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");
60 static unsigned long apst_primary_timeout_ms
= 100;
61 module_param(apst_primary_timeout_ms
, ulong
, 0644);
62 MODULE_PARM_DESC(apst_primary_timeout_ms
,
63 "primary APST timeout in ms");
65 static unsigned long apst_secondary_timeout_ms
= 2000;
66 module_param(apst_secondary_timeout_ms
, ulong
, 0644);
67 MODULE_PARM_DESC(apst_secondary_timeout_ms
,
68 "secondary APST timeout in ms");
70 static unsigned long apst_primary_latency_tol_us
= 15000;
71 module_param(apst_primary_latency_tol_us
, ulong
, 0644);
72 MODULE_PARM_DESC(apst_primary_latency_tol_us
,
73 "primary APST latency tolerance in us");
75 static unsigned long apst_secondary_latency_tol_us
= 100000;
76 module_param(apst_secondary_latency_tol_us
, ulong
, 0644);
77 MODULE_PARM_DESC(apst_secondary_latency_tol_us
,
78 "secondary APST latency tolerance in us");
81 module_param(streams
, bool, 0644);
82 MODULE_PARM_DESC(streams
, "turn on support for Streams write directives");
85 * nvme_wq - hosts nvme related works that are not reset or delete
86 * nvme_reset_wq - hosts nvme reset works
87 * nvme_delete_wq - hosts nvme delete works
89 * nvme_wq will host works such as scan, aen handling, fw activation,
90 * keep-alive, periodic reconnects etc. nvme_reset_wq
91 * runs reset works which also flush works hosted on nvme_wq for
92 * serialization purposes. nvme_delete_wq host controller deletion
93 * works which flush reset works for serialization.
95 struct workqueue_struct
*nvme_wq
;
96 EXPORT_SYMBOL_GPL(nvme_wq
);
98 struct workqueue_struct
*nvme_reset_wq
;
99 EXPORT_SYMBOL_GPL(nvme_reset_wq
);
101 struct workqueue_struct
*nvme_delete_wq
;
102 EXPORT_SYMBOL_GPL(nvme_delete_wq
);
104 static LIST_HEAD(nvme_subsystems
);
105 static DEFINE_MUTEX(nvme_subsystems_lock
);
107 static DEFINE_IDA(nvme_instance_ida
);
108 static dev_t nvme_ctrl_base_chr_devt
;
109 static struct class *nvme_class
;
110 static struct class *nvme_subsys_class
;
112 static DEFINE_IDA(nvme_ns_chr_minor_ida
);
113 static dev_t nvme_ns_chr_devt
;
114 static struct class *nvme_ns_chr_class
;
116 static void nvme_put_subsystem(struct nvme_subsystem
*subsys
);
117 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
121 * Prepare a queue for teardown.
123 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
124 * the capacity to 0 after that to avoid blocking dispatchers that may be
125 * holding bd_butex. This will end buffered writers dirtying pages that can't
128 static void nvme_set_queue_dying(struct nvme_ns
*ns
)
130 if (test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
133 blk_set_queue_dying(ns
->queue
);
134 blk_mq_unquiesce_queue(ns
->queue
);
136 set_capacity_and_notify(ns
->disk
, 0);
139 void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
142 * Only new queue scan work when admin and IO queues are both alive
144 if (ctrl
->state
== NVME_CTRL_LIVE
&& ctrl
->tagset
)
145 queue_work(nvme_wq
, &ctrl
->scan_work
);
149 * Use this function to proceed with scheduling reset_work for a controller
150 * that had previously been set to the resetting state. This is intended for
151 * code paths that can't be interrupted by other reset attempts. A hot removal
152 * may prevent this from succeeding.
154 int nvme_try_sched_reset(struct nvme_ctrl
*ctrl
)
156 if (ctrl
->state
!= NVME_CTRL_RESETTING
)
158 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
162 EXPORT_SYMBOL_GPL(nvme_try_sched_reset
);
164 static void nvme_failfast_work(struct work_struct
*work
)
166 struct nvme_ctrl
*ctrl
= container_of(to_delayed_work(work
),
167 struct nvme_ctrl
, failfast_work
);
169 if (ctrl
->state
!= NVME_CTRL_CONNECTING
)
172 set_bit(NVME_CTRL_FAILFAST_EXPIRED
, &ctrl
->flags
);
173 dev_info(ctrl
->device
, "failfast expired\n");
174 nvme_kick_requeue_lists(ctrl
);
177 static inline void nvme_start_failfast_work(struct nvme_ctrl
*ctrl
)
179 if (!ctrl
->opts
|| ctrl
->opts
->fast_io_fail_tmo
== -1)
182 schedule_delayed_work(&ctrl
->failfast_work
,
183 ctrl
->opts
->fast_io_fail_tmo
* HZ
);
186 static inline void nvme_stop_failfast_work(struct nvme_ctrl
*ctrl
)
191 cancel_delayed_work_sync(&ctrl
->failfast_work
);
192 clear_bit(NVME_CTRL_FAILFAST_EXPIRED
, &ctrl
->flags
);
196 int nvme_reset_ctrl(struct nvme_ctrl
*ctrl
)
198 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
200 if (!queue_work(nvme_reset_wq
, &ctrl
->reset_work
))
204 EXPORT_SYMBOL_GPL(nvme_reset_ctrl
);
206 int nvme_reset_ctrl_sync(struct nvme_ctrl
*ctrl
)
210 ret
= nvme_reset_ctrl(ctrl
);
212 flush_work(&ctrl
->reset_work
);
213 if (ctrl
->state
!= NVME_CTRL_LIVE
)
220 static void nvme_do_delete_ctrl(struct nvme_ctrl
*ctrl
)
222 dev_info(ctrl
->device
,
223 "Removing ctrl: NQN \"%s\"\n", ctrl
->opts
->subsysnqn
);
225 flush_work(&ctrl
->reset_work
);
226 nvme_stop_ctrl(ctrl
);
227 nvme_remove_namespaces(ctrl
);
228 ctrl
->ops
->delete_ctrl(ctrl
);
229 nvme_uninit_ctrl(ctrl
);
232 static void nvme_delete_ctrl_work(struct work_struct
*work
)
234 struct nvme_ctrl
*ctrl
=
235 container_of(work
, struct nvme_ctrl
, delete_work
);
237 nvme_do_delete_ctrl(ctrl
);
240 int nvme_delete_ctrl(struct nvme_ctrl
*ctrl
)
242 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
244 if (!queue_work(nvme_delete_wq
, &ctrl
->delete_work
))
248 EXPORT_SYMBOL_GPL(nvme_delete_ctrl
);
250 static void nvme_delete_ctrl_sync(struct nvme_ctrl
*ctrl
)
253 * Keep a reference until nvme_do_delete_ctrl() complete,
254 * since ->delete_ctrl can free the controller.
257 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING
))
258 nvme_do_delete_ctrl(ctrl
);
262 static blk_status_t
nvme_error_status(u16 status
)
264 switch (status
& 0x7ff) {
265 case NVME_SC_SUCCESS
:
267 case NVME_SC_CAP_EXCEEDED
:
268 return BLK_STS_NOSPC
;
269 case NVME_SC_LBA_RANGE
:
270 case NVME_SC_CMD_INTERRUPTED
:
271 case NVME_SC_NS_NOT_READY
:
272 return BLK_STS_TARGET
;
273 case NVME_SC_BAD_ATTRIBUTES
:
274 case NVME_SC_ONCS_NOT_SUPPORTED
:
275 case NVME_SC_INVALID_OPCODE
:
276 case NVME_SC_INVALID_FIELD
:
277 case NVME_SC_INVALID_NS
:
278 return BLK_STS_NOTSUPP
;
279 case NVME_SC_WRITE_FAULT
:
280 case NVME_SC_READ_ERROR
:
281 case NVME_SC_UNWRITTEN_BLOCK
:
282 case NVME_SC_ACCESS_DENIED
:
283 case NVME_SC_READ_ONLY
:
284 case NVME_SC_COMPARE_FAILED
:
285 return BLK_STS_MEDIUM
;
286 case NVME_SC_GUARD_CHECK
:
287 case NVME_SC_APPTAG_CHECK
:
288 case NVME_SC_REFTAG_CHECK
:
289 case NVME_SC_INVALID_PI
:
290 return BLK_STS_PROTECTION
;
291 case NVME_SC_RESERVATION_CONFLICT
:
292 return BLK_STS_NEXUS
;
293 case NVME_SC_HOST_PATH_ERROR
:
294 return BLK_STS_TRANSPORT
;
295 case NVME_SC_ZONE_TOO_MANY_ACTIVE
:
296 return BLK_STS_ZONE_ACTIVE_RESOURCE
;
297 case NVME_SC_ZONE_TOO_MANY_OPEN
:
298 return BLK_STS_ZONE_OPEN_RESOURCE
;
300 return BLK_STS_IOERR
;
304 static void nvme_retry_req(struct request
*req
)
306 unsigned long delay
= 0;
309 /* The mask and shift result must be <= 3 */
310 crd
= (nvme_req(req
)->status
& NVME_SC_CRD
) >> 11;
312 delay
= nvme_req(req
)->ctrl
->crdt
[crd
- 1] * 100;
314 nvme_req(req
)->retries
++;
315 blk_mq_requeue_request(req
, false);
316 blk_mq_delay_kick_requeue_list(req
->q
, delay
);
319 enum nvme_disposition
{
325 static inline enum nvme_disposition
nvme_decide_disposition(struct request
*req
)
327 if (likely(nvme_req(req
)->status
== 0))
330 if (blk_noretry_request(req
) ||
331 (nvme_req(req
)->status
& NVME_SC_DNR
) ||
332 nvme_req(req
)->retries
>= nvme_max_retries
)
335 if (req
->cmd_flags
& REQ_NVME_MPATH
) {
336 if (nvme_is_path_error(nvme_req(req
)->status
) ||
337 blk_queue_dying(req
->q
))
340 if (blk_queue_dying(req
->q
))
347 static inline void nvme_end_req(struct request
*req
)
349 blk_status_t status
= nvme_error_status(nvme_req(req
)->status
);
351 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED
) &&
352 req_op(req
) == REQ_OP_ZONE_APPEND
)
353 req
->__sector
= nvme_lba_to_sect(req
->q
->queuedata
,
354 le64_to_cpu(nvme_req(req
)->result
.u64
));
356 nvme_trace_bio_complete(req
);
357 blk_mq_end_request(req
, status
);
360 void nvme_complete_rq(struct request
*req
)
362 trace_nvme_complete_rq(req
);
363 nvme_cleanup_cmd(req
);
365 if (nvme_req(req
)->ctrl
->kas
)
366 nvme_req(req
)->ctrl
->comp_seen
= true;
368 switch (nvme_decide_disposition(req
)) {
376 nvme_failover_req(req
);
380 EXPORT_SYMBOL_GPL(nvme_complete_rq
);
383 * Called to unwind from ->queue_rq on a failed command submission so that the
384 * multipathing code gets called to potentially failover to another path.
385 * The caller needs to unwind all transport specific resource allocations and
386 * must return propagate the return value.
388 blk_status_t
nvme_host_path_error(struct request
*req
)
390 nvme_req(req
)->status
= NVME_SC_HOST_PATH_ERROR
;
391 blk_mq_set_request_complete(req
);
392 nvme_complete_rq(req
);
395 EXPORT_SYMBOL_GPL(nvme_host_path_error
);
397 bool nvme_cancel_request(struct request
*req
, void *data
, bool reserved
)
399 dev_dbg_ratelimited(((struct nvme_ctrl
*) data
)->device
,
400 "Cancelling I/O %d", req
->tag
);
402 /* don't abort one completed request */
403 if (blk_mq_request_completed(req
))
406 nvme_req(req
)->status
= NVME_SC_HOST_ABORTED_CMD
;
407 nvme_req(req
)->flags
|= NVME_REQ_CANCELLED
;
408 blk_mq_complete_request(req
);
411 EXPORT_SYMBOL_GPL(nvme_cancel_request
);
413 void nvme_cancel_tagset(struct nvme_ctrl
*ctrl
)
416 blk_mq_tagset_busy_iter(ctrl
->tagset
,
417 nvme_cancel_request
, ctrl
);
418 blk_mq_tagset_wait_completed_request(ctrl
->tagset
);
421 EXPORT_SYMBOL_GPL(nvme_cancel_tagset
);
423 void nvme_cancel_admin_tagset(struct nvme_ctrl
*ctrl
)
425 if (ctrl
->admin_tagset
) {
426 blk_mq_tagset_busy_iter(ctrl
->admin_tagset
,
427 nvme_cancel_request
, ctrl
);
428 blk_mq_tagset_wait_completed_request(ctrl
->admin_tagset
);
431 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset
);
433 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
434 enum nvme_ctrl_state new_state
)
436 enum nvme_ctrl_state old_state
;
438 bool changed
= false;
440 spin_lock_irqsave(&ctrl
->lock
, flags
);
442 old_state
= ctrl
->state
;
447 case NVME_CTRL_RESETTING
:
448 case NVME_CTRL_CONNECTING
:
455 case NVME_CTRL_RESETTING
:
465 case NVME_CTRL_CONNECTING
:
468 case NVME_CTRL_RESETTING
:
475 case NVME_CTRL_DELETING
:
478 case NVME_CTRL_RESETTING
:
479 case NVME_CTRL_CONNECTING
:
486 case NVME_CTRL_DELETING_NOIO
:
488 case NVME_CTRL_DELETING
:
498 case NVME_CTRL_DELETING
:
510 ctrl
->state
= new_state
;
511 wake_up_all(&ctrl
->state_wq
);
514 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
518 if (ctrl
->state
== NVME_CTRL_LIVE
) {
519 if (old_state
== NVME_CTRL_CONNECTING
)
520 nvme_stop_failfast_work(ctrl
);
521 nvme_kick_requeue_lists(ctrl
);
522 } else if (ctrl
->state
== NVME_CTRL_CONNECTING
&&
523 old_state
== NVME_CTRL_RESETTING
) {
524 nvme_start_failfast_work(ctrl
);
528 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
531 * Returns true for sink states that can't ever transition back to live.
533 static bool nvme_state_terminal(struct nvme_ctrl
*ctrl
)
535 switch (ctrl
->state
) {
538 case NVME_CTRL_RESETTING
:
539 case NVME_CTRL_CONNECTING
:
541 case NVME_CTRL_DELETING
:
542 case NVME_CTRL_DELETING_NOIO
:
546 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl
->state
);
552 * Waits for the controller state to be resetting, or returns false if it is
553 * not possible to ever transition to that state.
555 bool nvme_wait_reset(struct nvme_ctrl
*ctrl
)
557 wait_event(ctrl
->state_wq
,
558 nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
) ||
559 nvme_state_terminal(ctrl
));
560 return ctrl
->state
== NVME_CTRL_RESETTING
;
562 EXPORT_SYMBOL_GPL(nvme_wait_reset
);
564 static void nvme_free_ns_head(struct kref
*ref
)
566 struct nvme_ns_head
*head
=
567 container_of(ref
, struct nvme_ns_head
, ref
);
569 nvme_mpath_remove_disk(head
);
570 ida_simple_remove(&head
->subsys
->ns_ida
, head
->instance
);
571 cleanup_srcu_struct(&head
->srcu
);
572 nvme_put_subsystem(head
->subsys
);
576 bool nvme_tryget_ns_head(struct nvme_ns_head
*head
)
578 return kref_get_unless_zero(&head
->ref
);
581 void nvme_put_ns_head(struct nvme_ns_head
*head
)
583 kref_put(&head
->ref
, nvme_free_ns_head
);
586 static void nvme_free_ns(struct kref
*kref
)
588 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
591 nvme_nvm_unregister(ns
);
594 nvme_put_ns_head(ns
->head
);
595 nvme_put_ctrl(ns
->ctrl
);
599 static inline bool nvme_get_ns(struct nvme_ns
*ns
)
601 return kref_get_unless_zero(&ns
->kref
);
604 void nvme_put_ns(struct nvme_ns
*ns
)
606 kref_put(&ns
->kref
, nvme_free_ns
);
608 EXPORT_SYMBOL_NS_GPL(nvme_put_ns
, NVME_TARGET_PASSTHRU
);
610 static inline void nvme_clear_nvme_request(struct request
*req
)
612 nvme_req(req
)->retries
= 0;
613 nvme_req(req
)->flags
= 0;
614 req
->rq_flags
|= RQF_DONTPREP
;
617 static inline unsigned int nvme_req_op(struct nvme_command
*cmd
)
619 return nvme_is_write(cmd
) ? REQ_OP_DRV_OUT
: REQ_OP_DRV_IN
;
622 static inline void nvme_init_request(struct request
*req
,
623 struct nvme_command
*cmd
)
625 if (req
->q
->queuedata
)
626 req
->timeout
= NVME_IO_TIMEOUT
;
627 else /* no queuedata implies admin queue */
628 req
->timeout
= NVME_ADMIN_TIMEOUT
;
630 /* passthru commands should let the driver set the SGL flags */
631 cmd
->common
.flags
&= ~NVME_CMD_SGL_ALL
;
633 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
634 nvme_clear_nvme_request(req
);
635 memcpy(nvme_req(req
)->cmd
, cmd
, sizeof(*cmd
));
638 struct request
*nvme_alloc_request(struct request_queue
*q
,
639 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
)
643 req
= blk_mq_alloc_request(q
, nvme_req_op(cmd
), flags
);
645 nvme_init_request(req
, cmd
);
648 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
650 static struct request
*nvme_alloc_request_qid(struct request_queue
*q
,
651 struct nvme_command
*cmd
, blk_mq_req_flags_t flags
, int qid
)
655 req
= blk_mq_alloc_request_hctx(q
, nvme_req_op(cmd
), flags
,
658 nvme_init_request(req
, cmd
);
663 * For something we're not in a state to send to the device the default action
664 * is to busy it and retry it after the controller state is recovered. However,
665 * if the controller is deleting or if anything is marked for failfast or
666 * nvme multipath it is immediately failed.
668 * Note: commands used to initialize the controller will be marked for failfast.
669 * Note: nvme cli/ioctl commands are marked for failfast.
