2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
30 #include <asm/unaligned.h>
35 #define NVME_MINORS (1U << MINORBITS)
37 unsigned char admin_timeout
= 60;
38 module_param(admin_timeout
, byte
, 0644);
39 MODULE_PARM_DESC(admin_timeout
, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout
);
42 unsigned char nvme_io_timeout
= 30;
43 module_param_named(io_timeout
, nvme_io_timeout
, byte
, 0644);
44 MODULE_PARM_DESC(io_timeout
, "timeout in seconds for I/O");
45 EXPORT_SYMBOL_GPL(nvme_io_timeout
);
47 unsigned char shutdown_timeout
= 5;
48 module_param(shutdown_timeout
, byte
, 0644);
49 MODULE_PARM_DESC(shutdown_timeout
, "timeout in seconds for controller shutdown");
51 unsigned int nvme_max_retries
= 5;
52 module_param_named(max_retries
, nvme_max_retries
, uint
, 0644);
53 MODULE_PARM_DESC(max_retries
, "max number of retries a command may have");
54 EXPORT_SYMBOL_GPL(nvme_max_retries
);
56 static int nvme_char_major
;
57 module_param(nvme_char_major
, int, 0);
59 static LIST_HEAD(nvme_ctrl_list
);
60 static DEFINE_SPINLOCK(dev_list_lock
);
62 static struct class *nvme_class
;
64 void nvme_cancel_request(struct request
*req
, void *data
, bool reserved
)
68 if (!blk_mq_request_started(req
))
71 dev_dbg_ratelimited(((struct nvme_ctrl
*) data
)->device
,
72 "Cancelling I/O %d", req
->tag
);
74 status
= NVME_SC_ABORT_REQ
;
75 if (blk_queue_dying(req
->q
))
76 status
|= NVME_SC_DNR
;
77 blk_mq_complete_request(req
, status
);
79 EXPORT_SYMBOL_GPL(nvme_cancel_request
);
81 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
82 enum nvme_ctrl_state new_state
)
84 enum nvme_ctrl_state old_state
;
87 spin_lock_irq(&ctrl
->lock
);
89 old_state
= ctrl
->state
;
94 case NVME_CTRL_RESETTING
:
95 case NVME_CTRL_RECONNECTING
:
102 case NVME_CTRL_RESETTING
:
106 case NVME_CTRL_RECONNECTING
:
113 case NVME_CTRL_RECONNECTING
:
122 case NVME_CTRL_DELETING
:
125 case NVME_CTRL_RESETTING
:
126 case NVME_CTRL_RECONNECTING
:
135 case NVME_CTRL_DELETING
:
147 ctrl
->state
= new_state
;
149 spin_unlock_irq(&ctrl
->lock
);
153 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
155 static void nvme_free_ns(struct kref
*kref
)
157 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
160 nvme_nvm_unregister(ns
);
163 spin_lock(&dev_list_lock
);
164 ns
->disk
->private_data
= NULL
;
165 spin_unlock(&dev_list_lock
);
169 ida_simple_remove(&ns
->ctrl
->ns_ida
, ns
->instance
);
170 nvme_put_ctrl(ns
->ctrl
);
174 static void nvme_put_ns(struct nvme_ns
*ns
)
176 kref_put(&ns
->kref
, nvme_free_ns
);
179 static struct nvme_ns
*nvme_get_ns_from_disk(struct gendisk
*disk
)
183 spin_lock(&dev_list_lock
);
184 ns
= disk
->private_data
;
186 if (!kref_get_unless_zero(&ns
->kref
))
188 if (!try_module_get(ns
->ctrl
->ops
->module
))
191 spin_unlock(&dev_list_lock
);
196 kref_put(&ns
->kref
, nvme_free_ns
);
198 spin_unlock(&dev_list_lock
);
202 void nvme_requeue_req(struct request
*req
)
206 blk_mq_requeue_request(req
);
207 spin_lock_irqsave(req
->q
->queue_lock
, flags
);
208 if (!blk_queue_stopped(req
->q
))
209 blk_mq_kick_requeue_list(req
->q
);
210 spin_unlock_irqrestore(req
->q
->queue_lock
, flags
);
212 EXPORT_SYMBOL_GPL(nvme_requeue_req
);
214 struct request
*nvme_alloc_request(struct request_queue
*q
,
215 struct nvme_command
*cmd
, unsigned int flags
, int qid
)
219 if (qid
== NVME_QID_ANY
) {
220 req
= blk_mq_alloc_request(q
, nvme_is_write(cmd
), flags
);
222 req
= blk_mq_alloc_request_hctx(q
, nvme_is_write(cmd
), flags
,
228 req
->cmd_type
= REQ_TYPE_DRV_PRIV
;
229 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
230 req
->cmd
= (unsigned char *)cmd
;
231 req
->cmd_len
= sizeof(struct nvme_command
);
235 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
237 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
238 struct nvme_command
*cmnd
)
240 memset(cmnd
, 0, sizeof(*cmnd
));
241 cmnd
->common
.opcode
= nvme_cmd_flush
;
242 cmnd
->common
.nsid
= cpu_to_le32(ns
->ns_id
);
245 static inline int nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
246 struct nvme_command
*cmnd
)
248 struct nvme_dsm_range
*range
;
251 unsigned int nr_bytes
= blk_rq_bytes(req
);
253 range
= kmalloc(sizeof(*range
), GFP_ATOMIC
);
255 return BLK_MQ_RQ_QUEUE_BUSY
;
257 range
->cattr
= cpu_to_le32(0);
258 range
->nlb
= cpu_to_le32(nr_bytes
>> ns
->lba_shift
);
259 range
->slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
261 memset(cmnd
, 0, sizeof(*cmnd
));
262 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
263 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->ns_id
);
265 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
267 req
->completion_data
= range
;
268 page
= virt_to_page(range
);
269 offset
= offset_in_page(range
);
270 blk_add_request_payload(req
, page
, offset
, sizeof(*range
));
273 * we set __data_len back to the size of the area to be discarded
274 * on disk. This allows us to report completion on the full amount
275 * of blocks described by the request.
