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>
34 #define NVME_MINORS (1U << MINORBITS)
36 unsigned char admin_timeout
= 60;
37 module_param(admin_timeout
, byte
, 0644);
38 MODULE_PARM_DESC(admin_timeout
, "timeout in seconds for admin commands");
39 EXPORT_SYMBOL_GPL(admin_timeout
);
41 unsigned char nvme_io_timeout
= 30;
42 module_param_named(io_timeout
, nvme_io_timeout
, byte
, 0644);
43 MODULE_PARM_DESC(io_timeout
, "timeout in seconds for I/O");
44 EXPORT_SYMBOL_GPL(nvme_io_timeout
);
46 unsigned char shutdown_timeout
= 5;
47 module_param(shutdown_timeout
, byte
, 0644);
48 MODULE_PARM_DESC(shutdown_timeout
, "timeout in seconds for controller shutdown");
50 static int nvme_major
;
51 module_param(nvme_major
, int, 0);
53 static int nvme_char_major
;
54 module_param(nvme_char_major
, int, 0);
56 static LIST_HEAD(nvme_ctrl_list
);
57 static DEFINE_SPINLOCK(dev_list_lock
);
59 static struct class *nvme_class
;
61 void nvme_cancel_request(struct request
*req
, void *data
, bool reserved
)
65 if (!blk_mq_request_started(req
))
68 dev_dbg_ratelimited(((struct nvme_ctrl
*) data
)->device
,
69 "Cancelling I/O %d", req
->tag
);
71 status
= NVME_SC_ABORT_REQ
;
72 if (blk_queue_dying(req
->q
))
73 status
|= NVME_SC_DNR
;
74 blk_mq_complete_request(req
, status
);
76 EXPORT_SYMBOL_GPL(nvme_cancel_request
);
78 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
79 enum nvme_ctrl_state new_state
)
81 enum nvme_ctrl_state old_state
= ctrl
->state
;
84 spin_lock_irq(&ctrl
->lock
);
88 case NVME_CTRL_RESETTING
:
95 case NVME_CTRL_RESETTING
:
105 case NVME_CTRL_DELETING
:
108 case NVME_CTRL_RESETTING
:
117 case NVME_CTRL_DELETING
:
127 spin_unlock_irq(&ctrl
->lock
);
130 ctrl
->state
= new_state
;
134 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
136 static void nvme_free_ns(struct kref
*kref
)
138 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
140 if (ns
->type
== NVME_NS_LIGHTNVM
)
141 nvme_nvm_unregister(ns
->queue
, ns
->disk
->disk_name
);
143 spin_lock(&dev_list_lock
);
144 ns
->disk
->private_data
= NULL
;
145 spin_unlock(&dev_list_lock
);
148 ida_simple_remove(&ns
->ctrl
->ns_ida
, ns
->instance
);
149 nvme_put_ctrl(ns
->ctrl
);
153 static void nvme_put_ns(struct nvme_ns
*ns
)
155 kref_put(&ns
->kref
, nvme_free_ns
);
158 static struct nvme_ns
*nvme_get_ns_from_disk(struct gendisk
*disk
)
162 spin_lock(&dev_list_lock
);
163 ns
= disk
->private_data
;
165 if (!kref_get_unless_zero(&ns
->kref
))
167 if (!try_module_get(ns
->ctrl
->ops
->module
))
170 spin_unlock(&dev_list_lock
);
175 kref_put(&ns
->kref
, nvme_free_ns
);
177 spin_unlock(&dev_list_lock
);
181 void nvme_requeue_req(struct request
*req
)
185 blk_mq_requeue_request(req
);
186 spin_lock_irqsave(req
->q
->queue_lock
, flags
);
187 if (!blk_queue_stopped(req
->q
))
188 blk_mq_kick_requeue_list(req
->q
);
189 spin_unlock_irqrestore(req
->q
->queue_lock
, flags
);
191 EXPORT_SYMBOL_GPL(nvme_requeue_req
);
193 struct request
*nvme_alloc_request(struct request_queue
*q
,
194 struct nvme_command
*cmd
, unsigned int flags
)
196 bool write
= cmd
->common
.opcode
& 1;
199 req
= blk_mq_alloc_request(q
, write
, flags
);
203 req
->cmd_type
= REQ_TYPE_DRV_PRIV
;
204 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
206 req
->__sector
= (sector_t
) -1;
207 req
->bio
= req
->biotail
= NULL
;
209 req
->cmd
= (unsigned char *)cmd
;
210 req
->cmd_len
= sizeof(struct nvme_command
);
214 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
216 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
217 struct nvme_command
*cmnd
)
219 memset(cmnd
, 0, sizeof(*cmnd
));
220 cmnd
->common
.opcode
= nvme_cmd_flush
;
221 cmnd
->common
.nsid
= cpu_to_le32(ns
->ns_id
);
224 static inline int nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
225 struct nvme_command
*cmnd
)
227 struct nvme_dsm_range
*range
;
230 unsigned int nr_bytes
= blk_rq_bytes(req
);
232 range
= kmalloc(sizeof(*range
), GFP_ATOMIC
);
234 return BLK_MQ_RQ_QUEUE_BUSY
;
236 range
->cattr
= cpu_to_le32(0);
237 range
->nlb
= cpu_to_le32(nr_bytes
>> ns
->lba_shift
);
238 range
->slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
240 memset(cmnd
, 0, sizeof(*cmnd
));
241 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
242 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->ns_id
);
244 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
246 req
->completion_data
= range
;
247 page
= virt_to_page(range
);
248 offset
= offset_in_page(range
);
249 blk_add_request_payload(req
, page
, offset
, sizeof(*range
));
252 * we set __data_len back to the size of the area to be discarded
253 * on disk. This allows us to report completion on the full amount
254 * of blocks described by the request.
