
)
1910 ret
= nvme_get_stream_params(ctrl
, &s
, ns
->head
->ns_id
);
1914 ns
->sws
= le32_to_cpu(s
.sws
);
1915 ns
->sgs
= le16_to_cpu(s
.sgs
);
1918 *phys_bs
= ns
->sws
* (1 << ns
->lba_shift
);
1920 *io_opt
= *phys_bs
* ns
->sgs
;
1926 static int nvme_configure_metadata(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1928 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1931 * The PI implementation requires the metadata size to be equal to the
1932 * t10 pi tuple size.
1934 ns
->ms
= le16_to_cpu(id
->lbaf
[id
->flbas
& NVME_NS_FLBAS_LBA_MASK
].ms
);
1935 if (ns
->ms
== sizeof(struct t10_pi_tuple
))
1936 ns
->pi_type
= id
->dps
& NVME_NS_DPS_PI_MASK
;
1940 ns
->features
&= ~(NVME_NS_METADATA_SUPPORTED
| NVME_NS_EXT_LBAS
);
1941 if (!ns
->ms
|| !(ctrl
->ops
->flags
& NVME_F_METADATA_SUPPORTED
))
1943 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
1945 * The NVMe over Fabrics specification only supports metadata as
1946 * part of the extended data LBA. We rely on HCA/HBA support to
1947 * remap the separate metadata buffer from the block layer.
1949 if (WARN_ON_ONCE(!(id
->flbas
& NVME_NS_FLBAS_META_EXT
)))
1951 if (ctrl
->max_integrity_segments
)
1953 (NVME_NS_METADATA_SUPPORTED
| NVME_NS_EXT_LBAS
);
1956 * For PCIe controllers, we can't easily remap the separate
1957 * metadata buffer from the block layer and thus require a
1958 * separate metadata buffer for block layer metadata/PI support.
1959 * We allow extended LBAs for the passthrough interface, though.
1961 if (id
->flbas
& NVME_NS_FLBAS_META_EXT
)
1962 ns
->features
|= NVME_NS_EXT_LBAS
;
1964 ns
->features
|= NVME_NS_METADATA_SUPPORTED
;
1970 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
1971 struct request_queue
*q
)
1973 bool vwc
= ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
;
1975 if (ctrl
->max_hw_sectors
) {
1977 (ctrl
->max_hw_sectors
/ (NVME_CTRL_PAGE_SIZE
>> 9)) + 1;
1979 max_segments
= min_not_zero(max_segments
, ctrl
->max_segments
);
1980 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
1981 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
1983 blk_queue_virt_boundary(q
, NVME_CTRL_PAGE_SIZE
- 1);
1984 blk_queue_dma_alignment(q
, 7);
1985 blk_queue_write_cache(q
, vwc
, vwc
);
1988 static void nvme_update_disk_info(struct gendisk
*disk
,
1989 struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1991 sector_t capacity
= nvme_lba_to_sect(ns
, le64_to_cpu(id
->nsze
));
1992 unsigned short bs
= 1 << ns
->lba_shift
;
1993 u32 atomic_bs
, phys_bs
, io_opt
= 0;
1996 * The block layer can't support LBA sizes larger than the page size
1997 * yet, so catch this early and don't allow block I/O.
1999 if (ns
->lba_shift
> PAGE_SHIFT
) {
2004 blk_integrity_unregister(disk
);
2006 atomic_bs
= phys_bs
= bs
;
2007 nvme_setup_streams_ns(ns
->ctrl
, ns
, &phys_bs
, &io_opt
);
2008 if (id
->nabo
== 0) {
2010 * Bit 1 indicates whether NAWUPF is defined for this namespace
2011 * and whether it should be used instead of AWUPF. If NAWUPF ==
2012 * 0 then AWUPF must be used instead.
2014 if (id
->nsfeat
& NVME_NS_FEAT_ATOMICS
&& id
->nawupf
)
2015 atomic_bs
= (1 + le16_to_cpu(id
->nawupf
)) * bs
;
2017 atomic_bs
= (1 + ns
->ctrl
->subsys
->awupf
) * bs
;
2020 if (id
->nsfeat
& NVME_NS_FEAT_IO_OPT
) {
2021 /* NPWG = Namespace Preferred Write Granularity */
2022 phys_bs
= bs
* (1 + le16_to_cpu(id
->npwg
));
2023 /* NOWS = Namespace Optimal Write Size */
2024 io_opt
= bs
* (1 + le16_to_cpu(id
->nows
));
2027 blk_queue_logical_block_size(disk
->queue
, bs
);
2029 * Linux filesystems assume writing a single physical block is
2030 * an atomic operation. Hence limit the physical block size to the
2031 * value of the Atomic Write Unit Power Fail parameter.
2033 blk_queue_physical_block_size(disk
->queue
, min(phys_bs
, atomic_bs
));
2034 blk_queue_io_min(disk
->queue
, phys_bs
);
2035 blk_queue_io_opt(disk
->queue
, io_opt
);
2038 * Register a metadata profile for PI, or the plain non-integrity NVMe
2039 * metadata masquerading as Type 0 if supported, otherwise reject block
2040 * I/O to namespaces with metadata except when the namespace supports
2041 * PI, as it can strip/insert in that case.
2044 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY
) &&
2045 (ns
->features
& NVME_NS_METADATA_SUPPORTED
))
2046 nvme_init_integrity(disk
, ns
->ms
, ns
->pi_type
,
2047 ns
->ctrl
->max_integrity_segments
);
2048 else if (!nvme_ns_has_pi(ns
))
2052 set_capacity_revalidate_and_notify(disk
, capacity
, false);
2054 nvme_config_discard(disk
, ns
);
2055 nvme_config_write_zeroes(disk
, ns
);
2057 if (id
->nsattr
& NVME_NS_ATTR_RO
)
2058 set_disk_ro(disk
, true);
2060 set_disk_ro(disk
, false);
2063 static inline bool nvme_first_scan(struct gendisk
*disk
)
2065 /* nvme_alloc_ns() scans the disk prior to adding it */
2066 return !(disk
->flags
& GENHD_FL_UP
);
2069 static void nvme_set_chunk_sectors(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
2071 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
2074 if ((ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) &&
2075 is_power_of_2(ctrl
->max_hw_sectors
))
2076 iob
= ctrl
->max_hw_sectors
;
2078 iob
= nvme_lba_to_sect(ns
, le16_to_cpu(id
->noiob
));
2083 if (!is_power_of_2(iob
)) {
2084 if (nvme_first_scan(ns
->disk
))
2085 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2086 ns
->disk
->disk_name
, iob
);
2090 if (blk_queue_is_zoned(ns
->disk
->queue
)) {
2091 if (nvme_first_scan(ns
->disk
))
2092 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2093 ns
->disk
->disk_name
);
2097 blk_queue_chunk_sectors(ns
->queue
, iob
);
2100 static int nvme_update_ns_info(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
2102 unsigned lbaf
= id
->flbas
& NVME_NS_FLBAS_LBA_MASK
;
2105 blk_mq_freeze_queue(ns
->disk
->queue
);
2106 ns
->lba_shift
= id
->lbaf
[lbaf
].ds
;
2107 nvme_set_queue_limits(ns
->ctrl
, ns
->queue
);
2109 if (ns
->head
->ids
.csi
== NVME_CSI_ZNS
) {
2110 ret
= nvme_update_zone_info(ns
, lbaf
);
2115 ret
= nvme_configure_metadata(ns
, id
);
2118 nvme_set_chunk_sectors(ns
, id
);
2119 nvme_update_disk_info(ns
->disk
, ns
, id
);
2120 blk_mq_unfreeze_queue(ns
->disk
->queue
);
2122 if (blk_queue_is_zoned(ns
->queue
)) {
2123 ret
= nvme_revalidate_zones(ns
);
2128 #ifdef CONFIG_NVME_MULTIPATH
2129 if (ns
->head
->disk
) {
2130 blk_mq_freeze_queue(ns
->head
->disk
->queue
);
2131 nvme_update_disk_info(ns
->head
->disk
, ns
, id
);
2132 blk_stack_limits(&ns
->head
->disk
->queue
->limits
,
2133 &ns
->queue
->limits
, 0);
2134 blk_queue_update_readahead(ns
->head
->disk
->queue
);
2135 nvme_update_bdev_size(ns
->head
->disk
);
2136 blk_mq_unfreeze_queue(ns
->head
->disk
->queue
);
2142 blk_mq_unfreeze_queue(ns
->disk
->queue
);
2146 static char nvme_pr_type(enum pr_type type
)
2149 case PR_WRITE_EXCLUSIVE
:
2151 case PR_EXCLUSIVE_ACCESS
:
2153 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
2155 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
2157 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
2159 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
2166 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
2167 u64 key
, u64 sa_key
, u8 op
)
2169 struct nvme_ns_head
*head
= NULL
;
2171 struct nvme_command c
;
2173 u8 data
[16] = { 0, };
2175 ns
= nvme_get_ns_from_disk(bdev
->bd_disk
, &head
, &srcu_idx
);
2177 return -EWOULDBLOCK
;
2179 put_unaligned_le64(key
, &data
[0]);
2180 put_unaligned_le64(sa_key
, &data
[8]);
2182 memset(&c
, 0, sizeof(c
));
2183 c
.common
.opcode
= op
;
2184 c
.common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
2185 c
.common
.cdw10
= cpu_to_le32(cdw10
);
2187 ret
= nvme_submit_sync_cmd(ns
->queue
, &c
, data
, 16);
2188 nvme_put_ns_from_disk(head
, srcu_idx
);
2192 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
2193 u64
new, unsigned flags
)
2197 if (flags
& ~PR_FL_IGNORE_KEY
)
2200 cdw10
= old
? 2 : 0;
2201 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
2202 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
2203 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
2206 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
2207 enum pr_type type
, unsigned flags
)
2211 if (flags
& ~PR_FL_IGNORE_KEY
)
2214 cdw10
= nvme_pr_type(type
) << 8;
2215 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
2216 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
2219 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
2220 enum pr_type type
, bool abort
)
2222 u32 cdw10
= nvme_pr_type(type
) << 8 | (abort
? 2 : 1);
2223 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
2226 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
2228 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
2229 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
2232 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2234 u32 cdw10
= nvme_pr_type(type
) << 8 | (key
? 1 << 3 : 0);
2235 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
2238 static const struct pr_ops nvme_pr_ops
= {
2239 .pr_register
= nvme_pr_register
,
2240 .pr_reserve
= nvme_pr_reserve
,
2241 .pr_release
= nvme_pr_release
,
2242 .pr_preempt
= nvme_pr_preempt
,
2243 .pr_clear
= nvme_pr_clear
,
2246 #ifdef CONFIG_BLK_SED_OPAL
2247 int nvme_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
, size_t len
,
2250 struct nvme_ctrl
*ctrl
= data
;
2251 struct nvme_command cmd
;
2253 memset(&cmd
, 0, sizeof(cmd
));
2255 cmd
.common
.opcode
= nvme_admin_security_send
;
2257 cmd
.common
.opcode
= nvme_admin_security_recv
;
2258 cmd
.common
.nsid
= 0;
2259 cmd
.common
.cdw10
= cpu_to_le32(((u32
)secp
) << 24 | ((u32
)spsp
) << 8);
2260 cmd
.common
.cdw11
= cpu_to_le32(len
);
2262 return __nvme_submit_sync_cmd(ctrl
->admin_q
, &cmd
, NULL
, buffer
, len
,
2263 ADMIN_TIMEOUT
, NVME_QID_ANY
, 1, 0, false);
2265 EXPORT_SYMBOL_GPL(nvme_sec_submit
);
2266 #endif /* CONFIG_BLK_SED_OPAL */
2268 static const struct block_device_operations nvme_fops
= {
2269 .owner
= THIS_MODULE
,
2270 .ioctl
= nvme_ioctl
,
2271 .compat_ioctl
= nvme_compat_ioctl
,
2273 .release
= nvme_release
,
2274 .getgeo
= nvme_getgeo
,
2275 .report_zones
= nvme_report_zones
,
2276 .pr_ops
= &nvme_pr_ops
,
2279 #ifdef CONFIG_NVME_MULTIPATH
2280 static int nvme_ns_head_open(struct block_device
*bdev
, fmode_t mode
)
2282 struct nvme_ns_head
*head
= bdev
->bd_disk
->private_data
;
2284 if (!kref_get_unless_zero(&head
->ref
))
2289 static void nvme_ns_head_release(struct gendisk
*disk
, fmode_t mode
)
2291 nvme_put_ns_head(disk
->private_data
);
2294 const struct block_device_operations nvme_ns_head_ops
= {
2295 .owner
= THIS_MODULE
,
2296 .submit_bio
= nvme_ns_head_submit_bio
,
2297 .open
= nvme_ns_head_open
,
2298 .release
= nvme_ns_head_release
,
2299 .ioctl
= nvme_ioctl
,
2300 .compat_ioctl
= nvme_compat_ioctl
,
2301 .getgeo
= nvme_getgeo
,
2302 .report_zones
= nvme_report_zones
,
2303 .pr_ops
= &nvme_pr_ops
,
2305 #endif /* CONFIG_NVME_MULTIPATH */
2307 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
2309 unsigned long timeout
=
2310 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
2311 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
2314 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2317 if ((csts
& NVME_CSTS_RDY
) == bit
)
2320 usleep_range(1000, 2000);
2321 if (fatal_signal_pending(current
))
2323 if (time_after(jiffies
, timeout
)) {
2324 dev_err(ctrl
->device
,
2325 "Device not ready; aborting %s, CSTS=0x%x\n",
2326 enabled
? "initialisation" : "reset", csts
);
2335 * If the device has been passed off to us in an enabled state, just clear
2336 * the enabled bit. The spec says we should set the 'shutdown notification
2337 * bits', but doing so may cause the device to complete commands to the
2338 * admin queue ... and we don't know what memory that might be pointing at!