277 req
->__data_len
= nr_bytes
;
282 static inline void nvme_setup_rw(struct nvme_ns
*ns
, struct request
*req
,
283 struct nvme_command
*cmnd
)
288 if (req
->cmd_flags
& REQ_FUA
)
289 control
|= NVME_RW_FUA
;
290 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
291 control
|= NVME_RW_LR
;
293 if (req
->cmd_flags
& REQ_RAHEAD
)
294 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
296 memset(cmnd
, 0, sizeof(*cmnd
));
297 cmnd
->rw
.opcode
= (rq_data_dir(req
) ? nvme_cmd_write
: nvme_cmd_read
);
298 cmnd
->rw
.command_id
= req
->tag
;
299 cmnd
->rw
.nsid
= cpu_to_le32(ns
->ns_id
);
300 cmnd
->rw
.slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
301 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
304 switch (ns
->pi_type
) {
305 case NVME_NS_DPS_PI_TYPE3
:
306 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
308 case NVME_NS_DPS_PI_TYPE1
:
309 case NVME_NS_DPS_PI_TYPE2
:
310 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
311 NVME_RW_PRINFO_PRCHK_REF
;
312 cmnd
->rw
.reftag
= cpu_to_le32(
313 nvme_block_nr(ns
, blk_rq_pos(req
)));
316 if (!blk_integrity_rq(req
))
317 control
|= NVME_RW_PRINFO_PRACT
;
320 cmnd
->rw
.control
= cpu_to_le16(control
);
321 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
324 int nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
325 struct nvme_command
*cmd
)
329 if (req
->cmd_type
== REQ_TYPE_DRV_PRIV
)
330 memcpy(cmd
, req
->cmd
, sizeof(*cmd
));
331 else if (req_op(req
) == REQ_OP_FLUSH
)
332 nvme_setup_flush(ns
, cmd
);
333 else if (req_op(req
) == REQ_OP_DISCARD
)
334 ret
= nvme_setup_discard(ns
, req
, cmd
);
336 nvme_setup_rw(ns
, req
, cmd
);
340 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
343 * Returns 0 on success. If the result is negative, it's a Linux error code;
344 * if the result is positive, it's an NVM Express status code
346 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
347 struct nvme_completion
*cqe
, void *buffer
, unsigned bufflen
,
348 unsigned timeout
, int qid
, int at_head
, int flags
)
353 req
= nvme_alloc_request(q
, cmd
, flags
, qid
);
357 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
360 if (buffer
&& bufflen
) {
361 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
366 blk_execute_rq(req
->q
, NULL
, req
, at_head
);
369 blk_mq_free_request(req
);
372 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd
);
374 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
375 void *buffer
, unsigned bufflen
)
377 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0,
380 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
382 int __nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
383 void __user
*ubuffer
, unsigned bufflen
,
384 void __user
*meta_buffer
, unsigned meta_len
, u32 meta_seed
,
385 u32
*result
, unsigned timeout
)
387 bool write
= nvme_is_write(cmd
);
388 struct nvme_completion cqe
;
389 struct nvme_ns
*ns
= q
->queuedata
;
390 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
392 struct bio
*bio
= NULL
;
396 req
= nvme_alloc_request(q
, cmd
, 0, NVME_QID_ANY
);
400 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
403 if (ubuffer
&& bufflen
) {
404 ret
= blk_rq_map_user(q
, req
, NULL
, ubuffer
, bufflen
,
412 bio
->bi_bdev
= bdget_disk(disk
, 0);
418 if (meta_buffer
&& meta_len
) {
419 struct bio_integrity_payload
*bip
;
421 meta
= kmalloc(meta_len
, GFP_KERNEL
);
428 if (copy_from_user(meta
, meta_buffer
,
435 bip
= bio_integrity_alloc(bio
, GFP_KERNEL
, 1);
441 bip
->bip_iter
.bi_size
= meta_len
;
442 bip
->bip_iter
.bi_sector
= meta_seed
;
444 ret
= bio_integrity_add_page(bio
, virt_to_page(meta
),
445 meta_len
, offset_in_page(meta
));
446 if (ret
!= meta_len
) {
453 blk_execute_rq(req
->q
, disk
, req
, 0);
456 *result
= le32_to_cpu(cqe
.result
);
457 if (meta
&& !ret
&& !write
) {
458 if (copy_to_user(meta_buffer
, meta
, meta_len
))
465 if (disk
&& bio
->bi_bdev
)
467 blk_rq_unmap_user(bio
);
470 blk_mq_free_request(req
);
474 int nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
475 void __user
*ubuffer
, unsigned bufflen
, u32
*result
,
478 return __nvme_submit_user_cmd(q
, cmd
, ubuffer
, bufflen
, NULL
, 0, 0,
482 static void nvme_keep_alive_end_io(struct request
*rq
, int error
)
484 struct nvme_ctrl
*ctrl
= rq
->end_io_data
;
486 blk_mq_free_request(rq
);
489 dev_err(ctrl
->device
,
490 "failed nvme_keep_alive_end_io error=%d\n", error
);
494 schedule_delayed_work(&ctrl
->ka_work
, ctrl
->kato
* HZ
);
497 static int nvme_keep_alive(struct nvme_ctrl
*ctrl
)
499 struct nvme_command c
;
502 memset(&c
, 0, sizeof(c
));
503 c
.common
.opcode
= nvme_admin_keep_alive
;
505 rq
= nvme_alloc_request(ctrl
->admin_q
, &c
, BLK_MQ_REQ_RESERVED
,
510 rq
->timeout
= ctrl
->kato
* HZ
;
511 rq
->end_io_data
= ctrl
;
513 blk_execute_rq_nowait(rq
->q
, NULL
, rq
, 0, nvme_keep_alive_end_io
);
518 static void nvme_keep_alive_work(struct work_struct
*work
)
520 struct nvme_ctrl
*ctrl
= container_of(to_delayed_work(work
),
521 struct nvme_ctrl
, ka_work
);
523 if (nvme_keep_alive(ctrl
)) {
524 /* allocation failure, reset the controller */
525 dev_err(ctrl
->device
, "keep-alive failed\n");
526 ctrl
->ops
->reset_ctrl(ctrl
);
531 void nvme_start_keep_alive(struct nvme_ctrl
*ctrl
)
533 if (unlikely(ctrl
->kato
== 0))
536 INIT_DELAYED_WORK(&ctrl
->ka_work
, nvme_keep_alive_work
);
537 schedule_delayed_work(&ctrl
->ka_work
, ctrl
->kato
* HZ
);
539 EXPORT_SYMBOL_GPL(nvme_start_keep_alive
);
541 void nvme_stop_keep_alive(struct nvme_ctrl
*ctrl
)
543 if (unlikely(ctrl
->kato
== 0))
546 cancel_delayed_work_sync(&ctrl
->ka_work
);
548 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive
);
550 int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
552 struct nvme_command c
= { };
555 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
556 c
.identify
.opcode
= nvme_admin_identify
;
557 c
.identify
.cns
= cpu_to_le32(1);
559 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
563 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
564 sizeof(struct nvme_id_ctrl
));
570 static int nvme_identify_ns_list(struct nvme_ctrl
*dev
, unsigned nsid
, __le32
*ns_list
)
572 struct nvme_command c
= { };
574 c
.identify
.opcode
= nvme_admin_identify
;
575 c
.identify
.cns
= cpu_to_le32(2);
576 c
.identify
.nsid
= cpu_to_le32(nsid
);
577 return nvme_submit_sync_cmd(dev
->admin_q
, &c
, ns_list
, 0x1000);
580 int nvme_identify_ns(struct nvme_ctrl
*dev
, unsigned nsid
,
581 struct nvme_id_ns
**id
)
583 struct nvme_command c
= { };
586 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
587 c
.identify
.opcode
= nvme_admin_identify
,
588 c
.identify
.nsid
= cpu_to_le32(nsid
),
590 *id
= kmalloc(sizeof(struct nvme_id_ns
), GFP_KERNEL
);
594 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
595 sizeof(struct nvme_id_ns
));
601 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned nsid
,
602 void *buffer
, size_t buflen
, u32
*result
)
604 struct nvme_command c
;
605 struct nvme_completion cqe
;
608 memset(&c
, 0, sizeof(c
));
609 c
.features
.opcode
= nvme_admin_get_features
;
610 c
.features
.nsid
= cpu_to_le32(nsid
);
611 c
.features
.fid
= cpu_to_le32(fid
);
613 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
, buffer
, buflen
, 0,
615 if (ret
>= 0 && result
)
616 *result
= le32_to_cpu(cqe
.result
);
620 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned dword11
,
621 void *buffer
, size_t buflen
, u32
*result
)
623 struct nvme_command c
;
624 struct nvme_completion cqe
;
627 memset(&c
, 0, sizeof(c
));
628 c
.features
.opcode
= nvme_admin_set_features
;
629 c
.features
.fid
= cpu_to_le32(fid
);
630 c
.features
.dword11
= cpu_to_le32(dword11
);
632 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
,
633 buffer
, buflen
, 0, NVME_QID_ANY
, 0, 0);
634 if (ret
>= 0 && result
)
635 *result
= le32_to_cpu(cqe
.result
);
639 int nvme_get_log_page(struct nvme_ctrl
*dev
, struct nvme_smart_log
**log
)
641 struct nvme_command c
= { };
644 c
.common
.opcode
= nvme_admin_get_log_page
,
645 c
.common
.nsid
= cpu_to_le32(0xFFFFFFFF),
646 c
.common
.cdw10
[0] = cpu_to_le32(
647 (((sizeof(struct nvme_smart_log
) / 4) - 1) << 16) |
650 *log
= kmalloc(sizeof(struct nvme_smart_log
), GFP_KERNEL
);
654 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *log
,
655 sizeof(struct nvme_smart_log
));
661 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
663 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
665 int status
, nr_io_queues
;
667 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, NULL
, 0,
673 * Degraded controllers might return an error when setting the queue
674 * count. We still want to be able to bring them online and offer
675 * access to the admin queue, as that might be only way to fix them up.