256 req
->__data_len
= nr_bytes
;
261 static inline void nvme_setup_rw(struct nvme_ns
*ns
, struct request
*req
,
262 struct nvme_command
*cmnd
)
267 if (req
->cmd_flags
& REQ_FUA
)
268 control
|= NVME_RW_FUA
;
269 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
270 control
|= NVME_RW_LR
;
272 if (req
->cmd_flags
& REQ_RAHEAD
)
273 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
275 memset(cmnd
, 0, sizeof(*cmnd
));
276 cmnd
->rw
.opcode
= (rq_data_dir(req
) ? nvme_cmd_write
: nvme_cmd_read
);
277 cmnd
->rw
.command_id
= req
->tag
;
278 cmnd
->rw
.nsid
= cpu_to_le32(ns
->ns_id
);
279 cmnd
->rw
.slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
280 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
283 switch (ns
->pi_type
) {
284 case NVME_NS_DPS_PI_TYPE3
:
285 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
287 case NVME_NS_DPS_PI_TYPE1
:
288 case NVME_NS_DPS_PI_TYPE2
:
289 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
290 NVME_RW_PRINFO_PRCHK_REF
;
291 cmnd
->rw
.reftag
= cpu_to_le32(
292 nvme_block_nr(ns
, blk_rq_pos(req
)));
295 if (!blk_integrity_rq(req
))
296 control
|= NVME_RW_PRINFO_PRACT
;
299 cmnd
->rw
.control
= cpu_to_le16(control
);
300 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
303 int nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
304 struct nvme_command
*cmd
)
308 if (req
->cmd_type
== REQ_TYPE_DRV_PRIV
)
309 memcpy(cmd
, req
->cmd
, sizeof(*cmd
));
310 else if (req_op(req
) == REQ_OP_FLUSH
)
311 nvme_setup_flush(ns
, cmd
);
312 else if (req_op(req
) == REQ_OP_DISCARD
)
313 ret
= nvme_setup_discard(ns
, req
, cmd
);
315 nvme_setup_rw(ns
, req
, cmd
);
319 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
322 * Returns 0 on success. If the result is negative, it's a Linux error code;
323 * if the result is positive, it's an NVM Express status code
325 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
326 struct nvme_completion
*cqe
, void *buffer
, unsigned bufflen
,
332 req
= nvme_alloc_request(q
, cmd
, 0);
336 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
339 if (buffer
&& bufflen
) {
340 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
345 blk_execute_rq(req
->q
, NULL
, req
, 0);
348 blk_mq_free_request(req
);
352 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
353 void *buffer
, unsigned bufflen
)
355 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0);
357 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
359 int __nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
360 void __user
*ubuffer
, unsigned bufflen
,
361 void __user
*meta_buffer
, unsigned meta_len
, u32 meta_seed
,
362 u32
*result
, unsigned timeout
)
364 bool write
= cmd
->common
.opcode
& 1;
365 struct nvme_completion cqe
;
366 struct nvme_ns
*ns
= q
->queuedata
;
367 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
369 struct bio
*bio
= NULL
;
373 req
= nvme_alloc_request(q
, cmd
, 0);
377 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
380 if (ubuffer
&& bufflen
) {
381 ret
= blk_rq_map_user(q
, req
, NULL
, ubuffer
, bufflen
,
389 bio
->bi_bdev
= bdget_disk(disk
, 0);
395 if (meta_buffer
&& meta_len
) {
396 struct bio_integrity_payload
*bip
;
398 meta
= kmalloc(meta_len
, GFP_KERNEL
);
405 if (copy_from_user(meta
, meta_buffer
,
412 bip
= bio_integrity_alloc(bio
, GFP_KERNEL
, 1);
418 bip
->bip_iter
.bi_size
= meta_len
;
419 bip
->bip_iter
.bi_sector
= meta_seed
;
421 ret
= bio_integrity_add_page(bio
, virt_to_page(meta
),
422 meta_len
, offset_in_page(meta
));
423 if (ret
!= meta_len
) {
430 blk_execute_rq(req
->q
, disk
, req
, 0);
433 *result
= le32_to_cpu(cqe
.result
);
434 if (meta
&& !ret
&& !write
) {
435 if (copy_to_user(meta_buffer
, meta
, meta_len
))
442 if (disk
&& bio
->bi_bdev
)
444 blk_rq_unmap_user(bio
);
447 blk_mq_free_request(req
);
451 int nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
452 void __user
*ubuffer
, unsigned bufflen
, u32
*result
,
455 return __nvme_submit_user_cmd(q
, cmd
, ubuffer
, bufflen
, NULL
, 0, 0,
459 int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
461 struct nvme_command c
= { };
464 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
465 c
.identify
.opcode
= nvme_admin_identify
;
466 c
.identify
.cns
= cpu_to_le32(1);
468 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
472 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
473 sizeof(struct nvme_id_ctrl
));
479 static int nvme_identify_ns_list(struct nvme_ctrl
*dev
, unsigned nsid
, __le32
*ns_list
)
481 struct nvme_command c
= { };
483 c
.identify
.opcode
= nvme_admin_identify
;
484 c
.identify
.cns
= cpu_to_le32(2);
485 c
.identify
.nsid
= cpu_to_le32(nsid
);
486 return nvme_submit_sync_cmd(dev
->admin_q
, &c
, ns_list
, 0x1000);
489 int nvme_identify_ns(struct nvme_ctrl
*dev
, unsigned nsid
,
490 struct nvme_id_ns
**id
)
492 struct nvme_command c
= { };
495 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
496 c
.identify
.opcode
= nvme_admin_identify
,
497 c
.identify
.nsid
= cpu_to_le32(nsid
),
499 *id
= kmalloc(sizeof(struct nvme_id_ns
), GFP_KERNEL
);
503 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
504 sizeof(struct nvme_id_ns
));
510 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned nsid
,
511 dma_addr_t dma_addr
, u32
*result
)
513 struct nvme_command c
;
514 struct nvme_completion cqe
;
517 memset(&c
, 0, sizeof(c
));
518 c
.features
.opcode
= nvme_admin_get_features
;
519 c
.features
.nsid
= cpu_to_le32(nsid
);
520 c
.features
.prp1
= cpu_to_le64(dma_addr
);
521 c
.features
.fid
= cpu_to_le32(fid
);
523 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
, NULL
, 0, 0);
525 *result
= le32_to_cpu(cqe
.result
);
