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 bool nvme_change_ctrl_state(struct nvme_ctrl
*ctrl
,
62 enum nvme_ctrl_state new_state
)
64 enum nvme_ctrl_state old_state
= ctrl
->state
;
67 spin_lock_irq(&ctrl
->lock
);
71 case NVME_CTRL_RESETTING
:
78 case NVME_CTRL_RESETTING
:
88 case NVME_CTRL_DELETING
:
91 case NVME_CTRL_RESETTING
:
101 spin_unlock_irq(&ctrl
->lock
);
104 ctrl
->state
= new_state
;
108 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state
);
110 static void nvme_free_ns(struct kref
*kref
)
112 struct nvme_ns
*ns
= container_of(kref
, struct nvme_ns
, kref
);
114 if (ns
->type
== NVME_NS_LIGHTNVM
)
115 nvme_nvm_unregister(ns
->queue
, ns
->disk
->disk_name
);
117 spin_lock(&dev_list_lock
);
118 ns
->disk
->private_data
= NULL
;
119 spin_unlock(&dev_list_lock
);
122 ida_simple_remove(&ns
->ctrl
->ns_ida
, ns
->instance
);
123 nvme_put_ctrl(ns
->ctrl
);
127 static void nvme_put_ns(struct nvme_ns
*ns
)
129 kref_put(&ns
->kref
, nvme_free_ns
);
132 static struct nvme_ns
*nvme_get_ns_from_disk(struct gendisk
*disk
)
136 spin_lock(&dev_list_lock
);
137 ns
= disk
->private_data
;
139 if (!kref_get_unless_zero(&ns
->kref
))
141 if (!try_module_get(ns
->ctrl
->ops
->module
))
144 spin_unlock(&dev_list_lock
);
149 kref_put(&ns
->kref
, nvme_free_ns
);
151 spin_unlock(&dev_list_lock
);
155 void nvme_requeue_req(struct request
*req
)
159 blk_mq_requeue_request(req
);
160 spin_lock_irqsave(req
->q
->queue_lock
, flags
);
161 if (!blk_queue_stopped(req
->q
))
162 blk_mq_kick_requeue_list(req
->q
);
163 spin_unlock_irqrestore(req
->q
->queue_lock
, flags
);
165 EXPORT_SYMBOL_GPL(nvme_requeue_req
);
167 struct request
*nvme_alloc_request(struct request_queue
*q
,
168 struct nvme_command
*cmd
, unsigned int flags
)
170 bool write
= cmd
->common
.opcode
& 1;
173 req
= blk_mq_alloc_request(q
, write
, flags
);
177 req
->cmd_type
= REQ_TYPE_DRV_PRIV
;
178 req
->cmd_flags
|= REQ_FAILFAST_DRIVER
;
180 req
->__sector
= (sector_t
) -1;
181 req
->bio
= req
->biotail
= NULL
;
183 req
->cmd
= (unsigned char *)cmd
;
184 req
->cmd_len
= sizeof(struct nvme_command
);
188 EXPORT_SYMBOL_GPL(nvme_alloc_request
);
190 static inline void nvme_setup_flush(struct nvme_ns
*ns
,
191 struct nvme_command
*cmnd
)
193 memset(cmnd
, 0, sizeof(*cmnd
));
194 cmnd
->common
.opcode
= nvme_cmd_flush
;
195 cmnd
->common
.nsid
= cpu_to_le32(ns
->ns_id
);
198 static inline int nvme_setup_discard(struct nvme_ns
*ns
, struct request
*req
,
199 struct nvme_command
*cmnd
)
201 struct nvme_dsm_range
*range
;
204 unsigned int nr_bytes
= blk_rq_bytes(req
);
206 range
= kmalloc(sizeof(*range
), GFP_ATOMIC
);
208 return BLK_MQ_RQ_QUEUE_BUSY
;
210 range
->cattr
= cpu_to_le32(0);
211 range
->nlb
= cpu_to_le32(nr_bytes
>> ns
->lba_shift
);
212 range
->slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
214 memset(cmnd
, 0, sizeof(*cmnd
));
215 cmnd
->dsm
.opcode
= nvme_cmd_dsm
;
216 cmnd
->dsm
.nsid
= cpu_to_le32(ns
->ns_id
);
218 cmnd
->dsm
.attributes
= cpu_to_le32(NVME_DSMGMT_AD
);
220 req
->completion_data
= range
;
221 page
= virt_to_page(range
);
222 offset
= offset_in_page(range
);
223 blk_add_request_payload(req
, page
, offset
, sizeof(*range
));
226 * we set __data_len back to the size of the area to be discarded
227 * on disk. This allows us to report completion on the full amount
228 * of blocks described by the request.