2340 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
)
2344 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2345 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
2347 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2351 if (ctrl
->quirks
& NVME_QUIRK_DELAY_BEFORE_CHK_RDY
)
2352 msleep(NVME_QUIRK_DELAY_AMOUNT
);
2354 return nvme_wait_ready(ctrl
, ctrl
->cap
, false);
2356 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
2358 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
)
2360 unsigned dev_page_min
;
2363 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &ctrl
->cap
);
2365 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
2368 dev_page_min
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12;
2370 if (NVME_CTRL_PAGE_SHIFT
< dev_page_min
) {
2371 dev_err(ctrl
->device
,
2372 "Minimum device page size %u too large for host (%u)\n",
2373 1 << dev_page_min
, 1 << NVME_CTRL_PAGE_SHIFT
);
2377 if (NVME_CAP_CSS(ctrl
->cap
) & NVME_CAP_CSS_CSI
)
2378 ctrl
->ctrl_config
= NVME_CC_CSS_CSI
;
2380 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
2381 ctrl
->ctrl_config
|= (NVME_CTRL_PAGE_SHIFT
- 12) << NVME_CC_MPS_SHIFT
;
2382 ctrl
->ctrl_config
|= NVME_CC_AMS_RR
| NVME_CC_SHN_NONE
;
2383 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
2384 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
2386 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2389 return nvme_wait_ready(ctrl
, ctrl
->cap
, true);
2391 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
2393 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
2395 unsigned long timeout
= jiffies
+ (ctrl
->shutdown_timeout
* HZ
);
2399 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2400 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
2402 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2406 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2407 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
2411 if (fatal_signal_pending(current
))
2413 if (time_after(jiffies
, timeout
)) {
2414 dev_err(ctrl
->device
,
2415 "Device shutdown incomplete; abort shutdown\n");
2422 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
2424 static int nvme_configure_timestamp(struct nvme_ctrl
*ctrl
)
2429 if (!(ctrl
->oncs
& NVME_CTRL_ONCS_TIMESTAMP
))
2432 ts
= cpu_to_le64(ktime_to_ms(ktime_get_real()));
2433 ret
= nvme_set_features(ctrl
, NVME_FEAT_TIMESTAMP
, 0, &ts
, sizeof(ts
),
2436 dev_warn_once(ctrl
->device
,
2437 "could not set timestamp (%d)\n", ret
);
2441 static int nvme_configure_acre(struct nvme_ctrl
*ctrl
)
2443 struct nvme_feat_host_behavior
*host
;
2446 /* Don't bother enabling the feature if retry delay is not reported */
2450 host
= kzalloc(sizeof(*host
), GFP_KERNEL
);
2454 host
->acre
= NVME_ENABLE_ACRE
;
2455 ret
= nvme_set_features(ctrl
, NVME_FEAT_HOST_BEHAVIOR
, 0,
2456 host
, sizeof(*host
), NULL
);
2461 static int nvme_configure_apst(struct nvme_ctrl
*ctrl
)
2464 * APST (Autonomous Power State Transition) lets us program a
2465 * table of power state transitions that the controller will
2466 * perform automatically. We configure it with a simple
2467 * heuristic: we are willing to spend at most 2% of the time
2468 * transitioning between power states. Therefore, when running
2469 * in any given state, we will enter the next lower-power
2470 * non-operational state after waiting 50 * (enlat + exlat)
2471 * microseconds, as long as that state's exit latency is under
2472 * the requested maximum latency.
2474 * We will not autonomously enter any non-operational state for
2475 * which the total latency exceeds ps_max_latency_us. Users
2476 * can set ps_max_latency_us to zero to turn off APST.
2480 struct nvme_feat_auto_pst
*table
;
2486 * If APST isn't supported or if we haven't been initialized yet,
2487 * then don't do anything.
2492 if (ctrl
->npss
> 31) {
2493 dev_warn(ctrl
->device
, "NPSS is invalid; not using APST\n");
2497 table
= kzalloc(sizeof(*table
), GFP_KERNEL
);
2501 if (!ctrl
->apst_enabled
|| ctrl
->ps_max_latency_us
== 0) {
2502 /* Turn off APST. */
2504 dev_dbg(ctrl
->device
, "APST disabled\n");
2506 __le64 target
= cpu_to_le64(0);
2510 * Walk through all states from lowest- to highest-power.
2511 * According to the spec, lower-numbered states use more
2512 * power. NPSS, despite the name, is the index of the
2513 * lowest-power state, not the number of states.
2515 for (state
= (int)ctrl
->npss
; state
>= 0; state
--) {
2516 u64 total_latency_us
, exit_latency_us
, transition_ms
;
2519 table
->entries
[state
] = target
;
2522 * Don't allow transitions to the deepest state
2523 * if it's quirked off.
2525 if (state
== ctrl
->npss
&&
2526 (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
))
2530 * Is this state a useful non-operational state for
2531 * higher-power states to autonomously transition to?
2533 if (!(ctrl
->psd
[state
].flags
&
2534 NVME_PS_FLAGS_NON_OP_STATE
))
2538 (u64
)le32_to_cpu(ctrl
->psd
[state
].exit_lat
);
2539 if (exit_latency_us
> ctrl
->ps_max_latency_us
)
2544 le32_to_cpu(ctrl
->psd
[state
].entry_lat
);
2547 * This state is good. Use it as the APST idle
2548 * target for higher power states.