529 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned dword11
,
530 dma_addr_t dma_addr
, u32
*result
)
532 struct nvme_command c
;
533 struct nvme_completion cqe
;
536 memset(&c
, 0, sizeof(c
));
537 c
.features
.opcode
= nvme_admin_set_features
;
538 c
.features
.prp1
= cpu_to_le64(dma_addr
);
539 c
.features
.fid
= cpu_to_le32(fid
);
540 c
.features
.dword11
= cpu_to_le32(dword11
);
542 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
, NULL
, 0, 0);
544 *result
= le32_to_cpu(cqe
.result
);
548 int nvme_get_log_page(struct nvme_ctrl
*dev
, struct nvme_smart_log
**log
)
550 struct nvme_command c
= { };
553 c
.common
.opcode
= nvme_admin_get_log_page
,
554 c
.common
.nsid
= cpu_to_le32(0xFFFFFFFF),
555 c
.common
.cdw10
[0] = cpu_to_le32(
556 (((sizeof(struct nvme_smart_log
) / 4) - 1) << 16) |
559 *log
= kmalloc(sizeof(struct nvme_smart_log
), GFP_KERNEL
);
563 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *log
,
564 sizeof(struct nvme_smart_log
));
570 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
572 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
574 int status
, nr_io_queues
;
576 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, 0,
581 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
582 *count
= min(*count
, nr_io_queues
);
585 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
587 static int nvme_submit_io(struct nvme_ns
*ns
, struct nvme_user_io __user
*uio
)
589 struct nvme_user_io io
;
590 struct nvme_command c
;
591 unsigned length
, meta_len
;
592 void __user
*metadata
;
594 if (copy_from_user(&io
, uio
, sizeof(io
)))
602 case nvme_cmd_compare
:
608 length
= (io
.nblocks
+ 1) << ns
->lba_shift
;
609 meta_len
= (io
.nblocks
+ 1) * ns
->ms
;
610 metadata
= (void __user
*)(uintptr_t)io
.metadata
;
615 } else if (meta_len
) {
616 if ((io
.metadata
& 3) || !io
.metadata
)
620 memset(&c
, 0, sizeof(c
));
621 c
.rw
.opcode
= io
.opcode
;
622 c
.rw
.flags
= io
.flags
;
623 c
.rw
.nsid
= cpu_to_le32(ns
->ns_id
);
624 c
.rw
.slba
= cpu_to_le64(io
.slba
);
625 c
.rw
.length
= cpu_to_le16(io
.nblocks
);
626 c
.rw
.control
= cpu_to_le16(io
.control
);
627 c
.rw
.dsmgmt
= cpu_to_le32(io
.dsmgmt
);
628 c
.rw
.reftag
= cpu_to_le32(io
.reftag
);
629 c
.rw
.apptag
= cpu_to_le16(io
.apptag
);
630 c
.rw
.appmask
= cpu_to_le16(io
.appmask
);
632 return __nvme_submit_user_cmd(ns
->queue
, &c
,
633 (void __user
*)(uintptr_t)io
.addr
, length
,
634 metadata
, meta_len
, io
.slba
, NULL
, 0);
637 static int nvme_user_cmd(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
638 struct nvme_passthru_cmd __user
*ucmd
)
640 struct nvme_passthru_cmd cmd
;
641 struct nvme_command c
;
642 unsigned timeout
= 0;
645 if (!capable(CAP_SYS_ADMIN
))
647 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
652 memset(&c
, 0, sizeof(c
));
653 c
.common
.opcode
= cmd
.opcode
;
654 c
.common
.flags
= cmd
.flags
;
655 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
656 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
657 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
658 c
.common
.cdw10
[0] = cpu_to_le32(cmd
.cdw10
);
659 c
.common
.cdw10
[1] = cpu_to_le32(cmd
.cdw11
);
660 c
.common
.cdw10
[2] = cpu_to_le32(cmd
.cdw12
);
661 c
.common
.cdw10
[3] = cpu_to_le32(cmd
.cdw13
);
662 c
.common
.cdw10
[4] = cpu_to_le32(cmd
.cdw14
);
663 c
.common
.cdw10
[5] = cpu_to_le32(cmd
.cdw15
);
666 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
668 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
669 (void __user
*)(uintptr_t)cmd
.addr
, cmd
.data_len
,
670 &cmd
.result
, timeout
);
672 if (put_user(cmd
.result
, &ucmd
->result
))
679 static int nvme_ioctl(struct block_device
*bdev
, fmode_t mode
,
680 unsigned int cmd
, unsigned long arg
)
682 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
686 force_successful_syscall_return();
688 case NVME_IOCTL_ADMIN_CMD
:
689 return nvme_user_cmd(ns
->ctrl
, NULL
, (void __user
*)arg
);
690 case NVME_IOCTL_IO_CMD
:
691 return nvme_user_cmd(ns
->ctrl
, ns
, (void __user
*)arg
);
692 case NVME_IOCTL_SUBMIT_IO
:
693 return nvme_submit_io(ns
, (void __user
*)arg
);
694 #ifdef CONFIG_BLK_DEV_NVME_SCSI
695 case SG_GET_VERSION_NUM
:
696 return nvme_sg_get_version_num((void __user
*)arg
);
698 return nvme_sg_io(ns
, (void __user
*)arg
);
706 static int nvme_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
707 unsigned int cmd
, unsigned long arg
)
713 return nvme_ioctl(bdev
, mode
, cmd
, arg
);
716 #define nvme_compat_ioctl NULL
719 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
721 return nvme_get_ns_from_disk(bdev
->bd_disk
) ? 0 : -ENXIO
;
724 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
726 struct nvme_ns
*ns
= disk
->private_data
;
728 module_put(ns
->ctrl
->ops
->module
);
732 static int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
734 /* some standard values */
736 geo
->sectors
= 1 << 5;
737 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
741 #ifdef CONFIG_BLK_DEV_INTEGRITY
742 static void nvme_init_integrity(struct nvme_ns
*ns
)
744 struct blk_integrity integrity
;
746 switch (ns
->pi_type
) {
747 case NVME_NS_DPS_PI_TYPE3
:
748 integrity
.profile
= &t10_pi_type3_crc
;
749 integrity
.tag_size
= sizeof(u16
) + sizeof(u32
);
750 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
752 case NVME_NS_DPS_PI_TYPE1
:
753 case NVME_NS_DPS_PI_TYPE2
:
754 integrity
.profile
= &t10_pi_type1_crc
;
755 integrity
.tag_size
= sizeof(u16
);
756 integrity
.flags
|= BLK_INTEGRITY_DEVICE_CAPABLE
;
759 integrity
.profile
= NULL
;
762 integrity
.tuple_size
= ns
->ms
;
763 blk_integrity_register(ns
->disk
, &integrity
);
764 blk_queue_max_integrity_segments(ns
->queue
, 1);