678 dev_err(ctrl
->dev
, "Could not set queue count (%d)\n", status
);
681 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
682 *count
= min(*count
, nr_io_queues
);
687 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
689 static int nvme_submit_io(struct nvme_ns
*ns
, struct nvme_user_io __user
*uio
)
691 struct nvme_user_io io
;
692 struct nvme_command c
;
693 unsigned length
, meta_len
;
694 void __user
*metadata
;
696 if (copy_from_user(&io
, uio
, sizeof(io
)))
704 case nvme_cmd_compare
:
710 length
= (io
.nblocks
+ 1) << ns
->lba_shift
;
711 meta_len
= (io
.nblocks
+ 1) * ns
->ms
;
712 metadata
= (void __user
*)(uintptr_t)io
.metadata
;
717 } else if (meta_len
) {
718 if ((io
.metadata
& 3) || !io
.metadata
)
722 memset(&c
, 0, sizeof(c
));
723 c
.rw
.opcode
= io
.opcode
;
724 c
.rw
.flags
= io
.flags
;
725 c
.rw
.nsid
= cpu_to_le32(ns
->ns_id
);
726 c
.rw
.slba
= cpu_to_le64(io
.slba
);
727 c
.rw
.length
= cpu_to_le16(io
.nblocks
);
728 c
.rw
.control
= cpu_to_le16(io
.control
);
729 c
.rw
.dsmgmt
= cpu_to_le32(io
.dsmgmt
);
730 c
.rw
.reftag
= cpu_to_le32(io
.reftag
);
731 c
.rw
.apptag
= cpu_to_le16(io
.apptag
);
732 c
.rw
.appmask
= cpu_to_le16(io
.appmask
);
734 return __nvme_submit_user_cmd(ns
->queue
, &c
,
735 (void __user
*)(uintptr_t)io
.addr
, length
,
736 metadata
, meta_len
, io
.slba
, NULL
, 0);
739 static int nvme_user_cmd(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
740 struct nvme_passthru_cmd __user
*ucmd
)
742 struct nvme_passthru_cmd cmd
;
743 struct nvme_command c
;
744 unsigned timeout
= 0;
747 if (!capable(CAP_SYS_ADMIN
))
749 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
754 memset(&c
, 0, sizeof(c
));
755 c
.common
.opcode
= cmd
.opcode
;
756 c
.common
.flags
= cmd
.flags
;
757 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
758 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
759 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
760 c
.common
.cdw10
[0] = cpu_to_le32(cmd
.cdw10
);
761 c
.common
.cdw10
[1] = cpu_to_le32(cmd
.cdw11
);
762 c
.common
.cdw10
[2] = cpu_to_le32(cmd
.cdw12
);
763 c
.common
.cdw10
[3] = cpu_to_le32(cmd
.cdw13
);
764 c
.common
.cdw10
[4] = cpu_to_le32(cmd
.cdw14
);
765 c
.common
.cdw10
[5] = cpu_to_le32(cmd
.cdw15
);
768 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
770 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
771 (void __user
*)(uintptr_t)cmd
.addr
, cmd
.data_len
,
772 &cmd
.result
, timeout
);
774 if (put_user(cmd
.result
, &ucmd
->result
))
781 static int nvme_ioctl(struct block_device
*bdev
, fmode_t mode
,
782 unsigned int cmd
, unsigned long arg
)
784 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
788 force_successful_syscall_return();
790 case NVME_IOCTL_ADMIN_CMD
:
791 return nvme_user_cmd(ns
->ctrl
, NULL
, (void __user
*)arg
);
792 case NVME_IOCTL_IO_CMD
:
793 return nvme_user_cmd(ns
->ctrl
, ns
, (void __user
*)arg
);
794 case NVME_IOCTL_SUBMIT_IO
:
795 return nvme_submit_io(ns
, (void __user
*)arg
);
796 #ifdef CONFIG_BLK_DEV_NVME_SCSI
797 case SG_GET_VERSION_NUM
:
798 return nvme_sg_get_version_num((void __user
*)arg
);
800 return nvme_sg_io(ns
, (void __user
*)arg
);
808 static int nvme_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
809 unsigned int cmd
, unsigned long arg
)
815 return nvme_ioctl(bdev
, mode
, cmd
, arg
);
818 #define nvme_compat_ioctl NULL
821 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
823 return nvme_get_ns_from_disk(bdev
->bd_disk
) ? 0 : -ENXIO
;
826 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
828 struct nvme_ns
*ns
= disk
->private_data
;
830 module_put(ns
->ctrl
->ops
->module
);
834 static int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
836 /* some standard values */
838 geo
->sectors
= 1 << 5;
839 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
843 #ifdef CONFIG_BLK_DEV_INTEGRITY
844 static void nvme_init_integrity(struct nvme_ns
*ns
)
846 struct blk_integrity integrity
;
848 memset(&integrity
, 0, sizeof(integrity
));
849 switch (ns
->pi_type
) {
850 case NVME_NS_DPS_PI_TYPE3
:
851 integrity
.profile
= &t10_pi_type3_crc
;
852 integrity
.tag_size
= sizeof(u16
) + sizeof(u32
);
853 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
855 case NVME_NS_DPS_PI_TYPE1
:
856 case NVME_NS_DPS_PI_TYPE2
:
857 integrity
.profile
= &t10_pi_type1_crc
;
858 integrity
.tag_size
= sizeof(u16
);
859 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
862 integrity
.profile
= NULL
;
865 integrity
.tuple_size
= ns
->ms
;
866 blk_integrity_register(ns
->disk
, &integrity
);
867 blk_queue_max_integrity_segments(ns
->queue
, 1);
870 static void nvme_init_integrity(struct nvme_ns
*ns
)
873 #endif /* CONFIG_BLK_DEV_INTEGRITY */
875 static void nvme_config_discard(struct nvme_ns
*ns
)
877 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
878 u32 logical_block_size
= queue_logical_block_size(ns
->queue
);
880 if (ctrl
->quirks
& NVME_QUIRK_DISCARD_ZEROES
)
881 ns
->queue
->limits
.discard_zeroes_data
= 1;
883 ns
->queue
->limits
.discard_zeroes_data
= 0;
885 ns
->queue
->limits
.discard_alignment
= logical_block_size
;
886 ns
->queue
->limits
.discard_granularity
= logical_block_size
;
887 blk_queue_max_discard_sectors(ns
->queue
, UINT_MAX
);
888 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, ns
->queue
);
891 static int nvme_revalidate_ns(struct nvme_ns
*ns
, struct nvme_id_ns
**id
)
893 if (nvme_identify_ns(ns
->ctrl
, ns
->ns_id
, id
)) {
894 dev_warn(ns
->ctrl
->dev
, "%s: Identify failure\n", __func__
);
898 if ((*id
)->ncap
== 0) {
903 if (ns
->ctrl
->vs
>= NVME_VS(1, 1))
904 memcpy(ns
->eui
, (*id
)->eui64
, sizeof(ns
->eui
));
905 if (ns
->ctrl
->vs
>= NVME_VS(1, 2))
906 memcpy(ns
->uuid
, (*id
)->nguid
, sizeof(ns
->uuid
));
911 static void __nvme_revalidate_disk(struct gendisk
*disk
, struct nvme_id_ns
*id
)
913 struct nvme_ns
*ns
= disk
->private_data
;
919 lbaf
= id
->flbas
& NVME_NS_FLBAS_LBA_MASK
;
920 ns
->lba_shift
= id
->lbaf
[lbaf
].ds
;
921 ns
->ms
= le16_to_cpu(id
->lbaf
[lbaf
].ms
);
922 ns
->ext
= ns
->ms
&& (id
->flbas
& NVME_NS_FLBAS_META_EXT
);
925 * If identify namespace failed, use default 512 byte block size so
926 * block layer can use before failing read/write for 0 capacity.