671 blk_status_t
nvme_fail_nonready_command(struct nvme_ctrl
*ctrl
,
674 if (ctrl
->state
!= NVME_CTRL_DELETING_NOIO
&&
675 ctrl
->state
!= NVME_CTRL_DEAD
&&
676 !test_bit(NVME_CTRL_FAILFAST_EXPIRED
, &ctrl
->flags
) &&
677 !blk_noretry_request(rq
) && !(rq
->cmd_flags
& REQ_NVME_MPATH
))
678 return BLK_STS_RESOURCE
;
679 return nvme_host_path_error(rq
);
681 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command
);
683 bool __nvme_check_ready(struct nvme_ctrl
*ctrl
, struct request
*rq
,
686 struct nvme_request
*req
= nvme_req(rq
);
689 * currently we have a problem sending passthru commands
690 * on the admin_q if the controller is not LIVE because we can't
691 * make sure that they are going out after the admin connect,
692 * controller enable and/or other commands in the initialization
693 * sequence. until the controller will be LIVE, fail with
694 * BLK_STS_RESOURCE so that they will be rescheduled.
696 if (rq
->q
== ctrl
->admin_q
&& (req
->flags
& NVME_REQ_USERCMD
))
699 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
701 * Only allow commands on a live queue, except for the connect
702 * command, which is require to set the queue live in the
703 * appropinquate states.
705 switch (ctrl
->state
) {
706 case NVME_CTRL_CONNECTING
:
707 if (blk_rq_is_passthrough(rq
) && nvme_is_fabrics(req
->cmd
) &&
708 req
->cmd
->fabrics
.fctype
== nvme_fabrics_type_connect
)
720 EXPORT_SYMBOL_GPL(__nvme_check_ready
);
722 static int nvme_toggle_streams(struct nvme_ctrl
*ctrl
, bool enable
)
724 struct nvme_command c
= { };
726 c
.directive
.opcode
= nvme_admin_directive_send
;
727 c
.directive
.nsid
= cpu_to_le32(NVME_NSID_ALL
);
728 c
.directive
.doper
= NVME_DIR_SND_ID_OP_ENABLE
;
729 c
.directive
.dtype
= NVME_DIR_IDENTIFY
;
730 c
.directive
.tdtype
= NVME_DIR_STREAMS
;
731 c
.directive
.endir
= enable
? NVME_DIR_ENDIR
: 0;
733 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, NULL
, 0);
736 static int nvme_disable_streams(struct nvme_ctrl
*ctrl
)
738 return nvme_toggle_streams(ctrl
, false);
741 static int nvme_enable_streams(struct nvme_ctrl
*ctrl
)
743 return nvme_toggle_streams(ctrl
, true);
746 static int nvme_get_stream_params(struct nvme_ctrl
*ctrl
,
747 struct streams_directive_params
*s
, u32 nsid
)
749 struct nvme_command c
= { };
751 memset(s
, 0, sizeof(*s
));
753 c
.directive
.opcode
= nvme_admin_directive_recv
;
754 c
.directive
.nsid
= cpu_to_le32(nsid
);
755 c
.directive
.numd
= cpu_to_le32(nvme_bytes_to_numd(sizeof(*s
)));
756 c
.directive
.doper
= NVME_DIR_RCV_ST_OP_PARAM
;
757 c
.directive
.dtype
= NVME_DIR_STREAMS
;
759 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, s
, sizeof(*s
));
762 static int nvme_configure_directives(struct nvme_ctrl
*ctrl
)
764 struct streams_directive_params s
;
767 if (!(ctrl
->oacs
& NVME_CTRL_OACS_DIRECTIVES
))
772 ret
= nvme_enable_streams(ctrl
);
776 ret
= nvme_get_stream_params(ctrl
, &s
, NVME_NSID_ALL
);
778 goto out_disable_stream
;
780 ctrl
->nssa
= le16_to_cpu(s
.nssa
);
781 if (ctrl
->nssa
< BLK_MAX_WRITE_HINTS
- 1) {
782 dev_info(ctrl
->device
, "too few streams (%u) available\n",
784 goto out_disable_stream
;
787 ctrl
->nr_streams
= min_t(u16
, ctrl
->nssa
, BLK_MAX_WRITE_HINTS
- 1);
788 dev_info(ctrl
->device
, "Using %u streams\n", ctrl
->nr_streams
);
792 nvme_disable_streams(ctrl
);
797 * Check if 'req' has a write hint associated with it. If it does, assign
798 * a valid namespace stream to the write.
800 static void nvme_assign_write_stream(struct nvme_ctrl
*ctrl
,
801 struct request
*req
, u16
*control
,
804 enum rw_hint streamid
= req
->write_hint
;
806 if (streamid
== WRITE_LIFE_NOT_SET
|| streamid
== WRITE_LIFE_NONE
)
810 if (WARN_ON_ONCE(streamid
> ctrl
->nr_streams
))
813 *control
|= NVME_RW_DTYPE_STREAMS
;
814 *dsmgmt
|= streamid
<< 16;
817 if (streamid
< ARRAY_SIZE(req
->q
->write_hints
))
818 req
->q
->write_hints
[streamid
] += blk_rq_bytes(req
) >> 9;
821 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
822 struct nvme_command
*cmnd
)
824 cmnd
->common
.opcode
= nvme_cmd_flush
;
825 cmnd
->common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
828 static blk_status_t
nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
829 struct nvme_command
*cmnd
)
831 unsigned short segments
= blk_rq_nr_discard_segments(req
), n
= 0;
832 struct nvme_dsm_range
*range
;
836 * Some devices do not consider the DSM 'Number of Ranges' field when
837 * determining how much data to DMA. Always allocate memory for maximum
838 * number of segments to prevent device reading beyond end of buffer.
840 static const size_t alloc_size
= sizeof(*range
) * NVME_DSM_MAX_RANGES
;
842 range
= kzalloc(alloc_size
, GFP_ATOMIC
| __GFP_NOWARN
);
845 * If we fail allocation our range, fallback to the controller
846 * discard page. If that's also busy, it's safe to return
847 * busy, as we know we can make progress once that's freed.
849 if (test_and_set_bit_lock(0, &ns
->ctrl
->discard_page_busy
))
850 return BLK_STS_RESOURCE
;
852 range
= page_address(ns
->ctrl
->discard_page
);
855 __rq_for_each_bio(bio
, req
) {
856 u64 slba
= nvme_sect_to_lba(ns
, bio
->bi_iter
.bi_sector
);
857 u32 nlb
= bio
->bi_iter
.bi_size
>> ns
->lba_shift
;
860 range
[n
].cattr
= cpu_to_le32(0);
861 range
[n
].nlb
= cpu_to_le32(nlb
);
862 range
[n
].slba
= cpu_to_le64(slba
);
867 if (WARN_ON_ONCE(n
!= segments
)) {
868 if (virt_to_page(range
) == ns
->ctrl
->discard_page
)
869 clear_bit_unlock(0, &ns
->ctrl
->discard_page_busy
);
872 return BLK_STS_IOERR
;
875 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
876 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
877 cmnd
->dsm
.nr
= cpu_to_le32(segments
- 1);
878 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
880 req
->special_vec
.bv_page
= virt_to_page(range
);
881 req
->special_vec
.bv_offset
= offset_in_page(range
);
882 req
->special_vec
.bv_len
= alloc_size
;
883 req
->rq_flags
|= RQF_SPECIAL_PAYLOAD
;
888 static inline blk_status_t
nvme_setup_write_zeroes(struct nvme_ns
*ns
,
889 struct request
*req
, struct nvme_command
*cmnd
)
891 if (ns
->ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
892 return nvme_setup_discard(ns
, req
, cmnd
);
894 cmnd
->write_zeroes
.opcode
= nvme_cmd_write_zeroes
;
895 cmnd
->write_zeroes
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
896 cmnd
->write_zeroes
.slba
=
897 cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
898 cmnd
->write_zeroes
.length
=
899 cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
900 cmnd
->write_zeroes
.control
= 0;
904 static inline blk_status_t
nvme_setup_rw(struct nvme_ns
*ns
,
905 struct request
*req
, struct nvme_command
*cmnd
,
908 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
912 if (req
->cmd_flags
& REQ_FUA
)
913 control
|= NVME_RW_FUA
;
914 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
915 control
|= NVME_RW_LR
;
917 if (req
->cmd_flags
& REQ_RAHEAD
)
918 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
920 cmnd
->rw
.opcode
= op
;
921 cmnd
->rw
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
922 cmnd
->rw
.slba
= cpu_to_le64(nvme_sect_to_lba(ns
, blk_rq_pos(req
)));
923 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
925 if (req_op(req
) == REQ_OP_WRITE
&& ctrl
->nr_streams
)
926 nvme_assign_write_stream(ctrl
, req
, &control
, &dsmgmt
);
930 * If formated with metadata, the block layer always provides a
931 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
932 * we enable the PRACT bit for protection information or set the
933 * namespace capacity to zero to prevent any I/O.
935 if (!blk_integrity_rq(req
)) {
936 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns
)))
937 return BLK_STS_NOTSUPP
;
938 control
|= NVME_RW_PRINFO_PRACT
;
941 switch (ns
->pi_type
) {
942 case NVME_NS_DPS_PI_TYPE3
:
943 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
945 case NVME_NS_DPS_PI_TYPE1
:
946 case NVME_NS_DPS_PI_TYPE2
:
947 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
948 NVME_RW_PRINFO_PRCHK_REF
;
949 if (op
== nvme_cmd_zone_append
)
950 control
|= NVME_RW_APPEND_PIREMAP
;
951 cmnd
->rw
.reftag
= cpu_to_le32(t10_pi_ref_tag(req
));
956 cmnd
->rw
.control
= cpu_to_le16(control
);
957 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
961 void nvme_cleanup_cmd(struct request
*req
)
963 if (req
->rq_flags
& RQF_SPECIAL_PAYLOAD
) {
964 struct nvme_ctrl
*ctrl
= nvme_req(req
)->ctrl
;
965 struct page
*page
= req
->special_vec
.bv_page
;
967 if (page
== ctrl
->discard_page
)
968 clear_bit_unlock(0, &ctrl
->discard_page_busy
);
970 kfree(page_address(page
) + req
->special_vec
.bv_offset
);
973 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd
);
975 blk_status_t
nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
)
977 struct nvme_command
*cmd
= nvme_req(req
)->cmd
;
978 blk_status_t ret
= BLK_STS_OK
;
980 if (!(req
->rq_flags
& RQF_DONTPREP
)) {
981 nvme_clear_nvme_request(req
);
982 memset(cmd
, 0, sizeof(*cmd
));
985 switch (req_op(req
)) {
988 /* these are setup prior to execution in nvme_init_request() */
991 nvme_setup_flush(ns
, cmd
);
993 case REQ_OP_ZONE_RESET_ALL
:
994 case REQ_OP_ZONE_RESET
:
995 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_RESET
);
997 case REQ_OP_ZONE_OPEN
:
998 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_OPEN
);
1000 case REQ_OP_ZONE_CLOSE
:
1001 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_CLOSE
);
1003 case REQ_OP_ZONE_FINISH
:
1004 ret
= nvme_setup_zone_mgmt_send(ns
, req
, cmd
, NVME_ZONE_FINISH
);
1006 case REQ_OP_WRITE_ZEROES
:
1007 ret
= nvme_setup_write_zeroes(ns
, req
, cmd
);
1009 case REQ_OP_DISCARD
:
1010 ret
= nvme_setup_discard(ns
, req
, cmd
);
1013 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_read
);
1016 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_write
);
1018 case REQ_OP_ZONE_APPEND
:
1019 ret
= nvme_setup_rw(ns
, req
, cmd
, nvme_cmd_zone_append
);
1023 return BLK_STS_IOERR
;
1026 cmd
->common
.command_id
= req
->tag
;
1027 trace_nvme_setup_cmd(req
, cmd
);
1030 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
1032 static void nvme_end_sync_rq(struct request
*rq
, blk_status_t error
)
1034 struct completion
*waiting
= rq
->end_io_data
;
1036 rq
->end_io_data
= NULL
;
1040 static void nvme_execute_rq_polled(struct request_queue
*q
,
1041 struct gendisk
*bd_disk
, struct request
*rq
, int at_head
)
1043 DECLARE_COMPLETION_ONSTACK(wait
);
1045 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL
, &q
->queue_flags
));
1047 rq
->cmd_flags
|= REQ_HIPRI
;
1048 rq
->end_io_data
= &wait
;
1049 blk_execute_rq_nowait(bd_disk
, rq
, at_head
, nvme_end_sync_rq
);
1051 while (!completion_done(&wait
)) {
1052 blk_poll(q
, request_to_qc_t(rq
->mq_hctx
, rq
), true);
1058 * Returns 0 on success. If the result is negative, it's a Linux error code;
1059 * if the result is positive, it's an NVM Express status code
1061 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
1062 union nvme_result
*result
, void *buffer
, unsigned bufflen
,
1063 unsigned timeout
, int qid
, int at_head
,
1064 blk_mq_req_flags_t flags
, bool poll
)
1066 struct request
*req
;
1069 if (qid
== NVME_QID_ANY
)
1070 req
= nvme_alloc_request(q
, cmd
, flags
);
1072 req
= nvme_alloc_request_qid(q
, cmd
, flags
, qid
);
1074 return PTR_ERR(req
);
1077 req
->timeout
= timeout
;
1079 if (buffer
&& bufflen
) {
1080 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
1086 nvme_execute_rq_polled(req
->q
, NULL
, req
, at_head
);
1088 blk_execute_rq(NULL
, req
, at_head
);
1090 *result
= nvme_req(req
)->result
;
1091 if (nvme_req(req
)->flags
& NVME_REQ_CANCELLED
)
1094 ret
= nvme_req(req
)->status
;
1096 blk_mq_free_request(req
);
1099 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd
);
1101 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
1102 void *buffer
, unsigned bufflen
)
1104 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0,
1105 NVME_QID_ANY
, 0, 0, false);
1107 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
1109 static u32
nvme_known_admin_effects(u8 opcode
)
1112 case nvme_admin_format_nvm
:
1113 return NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_NCC
|
1114 NVME_CMD_EFFECTS_CSE_MASK
;
1115 case nvme_admin_sanitize_nvm
:
1116 return NVME_CMD_EFFECTS_LBCC
| NVME_CMD_EFFECTS_CSE_MASK
;
1123 u32
nvme_command_effects(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
, u8 opcode
)
1128 if (ns
->head
->effects
)
1129 effects
= le32_to_cpu(ns
->head
->effects
->iocs
[opcode
]);
1130 if (effects
& ~(NVME_CMD_EFFECTS_CSUPP
| NVME_CMD_EFFECTS_LBCC
))
1131 dev_warn_once(ctrl
->device
,
1132 "IO command:%02x has unhandled effects:%08x\n",
1138 effects
= le32_to_cpu(ctrl
->effects
->acs
[opcode
]);
1139 effects
|= nvme_known_admin_effects(opcode
);
1143 EXPORT_SYMBOL_NS_GPL(nvme_command_effects
, NVME_TARGET_PASSTHRU
);
1145 static u32
nvme_passthru_start(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1148 u32 effects
= nvme_command_effects(ctrl
, ns
, opcode
);
1151 * For simplicity, IO to all namespaces is quiesced even if the command
1152 * effects say only one namespace is affected.
1154 if (effects
& NVME_CMD_EFFECTS_CSE_MASK
) {
1155 mutex_lock(&ctrl
->scan_lock
);
1156 mutex_lock(&ctrl
->subsys
->lock
);
1157 nvme_mpath_start_freeze(ctrl
->subsys
);
1158 nvme_mpath_wait_freeze(ctrl
->subsys
);
1159 nvme_start_freeze(ctrl
);
1160 nvme_wait_freeze(ctrl
);
1165 static void nvme_passthru_end(struct nvme_ctrl
*ctrl
, u32 effects
)
1167 if (effects
& NVME_CMD_EFFECTS_CSE_MASK
) {
1168 nvme_unfreeze(ctrl
);
1169 nvme_mpath_unfreeze(ctrl
->subsys
);
1170 mutex_unlock(&ctrl
->subsys
->lock
);
1171 nvme_remove_invalid_namespaces(ctrl
, NVME_NSID_ALL
);
1172 mutex_unlock(&ctrl
->scan_lock
);
1174 if (effects
& NVME_CMD_EFFECTS_CCC
)
1175 nvme_init_ctrl_finish(ctrl
);
1176 if (effects
& (NVME_CMD_EFFECTS_NIC
| NVME_CMD_EFFECTS_NCC
)) {
1177 nvme_queue_scan(ctrl
);
1178 flush_work(&ctrl
->scan_work
);
1182 void nvme_execute_passthru_rq(struct request
*rq
)
1184 struct nvme_command
*cmd
= nvme_req(rq
)->cmd
;
1185 struct nvme_ctrl
*ctrl
= nvme_req(rq
)->ctrl
;
1186 struct nvme_ns
*ns
= rq
->q
->queuedata
;
1187 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
1190 effects
= nvme_passthru_start(ctrl
, ns
, cmd
->common
.opcode
);
1191 blk_execute_rq(disk
, rq
, 0);
1192 if (effects
) /* nothing to be done for zero cmd effects */
1193 nvme_passthru_end(ctrl
, effects
);
1195 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq
, NVME_TARGET_PASSTHRU
);
1198 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1200 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1201 * accounting for transport roundtrip times [..].
1203 static void nvme_queue_keep_alive_work(struct nvme_ctrl
*ctrl
)
1205 queue_delayed_work(nvme_wq
, &ctrl
->ka_work
, ctrl
->kato
* HZ
/ 2);
1208 static void nvme_keep_alive_end_io(struct request
*rq
, blk_status_t status
)
1210 struct nvme_ctrl
*ctrl
= rq
->end_io_data
;
1211 unsigned long flags
;
1212 bool startka
= false;
1214 blk_mq_free_request(rq
);
1217 dev_err(ctrl
->device
,
1218 "failed nvme_keep_alive_end_io error=%d\n",
1223 ctrl
->comp_seen
= false;
1224 spin_lock_irqsave(&ctrl
->lock
, flags
);
1225 if (ctrl
->state
== NVME_CTRL_LIVE
||
1226 ctrl
->state
== NVME_CTRL_CONNECTING
)
1228 spin_unlock_irqrestore(&ctrl
->lock
, flags
);
1230 nvme_queue_keep_alive_work(ctrl
);
1233 static void nvme_keep_alive_work(struct work_struct
*work
)
1235 struct nvme_ctrl
*ctrl
= container_of(to_delayed_work(work
),
1236 struct nvme_ctrl
, ka_work
);
1237 bool comp_seen
= ctrl
->comp_seen
;
1240 if ((ctrl
->ctratt
& NVME_CTRL_ATTR_TBKAS
) && comp_seen
) {
1241 dev_dbg(ctrl
->device
,
1242 "reschedule traffic based keep-alive timer\n");
1243 ctrl
->comp_seen
= false;
1244 nvme_queue_keep_alive_work(ctrl
);
1248 rq
= nvme_alloc_request(ctrl
->admin_q
, &ctrl
->ka_cmd
,
1249 BLK_MQ_REQ_RESERVED
| BLK_MQ_REQ_NOWAIT
);
1251 /* allocation failure, reset the controller */
1252 dev_err(ctrl
->device
, "keep-alive failed: %ld\n", PTR_ERR(rq
));
1253 nvme_reset_ctrl(ctrl
);
1257 rq
->timeout
= ctrl
->kato
* HZ
;
1258 rq
->end_io_data
= ctrl
;
1259 blk_execute_rq_nowait(NULL
, rq
, 0, nvme_keep_alive_end_io
);
1262 static void nvme_start_keep_alive(struct nvme_ctrl
*ctrl
)
1264 if (unlikely(ctrl
->kato
== 0))
1267 nvme_queue_keep_alive_work(ctrl
);
1270 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
)
1272 if (unlikely(ctrl
->kato
== 0))
1275 cancel_delayed_work_sync(&ctrl
->ka_work
);
1277 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive
);
1280 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1281 * flag, thus sending any new CNS opcodes has a big chance of not working.