230 req
->__data_len
= nr_bytes
;
235 static inline void nvme_setup_rw(struct nvme_ns
*ns
, struct request
*req
,
236 struct nvme_command
*cmnd
)
241 if (req
->cmd_flags
& REQ_FUA
)
242 control
|= NVME_RW_FUA
;
243 if (req
->cmd_flags
& (REQ_FAILFAST_DEV
| REQ_RAHEAD
))
244 control
|= NVME_RW_LR
;
246 if (req
->cmd_flags
& REQ_RAHEAD
)
247 dsmgmt
|= NVME_RW_DSM_FREQ_PREFETCH
;
249 memset(cmnd
, 0, sizeof(*cmnd
));
250 cmnd
->rw
.opcode
= (rq_data_dir(req
) ? nvme_cmd_write
: nvme_cmd_read
);
251 cmnd
->rw
.command_id
= req
->tag
;
252 cmnd
->rw
.nsid
= cpu_to_le32(ns
->ns_id
);
253 cmnd
->rw
.slba
= cpu_to_le64(nvme_block_nr(ns
, blk_rq_pos(req
)));
254 cmnd
->rw
.length
= cpu_to_le16((blk_rq_bytes(req
) >> ns
->lba_shift
) - 1);
257 switch (ns
->pi_type
) {
258 case NVME_NS_DPS_PI_TYPE3
:
259 control
|= NVME_RW_PRINFO_PRCHK_GUARD
;
261 case NVME_NS_DPS_PI_TYPE1
:
262 case NVME_NS_DPS_PI_TYPE2
:
263 control
|= NVME_RW_PRINFO_PRCHK_GUARD
|
264 NVME_RW_PRINFO_PRCHK_REF
;
265 cmnd
->rw
.reftag
= cpu_to_le32(
266 nvme_block_nr(ns
, blk_rq_pos(req
)));
269 if (!blk_integrity_rq(req
))
270 control
|= NVME_RW_PRINFO_PRACT
;
273 cmnd
->rw
.control
= cpu_to_le16(control
);
274 cmnd
->rw
.dsmgmt
= cpu_to_le32(dsmgmt
);
277 int nvme_setup_cmd(struct nvme_ns
*ns
, struct request
*req
,
278 struct nvme_command
*cmd
)
282 if (req
->cmd_type
== REQ_TYPE_DRV_PRIV
)
283 memcpy(cmd
, req
->cmd
, sizeof(*cmd
));
284 else if (req
->cmd_flags
& REQ_FLUSH
)
285 nvme_setup_flush(ns
, cmd
);
286 else if (req
->cmd_flags
& REQ_DISCARD
)
287 ret
= nvme_setup_discard(ns
, req
, cmd
);
289 nvme_setup_rw(ns
, req
, cmd
);
293 EXPORT_SYMBOL_GPL(nvme_setup_cmd
);
296 * Returns 0 on success. If the result is negative, it's a Linux error code;
297 * if the result is positive, it's an NVM Express status code
299 int __nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
300 struct nvme_completion
*cqe
, void *buffer
, unsigned bufflen
,
306 req
= nvme_alloc_request(q
, cmd
, 0);
310 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
313 if (buffer
&& bufflen
) {
314 ret
= blk_rq_map_kern(q
, req
, buffer
, bufflen
, GFP_KERNEL
);
319 blk_execute_rq(req
->q
, NULL
, req
, 0);
322 blk_mq_free_request(req
);
326 int nvme_submit_sync_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
327 void *buffer
, unsigned bufflen
)
329 return __nvme_submit_sync_cmd(q
, cmd
, NULL
, buffer
, bufflen
, 0);
331 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd
);
333 int __nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
334 void __user
*ubuffer
, unsigned bufflen
,
335 void __user
*meta_buffer
, unsigned meta_len
, u32 meta_seed