928 if (ns
->lba_shift
== 0)
930 bs
= 1 << ns
->lba_shift
;
931 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
932 pi_type
= ns
->ms
== sizeof(struct t10_pi_tuple
) ?
933 id
->dps
& NVME_NS_DPS_PI_MASK
: 0;
935 blk_mq_freeze_queue(disk
->queue
);
936 if (blk_get_integrity(disk
) && (ns
->pi_type
!= pi_type
||
938 bs
!= queue_logical_block_size(disk
->queue
) ||
939 (ns
->ms
&& ns
->ext
)))
940 blk_integrity_unregister(disk
);
942 ns
->pi_type
= pi_type
;
943 blk_queue_logical_block_size(ns
->queue
, bs
);
945 if (ns
->ms
&& !blk_get_integrity(disk
) && !ns
->ext
)
946 nvme_init_integrity(ns
);
947 if (ns
->ms
&& !(ns
->ms
== 8 && ns
->pi_type
) && !blk_get_integrity(disk
))
948 set_capacity(disk
, 0);
950 set_capacity(disk
, le64_to_cpup(&id
->nsze
) << (ns
->lba_shift
- 9));
952 if (ns
->ctrl
->oncs
& NVME_CTRL_ONCS_DSM
)
953 nvme_config_discard(ns
);
954 blk_mq_unfreeze_queue(disk
->queue
);
957 static int nvme_revalidate_disk(struct gendisk
*disk
)
959 struct nvme_ns
*ns
= disk
->private_data
;
960 struct nvme_id_ns
*id
= NULL
;
963 if (test_bit(NVME_NS_DEAD
, &ns
->flags
)) {
964 set_capacity(disk
, 0);
968 ret
= nvme_revalidate_ns(ns
, &id
);
972 __nvme_revalidate_disk(disk
, id
);
978 static char nvme_pr_type(enum pr_type type
)
981 case PR_WRITE_EXCLUSIVE
:
983 case PR_EXCLUSIVE_ACCESS
:
985 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
987 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
989 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
991 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
998 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
999 u64 key
, u64 sa_key
, u8 op
)
1001 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
1002 struct nvme_command c
;
1003 u8 data
[16] = { 0, };
1005 put_unaligned_le64(key
, &data
[0]);
1006 put_unaligned_le64(sa_key
, &data
[8]);
1008 memset(&c
, 0, sizeof(c
));
1009 c
.common
.opcode
= op
;
1010 c
.common
.nsid
= cpu_to_le32(ns
->ns_id
);
1011 c
.common
.cdw10
[0] = cpu_to_le32(cdw10
);
1013 return nvme_submit_sync_cmd(ns
->queue
, &c
, data
, 16);
1016 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
1017 u64
new, unsigned flags
)
1021 if (flags
& ~PR_FL_IGNORE_KEY
)
1024 cdw10
= old
? 2 : 0;
1025 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
1026 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
1027 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
1030 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
1031 enum pr_type type
, unsigned flags
)
1035 if (flags
& ~PR_FL_IGNORE_KEY
)
1038 cdw10
= nvme_pr_type(type
) << 8;
1039 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
1040 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
1043 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
1044 enum pr_type type
, bool abort
)
1046 u32 cdw10
= nvme_pr_type(type
) << 8 | abort
? 2 : 1;
1047 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
1050 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
1052 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
1053 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
1056 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
1058 u32 cdw10
= nvme_pr_type(type
) << 8 | key
? 1 << 3 : 0;
1059 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
1062 static const struct pr_ops nvme_pr_ops
= {
1063 .pr_register
= nvme_pr_register
,
1064 .pr_reserve
= nvme_pr_reserve
,
1065 .pr_release
= nvme_pr_release
,
1066 .pr_preempt
= nvme_pr_preempt
,
1067 .pr_clear
= nvme_pr_clear
,
1070 static const struct block_device_operations nvme_fops
= {
1071 .owner
= THIS_MODULE
,
1072 .ioctl
= nvme_ioctl
,
1073 .compat_ioctl
= nvme_compat_ioctl
,
1075 .release
= nvme_release
,
1076 .getgeo
= nvme_getgeo
,
1077 .revalidate_disk
= nvme_revalidate_disk
,
1078 .pr_ops
= &nvme_pr_ops
,
1081 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
1083 unsigned long timeout
=
1084 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
1085 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
1088 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
1089 if ((csts
& NVME_CSTS_RDY
) == bit
)
1093 if (fatal_signal_pending(current
))
1095 if (time_after(jiffies
, timeout
)) {
1096 dev_err(ctrl
->device
,
1097 "Device not ready; aborting %s\n", enabled
?
1098 "initialisation" : "reset");
1107 * If the device has been passed off to us in an enabled state, just clear
1108 * the enabled bit. The spec says we should set the 'shutdown notification
1109 * bits', but doing so may cause the device to complete commands to the
1110 * admin queue ... and we don't know what memory that might be pointing at!
1112 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
1116 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
1117 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
1119 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1123 /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1124 * load, since we only need the quirk on reset_controller. Notice
1125 * that the HGST device needs this delay only in firmware activation
1126 * procedure; unfortunately we have no (easy) way to verify this.