1282 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1283 * (but not for any later version).
1285 static bool nvme_ctrl_limited_cns(struct nvme_ctrl
*ctrl
)
1287 if (ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)
1288 return ctrl
->vs
< NVME_VS(1, 2, 0);
1289 return ctrl
->vs
< NVME_VS(1, 1, 0);
1292 static int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
1294 struct nvme_command c
= { };
1297 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1298 c
.identify
.opcode
= nvme_admin_identify
;
1299 c
.identify
.cns
= NVME_ID_CNS_CTRL
;
1301 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
1305 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
1306 sizeof(struct nvme_id_ctrl
));
1312 static bool nvme_multi_css(struct nvme_ctrl
*ctrl
)
1314 return (ctrl
->ctrl_config
& NVME_CC_CSS_MASK
) == NVME_CC_CSS_CSI
;
1317 static int nvme_process_ns_desc(struct nvme_ctrl
*ctrl
, struct nvme_ns_ids
*ids
,
1318 struct nvme_ns_id_desc
*cur
, bool *csi_seen
)
1320 const char *warn_str
= "ctrl returned bogus length:";
1323 switch (cur
->nidt
) {
1324 case NVME_NIDT_EUI64
:
1325 if (cur
->nidl
!= NVME_NIDT_EUI64_LEN
) {
1326 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_EUI64\n",
1327 warn_str
, cur
->nidl
);
1330 memcpy(ids
->eui64
, data
+ sizeof(*cur
), NVME_NIDT_EUI64_LEN
);
1331 return NVME_NIDT_EUI64_LEN
;
1332 case NVME_NIDT_NGUID
:
1333 if (cur
->nidl
!= NVME_NIDT_NGUID_LEN
) {
1334 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_NGUID\n",
1335 warn_str
, cur
->nidl
);
1338 memcpy(ids
->nguid
, data
+ sizeof(*cur
), NVME_NIDT_NGUID_LEN
);
1339 return NVME_NIDT_NGUID_LEN
;
1340 case NVME_NIDT_UUID
:
1341 if (cur
->nidl
!= NVME_NIDT_UUID_LEN
) {
1342 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_UUID\n",
1343 warn_str
, cur
->nidl
);
1346 uuid_copy(&ids
->uuid
, data
+ sizeof(*cur
));
1347 return NVME_NIDT_UUID_LEN
;
1349 if (cur
->nidl
!= NVME_NIDT_CSI_LEN
) {
1350 dev_warn(ctrl
->device
, "%s %d for NVME_NIDT_CSI\n",
1351 warn_str
, cur
->nidl
);
1354 memcpy(&ids
->csi
, data
+ sizeof(*cur
), NVME_NIDT_CSI_LEN
);
1356 return NVME_NIDT_CSI_LEN
;
1358 /* Skip unknown types */
1363 static int nvme_identify_ns_descs(struct nvme_ctrl
*ctrl
, unsigned nsid
,
1364 struct nvme_ns_ids
*ids
)
1366 struct nvme_command c
= { };
1367 bool csi_seen
= false;
1368 int status
, pos
, len
;
1371 if (ctrl
->vs
< NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl
))
1373 if (ctrl
->quirks
& NVME_QUIRK_NO_NS_DESC_LIST
)
1376 c
.identify
.opcode
= nvme_admin_identify
;
1377 c
.identify
.nsid
= cpu_to_le32(nsid
);
1378 c
.identify
.cns
= NVME_ID_CNS_NS_DESC_LIST
;
1380 data
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
1384 status
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, data
,
1385 NVME_IDENTIFY_DATA_SIZE
);
1387 dev_warn(ctrl
->device
,
1388 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1393 for (pos
= 0; pos
< NVME_IDENTIFY_DATA_SIZE
; pos
+= len
) {
1394 struct nvme_ns_id_desc
*cur
= data
+ pos
;
1399 len
= nvme_process_ns_desc(ctrl
, ids
, cur
, &csi_seen
);
1403 len
+= sizeof(*cur
);
1406 if (nvme_multi_css(ctrl
) && !csi_seen
) {
1407 dev_warn(ctrl
->device
, "Command set not reported for nsid:%d\n",
1417 static int nvme_identify_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
,
1418 struct nvme_ns_ids
*ids
, struct nvme_id_ns
**id
)
1420 struct nvme_command c
= { };
1423 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1424 c
.identify
.opcode
= nvme_admin_identify
;
1425 c
.identify
.nsid
= cpu_to_le32(nsid
);
1426 c
.identify
.cns
= NVME_ID_CNS_NS
;
1428 *id
= kmalloc(sizeof(**id
), GFP_KERNEL
);
1432 error
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, *id
, sizeof(**id
));
1434 dev_warn(ctrl
->device
, "Identify namespace failed (%d)\n", error
);
1438 error
= NVME_SC_INVALID_NS
| NVME_SC_DNR
;
1439 if ((*id
)->ncap
== 0) /* namespace not allocated or attached */
1442 if (ctrl
->vs
>= NVME_VS(1, 1, 0) &&
1443 !memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
1444 memcpy(ids
->eui64
, (*id
)->eui64
, sizeof(ids
->eui64
));
1445 if (ctrl
->vs
>= NVME_VS(1, 2, 0) &&
1446 !memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
1447 memcpy(ids
->nguid
, (*id
)->nguid
, sizeof(ids
->nguid
));
1456 static int nvme_features(struct nvme_ctrl
*dev
, u8 op
, unsigned int fid
,
1457 unsigned int dword11
, void *buffer
, size_t buflen
, u32
*result
)
1459 union nvme_result res
= { 0 };
1460 struct nvme_command c
= { };
1463 c
.features
.opcode
= op
;
1464 c
.features
.fid
= cpu_to_le32(fid
);
1465 c
.features
.dword11
= cpu_to_le32(dword11
);
1467 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &res
,
1468 buffer
, buflen
, 0, NVME_QID_ANY
, 0, 0, false);
1469 if (ret
>= 0 && result
)
1470 *result
= le32_to_cpu(res
.u32
);
1474 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1475 unsigned int dword11
, void *buffer
, size_t buflen
,
1478 return nvme_features(dev
, nvme_admin_set_features
, fid
, dword11
, buffer
,
1481 EXPORT_SYMBOL_GPL(nvme_set_features
);
1483 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned int fid
,
1484 unsigned int dword11
, void *buffer
, size_t buflen
,
1487 return nvme_features(dev
, nvme_admin_get_features
, fid
, dword11
, buffer
,
1490 EXPORT_SYMBOL_GPL(nvme_get_features
);
1492 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
1494 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
1496 int status
, nr_io_queues
;
1498 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, NULL
, 0,
1504 * Degraded controllers might return an error when setting the queue
1505 * count. We still want to be able to bring them online and offer
1506 * access to the admin queue, as that might be only way to fix them up.
1509 dev_err(ctrl
->device
, "Could not set queue count (%d)\n", status
);
1512 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
1513 *count
= min(*count
, nr_io_queues
);
1518 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
1520 #define NVME_AEN_SUPPORTED \
1521 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1522 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1524 static void nvme_enable_aen(struct nvme_ctrl
*ctrl
)
1526 u32 result
, supported_aens
= ctrl
->oaes
& NVME_AEN_SUPPORTED
;
1529 if (!supported_aens
)
1532 status
= nvme_set_features(ctrl
, NVME_FEAT_ASYNC_EVENT
, supported_aens
,
1535 dev_warn(ctrl
->device
, "Failed to configure AEN (cfg %x)\n",
1538 queue_work(nvme_wq
, &ctrl
->async_event_work
);
1541 static int nvme_ns_open(struct nvme_ns
*ns
)
1544 /* should never be called due to GENHD_FL_HIDDEN */
1545 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns
->head
)))
1547 if (!nvme_get_ns(ns
))
1549 if (!try_module_get(ns
->ctrl
->ops
->module
))
1560 static void nvme_ns_release(struct nvme_ns
*ns
)
1563 module_put(ns
->ctrl
->ops
->module
);
1567 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
1569 return nvme_ns_open(bdev
->bd_disk
->private_data
);
1572 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
1574 nvme_ns_release(disk
->private_data
);
1577 int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
1579 /* some standard values */
1580 geo
->heads
= 1 << 6;
1581 geo
->sectors
= 1 << 5;
1582 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
1586 #ifdef CONFIG_BLK_DEV_INTEGRITY
1587 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1588 u32 max_integrity_segments
)
1590 struct blk_integrity integrity
= { };
1593 case NVME_NS_DPS_PI_TYPE3
:
1594 integrity
.profile
= &t10_pi_type3_crc
;
1595 integrity
.tag_size
= sizeof(u16
) + sizeof(u32
);
1596 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1598 case NVME_NS_DPS_PI_TYPE1
:
1599 case NVME_NS_DPS_PI_TYPE2
:
1600 integrity
.profile
= &t10_pi_type1_crc
;
1601 integrity
.tag_size
= sizeof(u16
);
1602 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
1605 integrity
.profile
= NULL
;
1608 integrity
.tuple_size
= ms
;
1609 blk_integrity_register(disk
, &integrity
);
1610 blk_queue_max_integrity_segments(disk
->queue
, max_integrity_segments
);
1613 static void nvme_init_integrity(struct gendisk
*disk
, u16 ms
, u8 pi_type
,
1614 u32 max_integrity_segments
)
1617 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1619 static void nvme_config_discard(struct gendisk
*disk
, struct nvme_ns
*ns
)
1621 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1622 struct request_queue
*queue
= disk
->queue
;
1623 u32 size
= queue_logical_block_size(queue
);
1625 if (ctrl
->max_discard_sectors
== 0) {
1626 blk_queue_flag_clear(QUEUE_FLAG_DISCARD
, queue
);
1630 if (ctrl
->nr_streams
&& ns
->sws
&& ns
->sgs
)
1631 size
*= ns
->sws
* ns
->sgs
;
1633 BUILD_BUG_ON(PAGE_SIZE
/ sizeof(struct nvme_dsm_range
) <
1634 NVME_DSM_MAX_RANGES
);
1636 queue
->limits
.discard_alignment
= 0;
1637 queue
->limits
.discard_granularity
= size
;
1639 /* If discard is already enabled, don't reset queue limits */
1640 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD
, queue
))
1643 blk_queue_max_discard_sectors(queue
, ctrl
->max_discard_sectors
);
1644 blk_queue_max_discard_segments(queue
, ctrl
->max_discard_segments
);
1646 if (ctrl
->quirks
& NVME_QUIRK_DEALLOCATE_ZEROES
)
1647 blk_queue_max_write_zeroes_sectors(queue
, UINT_MAX
);
1650 static bool nvme_ns_ids_valid(struct nvme_ns_ids
*ids
)
1652 return !uuid_is_null(&ids
->uuid
) ||
1653 memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)) ||
1654 memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
));
1657 static bool nvme_ns_ids_equal(struct nvme_ns_ids
*a
, struct nvme_ns_ids
*b
)
1659 return uuid_equal(&a
->uuid
, &b
->uuid
) &&
1660 memcmp(&a
->nguid
, &b
->nguid
, sizeof(a
->nguid
)) == 0 &&
1661 memcmp(&a
->eui64
, &b
->eui64
, sizeof(a
->eui64
)) == 0 &&
1665 static int nvme_setup_streams_ns(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
1666 u32
*phys_bs
, u32
*io_opt
)
1668 struct streams_directive_params s
;
1671 if (!ctrl
->nr_streams
)
1674 ret
= nvme_get_stream_params(ctrl
, &s
, ns
->head
->ns_id
);
1678 ns
->sws
= le32_to_cpu(s
.sws
);
1679 ns
->sgs
= le16_to_cpu(s
.sgs
);
1682 *phys_bs
= ns
->sws
* (1 << ns
->lba_shift
);
1684 *io_opt
= *phys_bs
* ns
->sgs
;
1690 static int nvme_configure_metadata(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1692 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1695 * The PI implementation requires the metadata size to be equal to the
1696 * t10 pi tuple size.
1698 ns
->ms
= le16_to_cpu(id
->lbaf
[id
->flbas
& NVME_NS_FLBAS_LBA_MASK
].ms
);
1699 if (ns
->ms
== sizeof(struct t10_pi_tuple
))
1700 ns
->pi_type
= id
->dps
& NVME_NS_DPS_PI_MASK
;
1704 ns
->features
&= ~(NVME_NS_METADATA_SUPPORTED
| NVME_NS_EXT_LBAS
);
1705 if (!ns
->ms
|| !(ctrl
->ops
->flags
& NVME_F_METADATA_SUPPORTED
))
1707 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
1709 * The NVMe over Fabrics specification only supports metadata as
1710 * part of the extended data LBA. We rely on HCA/HBA support to
1711 * remap the separate metadata buffer from the block layer.
1713 if (WARN_ON_ONCE(!(id
->flbas
& NVME_NS_FLBAS_META_EXT
)))
1715 if (ctrl
->max_integrity_segments
)
1717 (NVME_NS_METADATA_SUPPORTED
| NVME_NS_EXT_LBAS
);
1720 * For PCIe controllers, we can't easily remap the separate
1721 * metadata buffer from the block layer and thus require a
1722 * separate metadata buffer for block layer metadata/PI support.
1723 * We allow extended LBAs for the passthrough interface, though.
1725 if (id
->flbas
& NVME_NS_FLBAS_META_EXT
)
1726 ns
->features
|= NVME_NS_EXT_LBAS
;
1728 ns
->features
|= NVME_NS_METADATA_SUPPORTED
;
1734 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
1735 struct request_queue
*q
)
1737 bool vwc
= ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
;
1739 if (ctrl
->max_hw_sectors
) {
1741 (ctrl
->max_hw_sectors
/ (NVME_CTRL_PAGE_SIZE
>> 9)) + 1;
1743 max_segments
= min_not_zero(max_segments
, ctrl
->max_segments
);
1744 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
1745 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
1747 blk_queue_virt_boundary(q
, NVME_CTRL_PAGE_SIZE
- 1);
1748 blk_queue_dma_alignment(q
, 7);
1749 blk_queue_write_cache(q
, vwc
, vwc
);
1752 static void nvme_update_disk_info(struct gendisk
*disk
,
1753 struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1755 sector_t capacity
= nvme_lba_to_sect(ns
, le64_to_cpu(id
->nsze
));
1756 unsigned short bs
= 1 << ns
->lba_shift
;
1757 u32 atomic_bs
, phys_bs
, io_opt
= 0;
1760 * The block layer can't support LBA sizes larger than the page size
1761 * yet, so catch this early and don't allow block I/O.
1763 if (ns
->lba_shift
> PAGE_SHIFT
) {
1768 blk_integrity_unregister(disk
);
1770 atomic_bs
= phys_bs
= bs
;
1771 nvme_setup_streams_ns(ns
->ctrl
, ns
, &phys_bs
, &io_opt
);
1772 if (id
->nabo
== 0) {
1774 * Bit 1 indicates whether NAWUPF is defined for this namespace
1775 * and whether it should be used instead of AWUPF. If NAWUPF ==
1776 * 0 then AWUPF must be used instead.
1778 if (id
->nsfeat
& NVME_NS_FEAT_ATOMICS
&& id
->nawupf
)
1779 atomic_bs
= (1 + le16_to_cpu(id
->nawupf
)) * bs
;
1781 atomic_bs
= (1 + ns
->ctrl
->subsys
->awupf
) * bs
;
1784 if (id
->nsfeat
& NVME_NS_FEAT_IO_OPT
) {
1785 /* NPWG = Namespace Preferred Write Granularity */
1786 phys_bs
= bs
* (1 + le16_to_cpu(id
->npwg
));
1787 /* NOWS = Namespace Optimal Write Size */
1788 io_opt
= bs
* (1 + le16_to_cpu(id
->nows
));
1791 blk_queue_logical_block_size(disk
->queue
, bs
);
1793 * Linux filesystems assume writing a single physical block is
1794 * an atomic operation. Hence limit the physical block size to the
1795 * value of the Atomic Write Unit Power Fail parameter.
1797 blk_queue_physical_block_size(disk
->queue
, min(phys_bs
, atomic_bs
));
1798 blk_queue_io_min(disk
->queue
, phys_bs
);
1799 blk_queue_io_opt(disk
->queue
, io_opt
);
1802 * Register a metadata profile for PI, or the plain non-integrity NVMe
1803 * metadata masquerading as Type 0 if supported, otherwise reject block
1804 * I/O to namespaces with metadata except when the namespace supports
1805 * PI, as it can strip/insert in that case.