2550 transition_ms
= total_latency_us
+ 19;
2551 do_div(transition_ms
, 20);
2552 if (transition_ms
> (1 << 24) - 1)
2553 transition_ms
= (1 << 24) - 1;
2555 target
= cpu_to_le64((state
<< 3) |
2556 (transition_ms
<< 8));
2561 if (total_latency_us
> max_lat_us
)
2562 max_lat_us
= total_latency_us
;
2568 dev_dbg(ctrl
->device
, "APST enabled but no non-operational states are available\n");
2570 dev_dbg(ctrl
->device
, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2571 max_ps
, max_lat_us
, (int)sizeof(*table
), table
);
2575 ret
= nvme_set_features(ctrl
, NVME_FEAT_AUTO_PST
, apste
,
2576 table
, sizeof(*table
), NULL
);
2578 dev_err(ctrl
->device
, "failed to set APST feature (%d)\n", ret
);
2584 static void nvme_set_latency_tolerance(struct device
*dev
, s32 val
)
2586 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
2590 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT
:
2591 case PM_QOS_LATENCY_ANY
:
2599 if (ctrl
->ps_max_latency_us
!= latency
) {
2600 ctrl
->ps_max_latency_us
= latency
;
2601 nvme_configure_apst(ctrl
);
2605 struct nvme_core_quirk_entry
{
2607 * NVMe model and firmware strings are padded with spaces. For
2608 * simplicity, strings in the quirk table are padded with NULLs
2614 unsigned long quirks
;
2617 static const struct nvme_core_quirk_entry core_quirks
[] = {
2620 * This Toshiba device seems to die using any APST states. See:
2621 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2624 .mn
= "THNSF5256GPUK TOSHIBA",
2625 .quirks
= NVME_QUIRK_NO_APST
,
2629 * This LiteON CL1-3D*-Q11 firmware version has a race
2630 * condition associated with actions related to suspend to idle
2631 * LiteON has resolved the problem in future firmware
2635 .quirks
= NVME_QUIRK_SIMPLE_SUSPEND
,
2639 /* match is null-terminated but idstr is space-padded. */
2640 static bool string_matches(const char *idstr
, const char *match
, size_t len
)
2647 matchlen
= strlen(match
);
2648 WARN_ON_ONCE(matchlen
> len
);
2650 if (memcmp(idstr
, match
, matchlen
))
2653 for (; matchlen
< len
; matchlen
++)
2654 if (idstr
[matchlen
] != ' ')
2660 static bool quirk_matches(const struct nvme_id_ctrl
*id
,
2661 const struct nvme_core_quirk_entry
*q
)
2663 return q
->vid
== le16_to_cpu(id
->vid
) &&
2664 string_matches(id
->mn
, q
->mn
, sizeof(id
->mn
)) &&
2665 string_matches(id
->fr
, q
->fr
, sizeof(id
->fr
));
2668 static void nvme_init_subnqn(struct nvme_subsystem
*subsys
, struct nvme_ctrl
*ctrl
,
2669 struct nvme_id_ctrl
*id
)
2674 if(!(ctrl
->quirks
& NVME_QUIRK_IGNORE_DEV_SUBNQN
)) {
2675 nqnlen
= strnlen(id
->subnqn
, NVMF_NQN_SIZE
);
2676 if (nqnlen
> 0 && nqnlen
< NVMF_NQN_SIZE
) {
2677 strlcpy(subsys
->subnqn
, id
->subnqn
, NVMF_NQN_SIZE
);
2681 if (ctrl
->vs
>= NVME_VS(1, 2, 1))
2682 dev_warn(ctrl
->device
, "missing or invalid SUBNQN field.\n");
2685 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2686 off
= snprintf(subsys
->subnqn
, NVMF_NQN_SIZE
,
2687 "nqn.2014.08.org.nvmexpress:%04x%04x",
2688 le16_to_cpu(id
->vid
), le16_to_cpu(id
->ssvid
));
2689 memcpy(subsys
->subnqn
+ off
, id
->sn
, sizeof(id
->sn
));
2690 off
+= sizeof(id
->sn
);
2691 memcpy(subsys
->subnqn
+ off
, id
->mn
, sizeof(id
->mn
));
2692 off
+= sizeof(id
->mn
);
2693 memset(subsys
->subnqn
+ off
, 0, sizeof(subsys
->subnqn
) - off
);
2696 static void nvme_release_subsystem(struct device
*dev
)
2698 struct nvme_subsystem
*subsys
=
2699 container_of(dev
, struct nvme_subsystem
, dev
);
2701 if (subsys
->instance
>= 0)
2702 ida_simple_remove(&nvme_instance_ida
, subsys
->instance
);
2706 static void nvme_destroy_subsystem(struct kref
*ref
)
2708 struct nvme_subsystem
*subsys
=
2709 container_of(ref
, struct nvme_subsystem
, ref
);
2711 mutex_lock(&nvme_subsystems_lock
);
2712 list_del(&subsys
->entry
);
2713 mutex_unlock(&nvme_subsystems_lock
);
2715 ida_destroy(&subsys
->ns_ida
);
2716 device_del(&subsys
->dev
);
2717 put_device(&subsys
->dev
);
2720 static void nvme_put_subsystem(struct nvme_subsystem
*subsys
)
2722 kref_put(&subsys
->ref
, nvme_destroy_subsystem
);
2725 static struct nvme_subsystem
*__nvme_find_get_subsystem(const char *subsysnqn
)
2727 struct nvme_subsystem
*subsys
;
2729 lockdep_assert_held(&nvme_subsystems_lock
);
2732 * Fail matches for discovery subsystems. This results
2733 * in each discovery controller bound to a unique subsystem.
2734 * This avoids issues with validating controller values
2735 * that can only be true when there is a single unique subsystem.
2736 * There may be multiple and completely independent entities
2737 * that provide discovery controllers.
2739 if (!strcmp(subsysnqn
, NVME_DISC_SUBSYS_NAME
))
2742 list_for_each_entry(subsys
, &nvme_subsystems
, entry
) {
2743 if (strcmp(subsys
->subnqn
, subsysnqn
))
2745 if (!kref_get_unless_zero(&subsys
->ref
))
2753 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2754 struct device_attribute subsys_attr_##_name = \
2755 __ATTR(_name, _mode, _show, NULL)
2757 static ssize_t
nvme_subsys_show_nqn(struct device
*dev
,
2758 struct device_attribute
*attr
,
2761 struct nvme_subsystem
*subsys
=
2762 container_of(dev
, struct nvme_subsystem
, dev
);
2764 return snprintf(buf
, PAGE_SIZE
, "%s\n", subsys
->subnqn
);
2766 static SUBSYS_ATTR_RO(subsysnqn
, S_IRUGO
, nvme_subsys_show_nqn
);
2768 #define nvme_subsys_show_str_function(field) \
2769 static ssize_t subsys_##field##_show(struct device *dev, \
2770 struct device_attribute *attr, char *buf) \
2772 struct nvme_subsystem *subsys = \
2773 container_of(dev, struct nvme_subsystem, dev); \
2774 return sprintf(buf, "%.*s\n", \
2775 (int)sizeof(subsys->field), subsys->field); \
2777 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2779 nvme_subsys_show_str_function(model
);
2780 nvme_subsys_show_str_function(serial
);
2781 nvme_subsys_show_str_function(firmware_rev
);
2783 static struct attribute
*nvme_subsys_attrs
[] = {
2784 &subsys_attr_model
.attr
,
2785 &subsys_attr_serial
.attr
,
2786 &subsys_attr_firmware_rev
.attr
,
2787 &subsys_attr_subsysnqn
.attr
,
2788 #ifdef CONFIG_NVME_MULTIPATH
2789 &subsys_attr_iopolicy
.attr
,
2794 static struct attribute_group nvme_subsys_attrs_group
= {
2795 .attrs
= nvme_subsys_attrs
,
2798 static const struct attribute_group
*nvme_subsys_attrs_groups
[] = {
2799 &nvme_subsys_attrs_group
,
2803 static bool nvme_validate_cntlid(struct nvme_subsystem
*subsys
,
2804 struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2806 struct nvme_ctrl
*tmp
;
2808 lockdep_assert_held(&nvme_subsystems_lock
);
2810 list_for_each_entry(tmp
, &subsys
->ctrls
, subsys_entry
) {
2811 if (nvme_state_terminal(tmp
))
2814 if (tmp
->cntlid
== ctrl
->cntlid
) {
2815 dev_err(ctrl
->device
,
2816 "Duplicate cntlid %u with %s, rejecting\n",
2817 ctrl
->cntlid
, dev_name(tmp
->device
));
2821 if ((id
->cmic
& NVME_CTRL_CMIC_MULTI_CTRL
) ||
2822 (ctrl
->opts
&& ctrl
->opts
->discovery_nqn
))
2825 dev_err(ctrl
->device
,
2826 "Subsystem does not support multiple controllers\n");
2833 static int nvme_init_subsystem(struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2835 struct nvme_subsystem
*subsys
, *found
;
2838 subsys
= kzalloc(sizeof(*subsys
), GFP_KERNEL
);
2842 subsys
->instance
= -1;
2843 mutex_init(&subsys
->lock
);
2844 kref_init(&subsys
->ref
);
2845 INIT_LIST_HEAD(&subsys
->ctrls
);
2846 INIT_LIST_HEAD(&subsys
->nsheads
);
2847 nvme_init_subnqn(subsys
, ctrl
, id
);
2848 memcpy(subsys
->serial
, id
->sn
, sizeof(subsys
->serial
));
2849 memcpy(subsys
->model
, id
->mn
, sizeof(subsys
->model
));
2850 memcpy(subsys
->firmware_rev
, id
->fr
, sizeof(subsys
->firmware_rev
));
2851 subsys
->vendor_id
= le16_to_cpu(id
->vid
);
2852 subsys
->cmic
= id
->cmic
;
2853 subsys
->awupf
= le16_to_cpu(id
->awupf
);
2854 #ifdef CONFIG_NVME_MULTIPATH
2855 subsys
->iopolicy
= NVME_IOPOLICY_NUMA
;
2858 subsys
->dev
.class = nvme_subsys_class
;
2859 subsys
->dev
.release
= nvme_release_subsystem
;
2860 subsys
->dev
.groups
= nvme_subsys_attrs_groups
;
2861 dev_set_name(&subsys
->dev
, "nvme-subsys%d", ctrl
->instance
);
2862 device_initialize(&subsys
->dev
);
2864 mutex_lock(&nvme_subsystems_lock
);
2865 found
= __nvme_find_get_subsystem(subsys
->subnqn
);
2867 put_device(&subsys
->dev
);
2870 if (!nvme_validate_cntlid(subsys
, ctrl
, id
)) {
2872 goto out_put_subsystem
;
2875 ret
= device_add(&subsys
->dev
);
2877 dev_err(ctrl
->device
,
2878 "failed to register subsystem device.\n");
2879 put_device(&subsys
->dev
);
2882 ida_init(&subsys
->ns_ida
);
2883 list_add_tail(&subsys
->entry
, &nvme_subsystems
);
2886 ret
= sysfs_create_link(&subsys
->dev
.kobj
, &ctrl
->device
->kobj
,
2887 dev_name(ctrl
->device
));
2889 dev_err(ctrl
->device
,
2890 "failed to create sysfs link from subsystem.\n");
2891 goto out_put_subsystem
;
2895 subsys
->instance
= ctrl
->instance
;
2896 ctrl
->subsys
= subsys
;
2897 list_add_tail(&ctrl
->subsys_entry
, &subsys
->ctrls
);
2898 mutex_unlock(&nvme_subsystems_lock
);
2902 nvme_put_subsystem(subsys
);
2904 mutex_unlock(&nvme_subsystems_lock
);
2908 int nvme_get_log(struct nvme_ctrl
*ctrl
, u32 nsid
, u8 log_page
, u8 lsp
, u8 csi
,
2909 void *log
, size_t size
, u64 offset
)
2911 struct nvme_command c
= { };
2912 u32 dwlen
= nvme_bytes_to_numd(size
);
2914 c
.get_log_page
.opcode
= nvme_admin_get_log_page
;
2915 c
.get_log_page
.nsid
= cpu_to_le32(nsid
);
2916 c
.get_log_page
.lid
= log_page
;
2917 c
.get_log_page
.lsp
= lsp
;
2918 c
.get_log_page
.numdl
= cpu_to_le16(dwlen
& ((1 << 16) - 1));
2919 c
.get_log_page
.numdu
= cpu_to_le16(dwlen
>> 16);
2920 c
.get_log_page
.lpol
= cpu_to_le32(lower_32_bits(offset
));
2921 c
.get_log_page
.lpou
= cpu_to_le32(upper_32_bits(offset
));
2922 c
.get_log_page
.csi
= csi
;
2924 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, log
, size
);
2927 static int nvme_get_effects_log(struct nvme_ctrl
*ctrl
, u8 csi
,
2928 struct nvme_effects_log
**log
)
2930 struct nvme_cel
*cel
= xa_load(&ctrl
->cels
, csi
);
2936 cel
= kzalloc(sizeof(*cel
), GFP_KERNEL
);
2940 ret
= nvme_get_log(ctrl
, 0x00, NVME_LOG_CMD_EFFECTS
, 0, csi
,
2941 &cel
->log
, sizeof(cel
->log
), 0);
2948 xa_store(&ctrl
->cels
, cel
->csi
, cel
, GFP_KERNEL
);
2955 * Initialize the cached copies of the Identify data and various controller
2956 * register in our nvme_ctrl structure. This should be called as soon as
2957 * the admin queue is fully up and running.