767 static void nvme_init_integrity(struct nvme_ns
*ns
)
770 #endif /* CONFIG_BLK_DEV_INTEGRITY */
772 static void nvme_config_discard(struct nvme_ns
*ns
)
774 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
775 u32 logical_block_size
= queue_logical_block_size(ns
->queue
);
777 if (ctrl
->quirks
& NVME_QUIRK_DISCARD_ZEROES
)
778 ns
->queue
->limits
.discard_zeroes_data
= 1;
780 ns
->queue
->limits
.discard_zeroes_data
= 0;
782 ns
->queue
->limits
.discard_alignment
= logical_block_size
;
783 ns
->queue
->limits
.discard_granularity
= logical_block_size
;
784 blk_queue_max_discard_sectors(ns
->queue
, 0xffffffff);
785 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, ns
->queue
);
788 static int nvme_revalidate_disk(struct gendisk
*disk
)
790 struct nvme_ns
*ns
= disk
->private_data
;
791 struct nvme_id_ns
*id
;
796 if (test_bit(NVME_NS_DEAD
, &ns
->flags
)) {
797 set_capacity(disk
, 0);
800 if (nvme_identify_ns(ns
->ctrl
, ns
->ns_id
, &id
)) {
801 dev_warn(disk_to_dev(ns
->disk
), "%s: Identify failure\n",
810 if (nvme_nvm_ns_supported(ns
, id
) && ns
->type
!= NVME_NS_LIGHTNVM
) {
811 if (nvme_nvm_register(ns
->queue
, disk
->disk_name
)) {
812 dev_warn(disk_to_dev(ns
->disk
),
813 "%s: LightNVM init failure\n", __func__
);
817 ns
->type
= NVME_NS_LIGHTNVM
;
820 if (ns
->ctrl
->vs
>= NVME_VS(1, 1))
821 memcpy(ns
->eui
, id
->eui64
, sizeof(ns
->eui
));
822 if (ns
->ctrl
->vs
>= NVME_VS(1, 2))
823 memcpy(ns
->uuid
, id
->nguid
, sizeof(ns
->uuid
));
826 lbaf
= id
->flbas
& NVME_NS_FLBAS_LBA_MASK
;
827 ns
->lba_shift
= id
->lbaf
[lbaf
].ds
;
828 ns
->ms
= le16_to_cpu(id
->lbaf
[lbaf
].ms
);
829 ns
->ext
= ns
->ms
&& (id
->flbas
& NVME_NS_FLBAS_META_EXT
);
832 * If identify namespace failed, use default 512 byte block size so
833 * block layer can use before failing read/write for 0 capacity.
835 if (ns
->lba_shift
== 0)
837 bs
= 1 << ns
->lba_shift
;
838 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
839 pi_type
= ns
->ms
== sizeof(struct t10_pi_tuple
) ?
840 id
->dps
& NVME_NS_DPS_PI_MASK
: 0;
842 blk_mq_freeze_queue(disk
->queue
);
843 if (blk_get_integrity(disk
) && (ns
->pi_type
!= pi_type
||
845 bs
!= queue_logical_block_size(disk
->queue
) ||
846 (ns
->ms
&& ns
->ext
)))
847 blk_integrity_unregister(disk
);
849 ns
->pi_type
= pi_type
;
850 blk_queue_logical_block_size(ns
->queue
, bs
);
852 if (ns
->ms
&& !blk_get_integrity(disk
) && !ns
->ext
)
853 nvme_init_integrity(ns
);
854 if (ns
->ms
&& !(ns
->ms
== 8 && ns
->pi_type
) && !blk_get_integrity(disk
))
855 set_capacity(disk
, 0);
857 set_capacity(disk
, le64_to_cpup(&id
->nsze
) << (ns
->lba_shift
- 9));
859 if (ns
->ctrl
->oncs
& NVME_CTRL_ONCS_DSM
)
860 nvme_config_discard(ns
);
861 blk_mq_unfreeze_queue(disk
->queue
);
867 static char nvme_pr_type(enum pr_type type
)
870 case PR_WRITE_EXCLUSIVE
:
872 case PR_EXCLUSIVE_ACCESS
:
874 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
876 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
878 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
880 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
887 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
888 u64 key
, u64 sa_key
, u8 op
)
890 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
891 struct nvme_command c
;
892 u8 data
[16] = { 0, };
894 put_unaligned_le64(key
, &data
[0]);
895 put_unaligned_le64(sa_key
, &data
[8]);
897 memset(&c
, 0, sizeof(c
));
898 c
.common
.opcode
= op
;
899 c
.common
.nsid
= cpu_to_le32(ns
->ns_id
);
900 c
.common
.cdw10
[0] = cpu_to_le32(cdw10
);
902 return nvme_submit_sync_cmd(ns
->queue
, &c
, data
, 16);
905 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
906 u64
new, unsigned flags
)
910 if (flags
& ~PR_FL_IGNORE_KEY
)
914 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
915 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
916 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
919 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
920 enum pr_type type
, unsigned flags
)
924 if (flags
& ~PR_FL_IGNORE_KEY
)
927 cdw10
= nvme_pr_type(type
) << 8;
928 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
929 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
932 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
933 enum pr_type type
, bool abort
)
935 u32 cdw10
= nvme_pr_type(type
) << 8 | abort
? 2 : 1;
936 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
939 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
941 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
942 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
945 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
947 u32 cdw10
= nvme_pr_type(type
) << 8 | key
? 1 << 3 : 0;
948 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
951 static const struct pr_ops nvme_pr_ops
= {
952 .pr_register
= nvme_pr_register
,
953 .pr_reserve
= nvme_pr_reserve
,
954 .pr_release
= nvme_pr_release
,
955 .pr_preempt
= nvme_pr_preempt
,
956 .pr_clear
= nvme_pr_clear
,
959 static const struct block_device_operations nvme_fops
= {
960 .owner
= THIS_MODULE
,
962 .compat_ioctl
= nvme_compat_ioctl
,
964 .release
= nvme_release
,
965 .getgeo
= nvme_getgeo
,
966 .revalidate_disk
= nvme_revalidate_disk
,
967 .pr_ops
= &nvme_pr_ops
,
970 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
972 unsigned long timeout
=
973 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
974 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
977 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
978 if ((csts
& NVME_CSTS_RDY
) == bit
)
982 if (fatal_signal_pending(current
))
984 if (time_after(jiffies
, timeout
)) {
985 dev_err(ctrl
->device
,
986 "Device not ready; aborting %s\n", enabled
?