1808 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY
) &&
1809 (ns
->features
& NVME_NS_METADATA_SUPPORTED
))
1810 nvme_init_integrity(disk
, ns
->ms
, ns
->pi_type
,
1811 ns
->ctrl
->max_integrity_segments
);
1812 else if (!nvme_ns_has_pi(ns
))
1816 set_capacity_and_notify(disk
, capacity
);
1818 nvme_config_discard(disk
, ns
);
1819 blk_queue_max_write_zeroes_sectors(disk
->queue
,
1820 ns
->ctrl
->max_zeroes_sectors
);
1822 set_disk_ro(disk
, (id
->nsattr
& NVME_NS_ATTR_RO
) ||
1823 test_bit(NVME_NS_FORCE_RO
, &ns
->flags
));
1826 static inline bool nvme_first_scan(struct gendisk
*disk
)
1828 /* nvme_alloc_ns() scans the disk prior to adding it */
1829 return !(disk
->flags
& GENHD_FL_UP
);
1832 static void nvme_set_chunk_sectors(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1834 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1837 if ((ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) &&
1838 is_power_of_2(ctrl
->max_hw_sectors
))
1839 iob
= ctrl
->max_hw_sectors
;
1841 iob
= nvme_lba_to_sect(ns
, le16_to_cpu(id
->noiob
));
1846 if (!is_power_of_2(iob
)) {
1847 if (nvme_first_scan(ns
->disk
))
1848 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1849 ns
->disk
->disk_name
, iob
);
1853 if (blk_queue_is_zoned(ns
->disk
->queue
)) {
1854 if (nvme_first_scan(ns
->disk
))
1855 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1856 ns
->disk
->disk_name
);
1860 blk_queue_chunk_sectors(ns
->queue
, iob
);
1863 static int nvme_update_ns_info(struct nvme_ns
*ns
, struct nvme_id_ns
*id
)
1865 unsigned lbaf
= id
->flbas
& NVME_NS_FLBAS_LBA_MASK
;
1868 blk_mq_freeze_queue(ns
->disk
->queue
);
1869 ns
->lba_shift
= id
->lbaf
[lbaf
].ds
;
1870 nvme_set_queue_limits(ns
->ctrl
, ns
->queue
);
1872 ret
= nvme_configure_metadata(ns
, id
);
1875 nvme_set_chunk_sectors(ns
, id
);
1876 nvme_update_disk_info(ns
->disk
, ns
, id
);
1878 if (ns
->head
->ids
.csi
== NVME_CSI_ZNS
) {
1879 ret
= nvme_update_zone_info(ns
, lbaf
);
1884 blk_mq_unfreeze_queue(ns
->disk
->queue
);
1886 if (blk_queue_is_zoned(ns
->queue
)) {
1887 ret
= nvme_revalidate_zones(ns
);
1888 if (ret
&& !nvme_first_scan(ns
->disk
))
1892 if (nvme_ns_head_multipath(ns
->head
)) {
1893 blk_mq_freeze_queue(ns
->head
->disk
->queue
);
1894 nvme_update_disk_info(ns
->head
->disk
, ns
, id
);
1895 blk_stack_limits(&ns
->head
->disk
->queue
->limits
,
1896 &ns
->queue
->limits
, 0);
1897 blk_queue_update_readahead(ns
->head
->disk
->queue
);
1898 blk_mq_unfreeze_queue(ns
->head
->disk
->queue
);
1903 blk_mq_unfreeze_queue(ns
->disk
->queue
);
1906 * If probing fails due an unsupported feature, hide the block device,
1907 * but still allow other access.
1909 if (ret
== -ENODEV
) {
1910 ns
->disk
->flags
|= GENHD_FL_HIDDEN
;
1916 static char nvme_pr_type(enum pr_type type
)
1919 case PR_WRITE_EXCLUSIVE
:
1921 case PR_EXCLUSIVE_ACCESS
:
1923 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
1925 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
1927 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
1929 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
1936 static int nvme_send_ns_head_pr_command(struct block_device
*bdev
,
1937 struct nvme_command
*c
, u8 data
[16])
1939 struct nvme_ns_head
*head
= bdev
->bd_disk
->private_data
;
1940 int srcu_idx
= srcu_read_lock(&head
->srcu
);
1941 struct nvme_ns
*ns
= nvme_find_path(head
);
1942 int ret
= -EWOULDBLOCK
;
1945 c
->common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
1946 ret
= nvme_submit_sync_cmd(ns
->queue
, c
, data
, 16);
1948 srcu_read_unlock(&head
->srcu
, srcu_idx
);
1952 static int nvme_send_ns_pr_command(struct nvme_ns
*ns
, struct nvme_command
*c
,
1955 c
->common
.nsid
= cpu_to_le32(ns
->head
->ns_id
);
1956 return nvme_submit_sync_cmd(ns
->queue
, c
, data
, 16);
1959 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
1960 u64 key
, u64 sa_key
, u8 op
)
1962 struct nvme_command c
= { };
1963 u8 data
[16] = { 0, };
1965 put_unaligned_le64(key
, &data
[0]);
1966 put_unaligned_le64(sa_key
, &data
[8]);
1968 c
.common
.opcode
= op
;
1969 c
.common
.cdw10
= cpu_to_le32(cdw10
);
1971 if (IS_ENABLED(CONFIG_NVME_MULTIPATH
) &&
1972 bdev
->bd_disk
->fops
== &nvme_ns_head_ops
)
1973 return nvme_send_ns_head_pr_command(bdev
, &c
, data
);
1974 return nvme_send_ns_pr_command(bdev
->bd_disk
->private_data
, &c
, data
);
1977 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
1978 u64
new, unsigned flags
)
1982 if (flags
& ~PR_FL_IGNORE_KEY
)
1985 cdw10
= old
? 2 : 0;
1986 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
1987 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
1988 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
1991 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
1992 enum pr_type type
, unsigned flags
)
1996 if (flags
& ~PR_FL_IGNORE_KEY
)
1999 cdw10
= nvme_pr_type(type
) << 8;
2000 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
2001 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
2004 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
2005 enum pr_type type
, bool abort
)
2007 u32 cdw10
= nvme_pr_type(type
) << 8 | (abort
? 2 : 1);
2009 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
2012 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
2014 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
2016 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
2019 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
2021 u32 cdw10
= nvme_pr_type(type
) << 8 | (key
? 1 << 3 : 0);
2023 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
2026 const struct pr_ops nvme_pr_ops
= {
2027 .pr_register
= nvme_pr_register
,
2028 .pr_reserve
= nvme_pr_reserve
,
2029 .pr_release
= nvme_pr_release
,
2030 .pr_preempt
= nvme_pr_preempt
,
2031 .pr_clear
= nvme_pr_clear
,
2034 #ifdef CONFIG_BLK_SED_OPAL
2035 int nvme_sec_submit(void *data
, u16 spsp
, u8 secp
, void *buffer
, size_t len
,
2038 struct nvme_ctrl
*ctrl
= data
;
2039 struct nvme_command cmd
= { };
2042 cmd
.common
.opcode
= nvme_admin_security_send
;
2044 cmd
.common
.opcode
= nvme_admin_security_recv
;
2045 cmd
.common
.nsid
= 0;
2046 cmd
.common
.cdw10
= cpu_to_le32(((u32
)secp
) << 24 | ((u32
)spsp
) << 8);
2047 cmd
.common
.cdw11
= cpu_to_le32(len
);
2049 return __nvme_submit_sync_cmd(ctrl
->admin_q
, &cmd
, NULL
, buffer
, len
, 0,
2050 NVME_QID_ANY
, 1, 0, false);
2052 EXPORT_SYMBOL_GPL(nvme_sec_submit
);
2053 #endif /* CONFIG_BLK_SED_OPAL */
2055 #ifdef CONFIG_BLK_DEV_ZONED
2056 static int nvme_report_zones(struct gendisk
*disk
, sector_t sector
,
2057 unsigned int nr_zones
, report_zones_cb cb
, void *data
)
2059 return nvme_ns_report_zones(disk
->private_data
, sector
, nr_zones
, cb
,
2063 #define nvme_report_zones NULL
2064 #endif /* CONFIG_BLK_DEV_ZONED */
2066 static const struct block_device_operations nvme_bdev_ops
= {
2067 .owner
= THIS_MODULE
,
2068 .ioctl
= nvme_ioctl
,
2070 .release
= nvme_release
,
2071 .getgeo
= nvme_getgeo
,
2072 .report_zones
= nvme_report_zones
,
2073 .pr_ops
= &nvme_pr_ops
,
2076 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
2078 unsigned long timeout
=
2079 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
2080 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
2083 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2086 if ((csts
& NVME_CSTS_RDY
) == bit
)
2089 usleep_range(1000, 2000);
2090 if (fatal_signal_pending(current
))
2092 if (time_after(jiffies
, timeout
)) {
2093 dev_err(ctrl
->device
,
2094 "Device not ready; aborting %s, CSTS=0x%x\n",
2095 enabled
? "initialisation" : "reset", csts
);
2104 * If the device has been passed off to us in an enabled state, just clear
2105 * the enabled bit. The spec says we should set the 'shutdown notification
2106 * bits', but doing so may cause the device to complete commands to the
2107 * admin queue ... and we don't know what memory that might be pointing at!
2109 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
)
2113 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2114 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
2116 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2120 if (ctrl
->quirks
& NVME_QUIRK_DELAY_BEFORE_CHK_RDY
)
2121 msleep(NVME_QUIRK_DELAY_AMOUNT
);
2123 return nvme_wait_ready(ctrl
, ctrl
->cap
, false);
2125 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
2127 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
)
2129 unsigned dev_page_min
;
2132 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &ctrl
->cap
);
2134 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
2137 dev_page_min
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12;
2139 if (NVME_CTRL_PAGE_SHIFT
< dev_page_min
) {
2140 dev_err(ctrl
->device
,
2141 "Minimum device page size %u too large for host (%u)\n",
2142 1 << dev_page_min
, 1 << NVME_CTRL_PAGE_SHIFT
);
2146 if (NVME_CAP_CSS(ctrl
->cap
) & NVME_CAP_CSS_CSI
)
2147 ctrl
->ctrl_config
= NVME_CC_CSS_CSI
;
2149 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
2150 ctrl
->ctrl_config
|= (NVME_CTRL_PAGE_SHIFT
- 12) << NVME_CC_MPS_SHIFT
;
2151 ctrl
->ctrl_config
|= NVME_CC_AMS_RR
| NVME_CC_SHN_NONE
;
2152 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
2153 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
2155 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2158 return nvme_wait_ready(ctrl
, ctrl
->cap
, true);
2160 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
2162 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
2164 unsigned long timeout
= jiffies
+ (ctrl
->shutdown_timeout
* HZ
);
2168 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
2169 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
2171 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
2175 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
2176 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
2180 if (fatal_signal_pending(current
))
2182 if (time_after(jiffies
, timeout
)) {
2183 dev_err(ctrl
->device
,
2184 "Device shutdown incomplete; abort shutdown\n");
2191 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
2193 static int nvme_configure_timestamp(struct nvme_ctrl
*ctrl
)
2198 if (!(ctrl
->oncs
& NVME_CTRL_ONCS_TIMESTAMP
))
2201 ts
= cpu_to_le64(ktime_to_ms(ktime_get_real()));
2202 ret
= nvme_set_features(ctrl
, NVME_FEAT_TIMESTAMP
, 0, &ts
, sizeof(ts
),
2205 dev_warn_once(ctrl
->device
,
2206 "could not set timestamp (%d)\n", ret
);
2210 static int nvme_configure_acre(struct nvme_ctrl
*ctrl
)
2212 struct nvme_feat_host_behavior
*host
;
2215 /* Don't bother enabling the feature if retry delay is not reported */
2219 host
= kzalloc(sizeof(*host
), GFP_KERNEL
);
2223 host
->acre
= NVME_ENABLE_ACRE
;
2224 ret
= nvme_set_features(ctrl
, NVME_FEAT_HOST_BEHAVIOR
, 0,
2225 host
, sizeof(*host
), NULL
);
2231 * The function checks whether the given total (exlat + enlat) latency of
2232 * a power state allows the latter to be used as an APST transition target.
2233 * It does so by comparing the latency to the primary and secondary latency
2234 * tolerances defined by module params. If there's a match, the corresponding
2235 * timeout value is returned and the matching tolerance index (1 or 2) is
2238 static bool nvme_apst_get_transition_time(u64 total_latency
,
2239 u64
*transition_time
, unsigned *last_index
)
2241 if (total_latency
<= apst_primary_latency_tol_us
) {
2242 if (*last_index
== 1)
2245 *transition_time
= apst_primary_timeout_ms
;
2248 if (apst_secondary_timeout_ms
&&
2249 total_latency
<= apst_secondary_latency_tol_us
) {
2250 if (*last_index
<= 2)
2253 *transition_time
= apst_secondary_timeout_ms
;
2260 * APST (Autonomous Power State Transition) lets us program a table of power
2261 * state transitions that the controller will perform automatically.
2263 * Depending on module params, one of the two supported techniques will be used:
2265 * - If the parameters provide explicit timeouts and tolerances, they will be
2266 * used to build a table with up to 2 non-operational states to transition to.
2267 * The default parameter values were selected based on the values used by
2268 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2269 * regeneration of the APST table in the event of switching between external
2270 * and battery power, the timeouts and tolerances reflect a compromise
2271 * between values used by Microsoft for AC and battery scenarios.
2272 * - If not, we'll configure the table with a simple heuristic: we are willing
2273 * to spend at most 2% of the time transitioning between power states.
2274 * Therefore, when running in any given state, we will enter the next
2275 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2276 * microseconds, as long as that state's exit latency is under the requested
2279 * We will not autonomously enter any non-operational state for which the total
2280 * latency exceeds ps_max_latency_us.
2282 * Users can set ps_max_latency_us to zero to turn off APST.
2284 static int nvme_configure_apst(struct nvme_ctrl
*ctrl
)
2286 struct nvme_feat_auto_pst
*table
;
2293 unsigned last_lt_index
= UINT_MAX
;
2296 * If APST isn't supported or if we haven't been initialized yet,
2297 * then don't do anything.
2302 if (ctrl
->npss
> 31) {
2303 dev_warn(ctrl
->device
, "NPSS is invalid; not using APST\n");
2307 table
= kzalloc(sizeof(*table
), GFP_KERNEL
);
2311 if (!ctrl
->apst_enabled
|| ctrl
->ps_max_latency_us
== 0) {
2312 /* Turn off APST. */
2313 dev_dbg(ctrl
->device
, "APST disabled\n");
2318 * Walk through all states from lowest- to highest-power.
2319 * According to the spec, lower-numbered states use more power. NPSS,
2320 * despite the name, is the index of the lowest-power state, not the
2323 for (state
= (int)ctrl
->npss
; state
>= 0; state
--) {
2324 u64 total_latency_us
, exit_latency_us
, transition_ms
;
2327 table
->entries
[state
] = target
;
2330 * Don't allow transitions to the deepest state if it's quirked
2333 if (state
== ctrl
->npss
&&
2334 (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
))
2338 * Is this state a useful non-operational state for higher-power
2339 * states to autonomously transition to?
2341 if (!(ctrl
->psd
[state
].flags
& NVME_PS_FLAGS_NON_OP_STATE
))
2344 exit_latency_us
= (u64
)le32_to_cpu(ctrl
->psd
[state
].exit_lat
);
2345 if (exit_latency_us
> ctrl
->ps_max_latency_us
)
2348 total_latency_us
= exit_latency_us
+
2349 le32_to_cpu(ctrl
->psd
[state
].entry_lat
);
2352 * This state is good. It can be used as the APST idle target
2353 * for higher power states.
2355 if (apst_primary_timeout_ms
&& apst_primary_latency_tol_us
) {
2356 if (!nvme_apst_get_transition_time(total_latency_us
,
2357 &transition_ms
, &last_lt_index
))
2360 transition_ms
= total_latency_us
+ 19;
2361 do_div(transition_ms
, 20);
2362 if (transition_ms
> (1 << 24) - 1)
2363 transition_ms
= (1 << 24) - 1;
2366 target
= cpu_to_le64((state
<< 3) | (transition_ms
<< 8));
2369 if (total_latency_us
> max_lat_us
)
2370 max_lat_us
= total_latency_us
;
2374 dev_dbg(ctrl
->device
, "APST enabled but no non-operational states are available\n");
2376 dev_dbg(ctrl
->device
, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2377 max_ps
, max_lat_us
, (int)sizeof(*table
), table
);
2381 ret
= nvme_set_features(ctrl
, NVME_FEAT_AUTO_PST
, apste
,
2382 table
, sizeof(*table
), NULL
);
2384 dev_err(ctrl
->device
, "failed to set APST feature (%d)\n", ret
);
2389 static void nvme_set_latency_tolerance(struct device
*dev
, s32 val
)
2391 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
2395 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT
:
2396 case PM_QOS_LATENCY_ANY
:
2404 if (ctrl
->ps_max_latency_us
!= latency
) {
2405 ctrl
->ps_max_latency_us
= latency
;
2406 if (ctrl
->state
== NVME_CTRL_LIVE
)
2407 nvme_configure_apst(ctrl
);
2411 struct nvme_core_quirk_entry
{
2413 * NVMe model and firmware strings are padded with spaces. For
2414 * simplicity, strings in the quirk table are padded with NULLs
2420 unsigned long quirks
;
2423 static const struct nvme_core_quirk_entry core_quirks
[] = {
2426 * This Toshiba device seems to die using any APST states. See:
2427 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2430 .mn
= "THNSF5256GPUK TOSHIBA",
2431 .quirks
= NVME_QUIRK_NO_APST
,
2435 * This LiteON CL1-3D*-Q11 firmware version has a race
2436 * condition associated with actions related to suspend to idle
2437 * LiteON has resolved the problem in future firmware
2441 .quirks
= NVME_QUIRK_SIMPLE_SUSPEND
,
2445 /* match is null-terminated but idstr is space-padded. */
2446 static bool string_matches(const char *idstr
, const char *match
, size_t len
)
2453 matchlen
= strlen(match
);
2454 WARN_ON_ONCE(matchlen
> len
);
2456 if (memcmp(idstr
, match
, matchlen
))
2459 for (; matchlen
< len
; matchlen
++)
2460 if (idstr
[matchlen
] != ' ')
2466 static bool quirk_matches(const struct nvme_id_ctrl
*id
,
2467 const struct nvme_core_quirk_entry
*q
)
2469 return q
->vid
== le16_to_cpu(id
->vid
) &&
2470 string_matches(id
->mn
, q
->mn
, sizeof(id
->mn
)) &&
2471 string_matches(id
->fr
, q
->fr
, sizeof(id
->fr
));
2474 static void nvme_init_subnqn(struct nvme_subsystem
*subsys
, struct nvme_ctrl
*ctrl
,
2475 struct nvme_id_ctrl
*id
)
2480 if(!(ctrl
->quirks
& NVME_QUIRK_IGNORE_DEV_SUBNQN
)) {
2481 nqnlen
= strnlen(id
->subnqn
, NVMF_NQN_SIZE
);
2482 if (nqnlen
> 0 && nqnlen
< NVMF_NQN_SIZE
) {
2483 strlcpy(subsys
->subnqn
, id
->subnqn
, NVMF_NQN_SIZE
);
2487 if (ctrl
->vs
>= NVME_VS(1, 2, 1))
2488 dev_warn(ctrl
->device
, "missing or invalid SUBNQN field.\n");
2491 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2492 off
= snprintf(subsys
->subnqn
, NVMF_NQN_SIZE
,
2493 "nqn.2014.08.org.nvmexpress:%04x%04x",
2494 le16_to_cpu(id
->vid
), le16_to_cpu(id
->ssvid
));
2495 memcpy(subsys
->subnqn
+ off
, id
->sn
, sizeof(id
->sn
));
2496 off
+= sizeof(id
->sn
);
2497 memcpy(subsys
->subnqn
+ off
, id
->mn
, sizeof(id
->mn
));
2498 off
+= sizeof(id
->mn
);
2499 memset(subsys
->subnqn
+ off
, 0, sizeof(subsys
->subnqn
) - off
);
2502 static void nvme_release_subsystem(struct device
*dev
)
2504 struct nvme_subsystem
*subsys
=
2505 container_of(dev
, struct nvme_subsystem
, dev
);
2507 if (subsys
->instance
>= 0)
2508 ida_simple_remove(&nvme_instance_ida
, subsys
->instance
);
2512 static void nvme_destroy_subsystem(struct kref
*ref
)
2514 struct nvme_subsystem
*subsys
=
2515 container_of(ref
, struct nvme_subsystem
, ref
);
2517 mutex_lock(&nvme_subsystems_lock
);
2518 list_del(&subsys
->entry
);
2519 mutex_unlock(&nvme_subsystems_lock
);
2521 ida_destroy(&subsys
->ns_ida
);
2522 device_del(&subsys
->dev
);
2523 put_device(&subsys
->dev
);
2526 static void nvme_put_subsystem(struct nvme_subsystem
*subsys
)
2528 kref_put(&subsys
->ref
, nvme_destroy_subsystem
);
2531 static struct nvme_subsystem
*__nvme_find_get_subsystem(const char *subsysnqn
)
2533 struct nvme_subsystem
*subsys
;
2535 lockdep_assert_held(&nvme_subsystems_lock
);
2538 * Fail matches for discovery subsystems. This results
2539 * in each discovery controller bound to a unique subsystem.