2959 int nvme_init_identify(struct nvme_ctrl
*ctrl
)
2961 struct nvme_id_ctrl
*id
;
2962 int ret
, page_shift
;
2964 bool prev_apst_enabled
;
2966 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
2968 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
2971 page_shift
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12;
2972 ctrl
->sqsize
= min_t(u16
, NVME_CAP_MQES(ctrl
->cap
), ctrl
->sqsize
);
2974 if (ctrl
->vs
>= NVME_VS(1, 1, 0))
2975 ctrl
->subsystem
= NVME_CAP_NSSRC(ctrl
->cap
);
2977 ret
= nvme_identify_ctrl(ctrl
, &id
);
2979 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
2983 if (id
->lpa
& NVME_CTRL_LPA_CMD_EFFECTS_LOG
) {
2984 ret
= nvme_get_effects_log(ctrl
, NVME_CSI_NVM
, &ctrl
->effects
);
2989 if (!(ctrl
->ops
->flags
& NVME_F_FABRICS
))
2990 ctrl
->cntlid
= le16_to_cpu(id
->cntlid
);
2992 if (!ctrl
->identified
) {
2995 ret
= nvme_init_subsystem(ctrl
, id
);
3000 * Check for quirks. Quirk can depend on firmware version,
3001 * so, in principle, the set of quirks present can change
3002 * across a reset. As a possible future enhancement, we
3003 * could re-scan for quirks every time we reinitialize
3004 * the device, but we'd have to make sure that the driver
3005 * behaves intelligently if the quirks change.
3007 for (i
= 0; i
< ARRAY_SIZE(core_quirks
); i
++) {
3008 if (quirk_matches(id
, &core_quirks
[i
]))
3009 ctrl
->quirks
|= core_quirks
[i
].quirks
;
3013 if (force_apst
&& (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
)) {
3014 dev_warn(ctrl
->device
, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3015 ctrl
->quirks
&= ~NVME_QUIRK_NO_DEEPEST_PS
;
3018 ctrl
->crdt
[0] = le16_to_cpu(id
->crdt1
);
3019 ctrl
->crdt
[1] = le16_to_cpu(id
->crdt2
);
3020 ctrl
->crdt
[2] = le16_to_cpu(id
->crdt3
);
3022 ctrl
->oacs
= le16_to_cpu(id
->oacs
);
3023 ctrl
->oncs
= le16_to_cpu(id
->oncs
);
3024 ctrl
->mtfa
= le16_to_cpu(id
->mtfa
);
3025 ctrl
->oaes
= le32_to_cpu(id
->oaes
);
3026 ctrl
->wctemp
= le16_to_cpu(id
->wctemp
);
3027 ctrl
->cctemp
= le16_to_cpu(id
->cctemp
);
3029 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
3030 ctrl
->vwc
= id
->vwc
;
3032 max_hw_sectors
= 1 << (id
->mdts
+ page_shift
- 9);
3034 max_hw_sectors
= UINT_MAX
;
3035 ctrl
->max_hw_sectors
=
3036 min_not_zero(ctrl
->max_hw_sectors
, max_hw_sectors
);
3038 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
3039 ctrl
->sgls
= le32_to_cpu(id
->sgls
);
3040 ctrl
->kas
= le16_to_cpu(id
->kas
);
3041 ctrl
->max_namespaces
= le32_to_cpu(id
->mnan
);
3042 ctrl
->ctratt
= le32_to_cpu(id
->ctratt
);
3046 u32 transition_time
= le32_to_cpu(id
->rtd3e
) / USEC_PER_SEC
;
3048 ctrl
->shutdown_timeout
= clamp_t(unsigned int, transition_time
,
3049 shutdown_timeout
, 60);
3051 if (ctrl
->shutdown_timeout
!= shutdown_timeout
)
3052 dev_info(ctrl
->device
,
3053 "Shutdown timeout set to %u seconds\n",
3054 ctrl
->shutdown_timeout
);
3056 ctrl
->shutdown_timeout
= shutdown_timeout
;
3058 ctrl
->npss
= id
->npss
;
3059 ctrl
->apsta
= id
->apsta
;
3060 prev_apst_enabled
= ctrl
->apst_enabled
;
3061 if (ctrl
->quirks
& NVME_QUIRK_NO_APST
) {
3062 if (force_apst
&& id
->apsta
) {
3063 dev_warn(ctrl
->device
, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3064 ctrl
->apst_enabled
= true;
3066 ctrl
->apst_enabled
= false;
3069 ctrl
->apst_enabled
= id
->apsta
;
3071 memcpy(ctrl
->psd
, id
->psd
, sizeof(ctrl
->psd
));
3073 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
3074 ctrl
->icdoff
= le16_to_cpu(id
->icdoff
);
3075 ctrl
->ioccsz
= le32_to_cpu(id
->ioccsz
);
3076 ctrl
->iorcsz
= le32_to_cpu(id
->iorcsz
);
3077 ctrl
->maxcmd
= le16_to_cpu(id
->maxcmd
);
3080 * In fabrics we need to verify the cntlid matches the
3083 if (ctrl
->cntlid
!= le16_to_cpu(id
->cntlid
)) {
3084 dev_err(ctrl
->device
,
3085 "Mismatching cntlid: Connect %u vs Identify "
3087 ctrl
->cntlid
, le16_to_cpu(id
->cntlid
));
3092 if (!ctrl
->opts
->discovery_nqn
&& !ctrl
->kas
) {
3093 dev_err(ctrl
->device
,
3094 "keep-alive support is mandatory for fabrics\n");
3099 ctrl
->hmpre
= le32_to_cpu(id
->hmpre
);
3100 ctrl
->hmmin
= le32_to_cpu(id
->hmmin
);
3101 ctrl
->hmminds
= le32_to_cpu(id
->hmminds
);
3102 ctrl
->hmmaxd
= le16_to_cpu(id
->hmmaxd
);
3105 ret
= nvme_mpath_init(ctrl
, id
);
3111 if (ctrl
->apst_enabled
&& !prev_apst_enabled
)
3112 dev_pm_qos_expose_latency_tolerance(ctrl
->device
);
3113 else if (!ctrl
->apst_enabled
&& prev_apst_enabled
)
3114 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
3116 ret
= nvme_configure_apst(ctrl
);
3120 ret
= nvme_configure_timestamp(ctrl
);
3124 ret
= nvme_configure_directives(ctrl
);
3128 ret
= nvme_configure_acre(ctrl
);
3132 if (!ctrl
->identified
) {
3133 ret
= nvme_hwmon_init(ctrl
);
3138 ctrl
->identified
= true;
3146 EXPORT_SYMBOL_GPL(nvme_init_identify
);
3148 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
3150 struct nvme_ctrl
*ctrl
=
3151 container_of(inode
->i_cdev
, struct nvme_ctrl
, cdev
);
3153 switch (ctrl
->state
) {
3154 case NVME_CTRL_LIVE
:
3157 return -EWOULDBLOCK
;
3160 nvme_get_ctrl(ctrl
);
3161 if (!try_module_get(ctrl
->ops
->module
)) {
3162 nvme_put_ctrl(ctrl
);
3166 file
->private_data
= ctrl
;
3170 static int nvme_dev_release(struct inode
*inode
, struct file
*file
)
3172 struct nvme_ctrl
*ctrl
=
3173 container_of(inode
->i_cdev
, struct nvme_ctrl
, cdev
);
3175 module_put(ctrl
->ops
->module
);
3176 nvme_put_ctrl(ctrl
);
3180 static int nvme_dev_user_cmd(struct nvme_ctrl
*ctrl
, void __user
*argp
)
3185 down_read(&ctrl
->namespaces_rwsem
);
3186 if (list_empty(&ctrl
->namespaces
)) {
3191 ns
= list_first_entry(&ctrl
->namespaces
, struct nvme_ns
, list
);
3192 if (ns
!= list_last_entry(&ctrl
->namespaces
, struct nvme_ns
, list
)) {
3193 dev_warn(ctrl
->device
,
3194 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3199 dev_warn(ctrl
->device
,
3200 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3201 kref_get(&ns
->kref
);
3202 up_read(&ctrl
->namespaces_rwsem
);
3204 ret
= nvme_user_cmd(ctrl
, ns
, argp
);
3209 up_read(&ctrl
->namespaces_rwsem
);
3213 static long nvme_dev_ioctl(struct file
*file
, unsigned int cmd
,
3216 struct nvme_ctrl
*ctrl
= file
->private_data
;
3217 void __user
*argp
= (void __user
*)arg
;
3220 case NVME_IOCTL_ADMIN_CMD
:
3221 return nvme_user_cmd(ctrl
, NULL
, argp
);
3222 case NVME_IOCTL_ADMIN64_CMD
:
3223 return nvme_user_cmd64(ctrl
, NULL
, argp
);
3224 case NVME_IOCTL_IO_CMD
:
3225 return nvme_dev_user_cmd(ctrl
, argp
);
3226 case NVME_IOCTL_RESET
:
3227 dev_warn(ctrl
->device
, "resetting controller\n");
3228 return nvme_reset_ctrl_sync(ctrl
);
3229 case NVME_IOCTL_SUBSYS_RESET
:
3230 return nvme_reset_subsystem(ctrl
);
3231 case NVME_IOCTL_RESCAN
:
3232 nvme_queue_scan(ctrl
);
3239 static const struct file_operations nvme_dev_fops
= {
3240 .owner
= THIS_MODULE
,
3241 .open
= nvme_dev_open
,
3242 .release
= nvme_dev_release
,
3243 .unlocked_ioctl
= nvme_dev_ioctl
,
3244 .compat_ioctl
= compat_ptr_ioctl
,
3247 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
3248 struct device_attribute
*attr
, const char *buf
,
3251 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3254 ret
= nvme_reset_ctrl_sync(ctrl
);
3259 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
3261 static ssize_t
nvme_sysfs_rescan(struct device
*dev
,
3262 struct device_attribute
*attr
, const char *buf
,
3265 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3267 nvme_queue_scan(ctrl
);
3270 static DEVICE_ATTR(rescan_controller
, S_IWUSR
, NULL
, nvme_sysfs_rescan
);
3272 static inline struct nvme_ns_head
*dev_to_ns_head(struct device
*dev
)
3274 struct gendisk
*disk
= dev_to_disk(dev
);
3276 if (disk
->fops
== &nvme_fops
)
3277 return nvme_get_ns_from_dev(dev
)->head
;
3279 return disk
->private_data
;
3282 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
3285 struct nvme_ns_head
*head
= dev_to_ns_head(dev
);
3286 struct nvme_ns_ids
*ids
= &head
->ids
;
3287 struct nvme_subsystem
*subsys
= head
->subsys
;
3288 int serial_len
= sizeof(subsys
->serial
);
3289 int model_len
= sizeof(subsys
->model
);
3291 if (!uuid_is_null(&ids
->uuid
))
3292 return sprintf(buf
, "uuid.%pU\n", &ids
->uuid
);
3294 if (memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3295 return sprintf(buf
, "eui.%16phN\n", ids
->nguid
);
3297 if (memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3298 return sprintf(buf
, "eui.%8phN\n", ids
->eui64
);
3300 while (serial_len
> 0 && (subsys
->serial
[serial_len
- 1] == ' ' ||
3301 subsys