,
336 u32
*result
, unsigned timeout
)
338 bool write
= cmd
->common
.opcode
& 1;
339 struct nvme_completion cqe
;
340 struct nvme_ns
*ns
= q
->queuedata
;
341 struct gendisk
*disk
= ns
? ns
->disk
: NULL
;
343 struct bio
*bio
= NULL
;
347 req
= nvme_alloc_request(q
, cmd
, 0);
351 req
->timeout
= timeout
? timeout
: ADMIN_TIMEOUT
;
354 if (ubuffer
&& bufflen
) {
355 ret
= blk_rq_map_user(q
, req
, NULL
, ubuffer
, bufflen
,
363 bio
->bi_bdev
= bdget_disk(disk
, 0);
369 if (meta_buffer
&& meta_len
) {
370 struct bio_integrity_payload
*bip
;
372 meta
= kmalloc(meta_len
, GFP_KERNEL
);
379 if (copy_from_user(meta
, meta_buffer
,
386 bip
= bio_integrity_alloc(bio
, GFP_KERNEL
, 1);
392 bip
->bip_iter
.bi_size
= meta_len
;
393 bip
->bip_iter
.bi_sector
= meta_seed
;
395 ret
= bio_integrity_add_page(bio
, virt_to_page(meta
),
396 meta_len
, offset_in_page(meta
));
397 if (ret
!= meta_len
) {
404 blk_execute_rq(req
->q
, disk
, req
, 0);
407 *result
= le32_to_cpu(cqe
.result
);
408 if (meta
&& !ret
&& !write
) {
409 if (copy_to_user(meta_buffer
, meta
, meta_len
))
416 if (disk
&& bio
->bi_bdev
)
418 blk_rq_unmap_user(bio
);
421 blk_mq_free_request(req
);
425 int nvme_submit_user_cmd(struct request_queue
*q
, struct nvme_command
*cmd
,
426 void __user
*ubuffer
, unsigned bufflen
, u32
*result
,
429 return __nvme_submit_user_cmd(q
, cmd
, ubuffer
, bufflen
, NULL
, 0, 0,
433 int nvme_identify_ctrl(struct nvme_ctrl
*dev
, struct nvme_id_ctrl
**id
)
435 struct nvme_command c
= { };
438 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
439 c
.identify
.opcode
= nvme_admin_identify
;
440 c
.identify
.cns
= cpu_to_le32(1);
442 *id
= kmalloc(sizeof(struct nvme_id_ctrl
), GFP_KERNEL
);
446 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
447 sizeof(struct nvme_id_ctrl
));
453 static int nvme_identify_ns_list(struct nvme_ctrl
*dev
, unsigned nsid
, __le32
*ns_list
)
455 struct nvme_command c
= { };
457 c
.identify
.opcode
= nvme_admin_identify
;
458 c
.identify
.cns
= cpu_to_le32(2);
459 c
.identify
.nsid
= cpu_to_le32(nsid
);
460 return nvme_submit_sync_cmd(dev
->admin_q
, &c
, ns_list
, 0x1000);
463 int nvme_identify_ns(struct nvme_ctrl
*dev
, unsigned nsid
,
464 struct nvme_id_ns
**id
)
466 struct nvme_command c
= { };
469 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
470 c
.identify
.opcode
= nvme_admin_identify
,
471 c
.identify
.nsid
= cpu_to_le32(nsid
),
473 *id
= kmalloc(sizeof(struct nvme_id_ns
), GFP_KERNEL
);
477 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *id
,
478 sizeof(struct nvme_id_ns
));
484 int nvme_get_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned nsid
,