2540 * This avoids issues with validating controller values
2541 * that can only be true when there is a single unique subsystem.
2542 * There may be multiple and completely independent entities
2543 * that provide discovery controllers.
2545 if (!strcmp(subsysnqn
, NVME_DISC_SUBSYS_NAME
))
2548 list_for_each_entry(subsys
, &nvme_subsystems
, entry
) {
2549 if (strcmp(subsys
->subnqn
, subsysnqn
))
2551 if (!kref_get_unless_zero(&subsys
->ref
))
2559 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2560 struct device_attribute subsys_attr_##_name = \
2561 __ATTR(_name, _mode, _show, NULL)
2563 static ssize_t
nvme_subsys_show_nqn(struct device
*dev
,
2564 struct device_attribute
*attr
,
2567 struct nvme_subsystem
*subsys
=
2568 container_of(dev
, struct nvme_subsystem
, dev
);
2570 return sysfs_emit(buf
, "%s\n", subsys
->subnqn
);
2572 static SUBSYS_ATTR_RO(subsysnqn
, S_IRUGO
, nvme_subsys_show_nqn
);
2574 #define nvme_subsys_show_str_function(field) \
2575 static ssize_t subsys_##field##_show(struct device *dev, \
2576 struct device_attribute *attr, char *buf) \
2578 struct nvme_subsystem *subsys = \
2579 container_of(dev, struct nvme_subsystem, dev); \
2580 return sysfs_emit(buf, "%.*s\n", \
2581 (int)sizeof(subsys->field), subsys->field); \
2583 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2585 nvme_subsys_show_str_function(model
);
2586 nvme_subsys_show_str_function(serial
);
2587 nvme_subsys_show_str_function(firmware_rev
);
2589 static struct attribute
*nvme_subsys_attrs
[] = {
2590 &subsys_attr_model
.attr
,
2591 &subsys_attr_serial
.attr
,
2592 &subsys_attr_firmware_rev
.attr
,
2593 &subsys_attr_subsysnqn
.attr
,
2594 #ifdef CONFIG_NVME_MULTIPATH
2595 &subsys_attr_iopolicy
.attr
,
2600 static const struct attribute_group nvme_subsys_attrs_group
= {
2601 .attrs
= nvme_subsys_attrs
,
2604 static const struct attribute_group
*nvme_subsys_attrs_groups
[] = {
2605 &nvme_subsys_attrs_group
,
2609 static inline bool nvme_discovery_ctrl(struct nvme_ctrl
*ctrl
)
2611 return ctrl
->opts
&& ctrl
->opts
->discovery_nqn
;
2614 static bool nvme_validate_cntlid(struct nvme_subsystem
*subsys
,
2615 struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2617 struct nvme_ctrl
*tmp
;
2619 lockdep_assert_held(&nvme_subsystems_lock
);
2621 list_for_each_entry(tmp
, &subsys
->ctrls
, subsys_entry
) {
2622 if (nvme_state_terminal(tmp
))
2625 if (tmp
->cntlid
== ctrl
->cntlid
) {
2626 dev_err(ctrl
->device
,
2627 "Duplicate cntlid %u with %s, rejecting\n",
2628 ctrl
->cntlid
, dev_name(tmp
->device
));
2632 if ((id
->cmic
& NVME_CTRL_CMIC_MULTI_CTRL
) ||
2633 nvme_discovery_ctrl(ctrl
))
2636 dev_err(ctrl
->device
,
2637 "Subsystem does not support multiple controllers\n");
2644 static int nvme_init_subsystem(struct nvme_ctrl
*ctrl
, struct nvme_id_ctrl
*id
)
2646 struct nvme_subsystem
*subsys
, *found
;
2649 subsys
= kzalloc(sizeof(*subsys
), GFP_KERNEL
);
2653 subsys
->instance
= -1;
2654 mutex_init(&subsys
->lock
);
2655 kref_init(&subsys
->ref
);
2656 INIT_LIST_HEAD(&subsys
->ctrls
);
2657 INIT_LIST_HEAD(&subsys
->nsheads
);
2658 nvme_init_subnqn(subsys
, ctrl
, id
);
2659 memcpy(subsys
->serial
, id
->sn
, sizeof(subsys
->serial
));
2660 memcpy(subsys
->model
, id
->mn
, sizeof(subsys
->model
));
2661 memcpy(subsys
->firmware_rev
, id
->fr
, sizeof(subsys
->firmware_rev
));
2662 subsys
->vendor_id
= le16_to_cpu(id
->vid
);
2663 subsys
->cmic
= id
->cmic
;
2664 subsys
->awupf
= le16_to_cpu(id
->awupf
);
2665 #ifdef CONFIG_NVME_MULTIPATH
2666 subsys
->iopolicy
= NVME_IOPOLICY_NUMA
;
2669 subsys
->dev
.class = nvme_subsys_class
;
2670 subsys
->dev
.release
= nvme_release_subsystem
;
2671 subsys
->dev
.groups
= nvme_subsys_attrs_groups
;
2672 dev_set_name(&subsys
->dev
, "nvme-subsys%d", ctrl
->instance
);
2673 device_initialize(&subsys
->dev
);
2675 mutex_lock(&nvme_subsystems_lock
);
2676 found
= __nvme_find_get_subsystem(subsys
->subnqn
);
2678 put_device(&subsys
->dev
);
2681 if (!nvme_validate_cntlid(subsys
, ctrl
, id
)) {
2683 goto out_put_subsystem
;
2686 ret
= device_add(&subsys
->dev
);
2688 dev_err(ctrl
->device
,
2689 "failed to register subsystem device.\n");
2690 put_device(&subsys
->dev
);
2693 ida_init(&subsys
->ns_ida
);
2694 list_add_tail(&subsys
->entry
, &nvme_subsystems
);
2697 ret
= sysfs_create_link(&subsys
->dev
.kobj
, &ctrl
->device
->kobj
,
2698 dev_name(ctrl
->device
));
2700 dev_err(ctrl
->device
,
2701 "failed to create sysfs link from subsystem.\n");
2702 goto out_put_subsystem
;
2706 subsys
->instance
= ctrl
->instance
;
2707 ctrl
->subsys
= subsys
;
2708 list_add_tail(&ctrl
->subsys_entry
, &subsys
->ctrls
);
2709 mutex_unlock(&nvme_subsystems_lock
);
2713 nvme_put_subsystem(subsys
);
2715 mutex_unlock(&nvme_subsystems_lock
);
2719 int nvme_get_log(struct nvme_ctrl
*ctrl
, u32 nsid
, u8 log_page
, u8 lsp
, u8 csi
,
2720 void *log
, size_t size
, u64 offset
)
2722 struct nvme_command c
= { };
2723 u32 dwlen
= nvme_bytes_to_numd(size
);
2725 c
.get_log_page
.opcode
= nvme_admin_get_log_page
;
2726 c
.get_log_page
.nsid
= cpu_to_le32(nsid
);
2727 c
.get_log_page
.lid
= log_page
;
2728 c
.get_log_page
.lsp
= lsp
;
2729 c
.get_log_page
.numdl
= cpu_to_le16(dwlen
& ((1 << 16) - 1));
2730 c
.get_log_page
.numdu
= cpu_to_le16(dwlen
>> 16);
2731 c
.get_log_page
.lpol
= cpu_to_le32(lower_32_bits(offset
));
2732 c
.get_log_page
.lpou
= cpu_to_le32(upper_32_bits(offset
));
2733 c
.get_log_page
.csi
= csi
;
2735 return nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, log
, size
);
2738 static int nvme_get_effects_log(struct nvme_ctrl
*ctrl
, u8 csi
,
2739 struct nvme_effects_log
**log
)
2741 struct nvme_effects_log
*cel
= xa_load(&ctrl
->cels
, csi
);
2747 cel
= kzalloc(sizeof(*cel
), GFP_KERNEL
);
2751 ret
= nvme_get_log(ctrl
, 0x00, NVME_LOG_CMD_EFFECTS
, 0, csi
,
2752 cel
, sizeof(*cel
), 0);
2758 xa_store(&ctrl
->cels
, csi
, cel
, GFP_KERNEL
);
2764 static inline u32
nvme_mps_to_sectors(struct nvme_ctrl
*ctrl
, u32 units
)
2766 u32 page_shift
= NVME_CAP_MPSMIN(ctrl
->cap
) + 12, val
;
2768 if (check_shl_overflow(1U, units
+ page_shift
- 9, &val
))
2773 static int nvme_init_non_mdts_limits(struct nvme_ctrl
*ctrl
)
2775 struct nvme_command c
= { };
2776 struct nvme_id_ctrl_nvm
*id
;
2779 if (ctrl
->oncs
& NVME_CTRL_ONCS_DSM
) {
2780 ctrl
->max_discard_sectors
= UINT_MAX
;
2781 ctrl
->max_discard_segments
= NVME_DSM_MAX_RANGES
;
2783 ctrl
->max_discard_sectors
= 0;
2784 ctrl
->max_discard_segments
= 0;
2788 * Even though NVMe spec explicitly states that MDTS is not applicable
2789 * to the write-zeroes, we are cautious and limit the size to the
2790 * controllers max_hw_sectors value, which is based on the MDTS field
2791 * and possibly other limiting factors.
2793 if ((ctrl
->oncs
& NVME_CTRL_ONCS_WRITE_ZEROES
) &&
2794 !(ctrl
->quirks
& NVME_QUIRK_DISABLE_WRITE_ZEROES
))
2795 ctrl
->max_zeroes_sectors
= ctrl
->max_hw_sectors
;
2797 ctrl
->max_zeroes_sectors
= 0;
2799 if (nvme_ctrl_limited_cns(ctrl
))
2802 id
= kzalloc(sizeof(*id
), GFP_KERNEL
);
2806 c
.identify
.opcode
= nvme_admin_identify
;
2807 c
.identify
.cns
= NVME_ID_CNS_CS_CTRL
;
2808 c
.identify
.csi
= NVME_CSI_NVM
;
2810 ret
= nvme_submit_sync_cmd(ctrl
->admin_q
, &c
, id
, sizeof(*id
));
2815 ctrl
->max_discard_segments
= id
->dmrl
;
2817 ctrl
->max_discard_sectors
= le32_to_cpu(id
->dmrsl
);
2819 ctrl
->max_zeroes_sectors
= nvme_mps_to_sectors(ctrl
, id
->wzsl
);
2826 static int nvme_init_identify(struct nvme_ctrl
*ctrl
)
2828 struct nvme_id_ctrl
*id
;
2830 bool prev_apst_enabled
;
2833 ret
= nvme_identify_ctrl(ctrl
, &id
);
2835 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
2839 if (id
->lpa
& NVME_CTRL_LPA_CMD_EFFECTS_LOG
) {
2840 ret
= nvme_get_effects_log(ctrl
, NVME_CSI_NVM
, &ctrl
->effects
);
2845 if (!(ctrl
->ops
->flags
& NVME_F_FABRICS
))
2846 ctrl
->cntlid
= le16_to_cpu(id
->cntlid
);
2848 if (!ctrl
->identified
) {
2851 ret
= nvme_init_subsystem(ctrl
, id
);
2856 * Check for quirks. Quirk can depend on firmware version,
2857 * so, in principle, the set of quirks present can change
2858 * across a reset. As a possible future enhancement, we
2859 * could re-scan for quirks every time we reinitialize
2860 * the device, but we'd have to make sure that the driver
2861 * behaves intelligently if the quirks change.
2863 for (i
= 0; i
< ARRAY_SIZE(core_quirks
); i
++) {
2864 if (quirk_matches(id
, &core_quirks
[i
]))
2865 ctrl
->quirks
|= core_quirks
[i
].quirks
;
2869 if (force_apst
&& (ctrl
->quirks
& NVME_QUIRK_NO_DEEPEST_PS
)) {
2870 dev_warn(ctrl
->device
, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2871 ctrl
->quirks
&= ~NVME_QUIRK_NO_DEEPEST_PS
;
2874 ctrl
->crdt
[0] = le16_to_cpu(id
->crdt1
);
2875 ctrl
->crdt
[1] = le16_to_cpu(id
->crdt2
);
2876 ctrl
->crdt
[2] = le16_to_cpu(id
->crdt3
);
2878 ctrl
->oacs
= le16_to_cpu(id
->oacs
);
2879 ctrl
->oncs
= le16_to_cpu(id
->oncs
);
2880 ctrl
->mtfa
= le16_to_cpu(id
->mtfa
);
2881 ctrl
->oaes
= le32_to_cpu(id
->oaes
);
2882 ctrl
->wctemp
= le16_to_cpu(id
->wctemp
);
2883 ctrl
->cctemp
= le16_to_cpu(id
->cctemp
);
2885 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
2886 ctrl
->vwc
= id
->vwc
;
2888 max_hw_sectors
= nvme_mps_to_sectors(ctrl
, id
->mdts
);
2890 max_hw_sectors
= UINT_MAX
;
2891 ctrl
->max_hw_sectors
=
2892 min_not_zero(ctrl
->max_hw_sectors
, max_hw_sectors
);
2894 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
2895 ctrl
->sgls
= le32_to_cpu(id
->sgls
);
2896 ctrl
->kas
= le16_to_cpu(id
->kas
);
2897 ctrl
->max_namespaces
= le32_to_cpu(id
->mnan
);
2898 ctrl
->ctratt
= le32_to_cpu(id
->ctratt
);
2902 u32 transition_time
= le32_to_cpu(id
->rtd3e
) / USEC_PER_SEC
;
2904 ctrl
->shutdown_timeout
= clamp_t(unsigned int, transition_time
,
2905 shutdown_timeout
, 60);
2907 if (ctrl
->shutdown_timeout
!= shutdown_timeout
)
2908 dev_info(ctrl
->device
,
2909 "Shutdown timeout set to %u seconds\n",
2910 ctrl
->shutdown_timeout
);
2912 ctrl
->shutdown_timeout
= shutdown_timeout
;
2914 ctrl
->npss
= id
->npss
;
2915 ctrl
->apsta
= id
->apsta
;
2916 prev_apst_enabled
= ctrl
->apst_enabled
;
2917 if (ctrl
->quirks
& NVME_QUIRK_NO_APST
) {
2918 if (force_apst
&& id
->apsta
) {
2919 dev_warn(ctrl
->device
, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2920 ctrl
->apst_enabled
= true;
2922 ctrl
->apst_enabled
= false;
2925 ctrl
->apst_enabled
= id
->apsta
;
2927 memcpy(ctrl
->psd
, id
->psd
, sizeof(ctrl
->psd
));
2929 if (ctrl
->ops
->flags
& NVME_F_FABRICS
) {
2930 ctrl
->icdoff
= le16_to_cpu(id
->icdoff
);
2931 ctrl
->ioccsz
= le32_to_cpu(id
->ioccsz
);
2932 ctrl
->iorcsz
= le32_to_cpu(id
->iorcsz
);
2933 ctrl
->maxcmd
= le16_to_cpu(id
->maxcmd
);
2936 * In fabrics we need to verify the cntlid matches the
2939 if (ctrl
->cntlid
!= le16_to_cpu(id
->cntlid
)) {
2940 dev_err(ctrl
->device
,
2941 "Mismatching cntlid: Connect %u vs Identify "
2943 ctrl
->cntlid
, le16_to_cpu(id
->cntlid
));
2948 if (!nvme_discovery_ctrl(ctrl
) && !ctrl
->kas
) {
2949 dev_err(ctrl
->device
,
2950 "keep-alive support is mandatory for fabrics\n");
2955 ctrl
->hmpre
= le32_to_cpu(id
->hmpre
);
2956 ctrl
->hmmin
= le32_to_cpu(id
->hmmin
);
2957 ctrl
->hmminds
= le32_to_cpu(id
->hmminds
);
2958 ctrl
->hmmaxd
= le16_to_cpu(id
->hmmaxd
);
2961 ret
= nvme_mpath_init_identify(ctrl
, id
);
2965 if (ctrl
->apst_enabled
&& !prev_apst_enabled
)
2966 dev_pm_qos_expose_latency_tolerance(ctrl
->device
);
2967 else if (!ctrl
->apst_enabled
&& prev_apst_enabled
)
2968 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
2976 * Initialize the cached copies of the Identify data and various controller
2977 * register in our nvme_ctrl structure. This should be called as soon as
2978 * the admin queue is fully up and running.