->serial
[serial_len
- 1] == '\0'))
3303 while (model_len
> 0 && (subsys
->model
[model_len
- 1] == ' ' ||
3304 subsys
->model
[model_len
- 1] == '\0'))
3307 return sprintf(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", subsys
->vendor_id
,
3308 serial_len
, subsys
->serial
, model_len
, subsys
->model
,
3311 static DEVICE_ATTR_RO(wwid
);
3313 static ssize_t
nguid_show(struct device
*dev
, struct device_attribute
*attr
,
3316 return sprintf(buf
, "%pU\n", dev_to_ns_head(dev
)->ids
.nguid
);
3318 static DEVICE_ATTR_RO(nguid
);
3320 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
3323 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3325 /* For backward compatibility expose the NGUID to userspace if
3326 * we have no UUID set
3328 if (uuid_is_null(&ids
->uuid
)) {
3329 printk_ratelimited(KERN_WARNING
3330 "No UUID available providing old NGUID\n");
3331 return sprintf(buf
, "%pU\n", ids
->nguid
);
3333 return sprintf(buf
, "%pU\n", &ids
->uuid
);
3335 static DEVICE_ATTR_RO(uuid
);
3337 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
3340 return sprintf(buf
, "%8ph\n", dev_to_ns_head(dev
)->ids
.eui64
);
3342 static DEVICE_ATTR_RO(eui
);
3344 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
3347 return sprintf(buf
, "%d\n", dev_to_ns_head(dev
)->ns_id
);
3349 static DEVICE_ATTR_RO(nsid
);
3351 static struct attribute
*nvme_ns_id_attrs
[] = {
3352 &dev_attr_wwid
.attr
,
3353 &dev_attr_uuid
.attr
,
3354 &dev_attr_nguid
.attr
,
3356 &dev_attr_nsid
.attr
,
3357 #ifdef CONFIG_NVME_MULTIPATH
3358 &dev_attr_ana_grpid
.attr
,
3359 &dev_attr_ana_state
.attr
,
3364 static umode_t
nvme_ns_id_attrs_are_visible(struct kobject
*kobj
,
3365 struct attribute
*a
, int n
)
3367 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3368 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3370 if (a
== &dev_attr_uuid
.attr
) {
3371 if (uuid_is_null(&ids
->uuid
) &&
3372 !memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3375 if (a
== &dev_attr_nguid
.attr
) {
3376 if (!memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3379 if (a
== &dev_attr_eui
.attr
) {
3380 if (!memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3383 #ifdef CONFIG_NVME_MULTIPATH
3384 if (a
== &dev_attr_ana_grpid
.attr
|| a
== &dev_attr_ana_state
.attr
) {
3385 if (dev_to_disk(dev
)->fops
!= &nvme_fops
) /* per-path attr */
3387 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev
)->ctrl
))
3394 static const struct attribute_group nvme_ns_id_attr_group
= {
3395 .attrs
= nvme_ns_id_attrs
,
3396 .is_visible
= nvme_ns_id_attrs_are_visible
,
3399 const struct attribute_group
*nvme_ns_id_attr_groups
[] = {
3400 &nvme_ns_id_attr_group
,
3402 &nvme_nvm_attr_group
,
3407 #define nvme_show_str_function(field) \
3408 static ssize_t field##_show(struct device *dev, \
3409 struct device_attribute *attr, char *buf) \
3411 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3412 return sprintf(buf, "%.*s\n", \
3413 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3415 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3417 nvme_show_str_function(model
);
3418 nvme_show_str_function(serial
);
3419 nvme_show_str_function(firmware_rev
);
3421 #define nvme_show_int_function(field) \
3422 static ssize_t field##_show(struct device *dev, \
3423 struct device_attribute *attr, char *buf) \
3425 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3426 return sprintf(buf, "%d\n", ctrl->field); \
3428 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3430 nvme_show_int_function(cntlid
);
3431 nvme_show_int_function(numa_node
);
3432 nvme_show_int_function(queue_count
);
3433 nvme_show_int_function(sqsize
);
3435 static ssize_t
nvme_sysfs_delete(struct device
*dev
,
3436 struct device_attribute
*attr
, const char *buf
,
3439 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3441 if (device_remove_file_self(dev
, attr
))
3442 nvme_delete_ctrl_sync(ctrl
);
3445 static DEVICE_ATTR(delete_controller
, S_IWUSR
, NULL
, nvme_sysfs_delete
);
3447 static ssize_t
nvme_sysfs_show_transport(struct device
*dev
,
3448 struct device_attribute
*attr
,
3451 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3453 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->ops
->name
);
3455 static DEVICE_ATTR(transport
, S_IRUGO
, nvme_sysfs_show_transport
, NULL
);
3457 static ssize_t
nvme_sysfs_show_state(struct device
*dev
,
3458 struct device_attribute
*attr
,
3461 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3462 static const char *const state_name
[] = {
3463 [NVME_CTRL_NEW
] = "new",
3464 [NVME_CTRL_LIVE
] = "live",
3465 [NVME_CTRL_RESETTING
] = "resetting",
3466 [NVME_CTRL_CONNECTING
] = "connecting",
3467 [NVME_CTRL_DELETING
] = "deleting",
3468 [NVME_CTRL_DELETING_NOIO
]= "deleting (no IO)",
3469 [NVME_CTRL_DEAD
] = "dead",
3472 if ((unsigned)ctrl
->state
< ARRAY_SIZE(state_name
) &&
3473 state_name
[ctrl
->state
])
3474 return sprintf(buf
, "%s\n", state_name
[ctrl
->state
]);
3476 return sprintf(buf
, "unknown state\n");
3479 static DEVICE_ATTR(state
, S_IRUGO
, nvme_sysfs_show_state
, NULL
);
3481 static ssize_t
nvme_sysfs_show_subsysnqn(struct device
*dev
,
3482 struct device_attribute
*attr
,
3485 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3487 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->subsys
->subnqn
);
3489 static DEVICE_ATTR(subsysnqn
, S_IRUGO
, nvme_sysfs_show_subsysnqn
, NULL
);
3491 static ssize_t
nvme_sysfs_show_hostnqn(struct device
*dev
,
3492 struct device_attribute
*attr
,
3495 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3497 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->opts
->host
->nqn
);
3499 static DEVICE_ATTR(hostnqn
, S_IRUGO
, nvme_sysfs_show_hostnqn
, NULL
);
3501 static ssize_t
nvme_sysfs_show_hostid(struct device
*dev
,
3502 struct device_attribute
*attr
,
3505 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3507 return snprintf(buf
, PAGE_SIZE
, "%pU\n", &ctrl
->opts
->host
->id
);
3509 static DEVICE_ATTR(hostid
, S_IRUGO
, nvme_sysfs_show_hostid
, NULL
);
3511 static ssize_t
nvme_sysfs_show_address(struct device
*dev
,
3512 struct device_attribute
*attr
,
3515 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3517 return ctrl
->ops
->get_address(ctrl
, buf
, PAGE_SIZE
);
3519 static DEVICE_ATTR(address
, S_IRUGO
, nvme_sysfs_show_address
, NULL
);
3521 static ssize_t
nvme_ctrl_loss_tmo_show(struct device
*dev
,
3522 struct device_attribute
*attr
, char *buf
)
3524 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3525 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3527 if (ctrl
->opts
->max_reconnects
== -1)
3528 return sprintf(buf
, "off\n");
3529 return sprintf(buf
, "%d\n",
3530 opts
->max_reconnects
* opts
->reconnect_delay
);
3533 static ssize_t
nvme_ctrl_loss_tmo_store(struct device
*dev
,
3534 struct device_attribute
*attr
, const char *buf
, size_t count
)
3536 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3537 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3538 int ctrl_loss_tmo
, err
;
3540 err
= kstrtoint(buf
, 10, &ctrl_loss_tmo
);
3544 else if (ctrl_loss_tmo
< 0)
3545 opts
->max_reconnects
= -1;
3547 opts
->max_reconnects
= DIV_ROUND_UP(ctrl_loss_tmo
,
3548 opts
->reconnect_delay
);
3551 static DEVICE_ATTR(ctrl_loss_tmo
, S_IRUGO
| S_IWUSR
,
3552 nvme_ctrl_loss_tmo_show
, nvme_ctrl_loss_tmo_store
);
3554 static ssize_t
nvme_ctrl_reconnect_delay_show(struct device
*dev
,
3555 struct device_attribute
*attr
, char *buf
)
3557 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3559 if (ctrl
->opts
->reconnect_delay
== -1)
3560 return sprintf(buf
, "off\n");
3561 return sprintf(buf
, "%d\n", ctrl
->opts
->reconnect_delay
);
3564 static ssize_t
nvme_ctrl_reconnect_delay_store(struct device
*dev
,
3565 struct device_attribute
*attr
, const char *buf
, size_t count
)
3567 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3571 err
= kstrtou32(buf
, 10, &v
);
3575 ctrl
->opts
->reconnect_delay
= v
;
3578 static DEVICE_ATTR(reconnect_delay
, S_IRUGO
| S_IWUSR
,
3579 nvme_ctrl_reconnect_delay_show
, nvme_ctrl_reconnect_delay_store
);
3581 static struct attribute
*nvme_dev_attrs
[] = {
3582 &dev_attr_reset_controller
.attr
,
3583 &dev_attr_rescan_controller
.attr
,
3584 &dev_attr_model
.attr
,
3585 &dev_attr_serial
.attr
,
3586 &dev_attr_firmware_rev
.attr
,
3587 &dev_attr_cntlid
.attr
,
3588 &dev_attr_delete_controller
.attr
,
3589 &dev_attr_transport
.attr
,
3590 &dev_attr_subsysnqn
.attr
,
3591 &dev_attr_address
.attr
,