485 dma_addr_t dma_addr
, u32
*result
)
487 struct nvme_command c
;
488 struct nvme_completion cqe
;
491 memset(&c
, 0, sizeof(c
));
492 c
.features
.opcode
= nvme_admin_get_features
;
493 c
.features
.nsid
= cpu_to_le32(nsid
);
494 c
.features
.prp1
= cpu_to_le64(dma_addr
);
495 c
.features
.fid
= cpu_to_le32(fid
);
497 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
, NULL
, 0, 0);
499 *result
= le32_to_cpu(cqe
.result
);
503 int nvme_set_features(struct nvme_ctrl
*dev
, unsigned fid
, unsigned dword11
,
504 dma_addr_t dma_addr
, u32
*result
)
506 struct nvme_command c
;
507 struct nvme_completion cqe
;
510 memset(&c
, 0, sizeof(c
));
511 c
.features
.opcode
= nvme_admin_set_features
;
512 c
.features
.prp1
= cpu_to_le64(dma_addr
);
513 c
.features
.fid
= cpu_to_le32(fid
);
514 c
.features
.dword11
= cpu_to_le32(dword11
);
516 ret
= __nvme_submit_sync_cmd(dev
->admin_q
, &c
, &cqe
, NULL
, 0, 0);
518 *result
= le32_to_cpu(cqe
.result
);
522 int nvme_get_log_page(struct nvme_ctrl
*dev
, struct nvme_smart_log
**log
)
524 struct nvme_command c
= { };
527 c
.common
.opcode
= nvme_admin_get_log_page
,
528 c
.common
.nsid
= cpu_to_le32(0xFFFFFFFF),
529 c
.common
.cdw10
[0] = cpu_to_le32(
530 (((sizeof(struct nvme_smart_log
) / 4) - 1) << 16) |
533 *log
= kmalloc(sizeof(struct nvme_smart_log
), GFP_KERNEL
);
537 error
= nvme_submit_sync_cmd(dev
->admin_q
, &c
, *log
,
538 sizeof(struct nvme_smart_log
));
544 int nvme_set_queue_count(struct nvme_ctrl
*ctrl
, int *count
)
546 u32 q_count
= (*count
- 1) | ((*count
- 1) << 16);
548 int status
, nr_io_queues
;
550 status
= nvme_set_features(ctrl
, NVME_FEAT_NUM_QUEUES
, q_count
, 0,
555 nr_io_queues
= min(result
& 0xffff, result
>> 16) + 1;
556 *count
= min(*count
, nr_io_queues
);
559 EXPORT_SYMBOL_GPL(nvme_set_queue_count
);
561 static int nvme_submit_io(struct nvme_ns
*ns
, struct nvme_user_io __user
*uio
)
563 struct nvme_user_io io
;
564 struct nvme_command c
;
565 unsigned length
, meta_len
;
566 void __user
*metadata
;
568 if (copy_from_user(&io
, uio
, sizeof(io
)))
576 case nvme_cmd_compare
:
582 length
= (io
.nblocks
+ 1) << ns
->lba_shift
;
583 meta_len
= (io
.nblocks
+ 1) * ns
->ms
;
584 metadata
= (void __user
*)(uintptr_t)io
.metadata
;
589 } else if (meta_len
) {
590 if ((io
.metadata
& 3) || !io
.metadata
)
594 memset(&c
, 0, sizeof(c
));
595 c
.rw
.opcode
= io
.opcode
;
596 c
.rw
.flags
= io
.flags
;
597 c
.rw
.nsid
= cpu_to_le32(ns
->ns_id
);
598 c
.rw
.slba
= cpu_to_le64(io
.slba
);
599 c
.rw
.length
= cpu_to_le16(io
.nblocks
);
600 c
.rw
.control
= cpu_to_le16(io
.control
);
601 c
.rw
.dsmgmt
= cpu_to_le32(io
.dsmgmt
);
602 c
.rw