2980 int nvme_init_ctrl_finish(struct nvme_ctrl
*ctrl
)
2984 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
2986 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
2990 ctrl
->sqsize
= min_t(u16
, NVME_CAP_MQES(ctrl
->cap
), ctrl
->sqsize
);
2992 if (ctrl
->vs
>= NVME_VS(1, 1, 0))
2993 ctrl
->subsystem
= NVME_CAP_NSSRC(ctrl
->cap
);
2995 ret
= nvme_init_identify(ctrl
);
2999 ret
= nvme_init_non_mdts_limits(ctrl
);
3003 ret
= nvme_configure_apst(ctrl
);
3007 ret
= nvme_configure_timestamp(ctrl
);
3011 ret
= nvme_configure_directives(ctrl
);
3015 ret
= nvme_configure_acre(ctrl
);
3019 if (!ctrl
->identified
&& !nvme_discovery_ctrl(ctrl
)) {
3020 ret
= nvme_hwmon_init(ctrl
);
3025 ctrl
->identified
= true;
3029 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish
);
3031 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
3033 struct nvme_ctrl
*ctrl
=
3034 container_of(inode
->i_cdev
, struct nvme_ctrl
, cdev
);
3036 switch (ctrl
->state
) {
3037 case NVME_CTRL_LIVE
:
3040 return -EWOULDBLOCK
;
3043 nvme_get_ctrl(ctrl
);
3044 if (!try_module_get(ctrl
->ops
->module
)) {
3045 nvme_put_ctrl(ctrl
);
3049 file
->private_data
= ctrl
;
3053 static int nvme_dev_release(struct inode
*inode
, struct file
*file
)
3055 struct nvme_ctrl
*ctrl
=
3056 container_of(inode
->i_cdev
, struct nvme_ctrl
, cdev
);
3058 module_put(ctrl
->ops
->module
);
3059 nvme_put_ctrl(ctrl
);
3063 static const struct file_operations nvme_dev_fops
= {
3064 .owner
= THIS_MODULE
,
3065 .open
= nvme_dev_open
,
3066 .release
= nvme_dev_release
,
3067 .unlocked_ioctl
= nvme_dev_ioctl
,
3068 .compat_ioctl
= compat_ptr_ioctl
,
3071 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
3072 struct device_attribute
*attr
, const char *buf
,
3075 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3078 ret
= nvme_reset_ctrl_sync(ctrl
);
3083 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
3085 static ssize_t
nvme_sysfs_rescan(struct device
*dev
,
3086 struct device_attribute
*attr
, const char *buf
,
3089 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3091 nvme_queue_scan(ctrl
);
3094 static DEVICE_ATTR(rescan_controller
, S_IWUSR
, NULL
, nvme_sysfs_rescan
);
3096 static inline struct nvme_ns_head
*dev_to_ns_head(struct device
*dev
)
3098 struct gendisk
*disk
= dev_to_disk(dev
);
3100 if (disk
->fops
== &nvme_bdev_ops
)
3101 return nvme_get_ns_from_dev(dev
)->head
;
3103 return disk
->private_data
;
3106 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
3109 struct nvme_ns_head
*head
= dev_to_ns_head(dev
);
3110 struct nvme_ns_ids
*ids
= &head
->ids
;
3111 struct nvme_subsystem
*subsys
= head
->subsys
;
3112 int serial_len
= sizeof(subsys
->serial
);
3113 int model_len
= sizeof(subsys
->model
);
3115 if (!uuid_is_null(&ids
->uuid
))
3116 return sysfs_emit(buf
, "uuid.%pU\n", &ids
->uuid
);
3118 if (memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3119 return sysfs_emit(buf
, "eui.%16phN\n", ids
->nguid
);
3121 if (memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3122 return sysfs_emit(buf
, "eui.%8phN\n", ids
->eui64
);
3124 while (serial_len
> 0 && (subsys
->serial
[serial_len
- 1] == ' ' ||
3125 subsys
->serial
[serial_len
- 1] == '\0'))
3127 while (model_len
> 0 && (subsys
->model
[model_len
- 1] == ' ' ||
3128 subsys
->model
[model_len
- 1] == '\0'))
3131 return sysfs_emit(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", subsys
->vendor_id
,
3132 serial_len
, subsys
->serial
, model_len
, subsys
->model
,
3135 static DEVICE_ATTR_RO(wwid
);
3137 static ssize_t
nguid_show(struct device
*dev
, struct device_attribute
*attr
,
3140 return sysfs_emit(buf
, "%pU\n", dev_to_ns_head(dev
)->ids
.nguid
);
3142 static DEVICE_ATTR_RO(nguid
);
3144 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
3147 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3149 /* For backward compatibility expose the NGUID to userspace if
3150 * we have no UUID set
3152 if (uuid_is_null(&ids
->uuid
)) {
3153 printk_ratelimited(KERN_WARNING
3154 "No UUID available providing old NGUID\n");
3155 return sysfs_emit(buf
, "%pU\n", ids
->nguid
);
3157 return sysfs_emit(buf
, "%pU\n", &ids
->uuid
);
3159 static DEVICE_ATTR_RO(uuid
);
3161 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
3164 return sysfs_emit(buf
, "%8ph\n", dev_to_ns_head(dev
)->ids
.eui64
);
3166 static DEVICE_ATTR_RO(eui
);
3168 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
3171 return sysfs_emit(buf
, "%d\n", dev_to_ns_head(dev
)->ns_id
);
3173 static DEVICE_ATTR_RO(nsid
);
3175 static struct attribute
*nvme_ns_id_attrs
[] = {
3176 &dev_attr_wwid
.attr
,
3177 &dev_attr_uuid
.attr
,
3178 &dev_attr_nguid
.attr
,
3180 &dev_attr_nsid
.attr
,
3181 #ifdef CONFIG_NVME_MULTIPATH
3182 &dev_attr_ana_grpid
.attr
,
3183 &dev_attr_ana_state
.attr
,
3188 static umode_t
nvme_ns_id_attrs_are_visible(struct kobject
*kobj
,
3189 struct attribute
*a
, int n
)
3191 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3192 struct nvme_ns_ids
*ids
= &dev_to_ns_head(dev
)->ids
;
3194 if (a
== &dev_attr_uuid
.attr
) {
3195 if (uuid_is_null(&ids
->uuid
) &&
3196 !memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3199 if (a
== &dev_attr_nguid
.attr
) {
3200 if (!memchr_inv(ids
->nguid
, 0, sizeof(ids
->nguid
)))
3203 if (a
== &dev_attr_eui
.attr
) {
3204 if (!memchr_inv(ids
->eui64
, 0, sizeof(ids
->eui64
)))
3207 #ifdef CONFIG_NVME_MULTIPATH
3208 if (a
== &dev_attr_ana_grpid
.attr
|| a
== &dev_attr_ana_state
.attr
) {
3209 if (dev_to_disk(dev
)->fops
!= &nvme_bdev_ops
) /* per-path attr */
3211 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev
)->ctrl
))
3218 static const struct attribute_group nvme_ns_id_attr_group
= {
3219 .attrs
= nvme_ns_id_attrs
,
3220 .is_visible
= nvme_ns_id_attrs_are_visible
,
3223 const struct attribute_group
*nvme_ns_id_attr_groups
[] = {
3224 &nvme_ns_id_attr_group
,
3226 &nvme_nvm_attr_group
,
3231 #define nvme_show_str_function(field) \
3232 static ssize_t field##_show(struct device *dev, \
3233 struct device_attribute *attr, char *buf) \
3235 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3236 return sysfs_emit(buf, "%.*s\n", \
3237 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3239 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3241 nvme_show_str_function(model
);
3242 nvme_show_str_function(serial
);
3243 nvme_show_str_function(firmware_rev
);
3245 #define nvme_show_int_function(field) \
3246 static ssize_t field##_show(struct device *dev, \
3247 struct device_attribute *attr, char *buf) \
3249 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3250 return sysfs_emit(buf, "%d\n", ctrl->field); \
3252 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3254 nvme_show_int_function(cntlid
);
3255 nvme_show_int_function(numa_node
);
3256 nvme_show_int_function(queue_count
);
3257 nvme_show_int_function(sqsize
);
3258 nvme_show_int_function(kato
);
3260 static ssize_t
nvme_sysfs_delete(struct device
*dev
,
3261 struct device_attribute
*attr
, const char *buf
,
3264 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3266 if (device_remove_file_self(dev
, attr
))
3267 nvme_delete_ctrl_sync(ctrl
);
3270 static DEVICE_ATTR(delete_controller
, S_IWUSR
, NULL
, nvme_sysfs_delete
);
3272 static ssize_t
nvme_sysfs_show_transport(struct device
*dev
,
3273 struct device_attribute
*attr
,
3276 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3278 return sysfs_emit(buf
, "%s\n", ctrl
->ops
->name
);
3280 static DEVICE_ATTR(transport
, S_IRUGO
, nvme_sysfs_show_transport
, NULL
);
3282 static ssize_t
nvme_sysfs_show_state(struct device
*dev
,
3283 struct device_attribute
*attr
,
3286 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3287 static const char *const state_name
[] = {
3288 [NVME_CTRL_NEW
] = "new",
3289 [NVME_CTRL_LIVE
] = "live",
3290 [NVME_CTRL_RESETTING
] = "resetting",
3291 [NVME_CTRL_CONNECTING
] = "connecting",
3292 [NVME_CTRL_DELETING
] = "deleting",
3293 [NVME_CTRL_DELETING_NOIO
]= "deleting (no IO)",
3294 [NVME_CTRL_DEAD
] = "dead",
3297 if ((unsigned)ctrl
->state
< ARRAY_SIZE(state_name
) &&
3298 state_name
[ctrl
->state
])
3299 return sysfs_emit(buf
, "%s\n", state_name
[ctrl
->state
]);
3301 return sysfs_emit(buf
, "unknown state\n");
3304 static DEVICE_ATTR(state
, S_IRUGO
, nvme_sysfs_show_state
, NULL
);
3306 static ssize_t
nvme_sysfs_show_subsysnqn(struct device
*dev
,
3307 struct device_attribute
*attr
,
3310 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3312 return sysfs_emit(buf
, "%s\n", ctrl
->subsys
->subnqn
);
3314 static DEVICE_ATTR(subsysnqn
, S_IRUGO
, nvme_sysfs_show_subsysnqn
, NULL
);
3316 static ssize_t
nvme_sysfs_show_hostnqn(struct device
*dev
,
3317 struct device_attribute
*attr
,
3320 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3322 return sysfs_emit(buf
, "%s\n", ctrl
->opts
->host
->nqn
);
3324 static DEVICE_ATTR(hostnqn
, S_IRUGO
, nvme_sysfs_show_hostnqn
, NULL
);
3326 static ssize_t
nvme_sysfs_show_hostid(struct device
*dev
,
3327 struct device_attribute
*attr
,
3330 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3332 return sysfs_emit(buf
, "%pU\n", &ctrl
->opts
->host
->id
);
3334 static DEVICE_ATTR(hostid
, S_IRUGO
, nvme_sysfs_show_hostid
, NULL
);
3336 static ssize_t
nvme_sysfs_show_address(struct device
*dev
,
3337 struct device_attribute
*attr
,
3340 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3342 return ctrl
->ops
->get_address(ctrl
, buf
, PAGE_SIZE
);
3344 static DEVICE_ATTR(address
, S_IRUGO
, nvme_sysfs_show_address
, NULL
);
3346 static ssize_t
nvme_ctrl_loss_tmo_show(struct device
*dev
,
3347 struct device_attribute
*attr
, char *buf
)
3349 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3350 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3352 if (ctrl
->opts
->max_reconnects
== -1)
3353 return sysfs_emit(buf
, "off\n");
3354 return sysfs_emit(buf
, "%d\n",
3355 opts
->max_reconnects
* opts
->reconnect_delay
);
3358 static ssize_t
nvme_ctrl_loss_tmo_store(struct device
*dev
,
3359 struct device_attribute
*attr
, const char *buf
, size_t count
)
3361 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3362 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3363 int ctrl_loss_tmo
, err
;
3365 err
= kstrtoint(buf
, 10, &ctrl_loss_tmo
);
3369 if (ctrl_loss_tmo
< 0)
3370 opts
->max_reconnects
= -1;
3372 opts
->max_reconnects
= DIV_ROUND_UP(ctrl_loss_tmo
,
3373 opts
->reconnect_delay
);
3376 static DEVICE_ATTR(ctrl_loss_tmo
, S_IRUGO
| S_IWUSR
,
3377 nvme_ctrl_loss_tmo_show
, nvme_ctrl_loss_tmo_store
);
3379 static ssize_t
nvme_ctrl_reconnect_delay_show(struct device
*dev
,
3380 struct device_attribute
*attr
, char *buf
)
3382 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3384 if (ctrl
->opts
->reconnect_delay
== -1)
3385 return sysfs_emit(buf
, "off\n");
3386 return sysfs_emit(buf
, "%d\n", ctrl
->opts
->reconnect_delay
);
3389 static ssize_t
nvme_ctrl_reconnect_delay_store(struct device
*dev
,
3390 struct device_attribute
*attr
, const char *buf
, size_t count
)
3392 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3396 err
= kstrtou32(buf
, 10, &v
);
3400 ctrl
->opts
->reconnect_delay
= v
;
3403 static DEVICE_ATTR(reconnect_delay
, S_IRUGO
| S_IWUSR
,
3404 nvme_ctrl_reconnect_delay_show
, nvme_ctrl_reconnect_delay_store
);
3406 static ssize_t
nvme_ctrl_fast_io_fail_tmo_show(struct device
*dev
,
3407 struct device_attribute
*attr
, char *buf
)
3409 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3411 if (ctrl
->opts
->fast_io_fail_tmo
== -1)
3412 return sysfs_emit(buf
, "off\n");
3413 return sysfs_emit(buf
, "%d\n", ctrl
->opts
->fast_io_fail_tmo
);
3416 static ssize_t
nvme_ctrl_fast_io_fail_tmo_store(struct device
*dev
,
3417 struct device_attribute
*attr
, const char *buf
, size_t count
)
3419 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3420 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
3421 int fast_io_fail_tmo
, err
;
3423 err
= kstrtoint(buf
, 10, &fast_io_fail_tmo
);
3427 if (fast_io_fail_tmo
< 0)
3428 opts
->fast_io_fail_tmo
= -1;
3430 opts
->fast_io_fail_tmo
= fast_io_fail_tmo
;
3433 static DEVICE_ATTR(fast_io_fail_tmo
, S_IRUGO
| S_IWUSR
,
3434 nvme_ctrl_fast_io_fail_tmo_show
, nvme_ctrl_fast_io_fail_tmo_store
);
3436 static struct attribute
*nvme_dev_attrs
[] = {
3437 &dev_attr_reset_controller
.attr
,
3438 &dev_attr_rescan_controller
.attr
,
3439 &dev_attr_model
.attr
,
3440 &dev_attr_serial
.attr
,
3441 &dev_attr_firmware_rev
.attr
,
3442 &dev_attr_cntlid
.attr
,
3443 &dev_attr_delete_controller
.attr
,
3444 &dev_attr_transport
.attr
,
3445 &dev_attr_subsysnqn
.attr
,
3446 &dev_attr_address
.attr
,
3447 &dev_attr_state
.attr
,
3448 &dev_attr_numa_node
.attr
,
3449 &dev_attr_queue_count
.attr
,
3450 &dev_attr_sqsize
.attr
,
3451 &dev_attr_hostnqn
.attr
,
3452 &dev_attr_hostid
.attr
,
3453 &dev_attr_ctrl_loss_tmo
.attr
,
3454 &dev_attr_reconnect_delay
.attr
,
3455 &dev_attr_fast_io_fail_tmo
.attr
,
3456 &dev_attr_kato
.attr
,
3460 static umode_t
nvme_dev_attrs_are_visible(struct kobject
*kobj
,
3461 struct attribute
*a
, int n
)
3463 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
3464 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
3466 if (a
== &dev_attr_delete_controller
.attr
&& !ctrl
->ops
->delete_ctrl
)
3468 if (a
== &dev_attr_address
.attr
&& !ctrl
->ops
->get_address
)
3470 if (a
== &dev_attr_hostnqn
.attr
&& !ctrl
->opts
)
3472 if (a
== &dev_attr_hostid
.attr
&& !ctrl
->opts
)
3474 if (a
== &dev_attr_ctrl_loss_tmo
.attr
&& !ctrl
->opts
)
3476 if (a
== &dev_attr_reconnect_delay
.attr
&& !ctrl
->opts
)
3478 if (a
== &dev_attr_fast_io_fail_tmo
.attr
&& !ctrl
->opts
)
3484 static const struct attribute_group nvme_dev_attrs_group
= {
3485 .attrs
= nvme_dev_attrs
,
3486 .is_visible
= nvme_dev_attrs_are_visible
,
3489 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
3490 &nvme_dev_attrs_group
,
3494 static struct nvme_ns_head
*nvme_find_ns_head(struct nvme_subsystem
*subsys
,