987 "initialisation" : "reset");
996 * If the device has been passed off to us in an enabled state, just clear
997 * the enabled bit. The spec says we should set the 'shutdown notification
998 * bits', but doing so may cause the device to complete commands to the
999 * admin queue ... and we don't know what memory that might be pointing at!
1001 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
1005 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
1006 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
1008 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1011 return nvme_wait_ready(ctrl
, cap
, false);
1013 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
1015 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
1018 * Default to a 4K page size, with the intention to update this
1019 * path in the future to accomodate architectures with differing
1020 * kernel and IO page sizes.
1022 unsigned dev_page_min
= NVME_CAP_MPSMIN(cap
) + 12, page_shift
= 12;
1025 if (page_shift
< dev_page_min
) {
1026 dev_err(ctrl
->device
,
1027 "Minimum device page size %u too large for host (%u)\n",
1028 1 << dev_page_min
, 1 << page_shift
);
1032 ctrl
->page_size
= 1 << page_shift
;
1034 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
1035 ctrl
->ctrl_config
|= (page_shift
- 12) << NVME_CC_MPS_SHIFT
;
1036 ctrl
->ctrl_config
|= NVME_CC_ARB_RR
| NVME_CC_SHN_NONE
;
1037 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
1038 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
1040 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1043 return nvme_wait_ready(ctrl
, cap
, true);
1045 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
1047 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
1049 unsigned long timeout
= SHUTDOWN_TIMEOUT
+ jiffies
;
1053 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
1054 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
1056 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1060 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
1061 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
1065 if (fatal_signal_pending(current
))
1067 if (time_after(jiffies
, timeout
)) {
1068 dev_err(ctrl
->device
,
1069 "Device shutdown incomplete; abort shutdown\n");
1076 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
1078 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
1079 struct request_queue
*q
)
1083 if (ctrl
->max_hw_sectors
) {
1085 (ctrl
->max_hw_sectors
/ (ctrl
->page_size
>> 9)) + 1;
1087 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
1088 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
1090 if (ctrl
->stripe_size
)
1091 blk_queue_chunk_sectors(q
, ctrl
->stripe_size
>> 9);
1092 blk_queue_virt_boundary(q
, ctrl
->page_size
- 1);
1093 if (ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
)
1095 blk_queue_write_cache(q
, vwc
, vwc
);
1099 * Initialize the cached copies of the Identify data and various controller
1100 * register in our nvme_ctrl structure. This should be called as soon as
1101 * the admin queue is fully up and running.
1103 int nvme_init_identify(struct nvme_ctrl
*ctrl
)
1105 struct nvme_id_ctrl
*id
;
1107 int ret
, page_shift
;
1109 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
1111 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
1115 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &cap
);
1117 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
1120 page_shift
= NVME_CAP_MPSMIN(cap
) + 12;
1122 if (ctrl
->vs
>= NVME_VS(1, 1))
1123 ctrl
->subsystem
= NVME_CAP_NSSRC(cap
);
1125 ret
= nvme_identify_ctrl(ctrl
, &id
);
1127 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
1131 ctrl
->vid
= le16_to_cpu(id
->vid
);
1132 ctrl
->oncs
= le16_to_cpup(&id
->oncs
);
1133 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
1134 ctrl
->vwc
= id
->vwc
;
1135 ctrl
->cntlid
= le16_to_cpup(&id
->cntlid
);
1136 memcpy(ctrl
->serial
, id
->sn
, sizeof(id
->sn
));
1137 memcpy(ctrl
->model
, id
->mn
, sizeof(id
->mn
));
1138 memcpy(ctrl
->firmware_rev
, id
->fr
, sizeof(id
->fr
));
1140 ctrl
->max_hw_sectors
= 1 << (id
->mdts
+ page_shift
- 9);
1142 ctrl
->max_hw_sectors
= UINT_MAX
;
1144 if ((ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) && id
->vs
[3]) {
1145 unsigned int max_hw_sectors
;
1147 ctrl
->stripe_size
= 1 << (id
->vs
[3] + page_shift
);
1148 max_hw_sectors
= ctrl
->stripe_size
>> (page_shift
- 9);
1149 if (ctrl
->max_hw_sectors
) {
1150 ctrl
->max_hw_sectors
= min(max_hw_sectors
,
1151 ctrl
->max_hw_sectors
);
1153 ctrl
->max_hw_sectors
= max_hw_sectors
;
1157 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
1162 EXPORT_SYMBOL_GPL(nvme_init_identify
);
1164 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
1166 struct nvme_ctrl
*ctrl
;
1167 int instance
= iminor(inode
);
1170 spin_lock(&dev_list_lock
);
1171 list_for_each_entry(ctrl
, &nvme_ctrl_list
, node
) {
1172 if (ctrl
->instance
!= instance
)
1175 if (!ctrl
->admin_q
) {
1179 if (!kref_get_unless_zero(&ctrl
->kref
))
1181 file
->private_data
= ctrl
;
1185 spin_unlock(&dev_list_lock
);
1190 static int nvme_dev_release(struct inode
*inode
, struct file
*file
)
1192 nvme_put_ctrl(file
->private_data
);
1196 static int nvme_dev_user_cmd(struct nvme_ctrl