.reftag
= cpu_to_le32(io
.reftag
);
603 c
.rw
.apptag
= cpu_to_le16(io
.apptag
);
604 c
.rw
.appmask
= cpu_to_le16(io
.appmask
);
606 return __nvme_submit_user_cmd(ns
->queue
, &c
,
607 (void __user
*)(uintptr_t)io
.addr
, length
,
608 metadata
, meta_len
, io
.slba
, NULL
, 0);
611 static int nvme_user_cmd(struct nvme_ctrl
*ctrl
, struct nvme_ns
*ns
,
612 struct nvme_passthru_cmd __user
*ucmd
)
614 struct nvme_passthru_cmd cmd
;
615 struct nvme_command c
;
616 unsigned timeout
= 0;
619 if (!capable(CAP_SYS_ADMIN
))
621 if (copy_from_user(&cmd
, ucmd
, sizeof(cmd
)))
626 memset(&c
, 0, sizeof(c
));
627 c
.common
.opcode
= cmd
.opcode
;
628 c
.common
.flags
= cmd
.flags
;
629 c
.common
.nsid
= cpu_to_le32(cmd
.nsid
);
630 c
.common
.cdw2
[0] = cpu_to_le32(cmd
.cdw2
);
631 c
.common
.cdw2
[1] = cpu_to_le32(cmd
.cdw3
);
632 c
.common
.cdw10
[0] = cpu_to_le32(cmd
.cdw10
);
633 c
.common
.cdw10
[1] = cpu_to_le32(cmd
.cdw11
);
634 c
.common
.cdw10
[2] = cpu_to_le32(cmd
.cdw12
);
635 c
.common
.cdw10
[3] = cpu_to_le32(cmd
.cdw13
);
636 c
.common
.cdw10
[4] = cpu_to_le32(cmd
.cdw14
);
637 c
.common
.cdw10
[5] = cpu_to_le32(cmd
.cdw15
);
640 timeout
= msecs_to_jiffies(cmd
.timeout_ms
);
642 status
= nvme_submit_user_cmd(ns
? ns
->queue
: ctrl
->admin_q
, &c
,
643 (void __user
*)(uintptr_t)cmd
.addr
, cmd
.data_len
,
644 &cmd
.result
, timeout
);
646 if (put_user(cmd
.result
, &ucmd
->result
))
653 static int nvme_ioctl(struct block_device
*bdev
, fmode_t mode
,
654 unsigned int cmd
, unsigned long arg
)
656 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
660 force_successful_syscall_return();
662 case NVME_IOCTL_ADMIN_CMD
:
663 return nvme_user_cmd(ns
->ctrl
, NULL
, (void __user
*)arg
);
664 case NVME_IOCTL_IO_CMD
:
665 return nvme_user_cmd(ns
->ctrl
, ns
, (void __user
*)arg
);
666 case NVME_IOCTL_SUBMIT_IO
:
667 return nvme_submit_io(ns
, (void __user
*)arg
);
668 #ifdef CONFIG_BLK_DEV_NVME_SCSI
669 case SG_GET_VERSION_NUM
:
670 return nvme_sg_get_version_num((void __user
*)arg
);
672 return nvme_sg_io(ns
, (void __user
*)arg
);
680 static int nvme_compat_ioctl(struct block_device
*bdev
, fmode_t mode
,
681 unsigned int cmd
, unsigned long arg
)
687 return nvme_ioctl(bdev
, mode
, cmd
, arg
);
690 #define nvme_compat_ioctl NULL
693 static int nvme_open(struct block_device
*bdev
, fmode_t mode
)
695 return nvme_get_ns_from_disk(bdev
->bd_disk
) ? 0 : -ENXIO
;
698 static void nvme_release(struct gendisk
*disk
, fmode_t mode
)
700 struct nvme_ns
*ns
= disk
->private_data
;
702 module_put(ns
->ctrl
->ops
->module
);
706 static int nvme_getgeo(struct block_device
*bdev
, struct hd_geometry
*geo
)