3497 struct nvme_ns_head
*h
;
3499 lockdep_assert_held(&subsys
->lock
);
3501 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3502 if (h
->ns_id
== nsid
&& nvme_tryget_ns_head(h
))
3509 static int __nvme_check_ids(struct nvme_subsystem
*subsys
,
3510 struct nvme_ns_head
*new)
3512 struct nvme_ns_head
*h
;
3514 lockdep_assert_held(&subsys
->lock
);
3516 list_for_each_entry(h
, &subsys
->nsheads
, entry
) {
3517 if (nvme_ns_ids_valid(&new->ids
) &&
3518 nvme_ns_ids_equal(&new->ids
, &h
->ids
))
3525 void nvme_cdev_del(struct cdev
*cdev
, struct device
*cdev_device
)
3527 cdev_device_del(cdev
, cdev_device
);
3528 ida_simple_remove(&nvme_ns_chr_minor_ida
, MINOR(cdev_device
->devt
));
3531 int nvme_cdev_add(struct cdev
*cdev
, struct device
*cdev_device
,
3532 const struct file_operations
*fops
, struct module
*owner
)
3536 minor
= ida_simple_get(&nvme_ns_chr_minor_ida
, 0, 0, GFP_KERNEL
);
3539 cdev_device
->devt
= MKDEV(MAJOR(nvme_ns_chr_devt
), minor
);
3540 cdev_device
->class = nvme_ns_chr_class
;
3541 device_initialize(cdev_device
);
3542 cdev_init(cdev
, fops
);
3543 cdev
->owner
= owner
;
3544 ret
= cdev_device_add(cdev
, cdev_device
);
3546 put_device(cdev_device
);
3547 ida_simple_remove(&nvme_ns_chr_minor_ida
, minor
);
3552 static int nvme_ns_chr_open(struct inode
*inode
, struct file
*file
)
3554 return nvme_ns_open(container_of(inode
->i_cdev
, struct nvme_ns
, cdev
));
3557 static int nvme_ns_chr_release(struct inode
*inode
, struct file
*file
)
3559 nvme_ns_release(container_of(inode
->i_cdev
, struct nvme_ns
, cdev
));
3563 static const struct file_operations nvme_ns_chr_fops
= {
3564 .owner
= THIS_MODULE
,
3565 .open
= nvme_ns_chr_open
,
3566 .release
= nvme_ns_chr_release
,
3567 .unlocked_ioctl
= nvme_ns_chr_ioctl
,
3568 .compat_ioctl
= compat_ptr_ioctl
,
3571 static int nvme_add_ns_cdev(struct nvme_ns
*ns
)
3575 ns
->cdev_device
.parent
= ns
->ctrl
->device
;
3576 ret
= dev_set_name(&ns
->cdev_device
, "ng%dn%d",
3577 ns
->ctrl
->instance
, ns
->head
->instance
);
3580 ret
= nvme_cdev_add(&ns
->cdev
, &ns
->cdev_device
, &nvme_ns_chr_fops
,
3581 ns
->ctrl
->ops
->module
);
3583 kfree_const(ns
->cdev_device
.kobj
.name
);
3587 static struct nvme_ns_head
*nvme_alloc_ns_head(struct nvme_ctrl
*ctrl
,
3588 unsigned nsid
, struct nvme_ns_ids
*ids
)
3590 struct nvme_ns_head
*head
;
3591 size_t size
= sizeof(*head
);
3594 #ifdef CONFIG_NVME_MULTIPATH
3595 size
+= num_possible_nodes() * sizeof(struct nvme_ns
*);
3598 head
= kzalloc(size
, GFP_KERNEL
);
3601 ret
= ida_simple_get(&ctrl
->subsys
->ns_ida
, 1, 0, GFP_KERNEL
);
3604 head
->instance
= ret
;
3605 INIT_LIST_HEAD(&head
->list
);
3606 ret
= init_srcu_struct(&head
->srcu
);
3608 goto out_ida_remove
;
3609 head
->subsys
= ctrl
->subsys
;
3612 kref_init(&head
->ref
);
3614 ret
= __nvme_check_ids(ctrl
->subsys
, head
);
3616 dev_err(ctrl
->device
,
3617 "duplicate IDs for nsid %d\n", nsid
);
3618 goto out_cleanup_srcu
;
3621 if (head
->ids
.csi
) {
3622 ret
= nvme_get_effects_log(ctrl
, head
->ids
.csi
, &head
->effects
);
3624 goto out_cleanup_srcu
;
3626 head
->effects
= ctrl
->effects
;
3628 ret
= nvme_mpath_alloc_disk(ctrl
, head
);
3630 goto out_cleanup_srcu
;
3632 list_add_tail(&head
->entry
, &ctrl
->subsys
->nsheads
);
3634 kref_get(&ctrl
->subsys
->ref
);
3638 cleanup_srcu_struct(&head
->srcu
);
3640 ida_simple_remove(&ctrl
->subsys
->ns_ida
, head
->instance
);
3645 ret
= blk_status_to_errno(nvme_error_status(ret
));
3646 return ERR_PTR(ret
);
3649 static int nvme_init_ns_head(struct nvme_ns
*ns
, unsigned nsid
,
3650 struct nvme_ns_ids
*ids
, bool is_shared
)
3652 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
3653 struct nvme_ns_head
*head
= NULL
;
3656 mutex_lock(&ctrl
->subsys
->lock
);
3657 head
= nvme_find_ns_head(ctrl
->subsys
, nsid
);
3659 head
= nvme_alloc_ns_head(ctrl
, nsid
, ids
);
3661 ret
= PTR_ERR(head
);
3664 head
->shared
= is_shared
;
3667 if (!is_shared
|| !head
->shared
) {
3668 dev_err(ctrl
->device
,
3669 "Duplicate unshared namespace %d\n", nsid
);
3670 goto out_put_ns_head
;
3672 if (!nvme_ns_ids_equal(&head
->ids
, ids
)) {
3673 dev_err(ctrl
->device
,
3674 "IDs don't match for shared namespace %d\n",
3676 goto out_put_ns_head
;
3680 list_add_tail_rcu(&ns
->siblings
, &head
->list
);
3682 mutex_unlock(&ctrl
->subsys
->lock
);
3686 nvme_put_ns_head(head
);
3688 mutex_unlock(&ctrl
->subsys
->lock
);
3692 static int ns_cmp(void *priv
, const struct list_head
*a
,
3693 const struct list_head
*b
)
3695 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
3696 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
3698 return nsa
->head
->ns_id
- nsb
->head
->ns_id
;
3701 struct nvme_ns
*nvme_find_get_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3703 struct nvme_ns
*ns
, *ret
= NULL
;
3705 down_read(&ctrl
->namespaces_rwsem
);
3706 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
3707 if (ns
->head
->ns_id
== nsid
) {
3708 if (!nvme_get_ns(ns
))
3713 if (ns
->head
->ns_id
> nsid
)
3716 up_read(&ctrl
->namespaces_rwsem
);
3719 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns
, NVME_TARGET_PASSTHRU
);
3721 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
,
3722 struct nvme_ns_ids
*ids
)
3725 struct gendisk
*disk
;
3726 struct nvme_id_ns
*id
;
3727 int node
= ctrl
->numa_node
;
3729 if (nvme_identify_ns(ctrl
, nsid
, ids
, &id
))
3732 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
3736 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
3737 if (IS_ERR(ns
->queue
))
3740 if (ctrl
->opts
&& ctrl
->opts
->data_digest
)
3741 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES
, ns
->queue
);
3743 blk_queue_flag_set(QUEUE_FLAG_NONROT
, ns
->queue
);
3744 if (ctrl
->ops
->flags
& NVME_F_PCI_P2PDMA
)
3745 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA
, ns
->queue
);
3747 ns
->queue
->queuedata
= ns
;
3749 kref_init(&ns
->kref
);
3751 if (nvme_init_ns_head(ns
, nsid
, ids
, id
->nmic
& NVME_NS_NMIC_SHARED
))
3752 goto out_free_queue
;
3754 disk
= alloc_disk_node(0, node
);
3758 disk
->fops
= &nvme_bdev_ops
;
3759 disk
->private_data
= ns
;
3760 disk
->queue
= ns
->queue
;
3762 * Without the multipath code enabled, multiple controller per
3763 * subsystems are visible as devices and thus we cannot use the
3764 * subsystem instance.
3766 if (!nvme_mpath_set_disk_name(ns
, disk
->disk_name
, &disk
->flags
))
3767 sprintf(disk
->disk_name
, "nvme%dn%d", ctrl
->instance
,
3768 ns
->head
->instance
);
3771 if (nvme_update_ns_info(ns
, id
))
3774 if ((ctrl
->quirks
& NVME_QUIRK_LIGHTNVM
) && id
->vs
[0] == 0x1) {
3775 if (nvme_nvm_register(ns
, disk
->disk_name
, node
)) {
3776 dev_warn(ctrl
->device
, "LightNVM init failure\n");
3781 down_write(&ctrl
->namespaces_rwsem
);
3782 list_add_tail(&ns
->list
, &ctrl
->namespaces
);
3783 up_write(&ctrl
->namespaces_rwsem
);
3785 nvme_get_ctrl(ctrl
);
3787 device_add_disk(ctrl
->device
, ns
->disk
, nvme_ns_id_attr_groups
);
3788 if (!nvme_ns_head_multipath(ns
->head
))
3789 nvme_add_ns_cdev(ns
);
3791 nvme_mpath_add_disk(ns
, id
);
3792 nvme_fault_inject_init(&ns
->fault_inject
, ns
->disk
->disk_name
);
3797 /* prevent double queue cleanup */
3798 ns
->disk
->queue
= NULL
;
3801 mutex_lock(&ctrl
->subsys
->lock
);
3802 list_del_rcu(&ns
->siblings
);
3803 if (list_empty(&ns
->head
->list
))
3804 list_del_init(&ns
->head
->entry
);
3805 mutex_unlock(&ctrl
->subsys
->lock
);
3806 nvme_put_ns_head(ns
->head
);
3808 blk_cleanup_queue(ns
->queue
);
3815 static void nvme_ns_remove(struct nvme_ns
*ns
)
3817 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
3820 set_capacity(ns
->disk
, 0);
3821 nvme_fault_inject_fini(&ns
->fault_inject
);
3823 mutex_lock(&ns
->ctrl
->subsys
->lock
);
3824 list_del_rcu(&ns
->siblings
);
3825 if (list_empty(&ns
->head
->list
))
3826 list_del_init(&ns
->head
->entry
);
3827 mutex_unlock(&ns
->ctrl
->subsys
->lock
);
3829 synchronize_rcu(); /* guarantee not available in head->list */
3830 nvme_mpath_clear_current_path(ns
);
3831 synchronize_srcu(&ns
->head
->srcu
); /* wait for concurrent submissions */
3833 if (ns
->disk
->flags
& GENHD_FL_UP
) {
3834 if (!nvme_ns_head_multipath(ns
->head
))
3835 nvme_cdev_del(&ns
->cdev
, &ns
->cdev_device
);
3836 del_gendisk(ns
->disk
);
3837 blk_cleanup_queue(ns
->queue
);
3838 if (blk_get_integrity(ns
->disk
))
3839 blk_integrity_unregister(ns
->disk
);
3842 down_write(&ns
->ctrl
->namespaces_rwsem
);
3843 list_del_init(&ns
->list
);
3844 up_write(&ns
->ctrl
->namespaces_rwsem
);
3846 nvme_mpath_check_last_path(ns
);
3850 static void nvme_ns_remove_by_nsid(struct nvme_ctrl
*ctrl
, u32 nsid
)
3852 struct nvme_ns
*ns
= nvme_find_get_ns(ctrl
, nsid
);
3860 static void nvme_validate_ns(struct nvme_ns
*ns
, struct nvme_ns_ids
*ids
)
3862 struct nvme_id_ns
*id
;
3863 int ret
= NVME_SC_INVALID_NS
| NVME_SC_DNR
;
3865 if (test_bit(NVME_NS_DEAD
, &ns
->flags
))
3868 ret
= nvme_identify_ns(ns
->ctrl
, ns
->head
->ns_id
, ids
, &id
);
3872 ret
= NVME_SC_INVALID_NS
| NVME_SC_DNR
;
3873 if (!nvme_ns_ids_equal(&ns
->head
->ids
, ids
)) {
3874 dev_err(ns
->ctrl
->device
,
3875 "identifiers changed for nsid %d\n", ns
->head
->ns_id
);
3879 ret
= nvme_update_ns_info(ns
, id
);
3885 * Only remove the namespace if we got a fatal error back from the
3886 * device, otherwise ignore the error and just move on.
3888 * TODO: we should probably schedule a delayed retry here.
3890 if (ret
> 0 && (ret
& NVME_SC_DNR
))
3894 static void nvme_validate_or_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
3896 struct nvme_ns_ids ids
= { };
3899 if (nvme_identify_ns_descs(ctrl
, nsid
, &ids
))
3902 ns
= nvme_find_get_ns(ctrl
, nsid
);
3904 nvme_validate_ns(ns
, &ids
);
3911 nvme_alloc_ns(ctrl
, nsid
, &ids
);
3914 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED
)) {
3915 dev_warn(ctrl
->device
,
3916 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3920 if (!nvme_multi_css(ctrl
)) {
3921 dev_warn(ctrl
->device
,
3922 "command set not reported for nsid: %d\n",
3926 nvme_alloc_ns(ctrl
, nsid
, &ids
);
3929 dev_warn(ctrl
->device
, "unknown csi %u for nsid %u\n",
3935 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
3938 struct nvme_ns
*ns
, *next
;
3941 down_write(&ctrl
->namespaces_rwsem
);
3942 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
3943 if (ns
->head
->ns_id
> nsid
|| test_bit(NVME_NS_DEAD
, &ns
->flags
))
3944 list_move_tail(&ns
->list
, &rm_list
);
3946 up_write(&ctrl
->namespaces_rwsem
);
3948 list_for_each_entry_safe(ns
, next
, &rm_list
, list
)
3953 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
)
3955 const int nr_entries
= NVME_IDENTIFY_DATA_SIZE
/ sizeof(__le32
);
3960 if (nvme_ctrl_limited_cns(ctrl
))
3963 ns_list
= kzalloc(NVME_IDENTIFY_DATA_SIZE
, GFP_KERNEL
);
3968 struct nvme_command cmd
= {
3969 .identify
.opcode
= nvme_admin_identify
,
3970 .identify
.cns
= NVME_ID_CNS_NS_ACTIVE_LIST
,
3971 .identify
.nsid
= cpu_to_le32(prev
),
3974 ret
= nvme_submit_sync_cmd(ctrl
->admin_q
, &cmd
, ns_list
,
3975 NVME_IDENTIFY_DATA_SIZE
);
3977 dev_warn(ctrl
->device
,
3978 "Identify NS List failed (status=0x%x)\n", ret
);
3982 for (i
= 0; i
< nr_entries
; i
++) {
3983 u32 nsid
= le32_to_cpu(ns_list
[i
]);
3985 if (!nsid
) /* end of the list? */
3987 nvme_validate_or_alloc_ns(ctrl
, nsid
);
3988 while (++prev
< nsid
)
3989 nvme_ns_remove_by_nsid(ctrl
, prev
);
3993 nvme_remove_invalid_namespaces(ctrl
, prev
);
3999 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
)
4001 struct nvme_id_ctrl
*id
;
4004 if (nvme_identify_ctrl(ctrl
, &id
))
4006 nn
= le32_to_cpu(id
->nn
);
4009 for (i
= 1; i
<= nn
; i
++)
4010 nvme_validate_or_alloc_ns(ctrl
, i
);
4012 nvme_remove_invalid_namespaces(ctrl
, nn
);
4015 static void nvme_clear_changed_ns_log(struct nvme_ctrl
*ctrl
)
4017 size_t log_size
= NVME_MAX_CHANGED_NAMESPACES
* sizeof(__le32
);
4021 log
= kzalloc(log_size
, GFP_KERNEL
);
4026 * We need to read the log to clear the AEN, but we don't want to rely
4027 * on it for the changed namespace information as userspace could have
4028 * raced with us in reading the log page, which could cause us to miss
4031 error
= nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_CHANGED_NS
, 0,
4032 NVME_CSI_NVM
, log
, log_size
, 0);
4034 dev_warn(ctrl
->device
,
4035 "reading changed ns log failed: %d\n", error
);
4040 static void nvme_scan_work(struct work_struct
*work
)
4042 struct nvme_ctrl
*ctrl
=
4043 container_of(work
, struct nvme_ctrl
, scan_work
);
4045 /* No tagset on a live ctrl means IO queues could not created */
4046 if (ctrl
->state
!= NVME_CTRL_LIVE
|| !ctrl
->tagset
)
4049 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
)) {
4050 dev_info(ctrl
->device
, "rescanning namespaces.\n");
4051 nvme_clear_changed_ns_log(ctrl
);
4054 mutex_lock(&ctrl
->scan_lock
);
4055 if (nvme_scan_ns_list(ctrl
) != 0)
4056 nvme_scan_ns_sequential(ctrl
);
4057 mutex_unlock(&ctrl
->scan_lock
);
4059 down_write(&ctrl
->namespaces_rwsem
);
4060 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
4061 up_write(&ctrl
->namespaces_rwsem
);
4065 * This function iterates the namespace list unlocked to allow recovery from
4066 * controller failure. It is up to the caller to ensure the namespace list is
4067 * not modified by scan work while this function is executing.
4069 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
4071 struct nvme_ns
*ns
, *next
;
4075 * make sure to requeue I/O to all namespaces as these
4076 * might result from the scan itself and must complete
4077 * for the scan_work to make progress
4079 nvme_mpath_clear_ctrl_paths(ctrl
);
4081 /* prevent racing with ns scanning */
4082 flush_work(&ctrl
->scan_work
);
4085 * The dead states indicates the controller was not gracefully
4086 * disconnected. In that case, we won't be able to flush any data while
4087 * removing the namespaces' disks; fail all the queues now to avoid
4088 * potentially having to clean up the failed sync later.