*ctrl
, void __user
*argp
)
1201 mutex_lock(&ctrl
->namespaces_mutex
);
1202 if (list_empty(&ctrl
->namespaces
)) {
1207 ns
= list_first_entry(&ctrl
->namespaces
, struct nvme_ns
, list
);
1208 if (ns
!= list_last_entry(&ctrl
->namespaces
, struct nvme_ns
, list
)) {
1209 dev_warn(ctrl
->device
,
1210 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1215 dev_warn(ctrl
->device
,
1216 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1217 kref_get(&ns
->kref
);
1218 mutex_unlock(&ctrl
->namespaces_mutex
);
1220 ret
= nvme_user_cmd(ctrl
, ns
, argp
);
1225 mutex_unlock(&ctrl
->namespaces_mutex
);
1229 static long nvme_dev_ioctl(struct file
*file
, unsigned int cmd
,
1232 struct nvme_ctrl
*ctrl
= file
->private_data
;
1233 void __user
*argp
= (void __user
*)arg
;
1236 case NVME_IOCTL_ADMIN_CMD
:
1237 return nvme_user_cmd(ctrl
, NULL
, argp
);
1238 case NVME_IOCTL_IO_CMD
:
1239 return nvme_dev_user_cmd(ctrl
, argp
);
1240 case NVME_IOCTL_RESET
:
1241 dev_warn(ctrl
->device
, "resetting controller\n");
1242 return ctrl
->ops
->reset_ctrl(ctrl
);
1243 case NVME_IOCTL_SUBSYS_RESET
:
1244 return nvme_reset_subsystem(ctrl
);
1245 case NVME_IOCTL_RESCAN
:
1246 nvme_queue_scan(ctrl
);
1253 static const struct file_operations nvme_dev_fops
= {
1254 .owner
= THIS_MODULE
,
1255 .open
= nvme_dev_open
,
1256 .release
= nvme_dev_release
,
1257 .unlocked_ioctl
= nvme_dev_ioctl
,
1258 .compat_ioctl
= nvme_dev_ioctl
,
1261 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
1262 struct device_attribute
*attr
, const char *buf
,
1265 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1268 ret
= ctrl
->ops
->reset_ctrl(ctrl
);
1273 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
1275 static ssize_t
nvme_sysfs_rescan(struct device
*dev
,
1276 struct device_attribute
*attr
, const char *buf
,
1279 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1281 nvme_queue_scan(ctrl
);
1284 static DEVICE_ATTR(rescan_controller
, S_IWUSR
, NULL
, nvme_sysfs_rescan
);
1286 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
1289 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1290 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1291 int serial_len
= sizeof(ctrl
->serial
);
1292 int model_len
= sizeof(ctrl
->model
);
1294 if (memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1295 return sprintf(buf
, "eui.%16phN\n", ns
->uuid
);
1297 if (memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1298 return sprintf(buf
, "eui.%8phN\n", ns
->eui
);
1300 while (ctrl
->serial
[serial_len
- 1] == ' ')
1302 while (ctrl
->model
[model_len
- 1] == ' ')
1305 return sprintf(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl
->vid
,
1306 serial_len
, ctrl
->serial
, model_len
, ctrl
->model
, ns
->ns_id
);
1308 static DEVICE_ATTR(wwid
, S_IRUGO
, wwid_show
, NULL
);
1310 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
1313 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1314 return sprintf(buf
, "%pU\n", ns
->uuid
);
1316 static DEVICE_ATTR(uuid
, S_IRUGO
, uuid_show
, NULL
);
1318 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
1321 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1322 return sprintf(buf
, "%8phd\n", ns
->eui
);
1324 static DEVICE_ATTR(eui
, S_IRUGO
, eui_show
, NULL
);
1326 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
1329 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1330 return sprintf(buf
, "%d\n", ns
->ns_id
);
1332 static DEVICE_ATTR(nsid
, S_IRUGO
, nsid_show
, NULL
);
1334 static struct attribute
*nvme_ns_attrs
[] = {
1335 &dev_attr_wwid
.attr
,
1336 &dev_attr_uuid
.attr
,
1338 &dev_attr_nsid
.attr
,
1342 static umode_t
nvme_attrs_are_visible(struct kobject
*kobj
,
1343 struct attribute
*a
, int n
)
1345 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
1346 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1348 if (a
== &dev_attr_uuid
.attr
) {
1349 if (!memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1352 if (a
== &dev_attr_eui
.attr
) {
1353 if (!memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1359 static const struct attribute_group nvme_ns_attr_group
= {
1360 .attrs
= nvme_ns_attrs
,
1361 .is_visible
= nvme_attrs_are_visible
,
1364 #define nvme_show_str_function(field) \
1365 static ssize_t field##_show(struct device *dev, \
1366 struct device_attribute *attr, char *buf) \
1368 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1369 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1371 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1373 #define nvme_show_int_function(field) \
1374 static ssize_t field##_show(struct device *dev, \
1375 struct device_attribute *attr, char *buf) \
1377 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1378 return sprintf(buf, "%d\n", ctrl->field); \
1380 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1382 nvme_show_str_function(model
);
1383 nvme_show_str_function(serial
);
1384 nvme_show_str_function(firmware_rev
);
1385 nvme_show_int_function(cntlid
);
1387 static struct attribute
*nvme_dev_attrs
[] = {
1388 &dev_attr_reset_controller
.attr
,
1389 &dev_attr_rescan_controller
.attr
,
1390 &dev_attr_model
.attr
,
1391 &dev_attr_serial
.attr
,
1392 &dev_attr_firmware_rev