708 /* some standard values */
710 geo
->sectors
= 1 << 5;
711 geo
->cylinders
= get_capacity(bdev
->bd_disk
) >> 11;
715 #ifdef CONFIG_BLK_DEV_INTEGRITY
716 static void nvme_init_integrity(struct nvme_ns
*ns
)
718 struct blk_integrity integrity
;
720 switch (ns
->pi_type
) {
721 case NVME_NS_DPS_PI_TYPE3
:
722 integrity
.profile
= &t10_pi_type3_crc
;
724 case NVME_NS_DPS_PI_TYPE1
:
725 case NVME_NS_DPS_PI_TYPE2
:
726 integrity
.profile
= &t10_pi_type1_crc
;
729 integrity
.profile
= NULL
;
732 integrity
.tuple_size
= ns
->ms
;
733 blk_integrity_register(ns
->disk
, &integrity
);
734 blk_queue_max_integrity_segments(ns
->queue
, 1);
737 static void nvme_init_integrity(struct nvme_ns
*ns
)
740 #endif /* CONFIG_BLK_DEV_INTEGRITY */
742 static void nvme_config_discard(struct nvme_ns
*ns
)
744 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
745 u32 logical_block_size
= queue_logical_block_size(ns
->queue
);
747 if (ctrl
->quirks
& NVME_QUIRK_DISCARD_ZEROES
)
748 ns
->queue
->limits
.discard_zeroes_data
= 1;
750 ns
->queue
->limits
.discard_zeroes_data
= 0;
752 ns
->queue
->limits
.discard_alignment
= logical_block_size
;
753 ns
->queue
->limits
.discard_granularity
= logical_block_size
;
754 blk_queue_max_discard_sectors(ns
->queue
, 0xffffffff);
755 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD
, ns
->queue
);
758 static int nvme_revalidate_disk(struct gendisk
*disk
)
760 struct nvme_ns
*ns
= disk
->private_data
;
761 struct nvme_id_ns
*id
;
766 if (test_bit(NVME_NS_DEAD
, &ns
->flags
)) {
767 set_capacity(disk
, 0);
770 if (nvme_identify_ns(ns
->ctrl
, ns
->ns_id
, &id
)) {
771 dev_warn(disk_to_dev(ns
->disk
), "%s: Identify failure\n",
780 if (nvme_nvm_ns_supported(ns
, id
) && ns
->type
!= NVME_NS_LIGHTNVM
) {
781 if (nvme_nvm_register(ns
->queue
, disk
->disk_name
)) {
782 dev_warn(disk_to_dev(ns
->disk
),
783 "%s: LightNVM init failure\n", __func__
);
787 ns
->type
= NVME_NS_LIGHTNVM
;
790 if (ns
->ctrl
->vs
>= NVME_VS(1, 1))
791 memcpy(ns
->eui
, id
->eui64
, sizeof(ns
->eui
));
792 if (ns
->ctrl
->vs
>= NVME_VS(1, 2))
793 memcpy(ns
->uuid
, id
->nguid
, sizeof(ns
->uuid
));
796 lbaf
= id
->flbas
& NVME_NS_FLBAS_LBA_MASK
;
797 ns
->lba_shift
= id
->lbaf
[lbaf
].ds
;
798 ns
->ms
= le16_to_cpu(id
->lbaf
[lbaf
].ms
);
799 ns
->ext
= ns
->ms
&& (id
->flbas
& NVME_NS_FLBAS_META_EXT
);
802 * If identify namespace failed, use default 512 byte block size so
803 * block layer can use before failing read/write for 0 capacity.
805 if (ns
->lba_shift
== 0)
807 bs
= 1 << ns
->lba_shift
;
808 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
809 pi_type
= ns
->ms
== sizeof(struct t10_pi_tuple
) ?