3592 &dev_attr_state
.attr
,
3593 &dev_attr_numa_node
.attr
,
3594 &dev_attr_queue_count
.attr
,
3595 &dev_attr_sqsize
.attr
,
3596 &dev_attr_hostnqn
.attr
,
3597 &dev_attr_hostid
.attr
,
3598 &dev_attr_ctrl_loss_tmo
.attr
,
3599 &dev_attr_reconnect_delay
.attr
,
3603 static umode_t
nvme_dev_attrs_are_visible(struct kobject
*kobj
,
3604 struct attribute
*a
, int n
)
3606 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3607 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3609 if (a
== &dev_attr_delete_controller
.attr
&& !ctrl
->ops
->delete_ctrl
)
3611 if (a
== &dev_attr_address
.attr
&& !ctrl
->ops
->get_address
)
3613 if (a
== &dev_attr_hostnqn
.attr
&& !ctrl
->opts
)
3615 if (a
== &dev_attr_hostid
.attr
&& !ctrl
->opts
)
3617 if (a
== &dev_attr_ctrl_loss_tmo
.attr
&& !ctrl
->opts
)
3619 if (a
== &dev_attr_reconnect_delay
.attr
&& !ctrl
->opts
)
3625 static struct attribute_group nvme_dev_attrs_group
= {
3626 .attrs
= nvme_dev_attrs
,
3627 .is_visible
= nvme_dev_attrs_are_visible
,
3630 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
3631 &nvme_dev_attrs_group
,
3635 static struct nvme_ns_head
*nvme_find_ns_head(struct nvme_subsystem
*subsys
,
3638 struct nvme_ns_head
*h
;
3640 lockdep_assert_held(&subsys
->lock
);
3642 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3643 if (h
->ns_id
== nsid
&& kref_get_unless_zero(&h
->ref
))
3650 static int __nvme_check_ids(struct nvme_subsystem
*subsys
,
3651 struct nvme_ns_head
*new)
3653 struct nvme_ns_head
*h
;
3655 lockdep_assert_held(&subsys
->lock
);
3657 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3658 if (nvme_ns_ids_valid(&new->ids
) &&
3659 nvme_ns_ids_equal(&new->ids
, &h
->ids
))
3666 static struct nvme_ns_head
*nvme_alloc_ns_head(struct nvme_ctrl
*ctrl
,
3667 unsigned nsid
, struct nvme_ns_ids
*ids
)
3669 struct nvme_ns_head
*head
;
3670 size_t size
= sizeof(*head
);
3673 #ifdef CONFIG_NVME_MULTIPATH
3674 size
+= num_possible_nodes() * sizeof(struct nvme_ns
*);
3677 head
= kzalloc(size
, GFP_KERNEL
);
3680 ret
= ida_simple_get(&ctrl
->subsys
->ns_ida
, 1, 0, GFP_KERNEL
);
3683 head
->instance
= ret
;
3684 INIT_LIST_HEAD(&head
->list
);
3685 ret
= init_srcu_struct(&head
->srcu
);
3687 goto out_ida_remove
;
3688 head
->subsys
= ctrl
->subsys
;
3691 kref_init(&head
->ref
);
3693 ret
= __nvme_check_ids(ctrl
->subsys
, head
);
3695 dev_err(ctrl
->device
,
3696 "duplicate IDs for nsid %d\n", nsid
);
3697 goto out_cleanup_srcu
;
3700 if (head
->ids
.csi
) {
3701 ret
= nvme_get_effects_log(ctrl
, head
->ids
.csi
, &head
->effects
);
3703 goto out_cleanup_srcu
;
3705 head
->effects
= ctrl
->effects
;
3707 ret
= nvme_mpath_alloc_disk(ctrl
, head
);
3709 goto out_cleanup_srcu
;
3711 list_add_tail(&head
->entry
, &ctrl
->subsys
->nsheads
);
3713 kref_get(&ctrl
->subsys
->ref
);
3717 cleanup_srcu_struct(&head
->srcu
);
3719 ida_simple_remove(&ctrl
->subsys
->ns_ida
, head
->instance
);
3724 ret
= blk_status_to_errno(nvme_error_status(ret
));
3725 return ERR_PTR(ret
);
3728 static int nvme_init_ns_head(struct nvme_ns
*ns
, unsigned nsid
,
3729 struct nvme_ns_ids
*ids
, bool is_shared
)
3731 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
3732 struct nvme_ns_head
*head
= NULL
;
3735 mutex_lock(&ctrl
->subsys
->lock
);
3736 head
= nvme_find_ns_head(ctrl
->subsys
, nsid
);
3738 head
= nvme_alloc_ns_head(ctrl
, nsid
, ids
);
3740 ret
= PTR_ERR(head
);
3743 head
->shared
= is_shared
;
3746 if (!is_shared
|| !head
->shared
) {
3747 dev_err(ctrl
->device
,
3748 "Duplicate unshared namespace %d\n", nsid
);
3749 goto out_put_ns_head
;
3751 if (!nvme_ns_ids_equal(&head
->ids
, ids
)) {
3752 dev_err(ctrl
->device
,
3753 "IDs don't match for shared namespace %d\n",
3755 goto out_put_ns_head
;
3759 list_add_tail(&ns
->siblings
, &head
->list
);
3761 mutex_unlock(&ctrl
->subsys
->lock
);
3765 nvme_put_ns_head(head
);
3767 mutex_unlock(&ctrl
->subsys
->lock
);
3771 static int ns_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
3773 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
3774 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
3776 return nsa
->head
->ns_id
- nsb
->head
->ns_id
;
3779 struct nvme_ns
*nvme_find_get_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3781 struct nvme_ns
*ns
, *ret
= NULL
;
3783 down_read(&ctrl
->namespaces_rwsem
);
3784 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
3785 if (ns
->head
->ns_id
== nsid
) {
3786 if (!kref_get_unless_zero(&ns
->kref
))
3791 if (ns
->head
->ns_id
> nsid
)
3794 up_read(&ctrl
->namespaces_rwsem
);
3797 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns
, NVME_TARGET_PASSTHRU
);
3799 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
,
3800 struct nvme_ns_ids
*ids
)
3803 struct gendisk
*disk
;
3804 struct nvme_id_ns
*id
;
3805 char disk_name
[DISK_NAME_LEN
];
3806 int node
= ctrl
->numa_node
, flags
= GENHD_FL_EXT_DEVT
, ret
;
3808 if (nvme_identify_ns(ctrl
, nsid
, ids
, &id
))
3811 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
3815 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
3816 if (IS_ERR(ns
->queue
))
3819 if (ctrl
->opts
&& ctrl
->opts
->data_digest
)
3820 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES
, ns
->queue
);
3822 blk_queue_flag_set(QUEUE_FLAG_NONROT
, ns
->queue
);
3823 if (ctrl
->ops
->flags
& NVME_F_PCI_P2PDMA
)
3824 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA
, ns
->queue
);
3826 ns
->queue
->queuedata
= ns
;
3828 kref_init(&ns
->kref
);
3830 ret
= nvme_init_ns_head(ns
, nsid
, ids
, id
->nmic
& NVME_NS_NMIC_SHARED
);
3832 goto out_free_queue
;
3833 nvme_set_disk_name(disk_name
, ns
, ctrl
, &flags
);
3835 disk
= alloc_disk_node(0, node
);
3839 disk
->fops
= &nvme_fops
;
3840 disk
->private_data
= ns
;
3841 disk
->queue
= ns
->queue
;
3842 disk
->flags
= flags
;
3843 memcpy(disk
->disk_name
, disk_name
, DISK_NAME_LEN
);
3846 if (nvme_update_ns_info(ns
, id
))
3849 if ((ctrl
->quirks
& NVME_QUIRK_LIGHTNVM
) && id
->vs
[0] == 0x1) {
3850 ret
= nvme_nvm_register(ns
, disk_name
, node
);
3852 dev_warn(ctrl
->device
, "LightNVM init failure\n");
3857 down_write(&ctrl
->namespaces_rwsem
);
3858 list_add_tail(&ns
->list
, &ctrl
->namespaces
);
3859 up_write(&ctrl
->namespaces_rwsem
);
3861 nvme_get_ctrl(ctrl
);
3863 device_add_disk(ctrl
->device
, ns
->disk
, nvme_ns_id_attr_groups
);
3865 nvme_mpath_add_disk(ns
, id
);
3866 nvme_fault_inject_init(&ns
->fault_inject
, ns
->disk
->disk_name
);
3871 /* prevent double queue cleanup */
3872 ns
->disk
->queue
= NULL
;
3875 mutex_lock(&ctrl
->subsys
->lock
);
3876 list_del_rcu(&ns
->siblings
);
3877 if (list_empty(&ns
->head
->list
))
3878 list_del_init(&ns
->head
->entry
);
3879 mutex_unlock(&ctrl
->subsys
->lock
);
3880 nvme_put_ns_head(ns
->head
);
3882 blk_cleanup_queue(ns
->queue
);
3889 static void nvme_ns_remove(struct nvme_ns
*ns
)
3891 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
3894 set_capacity(ns
->disk
, 0);
3895 nvme_fault_inject_fini(&ns
->fault_inject
);
3897 mutex_lock(&ns
->ctrl
->subsys
->lock
);
3898 list_del_rcu(&ns
->siblings
);
3899 if (list_empty(&ns
->head
->list
))
3900 list_del_init(&ns
->head
->entry
);
3901 mutex_unlock(&ns
->ctrl
->subsys
->lock
);
3903 synchronize_rcu(); /* guarantee not available in head->list */
3904 nvme_mpath_clear_current_path(ns
);
3905 synchronize_srcu(&ns
->head
->srcu
); /* wait for concurrent submissions */
3907 if (ns
->disk
->flags
& GENHD_FL_UP
) {
3908 del_gendisk(ns
->disk
);
3909 blk_cleanup_queue(ns
->queue
);
3910 if (blk_get_integrity(ns
->disk
))
3911 blk_integrity_unregister(ns
->disk
);
3914 down_write(&ns
->ctrl
->namespaces_rwsem
);
3915 list_del_init(&ns
->list
);
3916 up_write(&ns
->ctrl
->namespaces_rwsem
);
3918 nvme_mpath_check_last_path(ns
);
3922 static void nvme_ns_remove_by_nsid(struct nvme_ctrl
*ctrl
, u32 nsid
)
3924 struct nvme_ns
*ns
= nvme_find_get_ns(ctrl
, nsid
);
3932 static void nvme_validate_ns(struct nvme_ns
*ns
, struct nvme_ns_ids
*ids
)
3934 struct nvme_id_ns
*id
;
3937 if (test_bit(NVME_NS_DEAD
, &ns
->flags
))
3940 ret
= nvme_identify_ns(ns
->ctrl
, ns
->head
->ns_id
, ids
, &id
);
3945 if (!nvme_ns_ids_equal(&ns
->head
->ids
, ids
)) {
3946 dev_err(ns
->ctrl
->device
,
3947 "identifiers changed for nsid %d\n", ns
->head
->ns_id
);
3951 ret
= nvme_update_ns_info(ns
, id
);
3957 * Only remove the namespace if we got a fatal error back from the
3958 * device, otherwise ignore the error and just move on.
3960 * TODO: we should probably schedule a delayed retry here.