.attr
,
1393 &dev_attr_cntlid
.attr
,
1397 static struct attribute_group nvme_dev_attrs_group
= {
1398 .attrs
= nvme_dev_attrs
,
1401 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
1402 &nvme_dev_attrs_group
,
1406 static int ns_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1408 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
1409 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
1411 return nsa
->ns_id
- nsb
->ns_id
;
1414 static struct nvme_ns
*nvme_find_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1418 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1420 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
1421 if (ns
->ns_id
== nsid
)
1423 if (ns
->ns_id
> nsid
)
1429 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1432 struct gendisk
*disk
;
1433 int node
= dev_to_node(ctrl
->dev
);
1435 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1437 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
1441 ns
->instance
= ida_simple_get(&ctrl
->ns_ida
, 1, 0, GFP_KERNEL
);
1442 if (ns
->instance
< 0)
1445 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
1446 if (IS_ERR(ns
->queue
))
1447 goto out_release_instance
;
1448 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, ns
->queue
);
1449 ns
->queue
->queuedata
= ns
;
1452 disk
= alloc_disk_node(0, node
);
1454 goto out_free_queue
;
1456 kref_init(&ns
->kref
);
1459 ns
->lba_shift
= 9; /* set to a default value for 512 until disk is validated */
1462 blk_queue_logical_block_size(ns
->queue
, 1 << ns
->lba_shift
);
1463 nvme_set_queue_limits(ctrl
, ns
->queue
);
1465 disk
->major
= nvme_major
;
1466 disk
->first_minor
= 0;
1467 disk
->fops
= &nvme_fops
;
1468 disk
->private_data
= ns
;
1469 disk
->queue
= ns
->queue
;
1470 disk
->driverfs_dev
= ctrl
->device
;
1471 disk
->flags
= GENHD_FL_EXT_DEVT
;
1472 sprintf(disk
->disk_name
, "nvme%dn%d", ctrl
->instance
, ns
->instance
);
1474 if (nvme_revalidate_disk(ns
->disk
))
1477 list_add_tail_rcu(&ns
->list
, &ctrl
->namespaces
);
1478 kref_get(&ctrl
->kref
);
1479 if (ns
->type
== NVME_NS_LIGHTNVM
)
1483 if (sysfs_create_group(&disk_to_dev(ns
->disk
)->kobj
,
1484 &nvme_ns_attr_group
))
1485 pr_warn("%s: failed to create sysfs group for identification\n",
1486 ns
->disk
->disk_name
);
1491 blk_cleanup_queue(ns
->queue
);
1492 out_release_instance
:
1493 ida_simple_remove(&ctrl
->ns_ida
, ns
->instance
);
1498 static void nvme_ns_remove(struct nvme_ns
*ns
)
1500 lockdep_assert_held(&ns
->ctrl
->namespaces_mutex
);
1502 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
1505 if (ns
->disk
->flags
& GENHD_FL_UP
) {
1506 if (blk_get_integrity(ns
->disk
))
1507 blk_integrity_unregister(ns
->disk
);
1508 sysfs_remove_group(&disk_to_dev(ns
->disk
)->kobj
,
1509 &nvme_ns_attr_group
);
1510 del_gendisk(ns
->disk
);
1511 blk_mq_abort_requeue_list(ns
->queue
);
1512 blk_cleanup_queue(ns
->queue
);
1514 list_del_init(&ns
->list
);
1519 static void nvme_validate_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1523 ns
= nvme_find_ns(ctrl
, nsid
);
1525 if (revalidate_disk(ns
->disk
))
1528 nvme_alloc_ns(ctrl
, nsid
);
1531 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
, unsigned nn
)
1535 unsigned i
, j
, nsid
, prev
= 0, num_lists
= DIV_ROUND_UP(nn
, 1024);
1538 ns_list
= kzalloc(0x1000, GFP_KERNEL
);
1542 for (i
= 0; i
< num_lists
; i
++) {
1543 ret
= nvme_identify_ns_list(ctrl
, prev
, ns_list
);
1547 for (j
= 0; j
< min(nn
, 1024U); j
++) {
1548 nsid
= le32_to_cpu(ns_list
[j
]);
1552 nvme_validate_ns(ctrl
, nsid
);
1554 while (++prev
< nsid
) {
1555 ns
= nvme_find_ns(ctrl
, prev
);
1567 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
, unsigned nn
)
1569 struct nvme_ns
*ns
, *next
;
1572 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1574 for (i
= 1; i
<= nn
; i
++)
1575 nvme_validate_ns(ctrl
, i
);
1577 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
1583 static void nvme_scan_work(struct work_struct
*work
)
1585 struct nvme_ctrl
*ctrl
=
1586 container_of(work
, struct nvme_ctrl
, scan_work
);
1587 struct nvme_id_ctrl
*id
;
1590 if (ctrl
->state
!= NVME_CTRL_LIVE
)
1593 if (nvme_identify_ctrl(ctrl
, &id
))
1596 mutex_lock(&ctrl
->namespaces_mutex
);
1597 nn
= le32_to_cpu(id
->nn
);
1598 if (ctrl
->vs
>= NVME_VS(1, 1) &&
1599 !(ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)) {
1600 if (!nvme_scan_ns_list(ctrl
, nn
))
1603 nvme_scan_ns_sequential(ctrl
, nn
);
1605 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
1606 mutex_unlock(&ctrl
->namespaces_mutex
);
1609 if (ctrl
->ops
->post_scan
)
1610 ctrl
->ops
->post_scan(ctrl
);
1613 void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
1616 * Do not queue new scan work when a controller is reset during
1619 if (ctrl
->state
== NVME_CTRL_LIVE
)
1620 schedule_work(&ctrl
->scan_work
);
1622 EXPORT_SYMBOL_GPL(nvme_queue_scan
);
1624 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
1626 struct nvme_ns
*ns
, *next
;
1629 * The dead states indicates the controller was not gracefully
1630 * disconnected. In that case, we won't be able to flush any data while
1631 * removing the namespaces' disks; fail all the queues now to avoid
1632 * potentially having to clean up the failed sync later.