1128 if ((ctrl
->quirks
& NVME_QUIRK_DELAY_BEFORE_CHK_RDY
) && ctrl
->tagset
)
1129 msleep(NVME_QUIRK_DELAY_AMOUNT
);
1131 return nvme_wait_ready(ctrl
, cap
, false);
1133 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
1135 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
1138 * Default to a 4K page size, with the intention to update this
1139 * path in the future to accomodate architectures with differing
1140 * kernel and IO page sizes.
1142 unsigned dev_page_min
= NVME_CAP_MPSMIN(cap
) + 12, page_shift
= 12;
1145 if (page_shift
< dev_page_min
) {
1146 dev_err(ctrl
->device
,
1147 "Minimum device page size %u too large for host (%u)\n",
1148 1 << dev_page_min
, 1 << page_shift
);
1152 ctrl
->page_size
= 1 << page_shift
;
1154 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
1155 ctrl
->ctrl_config
|= (page_shift
- 12) << NVME_CC_MPS_SHIFT
;
1156 ctrl
->ctrl_config
|= NVME_CC_ARB_RR
| NVME_CC_SHN_NONE
;
1157 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
1158 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
1160 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1163 return nvme_wait_ready(ctrl
, cap
, true);
1165 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
1167 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
1169 unsigned long timeout
= SHUTDOWN_TIMEOUT
+ jiffies
;
1173 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
1174 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
1176 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1180 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
1181 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
1185 if (fatal_signal_pending(current
))
1187 if (time_after(jiffies
, timeout
)) {
1188 dev_err(ctrl
->device
,
1189 "Device shutdown incomplete; abort shutdown\n");
1196 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
1198 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
1199 struct request_queue
*q
)
1203 if (ctrl
->max_hw_sectors
) {
1205 (ctrl
->max_hw_sectors
/ (ctrl
->page_size
>> 9)) + 1;
1207 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
1208 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
1210 if (ctrl
->stripe_size
)
1211 blk_queue_chunk_sectors(q
, ctrl
->stripe_size
>> 9);
1212 blk_queue_virt_boundary(q
, ctrl
->page_size
- 1);
1213 if (ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
)
1215 blk_queue_write_cache(q
, vwc
, vwc
);
1219 * Initialize the cached copies of the Identify data and various controller
1220 * register in our nvme_ctrl structure. This should be called as soon as
1221 * the admin queue is fully up and running.
1223 int nvme_init_identify(struct nvme_ctrl
*ctrl
)
1225 struct nvme_id_ctrl
*id
;
1227 int ret
, page_shift
;
1230 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
1232 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
1236 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &cap
);
1238 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
1241 page_shift
= NVME_CAP_MPSMIN(cap
) + 12;
1243 if (ctrl
->vs
>= NVME_VS(1, 1))
1244 ctrl
->subsystem
= NVME_CAP_NSSRC(cap
);
1246 ret
= nvme_identify_ctrl(ctrl
, &id
);
1248 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
1252 ctrl
->vid
= le16_to_cpu(id
->vid
);
1253 ctrl
->oncs
= le16_to_cpup(&id
->oncs
);
1254 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
1255 ctrl
->vwc
= id
->vwc
;
1256 ctrl
->cntlid
= le16_to_cpup(&id
->cntlid
);
1257 memcpy(ctrl
->serial
, id
->sn
, sizeof(id
->sn
));
1258 memcpy(ctrl
->model
, id
->mn
, sizeof(id
->mn
));
1259 memcpy(ctrl
->firmware_rev
, id
->fr
, sizeof(id
->fr
));
1261 max_hw_sectors
= 1 << (id
->mdts
+ page_shift
- 9);
1263 max_hw_sectors
= UINT_MAX
;
1264 ctrl
->max_hw_sectors
=
1265 min_not_zero(ctrl
->max_hw_sectors
, max_hw_sectors
);
1267 if ((ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) && id
->vs
[3]) {
1268 unsigned int max_hw_sectors
;
1270 ctrl
->stripe_size
= 1 << (id
->vs
[3] + page_shift
);
1271 max_hw_sectors
= ctrl
->stripe_size
>> (page_shift
- 9);
1272 if (ctrl
->max_hw_sectors
) {
1273 ctrl
->max_hw_sectors
= min(max_hw_sectors
,
1274 ctrl
->max_hw_sectors
);
1276 ctrl
->max_hw_sectors
= max_hw_sectors
;
1280 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
1281 ctrl
->sgls
= le32_to_cpu(id
->sgls
);
1282 ctrl
->kas
= le16_to_cpu(id
->kas
);
1284 if (ctrl
->ops
->is_fabrics
) {
1285 ctrl
->icdoff
= le16_to_cpu(id
->icdoff
);
1286 ctrl
->ioccsz
= le32_to_cpu(id
->ioccsz
);
1287 ctrl
->iorcsz
= le32_to_cpu(id
->iorcsz
);
1288 ctrl
->maxcmd
= le16_to_cpu(id
->maxcmd
);
1291 * In fabrics we need to verify the cntlid matches the
1294 if (ctrl
->cntlid
!= le16_to_cpu(id
->cntlid
))
1297 if (!ctrl
->opts
->discovery_nqn
&& !ctrl
->kas
) {
1299 "keep-alive support is mandatory for fabrics\n");
1303 ctrl
->cntlid
= le16_to_cpu(id
->cntlid
);
1309 EXPORT_SYMBOL_GPL(nvme_init_identify
);
1311 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
1313 struct nvme_ctrl
*ctrl
;
1314 int instance
= iminor(inode
);
1317 spin_lock(&dev_list_lock
);
1318 list_for_each_entry(ctrl
, &nvme_ctrl_list
, node
) {
1319 if (ctrl
->instance
!= instance
)
1322 if (!ctrl
->admin_q
) {
1326 if (!kref_get_unless_zero(&ctrl
->kref
))
1328 file
->private_data
= ctrl
;
1332 spin_unlock(&dev_list_lock
);
1337 static int nvme_dev_release(struct inode
*inode
, struct file
*file
)
1339 nvme_put_ctrl(file
->private_data
);
1343 static int nvme_dev_user_cmd(struct nvme_ctrl
*ctrl
, void __user
*argp
)
1348 mutex_lock(&ctrl
->namespaces_mutex
);
1349 if (list_empty(&ctrl
->namespaces
)) {
1354 ns
= list_first_entry(&ctrl
->namespaces
, struct nvme_ns
, list
);
1355 if (ns
!= list_last_entry(&ctrl
->namespaces
, struct nvme_ns
, list
)) {
1356 dev_warn(ctrl
->device
,
1357 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1362 dev_warn(ctrl
->device
,
1363 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1364 kref_get(&ns
->kref
);
1365 mutex_unlock(&ctrl
->namespaces_mutex
);
1367 ret
= nvme_user_cmd(ctrl
, ns
, argp
);
1372 mutex_unlock(&ctrl
->namespaces_mutex
);
1376 static long nvme_dev_ioctl(struct file
*file
, unsigned int cmd
,
1379 struct nvme_ctrl
*ctrl
= file
->private_data
;
1380 void __user