810 id
->dps
& NVME_NS_DPS_PI_MASK
: 0;
812 blk_mq_freeze_queue(disk
->queue
);
813 if (blk_get_integrity(disk
) && (ns
->pi_type
!= pi_type
||
815 bs
!= queue_logical_block_size(disk
->queue
) ||
816 (ns
->ms
&& ns
->ext
)))
817 blk_integrity_unregister(disk
);
819 ns
->pi_type
= pi_type
;
820 blk_queue_logical_block_size(ns
->queue
, bs
);
822 if (ns
->ms
&& !blk_get_integrity(disk
) && !ns
->ext
)
823 nvme_init_integrity(ns
);
824 if (ns
->ms
&& !(ns
->ms
== 8 && ns
->pi_type
) && !blk_get_integrity(disk
))
825 set_capacity(disk
, 0);
827 set_capacity(disk
, le64_to_cpup(&id
->nsze
) << (ns
->lba_shift
- 9));
829 if (ns
->ctrl
->oncs
& NVME_CTRL_ONCS_DSM
)
830 nvme_config_discard(ns
);
831 blk_mq_unfreeze_queue(disk
->queue
);
837 static char nvme_pr_type(enum pr_type type
)
840 case PR_WRITE_EXCLUSIVE
:
842 case PR_EXCLUSIVE_ACCESS
:
844 case PR_WRITE_EXCLUSIVE_REG_ONLY
:
846 case PR_EXCLUSIVE_ACCESS_REG_ONLY
:
848 case PR_WRITE_EXCLUSIVE_ALL_REGS
:
850 case PR_EXCLUSIVE_ACCESS_ALL_REGS
:
857 static int nvme_pr_command(struct block_device
*bdev
, u32 cdw10
,
858 u64 key
, u64 sa_key
, u8 op
)
860 struct nvme_ns
*ns
= bdev
->bd_disk
->private_data
;
861 struct nvme_command c
;
862 u8 data
[16] = { 0, };
864 put_unaligned_le64(key
, &data
[0]);
865 put_unaligned_le64(sa_key
, &data
[8]);
867 memset(&c
, 0, sizeof(c
));
868 c
.common
.opcode
= op
;
869 c
.common
.nsid
= cpu_to_le32(ns
->ns_id
);
870 c
.common
.cdw10
[0] = cpu_to_le32(cdw10
);
872 return nvme_submit_sync_cmd(ns
->queue
, &c
, data
, 16);
875 static int nvme_pr_register(struct block_device
*bdev
, u64 old
,
876 u64
new, unsigned flags
)
880 if (flags
& ~PR_FL_IGNORE_KEY
)
884 cdw10
|= (flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0;
885 cdw10
|= (1 << 30) | (1 << 31); /* PTPL=1 */
886 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_register
);
889 static int nvme_pr_reserve(struct block_device
*bdev
, u64 key
,
890 enum pr_type type
, unsigned flags
)
894 if (flags
& ~PR_FL_IGNORE_KEY
)
897 cdw10
= nvme_pr_type(type
) << 8;
898 cdw10
|= ((flags
& PR_FL_IGNORE_KEY
) ? 1 << 3 : 0);
899 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_acquire
);
902 static int nvme_pr_preempt(struct block_device
*bdev
, u64 old
, u64
new,
903 enum pr_type type
, bool abort
)
905 u32 cdw10
= nvme_pr_type(type
) << 8 | abort
? 2 : 1;
906 return nvme_pr_command(bdev
, cdw10
, old
, new, nvme_cmd_resv_acquire
);
909 static int nvme_pr_clear(struct block_device
*bdev
, u64 key
)
911 u32 cdw10
= 1 | (key
? 1 << 3 : 0);
912 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_register
);
915 static int nvme_pr_release(struct block_device
*bdev
, u64 key
, enum pr_type type
)
917 u32 cdw10
= nvme_pr_type(type
) << 8 | key
? 1 << 3 : 0;
918 return nvme_pr_command(bdev
, cdw10
, key
, 0, nvme_cmd_resv_release
);
921 static const struct pr_ops nvme_pr_ops
= {
922 .pr_register
= nvme_pr_register
,
923 .pr_reserve
= nvme_pr_reserve
,
924 .pr_release
= nvme_pr_release
,
925 .pr_preempt
= nvme_pr_preempt
,
926 .pr_clear
= nvme_pr_clear
,
929 static const struct block_device_operations nvme_fops
= {
930 .owner
= THIS_MODULE
,
932 .compat_ioctl
= nvme_compat_ioctl
,
934 .release
= nvme_release
,
935 .getgeo
= nvme_getgeo
,
936 .revalidate_disk
= nvme_revalidate_disk
,
937 .pr_ops
= &nvme_pr_ops
,
940 static int nvme_wait_ready(struct nvme_ctrl
*ctrl
, u64 cap
, bool enabled
)
942 unsigned long timeout
=
943 ((NVME_CAP_TIMEOUT(cap
) + 1) * HZ
/ 2) + jiffies
;
944 u32 csts
, bit
= enabled
? NVME_CSTS_RDY
: 0;
947 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
948 if ((csts
& NVME_CSTS_RDY
) == bit
)
952 if (fatal_signal_pending(current
))
954 if (time_after(jiffies
, timeout
)) {
955 dev_err(ctrl
->device
,
956 "Device not ready; aborting %s\n", enabled
?