1634 if (ctrl
->state
== NVME_CTRL_DEAD
)
1635 nvme_kill_queues(ctrl
);
1637 mutex_lock(&ctrl
->namespaces_mutex
);
1638 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
)
1640 mutex_unlock(&ctrl
->namespaces_mutex
);
1642 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
1644 static void nvme_async_event_work(struct work_struct
*work
)
1646 struct nvme_ctrl
*ctrl
=
1647 container_of(work
, struct nvme_ctrl
, async_event_work
);
1649 spin_lock_irq(&ctrl
->lock
);
1650 while (ctrl
->event_limit
> 0) {
1651 int aer_idx
= --ctrl
->event_limit
;
1653 spin_unlock_irq(&ctrl
->lock
);
1654 ctrl
->ops
->submit_async_event(ctrl
, aer_idx
);
1655 spin_lock_irq(&ctrl
->lock
);
1657 spin_unlock_irq(&ctrl
->lock
);
1660 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
,
1661 struct nvme_completion
*cqe
)
1663 u16 status
= le16_to_cpu(cqe
->status
) >> 1;
1664 u32 result
= le32_to_cpu(cqe
->result
);
1666 if (status
== NVME_SC_SUCCESS
|| status
== NVME_SC_ABORT_REQ
) {
1667 ++ctrl
->event_limit
;
1668 schedule_work(&ctrl
->async_event_work
);
1671 if (status
!= NVME_SC_SUCCESS
)
1674 switch (result
& 0xff07) {
1675 case NVME_AER_NOTICE_NS_CHANGED
:
1676 dev_info(ctrl
->device
, "rescanning\n");
1677 nvme_queue_scan(ctrl
);
1680 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
1683 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
1685 void nvme_queue_async_events(struct nvme_ctrl
*ctrl
)
1687 ctrl
->event_limit
= NVME_NR_AERS
;
1688 schedule_work(&ctrl
->async_event_work
);
1690 EXPORT_SYMBOL_GPL(nvme_queue_async_events
);
1692 static DEFINE_IDA(nvme_instance_ida
);
1694 static int nvme_set_instance(struct nvme_ctrl
*ctrl
)
1696 int instance
, error
;
1699 if (!ida_pre_get(&nvme_instance_ida
, GFP_KERNEL
))
1702 spin_lock(&dev_list_lock
);
1703 error
= ida_get_new(&nvme_instance_ida
, &instance
);
1704 spin_unlock(&dev_list_lock
);
1705 } while (error
== -EAGAIN
);
1710 ctrl
->instance
= instance
;
1714 static void nvme_release_instance(struct nvme_ctrl
*ctrl
)
1716 spin_lock(&dev_list_lock
);
1717 ida_remove(&nvme_instance_ida
, ctrl
->instance
);
1718 spin_unlock(&dev_list_lock
);
1721 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
1723 flush_work(&ctrl
->async_event_work
);
1724 flush_work(&ctrl
->scan_work
);
1725 nvme_remove_namespaces(ctrl
);
1727 device_destroy(nvme_class
, MKDEV(nvme_char_major
, ctrl
->instance
));
1729 spin_lock(&dev_list_lock
);
1730 list_del(&ctrl
->node
);
1731 spin_unlock(&dev_list_lock
);
1733 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
1735 static void nvme_free_ctrl(struct kref
*kref
)
1737 struct nvme_ctrl
*ctrl
= container_of(kref
, struct nvme_ctrl
, kref
);
1739 put_device(ctrl
->device
);
1740 nvme_release_instance(ctrl
);
1741 ida_destroy(&ctrl
->ns_ida
);
1743 ctrl
->ops
->free_ctrl(ctrl
);
1746 void nvme_put_ctrl(struct nvme_ctrl
*ctrl
)
1748 kref_put(&ctrl
->kref
, nvme_free_ctrl
);
1750 EXPORT_SYMBOL_GPL(nvme_put_ctrl
);
1753 * Initialize a NVMe controller structures. This needs to be called during
1754 * earliest initialization so that we have the initialized structured around
1757 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
1758 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
1762 ctrl
->state
= NVME_CTRL_NEW
;
1763 spin_lock_init(&ctrl
->lock
);
1764 INIT_LIST_HEAD(&ctrl
->namespaces
);
1765 mutex_init(&ctrl
->namespaces_mutex
);
1766 kref_init(&ctrl
->kref
);
1769 ctrl
->quirks
= quirks
;
1770 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
1771 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
1773 ret
= nvme_set_instance(ctrl
);
1777 ctrl
->device
= device_create_with_groups(nvme_class
, ctrl
->dev
,
1778 MKDEV(nvme_char_major
, ctrl
->instance
),
1779 ctrl
, nvme_dev_attr_groups
,
1780 "nvme%d", ctrl
->instance
);
1781 if (IS_ERR(ctrl
->device
)) {
1782 ret
= PTR_ERR(ctrl
->device
);
1783 goto out_release_instance
;
1785 get_device(ctrl
->device
);
1786 ida_init(&ctrl
->ns_ida
);
1788 spin_lock(&dev_list_lock
);
1789 list_add_tail(&ctrl
->node
, &nvme_ctrl_list
);
1790 spin_unlock(&dev_list_lock
);
1793 out_release_instance
:
1794 nvme_release_instance(ctrl
);
1798 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
1801 * nvme_kill_queues(): Ends all namespace queues
1802 * @ctrl: the dead controller that needs to end
1804 * Call this function when the driver determines it is unable to get the
1805 * controller in a state capable of servicing IO.
1807 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
1812 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1813 if (!kref_get_unless_zero(&ns
->kref
))
1817 * Revalidating a dead namespace sets capacity to 0. This will
1818 * end buffered writers dirtying pages that can't be synced.
1820 if (!test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
1821 revalidate_disk(ns
->disk
);
1823 blk_set_queue_dying(ns
->queue
);
1824 blk_mq_abort_requeue_list(ns
->queue
);
1825 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
1831 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
1833 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
1838 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1839 spin_lock_irq(ns
->queue
->queue_lock
);
1840 queue_flag_set(QUEUE_FLAG_STOPPED
, ns
->queue
);
1841 spin_unlock_irq(ns
->queue
->queue_lock
);
1843 blk_mq_cancel_requeue_work(ns
->queue
);
1844 blk_mq_stop_hw_queues(ns
->queue
);
1848 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
1850 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
1855 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1856 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED
, ns
->queue
);
1857 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
1858 blk_mq_kick_requeue_list(ns
->queue
);
1862 EXPORT_SYMBOL_GPL(nvme_start_queues
);
1864 int __init
nvme_core_init(void)
1868 result
= register_blkdev(nvme_major
, "nvme");
1871 else if (result
> 0)
1872 nvme_major
= result
;
1874 result
= __register_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme",
1877 goto unregister_blkdev
;
1878 else if (result
> 0)
1879 nvme_char_major
= result
;
1881 nvme_class
= class_create(THIS_MODULE
, "nvme");
1882 if (IS_ERR(nvme_class
)) {
1883 result
= PTR_ERR(nvme_class
);
1884 goto unregister_chrdev
;
1890 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
1892 unregister_blkdev(nvme_major
, "nvme");
1896 void nvme_core_exit(void)
1898 class_destroy(nvme_class
);
1899 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
1900 unregister_blkdev(nvme_major
, "nvme");
1903 MODULE_LICENSE("GPL");
1904 MODULE_VERSION("1.0");
1905 module_init(nvme_core_init
);
1906 module_exit(nvme_core_exit
);