3962 if (ret
&& ret
!= -ENOMEM
&& !(ret
> 0 && !(ret
& NVME_SC_DNR
)))
3965 revalidate_disk_size(ns
->disk
, true);
3968 static void nvme_validate_or_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3970 struct nvme_ns_ids ids
= { };
3973 if (nvme_identify_ns_descs(ctrl
, nsid
, &ids
))
3976 ns
= nvme_find_get_ns(ctrl
, nsid
);
3978 nvme_validate_ns(ns
, &ids
);
3985 nvme_alloc_ns(ctrl
, nsid
, &ids
);
3988 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED
)) {
3989 dev_warn(ctrl
->device
,
3990 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3994 nvme_alloc_ns(ctrl
, nsid
, &ids
);
3997 dev_warn(ctrl
->device
, "unknown csi %u for nsid %u\n",
4003 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
4006 struct nvme_ns
*ns
, *next
;
4009 down_write(&ctrl
->namespaces_rwsem
);
4010 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
4011 if (ns
->head
->ns_id
> nsid
|| test_bit(NVME_NS_DEAD
, &ns
->flags
))
4012 list_move_tail(&ns
->list
, &rm_list
);
4014 up_write(&ctrl
->namespaces_rwsem
);
4016 list_for_each_entry_safe(ns
, next
, &rm_list
, list
)
4021 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
)
4023 const int nr_entries
= NVME_IDENTIFY_DATA_SIZE
/ sizeof(__le32
);
4028 if (nvme_ctrl_limited_cns(ctrl
))
4031 ns_list
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
4036 struct nvme_command cmd
= {
4037 .identify
.opcode
= nvme_admin_identify
,
4038 .identify
.cns
= NVME_ID_CNS_NS_ACTIVE_LIST
,
4039 .identify
.nsid
= cpu_to_le32(prev
),
4042 ret
= nvme_submit_sync_cmd(ctrl
->admin_q
, &cmd
, ns_list
,
4043 NVME_IDENTIFY_DATA_SIZE
);
4047 for (i
= 0; i
< nr_entries
; i
++) {
4048 u32 nsid
= le32_to_cpu(ns_list
[i
]);
4050 if (!nsid
) /* end of the list? */
4052 nvme_validate_or_alloc_ns(ctrl
, nsid
);
4053 while (++prev
< nsid
)
4054 nvme_ns_remove_by_nsid(ctrl
, prev
);
4058 nvme_remove_invalid_namespaces(ctrl
, prev
);
4064 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
)
4066 struct nvme_id_ctrl
*id
;
4069 if (nvme_identify_ctrl(ctrl
, &id
))
4071 nn
= le32_to_cpu(id
->nn
);
4074 for (i
= 1; i
<= nn
; i
++)
4075 nvme_validate_or_alloc_ns(ctrl
, i
);
4077 nvme_remove_invalid_namespaces(ctrl
, nn
);
4080 static void nvme_clear_changed_ns_log(struct nvme_ctrl
*ctrl
)
4082 size_t log_size
= NVME_MAX_CHANGED_NAMESPACES
* sizeof(__le32
);
4086 log
= kzalloc(log_size
, GFP_KERNEL
);
4091 * We need to read the log to clear the AEN, but we don't want to rely
4092 * on it for the changed namespace information as userspace could have
4093 * raced with us in reading the log page, which could cause us to miss
4096 error
= nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_CHANGED_NS
, 0,
4097 NVME_CSI_NVM
, log
, log_size
, 0);
4099 dev_warn(ctrl
->device
,
4100 "reading changed ns log failed: %d\n", error
);
4105 static void nvme_scan_work(struct work_struct
*work
)
4107 struct nvme_ctrl
*ctrl
=
4108 container_of(work
, struct nvme_ctrl
, scan_work
);
4110 /* No tagset on a live ctrl means IO queues could not created */
4111 if (ctrl
->state
!= NVME_CTRL_LIVE
|| !ctrl
->tagset
)
4114 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
)) {
4115 dev_info(ctrl
->device
, "rescanning namespaces.\n");
4116 nvme_clear_changed_ns_log(ctrl
);
4119 mutex_lock(&ctrl
->scan_lock
);
4120 if (nvme_scan_ns_list(ctrl
) != 0)
4121 nvme_scan_ns_sequential(ctrl
);
4122 mutex_unlock(&ctrl
->scan_lock
);
4124 down_write(&ctrl
->namespaces_rwsem
);
4125 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
4126 up_write(&ctrl
->namespaces_rwsem
);
4130 * This function iterates the namespace list unlocked to allow recovery from
4131 * controller failure. It is up to the caller to ensure the namespace list is
4132 * not modified by scan work while this function is executing.
4134 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
4136 struct nvme_ns
*ns
, *next
;
4140 * make sure to requeue I/O to all namespaces as these
4141 * might result from the scan itself and must complete
4142 * for the scan_work to make progress
4144 nvme_mpath_clear_ctrl_paths(ctrl
);
4146 /* prevent racing with ns scanning */
4147 flush_work(&ctrl
->scan_work
);
4150 * The dead states indicates the controller was not gracefully
4151 * disconnected. In that case, we won't be able to flush any data while
4152 * removing the namespaces' disks; fail all the queues now to avoid
4153 * potentially having to clean up the failed sync later.
4155 if (ctrl
->state
== NVME_CTRL_DEAD
)
4156 nvme_kill_queues(ctrl
);
4158 /* this is a no-op when called from the controller reset handler */
4159 nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING_NOIO
);
4161 down_write(&ctrl
->namespaces_rwsem
);
4162 list_splice_init(&ctrl
->namespaces
, &ns_list
);
4163 up_write(&ctrl
->namespaces_rwsem
);
4165 list_for_each_entry_safe(ns
, next
, &ns_list
, list
)
4168 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
4170 static int nvme_class_uevent(struct device
*dev
, struct kobj_uevent_env
*env
)
4172 struct nvme_ctrl
*ctrl
=
4173 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
4174 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
4177 ret
= add_uevent_var(env
, "NVME_TRTYPE=%s", ctrl
->ops
->name
);
4182 ret
= add_uevent_var(env
, "NVME_TRADDR=%s", opts
->traddr
);
4186 ret
= add_uevent_var(env
, "NVME_TRSVCID=%s",
4187 opts
->trsvcid
?: "none");
4191 ret
= add_uevent_var(env
, "NVME_HOST_TRADDR=%s",
4192 opts
->host_traddr
?: "none");
4197 static void nvme_aen_uevent(struct nvme_ctrl
*ctrl
)
4199 char *envp
[2] = { NULL
, NULL
};
4200 u32 aen_result
= ctrl
->aen_result
;
4202 ctrl
->aen_result
= 0;
4206 envp
[0] = kasprintf(GFP_KERNEL
, "NVME_AEN=%#08x", aen_result
);
4209 kobject_uevent_env(&ctrl
->device
->kobj
, KOBJ_CHANGE
, envp
);
4213 static void nvme_async_event_work(struct work_struct
*work
)
4215 struct nvme_ctrl
*ctrl
=
4216 container_of(work
, struct nvme_ctrl
, async_event_work
);
4218 nvme_aen_uevent(ctrl
);
4219 ctrl
->ops
->submit_async_event(ctrl
);
4222 static bool nvme_ctrl_pp_status(struct nvme_ctrl
*ctrl
)
4227 if (ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
))
4233 return ((ctrl
->ctrl_config
& NVME_CC_ENABLE
) && (csts
& NVME_CSTS_PP
));
4236 static void nvme_get_fw_slot_info(struct nvme_ctrl
*ctrl
)
4238 struct nvme_fw_slot_info_log
*log
;
4240 log
= kmalloc(sizeof(*log
), GFP_KERNEL
);
4244 if (nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_FW_SLOT
, 0, NVME_CSI_NVM
,
4245 log
, sizeof(*log
), 0))
4246 dev_warn(ctrl
->device
, "Get FW SLOT INFO log error\n");
4250 static void nvme_fw_act_work(struct work_struct
*work
)
4252 struct nvme_ctrl
*ctrl
= container_of(work
,
4253 struct nvme_ctrl
, fw_act_work
);
4254 unsigned long fw_act_timeout
;
4257 fw_act_timeout
= jiffies
+
4258 msecs_to_jiffies(ctrl
->mtfa
* 100);
4260 fw_act_timeout
= jiffies
+
4261 msecs_to_jiffies(admin_timeout
* 1000);
4263 nvme_stop_queues(ctrl
);
4264 while (nvme_ctrl_pp_status(ctrl
)) {
4265 if (time_after(jiffies
, fw_act_timeout
)) {
4266 dev_warn(ctrl
->device
,
4267 "Fw activation timeout, reset controller\n");
4268 nvme_try_sched_reset(ctrl
);
4274 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_LIVE
))
4277 nvme_start_queues(ctrl
);
4278 /* read FW slot information to clear the AER */
4279 nvme_get_fw_slot_info(ctrl
);
4282 static void nvme_handle_aen_notice(struct nvme_ctrl
*ctrl
, u32 result
)
4284 u32 aer_notice_type
= (result
& 0xff00) >> 8;
4286 trace_nvme_async_event(ctrl
, aer_notice_type
);
4288 switch (aer_notice_type
) {
4289 case NVME_AER_NOTICE_NS_CHANGED
:
4290 set_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
);
4291 nvme_queue_scan(ctrl
);
4293 case NVME_AER_NOTICE_FW_ACT_STARTING
:
4295 * We are (ab)using the RESETTING state to prevent subsequent
4296 * recovery actions from interfering with the controller's
4297 * firmware activation.