*argp
= (void __user
*)arg
;
1383 case NVME_IOCTL_ADMIN_CMD
:
1384 return nvme_user_cmd(ctrl
, NULL
, argp
);
1385 case NVME_IOCTL_IO_CMD
:
1386 return nvme_dev_user_cmd(ctrl
, argp
);
1387 case NVME_IOCTL_RESET
:
1388 dev_warn(ctrl
->device
, "resetting controller\n");
1389 return ctrl
->ops
->reset_ctrl(ctrl
);
1390 case NVME_IOCTL_SUBSYS_RESET
:
1391 return nvme_reset_subsystem(ctrl
);
1392 case NVME_IOCTL_RESCAN
:
1393 nvme_queue_scan(ctrl
);
1400 static const struct file_operations nvme_dev_fops
= {
1401 .owner
= THIS_MODULE
,
1402 .open
= nvme_dev_open
,
1403 .release
= nvme_dev_release
,
1404 .unlocked_ioctl
= nvme_dev_ioctl
,
1405 .compat_ioctl
= nvme_dev_ioctl
,
1408 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
1409 struct device_attribute
*attr
, const char *buf
,
1412 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1415 ret
= ctrl
->ops
->reset_ctrl(ctrl
);
1420 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
1422 static ssize_t
nvme_sysfs_rescan(struct device
*dev
,
1423 struct device_attribute
*attr
, const char *buf
,
1426 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1428 nvme_queue_scan(ctrl
);
1431 static DEVICE_ATTR(rescan_controller
, S_IWUSR
, NULL
, nvme_sysfs_rescan
);
1433 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
1436 struct nvme_ns
*ns
= nvme_get_ns_from_dev(dev
);
1437 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1438 int serial_len
= sizeof(ctrl
->serial
);
1439 int model_len
= sizeof(ctrl
->model
);
1441 if (memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1442 return sprintf(buf
, "eui.%16phN\n", ns
->uuid
);
1444 if (memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1445 return sprintf(buf
, "eui.%8phN\n", ns
->eui
);
1447 while (ctrl
->serial
[serial_len
- 1] == ' ')
1449 while (ctrl
->model
[model_len
- 1] == ' ')
1452 return sprintf(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl
->vid
,
1453 serial_len
, ctrl
->serial
, model_len
, ctrl
->model
, ns
->ns_id
);
1455 static DEVICE_ATTR(wwid
, S_IRUGO
, wwid_show
, NULL
);
1457 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
1460 struct nvme_ns
*ns
= nvme_get_ns_from_dev(dev
);
1461 return sprintf(buf
, "%pU\n", ns
->uuid
);
1463 static DEVICE_ATTR(uuid
, S_IRUGO
, uuid_show
, NULL
);
1465 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
1468 struct nvme_ns
*ns
= nvme_get_ns_from_dev(dev
);
1469 return sprintf(buf
, "%8phd\n", ns
->eui
);
1471 static DEVICE_ATTR(eui
, S_IRUGO
, eui_show
, NULL
);
1473 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
1476 struct nvme_ns
*ns
= nvme_get_ns_from_dev(dev
);
1477 return sprintf(buf
, "%d\n", ns
->ns_id
);
1479 static DEVICE_ATTR(nsid
, S_IRUGO
, nsid_show
, NULL
);
1481 static struct attribute
*nvme_ns_attrs
[] = {
1482 &dev_attr_wwid
.attr
,
1483 &dev_attr_uuid
.attr
,
1485 &dev_attr_nsid
.attr
,
1489 static umode_t
nvme_ns_attrs_are_visible(struct kobject
*kobj
,
1490 struct attribute
*a
, int n
)
1492 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
1493 struct nvme_ns
*ns
= nvme_get_ns_from_dev(dev
);
1495 if (a
== &dev_attr_uuid
.attr
) {
1496 if (!memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1499 if (a
== &dev_attr_eui
.attr
) {
1500 if (!memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1506 static const struct attribute_group nvme_ns_attr_group
= {
1507 .attrs
= nvme_ns_attrs
,
1508 .is_visible
= nvme_ns_attrs_are_visible
,
1511 #define nvme_show_str_function(field) \
1512 static ssize_t field##_show(struct device *dev, \
1513 struct device_attribute *attr, char *buf) \
1515 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1516 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1518 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1520 #define nvme_show_int_function(field) \
1521 static ssize_t field##_show(struct device *dev, \
1522 struct device_attribute *attr, char *buf) \
1524 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1525 return sprintf(buf, "%d\n", ctrl->field); \
1527 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1529 nvme_show_str_function(model
);
1530 nvme_show_str_function(serial
);
1531 nvme_show_str_function(firmware_rev
);
1532 nvme_show_int_function(cntlid
);
1534 static ssize_t
nvme_sysfs_delete(struct device
*dev
,
1535 struct device_attribute
*attr
, const char *buf
,
1538 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1540 if (device_remove_file_self(dev
, attr
))
1541 ctrl
->ops
->delete_ctrl(ctrl
);
1544 static DEVICE_ATTR(delete_controller
, S_IWUSR
, NULL
, nvme_sysfs_delete
);
1546 static ssize_t
nvme_sysfs_show_transport(struct device
*dev
,
1547 struct device_attribute
*attr
,
1550 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1552 return snprintf(buf
, PAGE_SIZE
, "%s\n", ctrl
->ops
->name
);
1554 static DEVICE_ATTR(transport
, S_IRUGO
, nvme_sysfs_show_transport
, NULL
);
1556 static ssize_t
nvme_sysfs_show_subsysnqn(struct device
*dev
,
1557 struct device_attribute
*attr
,
1560 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1562 return snprintf(buf
, PAGE_SIZE
, "%s\n",
1563 ctrl
->ops
->get_subsysnqn(ctrl
));
1565 static DEVICE_ATTR(subsysnqn
, S_IRUGO
, nvme_sysfs_show_subsysnqn
, NULL
);
1567 static ssize_t
nvme_sysfs_show_address(struct device
*dev
,
1568 struct device_attribute
*attr
,
1571 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1573 return ctrl
->ops
->get_address(ctrl
, buf
, PAGE_SIZE
);
1575 static DEVICE_ATTR(address
, S_IRUGO
, nvme_sysfs_show_address
, NULL
);
1577 static struct attribute
*nvme_dev_attrs
[] = {
1578 &dev_attr_reset_controller
.attr
,
1579 &dev_attr_rescan_controller
.attr
,
1580 &dev_attr_model
.attr
,
1581 &dev_attr_serial
.attr
,
1582 &dev_attr_firmware_rev
.attr
,
1583 &dev_attr_cntlid
.attr
,
1584 &dev_attr_delete_controller
.attr
,
1585 &dev_attr_transport
.attr
,
1586 &dev_attr_subsysnqn
.attr
,
1587 &dev_attr_address
.attr
,
1591 #define CHECK_ATTR(ctrl, a, name) \
1592 if ((a) == &dev_attr_##name.attr && \
1593 !(ctrl)->ops->get_##name) \
1596 static umode_t
nvme_dev_attrs_are_visible(struct kobject
*kobj
,
1597 struct attribute
*a
, int n
)
1599 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