4090 if (ctrl
->state
== NVME_CTRL_DEAD
)
4091 nvme_kill_queues(ctrl
);
4093 /* this is a no-op when called from the controller reset handler */
4094 nvme_change_ctrl_state(ctrl
, NVME_CTRL_DELETING_NOIO
);
4096 down_write(&ctrl
->namespaces_rwsem
);
4097 list_splice_init(&ctrl
->namespaces
, &ns_list
);
4098 up_write(&ctrl
->namespaces_rwsem
);
4100 list_for_each_entry_safe(ns
, next
, &ns_list
, list
)
4103 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
4105 static int nvme_class_uevent(struct device
*dev
, struct kobj_uevent_env
*env
)
4107 struct nvme_ctrl
*ctrl
=
4108 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
4109 struct nvmf_ctrl_options
*opts
= ctrl
->opts
;
4112 ret
= add_uevent_var(env
, "NVME_TRTYPE=%s", ctrl
->ops
->name
);
4117 ret
= add_uevent_var(env
, "NVME_TRADDR=%s", opts
->traddr
);
4121 ret
= add_uevent_var(env
, "NVME_TRSVCID=%s",
4122 opts
->trsvcid
?: "none");
4126 ret
= add_uevent_var(env
, "NVME_HOST_TRADDR=%s",
4127 opts
->host_traddr
?: "none");
4131 ret
= add_uevent_var(env
, "NVME_HOST_IFACE=%s",
4132 opts
->host_iface
?: "none");
4137 static void nvme_aen_uevent(struct nvme_ctrl
*ctrl
)
4139 char *envp
[2] = { NULL
, NULL
};
4140 u32 aen_result
= ctrl
->aen_result
;
4142 ctrl
->aen_result
= 0;
4146 envp
[0] = kasprintf(GFP_KERNEL
, "NVME_AEN=%#08x", aen_result
);
4149 kobject_uevent_env(&ctrl
->device
->kobj
, KOBJ_CHANGE
, envp
);
4153 static void nvme_async_event_work(struct work_struct
*work
)
4155 struct nvme_ctrl
*ctrl
=
4156 container_of(work
, struct nvme_ctrl
, async_event_work
);
4158 nvme_aen_uevent(ctrl
);
4159 ctrl
->ops
->submit_async_event(ctrl
);
4162 static bool nvme_ctrl_pp_status(struct nvme_ctrl
*ctrl
)
4167 if (ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
))
4173 return ((ctrl
->ctrl_config
& NVME_CC_ENABLE
) && (csts
& NVME_CSTS_PP
));
4176 static void nvme_get_fw_slot_info(struct nvme_ctrl
*ctrl
)
4178 struct nvme_fw_slot_info_log
*log
;
4180 log
= kmalloc(sizeof(*log
), GFP_KERNEL
);
4184 if (nvme_get_log(ctrl
, NVME_NSID_ALL
, NVME_LOG_FW_SLOT
, 0, NVME_CSI_NVM
,
4185 log
, sizeof(*log
), 0))
4186 dev_warn(ctrl
->device
, "Get FW SLOT INFO log error\n");
4190 static void nvme_fw_act_work(struct work_struct
*work
)
4192 struct nvme_ctrl
*ctrl
= container_of(work
,
4193 struct nvme_ctrl
, fw_act_work
);
4194 unsigned long fw_act_timeout
;
4197 fw_act_timeout
= jiffies
+
4198 msecs_to_jiffies(ctrl
->mtfa
* 100);
4200 fw_act_timeout
= jiffies
+
4201 msecs_to_jiffies(admin_timeout
* 1000);
4203 nvme_stop_queues(ctrl
);
4204 while (nvme_ctrl_pp_status(ctrl
)) {
4205 if (time_after(jiffies
, fw_act_timeout
)) {
4206 dev_warn(ctrl
->device
,
4207 "Fw activation timeout, reset controller\n");
4208 nvme_try_sched_reset(ctrl
);
4214 if (!nvme_change_ctrl_state(ctrl
, NVME_CTRL_LIVE
))
4217 nvme_start_queues(ctrl
);
4218 /* read FW slot information to clear the AER */
4219 nvme_get_fw_slot_info(ctrl
);
4222 static void nvme_handle_aen_notice(struct nvme_ctrl
*ctrl
, u32 result
)
4224 u32 aer_notice_type
= (result
& 0xff00) >> 8;
4226 trace_nvme_async_event(ctrl
, aer_notice_type
);
4228 switch (aer_notice_type
) {
4229 case NVME_AER_NOTICE_NS_CHANGED
:
4230 set_bit(NVME_AER_NOTICE_NS_CHANGED
, &ctrl
->events
);
4231 nvme_queue_scan(ctrl
);
4233 case NVME_AER_NOTICE_FW_ACT_STARTING
:
4235 * We are (ab)using the RESETTING state to prevent subsequent
4236 * recovery actions from interfering with the controller's
4237 * firmware activation.
4239 if (nvme_change_ctrl_state(ctrl
, NVME_CTRL_RESETTING
))
4240 queue_work(nvme_wq
, &ctrl
->fw_act_work
);
4242 #ifdef CONFIG_NVME_MULTIPATH
4243 case NVME_AER_NOTICE_ANA
:
4244 if (!ctrl
->ana_log_buf
)
4246 queue_work(nvme_wq
, &ctrl
->ana_work
);
4249 case NVME_AER_NOTICE_DISC_CHANGED
:
4250 ctrl
->aen_result
= result
;
4253 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
4257 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
, __le16 status
,
4258 volatile union nvme_result
*res
)
4260 u32 result
= le32_to_cpu(res
->u32
);
4261 u32 aer_type
= result
& 0x07;
4263 if (le16_to_cpu(status
) >> 1 != NVME_SC_SUCCESS
)
4267 case NVME_AER_NOTICE
:
4268 nvme_handle_aen_notice(ctrl
, result
);
4270 case NVME_AER_ERROR
:
4271 case NVME_AER_SMART
:
4274 trace_nvme_async_event(ctrl
, aer_type
);
4275 ctrl
->aen_result
= result
;
4280 queue_work(nvme_wq
, &ctrl
->async_event_work
);
4282 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
4284 void nvme_stop_ctrl(struct nvme_ctrl
*ctrl
)
4286 nvme_mpath_stop(ctrl
);
4287 nvme_stop_keep_alive(ctrl
);
4288 nvme_stop_failfast_work(ctrl
);
4289 flush_work(&ctrl
->async_event_work
);
4290 cancel_work_sync(&ctrl
->fw_act_work
);
4292 EXPORT_SYMBOL_GPL(nvme_stop_ctrl
);
4294 void nvme_start_ctrl(struct nvme_ctrl
*ctrl
)
4296 nvme_start_keep_alive(ctrl
);
4298 nvme_enable_aen(ctrl
);
4300 if (ctrl
->queue_count
> 1) {
4301 nvme_queue_scan(ctrl
);
4302 nvme_start_queues(ctrl
);
4305 EXPORT_SYMBOL_GPL(nvme_start_ctrl
);
4307 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
4309 nvme_hwmon_exit(ctrl
);
4310 nvme_fault_inject_fini(&ctrl
->fault_inject
);
4311 dev_pm_qos_hide_latency_tolerance(ctrl
->device
);
4312 cdev_device_del(&ctrl
->cdev
, ctrl
->device
);
4313 nvme_put_ctrl(ctrl
);
4315 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
4317 static void nvme_free_cels(struct nvme_ctrl
*ctrl
)
4319 struct nvme_effects_log
*cel
;
4322 xa_for_each(&ctrl
->cels
, i
, cel
) {
4323 xa_erase(&ctrl
->cels
, i
);
4327 xa_destroy(&ctrl
->cels
);
4330 static void nvme_free_ctrl(struct device
*dev
)
4332 struct nvme_ctrl
*ctrl
=
4333 container_of(dev
, struct nvme_ctrl
, ctrl_device
);
4334 struct nvme_subsystem
*subsys
= ctrl
->subsys
;
4336 if (!subsys
|| ctrl
->instance
!= subsys
->instance
)
4337 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4339 nvme_free_cels(ctrl
);
4340 nvme_mpath_uninit(ctrl
);
4341 __free_page(ctrl
->discard_page
);
4344 mutex_lock(&nvme_subsystems_lock
);
4345 list_del(&ctrl
->subsys_entry
);
4346 sysfs_remove_link(&subsys
->dev
.kobj
, dev_name(ctrl
->device
));
4347 mutex_unlock(&nvme_subsystems_lock
);
4350 ctrl
->ops
->free_ctrl(ctrl
);
4353 nvme_put_subsystem(subsys
);
4357 * Initialize a NVMe controller structures. This needs to be called during
4358 * earliest initialization so that we have the initialized structured around
4361 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
4362 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
4366 ctrl
->state
= NVME_CTRL_NEW
;
4367 clear_bit(NVME_CTRL_FAILFAST_EXPIRED
, &ctrl
->flags
);
4368 spin_lock_init(&ctrl
->lock
);
4369 mutex_init(&ctrl
->scan_lock
);
4370 INIT_LIST_HEAD(&ctrl
->namespaces
);
4371 xa_init(&ctrl
->cels
);
4372 init_rwsem(&ctrl
->namespaces_rwsem
);
4375 ctrl
->quirks
= quirks
;
4376 ctrl
->numa_node
= NUMA_NO_NODE
;
4377 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
4378 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
4379 INIT_WORK(&ctrl
->fw_act_work
, nvme_fw_act_work
);
4380 INIT_WORK(&ctrl
->delete_work
, nvme_delete_ctrl_work
);
4381 init_waitqueue_head(&ctrl
->state_wq
);
4383 INIT_DELAYED_WORK(&ctrl
->ka_work
, nvme_keep_alive_work
);
4384 INIT_DELAYED_WORK(&ctrl
->failfast_work
, nvme_failfast_work
);
4385 memset(&ctrl
->ka_cmd
, 0, sizeof(ctrl
->ka_cmd
));
4386 ctrl
->ka_cmd
.common
.opcode
= nvme_admin_keep_alive
;
4388 BUILD_BUG_ON(NVME_DSM_MAX_RANGES
* sizeof(struct nvme_dsm_range
) >
4390 ctrl
->discard_page
= alloc_page(GFP_KERNEL
);
4391 if (!ctrl
->discard_page
) {
4396 ret
= ida_simple_get(&nvme_instance_ida
, 0, 0, GFP_KERNEL
);
4399 ctrl
->instance
= ret
;
4401 device_initialize(&ctrl
->ctrl_device
);
4402 ctrl
->device
= &ctrl
->ctrl_device
;
4403 ctrl
->device
->devt
= MKDEV(MAJOR(nvme_ctrl_base_chr_devt
),
4405 ctrl
->device
->class = nvme_class
;
4406 ctrl
->device
->parent
= ctrl
->dev
;
4407 ctrl
->device
->groups
= nvme_dev_attr_groups
;
4408 ctrl
->device
->release
= nvme_free_ctrl
;
4409 dev_set_drvdata(ctrl
->device
, ctrl
);
4410 ret
= dev_set_name(ctrl
->device
, "nvme%d", ctrl
->instance
);
4412 goto out_release_instance
;
4414 nvme_get_ctrl(ctrl
);
4415 cdev_init(&ctrl
->cdev
, &nvme_dev_fops
);
4416 ctrl
->cdev
.owner
= ops
->module
;
4417 ret
= cdev_device_add(&ctrl
->cdev
, ctrl
->device
);
4422 * Initialize latency tolerance controls. The sysfs files won't
4423 * be visible to userspace unless the device actually supports APST.
4425 ctrl
->device
->power
.set_latency_tolerance
= nvme_set_latency_tolerance
;
4426 dev_pm_qos_update_user_latency_tolerance(ctrl
->device
,
4427 min(default_ps_max_latency_us
, (unsigned long)S32_MAX
));
4429 nvme_fault_inject_init(&ctrl
->fault_inject
, dev_name(ctrl
->device
));
4430 nvme_mpath_init_ctrl(ctrl
);
4434 nvme_put_ctrl(ctrl
);
4435 kfree_const(ctrl
->device
->kobj
.name
);
4436 out_release_instance
:
4437 ida_simple_remove(&nvme_instance_ida
, ctrl
->instance
);
4439 if (ctrl
->discard_page
)
4440 __free_page(ctrl
->discard_page
);
4443 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
4446 * nvme_kill_queues(): Ends all namespace queues
4447 * @ctrl: the dead controller that needs to end
4449 * Call this function when the driver determines it is unable to get the
4450 * controller in a state capable of servicing IO.
4452 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
4456 down_read(&ctrl
->namespaces_rwsem
);
4458 /* Forcibly unquiesce queues to avoid blocking dispatch */
4459 if (ctrl
->admin_q
&& !blk_queue_dying(ctrl
->admin_q
))
4460 blk_mq_unquiesce_queue(ctrl
->admin_q
);
4462 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4463 nvme_set_queue_dying(ns
);
4465 up_read(&ctrl
->namespaces_rwsem
);
4467 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
4469 void nvme_unfreeze(struct nvme_ctrl
*ctrl
)
4473 down_read(&ctrl
->namespaces_rwsem
);
4474 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4475 blk_mq_unfreeze_queue(ns
->queue
);
4476 up_read(&ctrl
->namespaces_rwsem
);
4478 EXPORT_SYMBOL_GPL(nvme_unfreeze
);
4480 int nvme_wait_freeze_timeout(struct nvme_ctrl
*ctrl
, long timeout
)
4484 down_read(&ctrl
->namespaces_rwsem
);
4485 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
4486 timeout
= blk_mq_freeze_queue_wait_timeout(ns
->queue
, timeout
);
4490 up_read(&ctrl
->namespaces_rwsem
);
4493 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout
);
4495 void nvme_wait_freeze(struct nvme_ctrl
*ctrl
)
4499 down_read(&ctrl
->namespaces_rwsem
);
4500 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4501 blk_mq_freeze_queue_wait(ns
->queue
);
4502 up_read(&ctrl
->namespaces_rwsem
);
4504 EXPORT_SYMBOL_GPL(nvme_wait_freeze
);
4506 void nvme_start_freeze(struct nvme_ctrl
*ctrl
)
4510 down_read(&ctrl
->namespaces_rwsem
);
4511 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4512 blk_freeze_queue_start(ns
->queue
);
4513 up_read(&ctrl
->namespaces_rwsem
);
4515 EXPORT_SYMBOL_GPL(nvme_start_freeze
);
4517 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
4521 down_read(&ctrl
->namespaces_rwsem
);
4522 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4523 blk_mq_quiesce_queue(ns
->queue
);
4524 up_read(&ctrl
->namespaces_rwsem
);
4526 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
4528 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
4532 down_read(&ctrl
->namespaces_rwsem
);
4533 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4534 blk_mq_unquiesce_queue(ns
->queue
);
4535 up_read(&ctrl
->namespaces_rwsem
);
4537 EXPORT_SYMBOL_GPL(nvme_start_queues
);
4539 void nvme_sync_io_queues(struct nvme_ctrl
*ctrl
)
4543 down_read(&ctrl
->namespaces_rwsem
);
4544 list_for_each_entry(ns
, &ctrl
->namespaces
, list
)
4545 blk_sync_queue(ns
->queue
);
4546 up_read(&ctrl
->namespaces_rwsem
);
4548 EXPORT_SYMBOL_GPL(nvme_sync_io_queues
);
4550 void nvme_sync_queues(struct nvme_ctrl
*ctrl
)
4552 nvme_sync_io_queues(ctrl
);
4554 blk_sync_queue(ctrl
->admin_q
);
4556 EXPORT_SYMBOL_GPL(nvme_sync_queues
);
4558 struct nvme_ctrl
*nvme_ctrl_from_file(struct file
*file
)
4560 if (file
->f_op
!= &nvme_dev_fops
)
4562 return file
->private_data
;
4564 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file
, NVME_TARGET_PASSTHRU
);
4567 * Check we didn't inadvertently grow the command structure sizes:
4569 static inline void _nvme_check_size(void)
4571 BUILD_BUG_ON(sizeof(struct nvme_common_command
) != 64);
4572 BUILD_BUG_ON(sizeof(struct nvme_rw_command
) != 64);
4573 BUILD_BUG_ON(sizeof(struct nvme_identify
) != 64);
4574 BUILD_BUG_ON(sizeof(struct nvme_features
) != 64);
4575 BUILD_BUG_ON(sizeof(struct nvme_download_firmware
) != 64);
4576 BUILD_BUG_ON(sizeof(struct nvme_format_cmd
) != 64);
4577 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd
) != 64);
4578 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd
) != 64);
4579 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd
) != 64);
4580 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command
) != 64);
4581 BUILD_BUG_ON(sizeof(struct nvme_command
) != 64);
4582 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl
) != NVME_IDENTIFY_DATA_SIZE
);
4583 BUILD_BUG_ON(sizeof(struct nvme_id_ns
) != NVME_IDENTIFY_DATA_SIZE
);
4584 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns
) != NVME_IDENTIFY_DATA_SIZE
);
4585 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns
) != NVME_IDENTIFY_DATA_SIZE
);
4586 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm
) != NVME_IDENTIFY_DATA_SIZE
);
4587 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type
) != 64);
4588 BUILD_BUG_ON(sizeof(struct nvme_smart_log
) != 512);
4589 BUILD_BUG_ON(sizeof(struct nvme_dbbuf
) != 64);
4590 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd
) != 64);
4594 static int __init
nvme_core_init(void)
4596 int result
= -ENOMEM
;
4600 nvme_wq
= alloc_workqueue("nvme-wq",
4601 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4605 nvme_reset_wq
= alloc_workqueue("nvme-reset-wq",
4606 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4610 nvme_delete_wq
= alloc_workqueue("nvme-delete-wq",
4611 WQ_UNBOUND
| WQ_MEM_RECLAIM
| WQ_SYSFS
, 0);
4612 if (!nvme_delete_wq
)
4613 goto destroy_reset_wq
;
4615 result
= alloc_chrdev_region(&nvme_ctrl_base_chr_devt
, 0,
4616 NVME_MINORS
, "nvme");
4618 goto destroy_delete_wq
;
4620 nvme_class
= class_create(THIS_MODULE
, "nvme");
4621 if (IS_ERR(nvme_class
)) {
4622 result
= PTR_ERR(nvme_class
);
4623 goto unregister_chrdev
;
4625 nvme_class
->dev_uevent
= nvme_class_uevent
;
4627 nvme_subsys_class
= class_create(THIS_MODULE
, "nvme-subsystem");
4628 if (IS_ERR(nvme_subsys_class
)) {
4629 result
= PTR_ERR(nvme_subsys_class
);
4633 result
= alloc_chrdev_region(&nvme_ns_chr_devt
, 0, NVME_MINORS
,
4636 goto destroy_subsys_class
;
4638 nvme_ns_chr_class
= class_create(THIS_MODULE
, "nvme-generic");
4639 if (IS_ERR(nvme_ns_chr_class
)) {
4640 result
= PTR_ERR(nvme_ns_chr_class
);
4641 goto unregister_generic_ns
;
4646 unregister_generic_ns
:
4647 unregister_chrdev_region(nvme_ns_chr_devt
, NVME_MINORS
);
4648 destroy_subsys_class
:
4649 class_destroy(nvme_subsys_class
);
4651 class_destroy(nvme_class
);
4653 unregister_chrdev_region(nvme_ctrl_base_chr_devt
, NVME_MINORS
);
4655 destroy_workqueue(nvme_delete_wq
);
4657 destroy_workqueue(nvme_reset_wq
);
4659 destroy_workqueue(nvme_wq
);
4664 static void __exit
nvme_core_exit(void)
4666 class_destroy(nvme_ns_chr_class
);
4667 class_destroy(nvme_subsys_class
);
4668 class_destroy(nvme_class
);
4669 unregister_chrdev_region(nvme_ns_chr_devt
, NVME_MINORS
);
4670 unregister_chrdev_region(nvme_ctrl_base_chr_devt
, NVME_MINORS
);
4671 destroy_workqueue(nvme_delete_wq
);
4672 destroy_workqueue(nvme_reset_wq
);
4673 destroy_workqueue(nvme_wq
);
4674 ida_destroy(&nvme_ns_chr_minor_ida
);
4675 ida_destroy(&nvme_instance_ida
);
4678 MODULE_LICENSE("GPL");
4679 MODULE_VERSION("1.0");
4680 module_init(nvme_core_init
);
4681 module_exit(nvme_core_exit
);