4299 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
4300 queue_work(nvme_wq
, &ctrl
->fw_act_work
);
4302 #ifdef CONFIG_NVME_MULTIPATH
4303 case NVME_AER_NOTICE_ANA
:
4304 if (!ctrl
->ana_log_buf
)
4306 queue_work(nvme_wq
, &ctrl
->ana_work
);
4309 case NVME_AER_NOTICE_DISC_CHANGED
:
4310 ctrl
->aen_result
= result
;
4313 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
4317 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
, __le16 status
,
4318 volatile union nvme_result
*res
)
4320 u32 result
= le32_to_cpu(res
->u32
);
4321 u32 aer_type
= result
& 0x07;
4323 if (le16_to_cpu(status
) >> 1 != NVME_SC_SUCCESS
)
4327 case NVME_AER_NOTICE
:
4328 nvme_handle_aen_notice(ctrl
, result
);
4330 case NVME_AER_ERROR
:
4331 case NVME_AER_SMART
:
4334 trace_nvme_async_event(ctrl
, aer_type
);
4335 ctrl
->aen_result
= result
;
4340 queue_work(nvme_wq
, &ctrl
->async_event_work
);
4342 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
4344 void nvme_stop_ctrl(struct nvme_ctrl
*ctrl
)
4346 nvme_mpath_stop(ctrl
);
4347 nvme_stop_keep_alive(ctrl
);
4348 flush_work(&ctrl
->async_event_work
);
4349 cancel_work_sync(&ctrl
->fw_act_work
);
4351 EXPORT_SYMBOL_GPL(nvme_stop_ctrl
);
4353 void nvme_start_ctrl(struct nvme_ctrl
*ctrl
)
4355 nvme_start_keep_alive(ctrl
);
4357 nvme_enable_aen(ctrl
);
4359 if (ctrl
->queue_count
> 1) {
4360 nvme_queue_scan(ctrl
);
4361 nvme_start_queues(ctrl
);
4364 EXPORT_SYMBOL_GPL(nvme_start_ctrl
);
4366 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
4368 nvme_fault_inject_fini(&ctrl
->fault_inject
);
4369 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
4370 cdev_device_del(&ctrl
->cdev
, ctrl
->device
);
4371 nvme_put_ctrl(ctrl
);
4373 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
4375 static void nvme_free_ctrl(struct device
*dev
)
4377 struct nvme_ctrl
*ctrl
=
4378 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
4379 struct nvme_subsystem
*subsys
= ctrl
->subsys
;
4381 if (!subsys
|| ctrl
->instance
!= subsys
->instance
)
4382 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4384 xa_destroy(&ctrl
->cels
);
4386 nvme_mpath_uninit(ctrl
);
4387 __free_page(ctrl
->discard_page
);
4390 mutex_lock(&nvme_subsystems_lock
);
4391 list_del(&ctrl
->subsys_entry
);
4392 sysfs_remove_link(&subsys
->dev
.kobj
, dev_name(ctrl
->device
));
4393 mutex_unlock(&nvme_subsystems_lock
);
4396 ctrl
->ops
->free_ctrl(ctrl
);
4399 nvme_put_subsystem(subsys
);
4403 * Initialize a NVMe controller structures. This needs to be called during
4404 * earliest initialization so that we have the initialized structured around
4407 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
4408 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
4412 ctrl
->state
= NVME_CTRL_NEW
;
4413 spin_lock_init(&ctrl
->lock
);
4414 mutex_init(&ctrl
->scan_lock
);
4415 INIT_LIST_HEAD(&ctrl
->namespaces
);
4416 xa_init(&ctrl
->cels
);
4417 init_rwsem(&ctrl
->namespaces_rwsem
);
4420 ctrl
->quirks
= quirks
;
4421 ctrl
->numa_node
= NUMA_NO_NODE
;
4422 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
4423 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
4424 INIT_WORK(&ctrl
->fw_act_work
, nvme_fw_act_work
);
4425 INIT_WORK(&ctrl
->delete_work
, nvme_delete_ctrl_work
);
4426 init_waitqueue_head(&ctrl
->state_wq
);
4428 INIT_DELAYED_WORK(&ctrl
->ka_work
, nvme_keep_alive_work
);
4429 memset(&ctrl
->ka_cmd
, 0, sizeof(ctrl
->ka_cmd
));
4430 ctrl
->ka_cmd
.common
.opcode
= nvme_admin_keep_alive
;
4432 BUILD_BUG_ON(NVME_DSM_MAX_RANGES
* sizeof(struct nvme_dsm_range
) >
4434 ctrl
->discard_page
= alloc_page(GFP_KERNEL
);
4435 if (!ctrl
->discard_page
) {
4440 ret
= ida_simple_get(&nvme_instance_ida
, 0, 0, GFP_KERNEL
);
4443 ctrl
->instance
= ret
;
4445 device_initialize(&ctrl
->ctrl_device
);
4446 ctrl
->device
= &ctrl
->ctrl_device
;
4447 ctrl
->device
->devt
= MKDEV(MAJOR(nvme_chr_devt
), ctrl
->instance
);
4448 ctrl
->device
->class = nvme_class
;
4449 ctrl
->device
->parent
= ctrl
->dev
;
4450 ctrl
->device
->groups
= nvme_dev_attr_groups
;
4451 ctrl
->device
->release
= nvme_free_ctrl
;
4452 dev_set_drvdata(ctrl
->device
, ctrl
);
4453 ret
= dev_set_name(ctrl
->device
, "nvme%d", ctrl
->instance
);
4455 goto out_release_instance
;
4457 nvme_get_ctrl(ctrl
);
4458 cdev_init(&ctrl
->cdev
, &nvme_dev_fops
);
4459 ctrl
->cdev
.owner
= ops
->module
;
4460 ret
= cdev_device_add(&ctrl
->cdev
, ctrl
->device
);
4465 * Initialize latency tolerance controls. The sysfs files won't
4466 * be visible to userspace unless the device actually supports APST.
4468 ctrl
->device
->power
.set_latency_tolerance
= nvme_set_latency_tolerance
;
4469 dev_pm_qos_update_user_latency_tolerance(ctrl
->device
,
4470 min(default_ps_max_latency_us
, (unsigned long)S32_MAX
));
4472 nvme_fault_inject_init(&ctrl
->fault_inject
, dev_name(ctrl
->device
));
4476 nvme_put_ctrl(ctrl
);
4477 kfree_const(ctrl
->device
->kobj
.name
);
4478 out_release_instance
:
4479 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4481 if (ctrl
->discard_page
)
4482 __free_page(ctrl
->discard_page
);
4485 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
4488 * nvme_kill_queues(): Ends all namespace queues
4489 * @ctrl: the dead controller that needs to end
4491 * Call this function when the driver determines it is unable to get the
4492 * controller in a state capable of servicing IO.
4494 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
4498 down_read(&ctrl
->namespaces_rwsem
);
4500 /* Forcibly unquiesce queues to avoid blocking dispatch */
4501 if (ctrl
->admin_q
&& !blk_queue_dying(ctrl
->admin_q
))
4502 blk_mq_unquiesce_queue(ctrl
->admin_q
);
4504 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4505 nvme_set_queue_dying(ns
);
4507 up_read(&ctrl
->namespaces_rwsem
);
4509 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
4511 void nvme_unfreeze(struct nvme_ctrl
*ctrl
)
4515 down_read(&ctrl
->namespaces_rwsem
);
4516 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4517 blk_mq_unfreeze_queue(ns
->queue
);
4518 up_read(&ctrl
->namespaces_rwsem
);
4520 EXPORT_SYMBOL_GPL(nvme_unfreeze
);
4522 int nvme_wait_freeze_timeout(struct nvme_ctrl
*ctrl
, long timeout
)
4526 down_read(&ctrl
->namespaces_rwsem
);
4527 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
4528 timeout
= blk_mq_freeze_queue_wait_timeout(ns
->queue
, timeout
);
4532 up_read(&ctrl
->namespaces_rwsem
);
4535 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout
);
4537 void nvme_wait_freeze(struct nvme_ctrl
*ctrl
)
4541 down_read(&ctrl
->namespaces_rwsem
);
4542 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4543 blk_mq_freeze_queue_wait(ns
->queue
);
4544 up_read(&ctrl
->namespaces_rwsem
);
4546 EXPORT_SYMBOL_GPL(nvme_wait_freeze
);
4548 void nvme_start_freeze(struct nvme_ctrl
*ctrl
)
4552 down_read(&ctrl
->namespaces_rwsem
);
4553 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4554 blk_freeze_queue_start(ns
->queue
);
4555 up_read(&ctrl
->namespaces_rwsem
);
4557 EXPORT_SYMBOL_GPL(nvme_start_freeze
);
4559 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
4563 down_read(&ctrl
->namespaces_rwsem
);
4564 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4565 blk_mq_quiesce_queue(ns
->queue
);
4566 up_read(&ctrl
->namespaces_rwsem
);
4568 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
4570 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
4574 down_read(&ctrl
->namespaces_rwsem
);
4575 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4576 blk_mq_unquiesce_queue(ns
->queue
);
4577 up_read(&ctrl
->namespaces_rwsem
);
4579 EXPORT_SYMBOL_GPL(nvme_start_queues
);
4582 void nvme_sync_queues(struct nvme_ctrl
*ctrl
)
4586 down_read(&ctrl
->namespaces_rwsem
);
4587 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4588 blk_sync_queue(ns
->queue
);
4589 up_read(&ctrl
->namespaces_rwsem
);
4592 blk_sync_queue(ctrl
->admin_q
);
4594 EXPORT_SYMBOL_GPL(nvme_sync_queues
);
4596 struct nvme_ctrl
*nvme_ctrl_from_file(struct file
*file
)
4598 if (file
->f_op
!= &nvme_dev_fops
)
4600 return file
->private_data
;
4602 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file
, NVME_TARGET_PASSTHRU
);
4605 * Check we didn't inadvertently grow the command structure sizes:
4607 static inline void _nvme_check_size(void)
4609 BUILD_BUG_ON(sizeof(struct nvme_common_command
) != 64);
4610 BUILD_BUG_ON(sizeof(struct nvme_rw_command
) != 64);
4611 BUILD_BUG_ON(sizeof(struct nvme_identify
) != 64);
4612 BUILD_BUG_ON(sizeof(struct nvme_features
) != 64);
4613 BUILD_BUG_ON(sizeof(struct nvme_download_firmware
) != 64);
4614 BUILD_BUG_ON(sizeof(struct nvme_format_cmd
) != 64);
4615 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd
) != 64);
4616 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd
) != 64);
4617 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd
) != 64);
4618 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command
) != 64);
4619 BUILD_BUG_ON(sizeof(struct nvme_command
) != 64);
4620 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl
) != NVME_IDENTIFY_DATA_SIZE
);
4621 BUILD_BUG_ON(sizeof(struct nvme_id_ns
) != NVME_IDENTIFY_DATA_SIZE
);
4622 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns
) != NVME_IDENTIFY_DATA_SIZE
);
4623 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns
) != NVME_IDENTIFY_DATA_SIZE
);
4624 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type
) != 64);
4625 BUILD_BUG_ON(sizeof(struct nvme_smart_log
) != 512);
4626 BUILD_BUG_ON(sizeof(struct nvme_dbbuf
) != 64);
4627 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd
) != 64);
4631 static int __init
nvme_core_init(void)
4633 int result
= -ENOMEM
;
4637 nvme_wq
= alloc_workqueue("nvme-wq",
4638 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4642 nvme_reset_wq
= alloc_workqueue("nvme-reset-wq",
4643 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4647 nvme_delete_wq
= alloc_workqueue("nvme-delete-wq",
4648 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4649 if (!nvme_delete_wq
)
4650 goto destroy_reset_wq
;
4652 result
= alloc_chrdev_region(&nvme_chr_devt
, 0, NVME_MINORS
, "nvme");
4654 goto destroy_delete_wq
;
4656 nvme_class
= class_create(THIS_MODULE
, "nvme");
4657 if (IS_ERR(nvme_class
)) {
4658 result
= PTR_ERR(nvme_class
);
4659 goto unregister_chrdev
;
4661 nvme_class
->dev_uevent
= nvme_class_uevent
;
4663 nvme_subsys_class
= class_create(THIS_MODULE
, "nvme-subsystem");
4664 if (IS_ERR(nvme_subsys_class
)) {
4665 result
= PTR_ERR(nvme_subsys_class
);
4671 class_destroy(nvme_class
);
4673 unregister_chrdev_region(nvme_chr_devt
, NVME_MINORS
);
4675 destroy_workqueue(nvme_delete_wq
);
4677 destroy_workqueue(nvme_reset_wq
);
4679 destroy_workqueue(nvme_wq
);
4684 static void __exit
nvme_core_exit(void)
4686 class_destroy(nvme_subsys_class
);
4687 class_destroy(nvme_class
);
4688 unregister_chrdev_region(nvme_chr_devt
, NVME_MINORS
);
4689 destroy_workqueue(nvme_delete_wq
);
4690 destroy_workqueue(nvme_reset_wq
);
4691 destroy_workqueue(nvme_wq
);
4692 ida_destroy(&nvme_instance_ida
);
4695 MODULE_LICENSE("GPL");
4696 MODULE_VERSION("1.0");
4697 module_init(nvme_core_init
);
4698 module_exit(nvme_core_exit
);