1600 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1602 if (a
== &dev_attr_delete_controller
.attr
) {
1603 if (!ctrl
->ops
->delete_ctrl
)
1607 CHECK_ATTR(ctrl
, a
, subsysnqn
);
1608 CHECK_ATTR(ctrl
, a
, address
);
1613 static struct attribute_group nvme_dev_attrs_group
= {
1614 .attrs
= nvme_dev_attrs
,
1615 .is_visible
= nvme_dev_attrs_are_visible
,
1618 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
1619 &nvme_dev_attrs_group
,
1623 static int ns_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1625 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
1626 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
1628 return nsa
->ns_id
- nsb
->ns_id
;
1631 static struct nvme_ns
*nvme_find_get_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1633 struct nvme_ns
*ns
, *ret
= NULL
;
1635 mutex_lock(&ctrl
->namespaces_mutex
);
1636 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
1637 if (ns
->ns_id
== nsid
) {
1638 kref_get(&ns
->kref
);
1642 if (ns
->ns_id
> nsid
)
1645 mutex_unlock(&ctrl
->namespaces_mutex
);
1649 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1652 struct gendisk
*disk
;
1653 struct nvme_id_ns
*id
;
1654 char disk_name
[DISK_NAME_LEN
];
1655 int node
= dev_to_node(ctrl
->dev
);
1657 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
1661 ns
->instance
= ida_simple_get(&ctrl
->ns_ida
, 1, 0, GFP_KERNEL
);
1662 if (ns
->instance
< 0)
1665 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
1666 if (IS_ERR(ns
->queue
))
1667 goto out_release_instance
;
1668 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, ns
->queue
);
1669 ns
->queue
->queuedata
= ns
;
1672 kref_init(&ns
->kref
);
1674 ns
->lba_shift
= 9; /* set to a default value for 512 until disk is validated */
1676 blk_queue_logical_block_size(ns
->queue
, 1 << ns
->lba_shift
);
1677 nvme_set_queue_limits(ctrl
, ns
->queue
);
1679 sprintf(disk_name
, "nvme%dn%d", ctrl
->instance
, ns
->instance
);
1681 if (nvme_revalidate_ns(ns
, &id
))
1682 goto out_free_queue
;
1684 if (nvme_nvm_ns_supported(ns
, id
)) {
1685 if (nvme_nvm_register(ns
, disk_name
, node
,
1686 &nvme_ns_attr_group
)) {
1687 dev_warn(ctrl
->dev
, "%s: LightNVM init failure\n",
1692 disk
= alloc_disk_node(0, node
);
1696 disk
->fops
= &nvme_fops
;
1697 disk
->private_data
= ns
;
1698 disk
->queue
= ns
->queue
;
1699 disk
->flags
= GENHD_FL_EXT_DEVT
;
1700 memcpy(disk
->disk_name
, disk_name
, DISK_NAME_LEN
);
1703 __nvme_revalidate_disk(disk
, id
);
1706 mutex_lock(&ctrl
->namespaces_mutex
);
1707 list_add_tail(&ns
->list
, &ctrl
->namespaces
);
1708 mutex_unlock(&ctrl
->namespaces_mutex
);
1710 kref_get(&ctrl
->kref
);
1717 device_add_disk(ctrl
->device
, ns
->disk
);
1718 if (sysfs_create_group(&disk_to_dev(ns
->disk
)->kobj
,
1719 &nvme_ns_attr_group
))
1720 pr_warn("%s: failed to create sysfs group for identification\n",
1721 ns
->disk
->disk_name
);
1726 blk_cleanup_queue(ns
->queue
);
1727 out_release_instance
:
1728 ida_simple_remove(&ctrl
->ns_ida
, ns
->instance
);
1733 static void nvme_ns_remove(struct nvme_ns
*ns
)
1735 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
1738 if (ns
->disk
&& ns
->disk
->flags
& GENHD_FL_UP
) {
1739 if (blk_get_integrity(ns
->disk
))
1740 blk_integrity_unregister(ns
->disk
);
1741 sysfs_remove_group(&disk_to_dev(ns
->disk
)->kobj
,
1742 &nvme_ns_attr_group
);
1743 del_gendisk(ns
->disk
);
1744 blk_mq_abort_requeue_list(ns
->queue
);
1745 blk_cleanup_queue(ns
->queue
);
1748 mutex_lock(&ns
->ctrl
->namespaces_mutex
);
1749 list_del_init(&ns
->list
);
1750 mutex_unlock(&ns
->ctrl
->namespaces_mutex
);
1755 static void nvme_validate_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1759 ns
= nvme_find_get_ns(ctrl
, nsid
);
1761 if (ns
->disk
&& revalidate_disk(ns
->disk
))
1765 nvme_alloc_ns(ctrl
, nsid
);
1768 static void nvme_remove_invalid_namespaces(struct nvme_ctrl
*ctrl
,
1771 struct nvme_ns
*ns
, *next
;
1773 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
1774 if (ns
->ns_id
> nsid
)
1779 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
, unsigned nn
)
1783 unsigned i
, j
, nsid
, prev
= 0, num_lists
= DIV_ROUND_UP(nn
, 1024);
1786 ns_list
= kzalloc(0x1000, GFP_KERNEL
);
1790 for (i
= 0; i
< num_lists
; i
++) {
1791 ret
= nvme_identify_ns_list(ctrl
, prev
, ns_list
);
1795 for (j
= 0; j
< min(nn
, 1024U); j
++) {
1796 nsid
= le32_to_cpu(ns_list
[j
]);
1800 nvme_validate_ns(ctrl
, nsid
);
1802 while (++prev
< nsid
) {
1803 ns
= nvme_find_get_ns(ctrl
, prev
);
1813 nvme_remove_invalid_namespaces(ctrl
, prev
);
1819 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
, unsigned nn
)
1823 for (i
= 1; i
<= nn
; i
++)
1824 nvme_validate_ns(ctrl
, i
);
1826 nvme_remove_invalid_namespaces(ctrl
, nn
);
1829 static void nvme_scan_work(struct work_struct
*work
)
1831 struct nvme_ctrl
*ctrl
=
1832 container_of(work
, struct nvme_ctrl
, scan_work
);
1833 struct nvme_id_ctrl
*id
;
1836 if (ctrl
->state
!= NVME_CTRL_LIVE
)
1839 if (nvme_identify_ctrl(ctrl
, &id
))
1842 nn
= le32_to_cpu(id
->nn
);
1843 if (ctrl
->vs
>= NVME_VS(1, 1) &&
1844 !(ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)) {
1845 if (!nvme_scan_ns_list(ctrl
, nn
))
1848 nvme_scan_ns_sequential(ctrl
, nn
);
1850 mutex_lock(&ctrl
->namespaces_mutex
);
1851 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
1852 mutex_unlock(&ctrl
->namespaces_mutex
);
1856 void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
1859 * Do not queue new scan work when a controller is reset during
1862 if (ctrl
->state
== NVME_CTRL_LIVE
)
1863 schedule_work(&ctrl
->scan_work
);
1865 EXPORT_SYMBOL_GPL(nvme_queue_scan
);
1868 * This function iterates the namespace list unlocked to allow recovery from
1869 * controller failure. It is up to the caller to ensure the namespace list is
1870 * not modified by scan work while this function is executing.
1872 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
1874 struct nvme_ns
*ns
, *next
;
1877 * The dead states indicates the controller was not gracefully
1878 * disconnected. In that case, we won't be able to flush any data while
1879 * removing the namespaces' disks; fail all the queues now to avoid
1880 * potentially having to clean up the failed sync later.
1882 if (ctrl
->state
== NVME_CTRL_DEAD
)
1883 nvme_kill_queues(ctrl
);
1885 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
)
1888 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
1890 static void nvme_async_event_work(struct work_struct
*work
)
1892 struct nvme_ctrl
*ctrl
=
1893 container_of(work
, struct nvme_ctrl
, async_event_work
);
1895 spin_lock_irq(&ctrl
->lock
);
1896 while (ctrl
->event_limit
> 0) {
1897 int aer_idx
= --ctrl
->event_limit
;
1899 spin_unlock_irq(&ctrl
->lock
);
1900 ctrl
->ops
->submit_async_event(ctrl
, aer_idx
);
1901 spin_lock_irq(&ctrl
->lock
);
1903 spin_unlock_irq(&ctrl
->lock
);
1906 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
,
1907 struct nvme_completion
*cqe
)
1909 u16 status
= le16_to_cpu(cqe
->status
) >> 1;
1910 u32 result
= le32_to_cpu(cqe
->result
);
1912 if (status
== NVME_SC_SUCCESS
|| status
== NVME_SC_ABORT_REQ
) {
1913 ++ctrl
->event_limit
;
1914 schedule_work(&ctrl
->async_event_work
);
1917 if (status
!= NVME_SC_SUCCESS
)
1920 switch (result
& 0xff07) {
1921 case NVME_AER_NOTICE_NS_CHANGED
:
1922 dev_info(ctrl
->device
, "rescanning\n");
1923 nvme_queue_scan(ctrl
);
1926 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
1929 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
1931 void nvme_queue_async_events(struct nvme_ctrl
*ctrl
)
1933 ctrl
->event_limit
= NVME_NR_AERS
;
1934 schedule_work(&ctrl
->async_event_work
);
1936 EXPORT_SYMBOL_GPL(nvme_queue_async_events
);
1938 static DEFINE_IDA(nvme_instance_ida
);
1940 static int nvme_set_instance(struct nvme_ctrl
*ctrl
)
1942 int instance
, error
;
1945 if (!ida_pre_get(&nvme_instance_ida
, GFP_KERNEL
))
1948 spin_lock(&dev_list_lock
);
1949 error
= ida_get_new(&nvme_instance_ida
, &instance
);
1950 spin_unlock(&dev_list_lock
);
1951 } while (error
== -EAGAIN
);
1956 ctrl
->instance
= instance
;
1960 static void nvme_release_instance(struct nvme_ctrl
*ctrl
)
1962 spin_lock(&dev_list_lock
);
1963 ida_remove(&nvme_instance_ida
, ctrl
->instance
);
1964 spin_unlock(&dev_list_lock
);
1967 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
1969 flush_work(&ctrl
->async_event_work
);
1970 flush_work(&ctrl
->scan_work
);
1971 nvme_remove_namespaces(ctrl
);
1973 device_destroy(nvme_class
, MKDEV(nvme_char_major
, ctrl
->instance
));
1975 spin_lock(&dev_list_lock
);
1976 list_del(&ctrl
->node
);
1977 spin_unlock(&dev_list_lock
);
1979 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
1981 static void nvme_free_ctrl(struct kref
*kref
)
1983 struct nvme_ctrl
*ctrl
= container_of(kref
, struct nvme_ctrl
, kref
);
1985 put_device(ctrl
->device
);
1986 nvme_release_instance(ctrl
);
1987 ida_destroy(&ctrl
->ns_ida
);
1989 ctrl
->ops
->free_ctrl(ctrl
);
1992 void nvme_put_ctrl(struct nvme_ctrl
*ctrl
)
1994 kref_put(&ctrl
->kref
, nvme_free_ctrl
);
1996 EXPORT_SYMBOL_GPL(nvme_put_ctrl
);
1999 * Initialize a NVMe controller structures. This needs to be called during
2000 * earliest initialization so that we have the initialized structured around
2003 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
2004 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
2008 ctrl
->state
= NVME_CTRL_NEW
;
2009 spin_lock_init(&ctrl
->lock
);
2010 INIT_LIST_HEAD(&ctrl
->namespaces
);
2011 mutex_init(&ctrl
->namespaces_mutex
);
2012 kref_init(&ctrl
->kref
);
2015 ctrl
->quirks
= quirks
;
2016 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
2017 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
2019 ret
= nvme_set_instance(ctrl
);
2023 ctrl
->device
= device_create_with_groups(nvme_class
, ctrl
->dev
,
2024 MKDEV(nvme_char_major
, ctrl
->instance
),
2025 ctrl
, nvme_dev_attr_groups
,
2026 "nvme%d", ctrl
->instance
);
2027 if (IS_ERR(ctrl
->device
)) {
2028 ret
= PTR_ERR(ctrl
->device
);
2029 goto out_release_instance
;
2031 get_device(ctrl
->device
);
2032 ida_init(&ctrl
->ns_ida
);
2034 spin_lock(&dev_list_lock
);
2035 list_add_tail(&ctrl
->node
, &nvme_ctrl_list
);
2036 spin_unlock(&dev_list_lock
);
2039 out_release_instance
:
2040 nvme_release_instance(ctrl
);
2044 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
2047 * nvme_kill_queues(): Ends all namespace queues
2048 * @ctrl: the dead controller that needs to end
2050 * Call this function when the driver determines it is unable to get the
2051 * controller in a state capable of servicing IO.
2053 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
2057 mutex_lock(&ctrl
->namespaces_mutex
);
2058 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
2060 * Revalidating a dead namespace sets capacity to 0. This will
2061 * end buffered writers dirtying pages that can't be synced.
2063 if (ns
->disk
&& !test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
2064 revalidate_disk(ns
->disk
);
2066 blk_set_queue_dying(ns
->queue
);
2067 blk_mq_abort_requeue_list(ns
->queue
);
2068 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
2070 mutex_unlock(&ctrl
->namespaces_mutex
);
2072 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
2074 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
2078 mutex_lock(&ctrl
->namespaces_mutex
);
2079 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
2080 spin_lock_irq(ns
->queue
->queue_lock
);
2081 queue_flag_set(QUEUE_FLAG_STOPPED
, ns
->queue
);
2082 spin_unlock_irq(ns
->queue
->queue_lock
);
2084 blk_mq_cancel_requeue_work(ns
->queue
);
2085 blk_mq_stop_hw_queues(ns
->queue
);
2087 mutex_unlock(&ctrl
->namespaces_mutex
);
2089 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
2091 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
2095 mutex_lock(&ctrl
->namespaces_mutex
);
2096 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
2097 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED
, ns
->queue
);
2098 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
2099 blk_mq_kick_requeue_list(ns
->queue
);
2101 mutex_unlock(&ctrl
->namespaces_mutex
);
2103 EXPORT_SYMBOL_GPL(nvme_start_queues
);
2105 int __init
nvme_core_init(void)
2109 result
= __register_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme",
2113 else if (result
> 0)
2114 nvme_char_major
= result
;
2116 nvme_class
= class_create(THIS_MODULE
, "nvme");
2117 if (IS_ERR(nvme_class
)) {
2118 result
= PTR_ERR(nvme_class
);
2119 goto unregister_chrdev
;
2125 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
2129 void nvme_core_exit(void)
2131 class_destroy(nvme_class
);
2132 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
2135 MODULE_LICENSE("GPL");
2136 MODULE_VERSION("1.0");
2137 module_init(nvme_core_init
);
2138 module_exit(nvme_core_exit
);