957 "initialisation" : "reset");
966 * If the device has been passed off to us in an enabled state, just clear
967 * the enabled bit. The spec says we should set the 'shutdown notification
968 * bits', but doing so may cause the device to complete commands to the
969 * admin queue ... and we don't know what memory that might be pointing at!
971 int nvme_disable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
975 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
976 ctrl
->ctrl_config
&= ~NVME_CC_ENABLE
;
978 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
981 return nvme_wait_ready(ctrl
, cap
, false);
983 EXPORT_SYMBOL_GPL(nvme_disable_ctrl
);
985 int nvme_enable_ctrl(struct nvme_ctrl
*ctrl
, u64 cap
)
988 * Default to a 4K page size, with the intention to update this
989 * path in the future to accomodate architectures with differing
990 * kernel and IO page sizes.
992 unsigned dev_page_min
= NVME_CAP_MPSMIN(cap
) + 12, page_shift
= 12;
995 if (page_shift
< dev_page_min
) {
996 dev_err(ctrl
->device
,
997 "Minimum device page size %u too large for host (%u)\n",
998 1 << dev_page_min
, 1 << page_shift
);
1002 ctrl
->page_size
= 1 << page_shift
;
1004 ctrl
->ctrl_config
= NVME_CC_CSS_NVM
;
1005 ctrl
->ctrl_config
|= (page_shift
- 12) << NVME_CC_MPS_SHIFT
;
1006 ctrl
->ctrl_config
|= NVME_CC_ARB_RR
| NVME_CC_SHN_NONE
;
1007 ctrl
->ctrl_config
|= NVME_CC_IOSQES
| NVME_CC_IOCQES
;
1008 ctrl
->ctrl_config
|= NVME_CC_ENABLE
;
1010 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1013 return nvme_wait_ready(ctrl
, cap
, true);
1015 EXPORT_SYMBOL_GPL(nvme_enable_ctrl
);
1017 int nvme_shutdown_ctrl(struct nvme_ctrl
*ctrl
)
1019 unsigned long timeout
= SHUTDOWN_TIMEOUT
+ jiffies
;
1023 ctrl
->ctrl_config
&= ~NVME_CC_SHN_MASK
;
1024 ctrl
->ctrl_config
|= NVME_CC_SHN_NORMAL
;
1026 ret
= ctrl
->ops
->reg_write32(ctrl
, NVME_REG_CC
, ctrl
->ctrl_config
);
1030 while ((ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_CSTS
, &csts
)) == 0) {
1031 if ((csts
& NVME_CSTS_SHST_MASK
) == NVME_CSTS_SHST_CMPLT
)
1035 if (fatal_signal_pending(current
))
1037 if (time_after(jiffies
, timeout
)) {
1038 dev_err(ctrl
->device
,
1039 "Device shutdown incomplete; abort shutdown\n");
1046 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl
);
1048 static void nvme_set_queue_limits(struct nvme_ctrl
*ctrl
,
1049 struct request_queue
*q
)
1053 if (ctrl
->max_hw_sectors
) {
1055 (ctrl
->max_hw_sectors
/ (ctrl
->page_size
>> 9)) + 1;
1057 blk_queue_max_hw_sectors(q
, ctrl
->max_hw_sectors
);
1058 blk_queue_max_segments(q
, min_t(u32
, max_segments
, USHRT_MAX
));
1060 if (ctrl
->stripe_size
)
1061 blk_queue_chunk_sectors(q
, ctrl
->stripe_size
>> 9);
1062 blk_queue_virt_boundary(q
, ctrl
->page_size
- 1);
1063 if (ctrl
->vwc
& NVME_CTRL_VWC_PRESENT
)
1065 blk_queue_write_cache(q
, vwc
, vwc
);
1069 * Initialize the cached copies of the Identify data and various controller
1070 * register in our nvme_ctrl structure. This should be called as soon as
1071 * the admin queue is fully up and running.
1073 int nvme_init_identify(struct nvme_ctrl
*ctrl
)
1075 struct nvme_id_ctrl
*id
;
1077 int ret
, page_shift
;
1079 ret
= ctrl
->ops
->reg_read32(ctrl
, NVME_REG_VS
, &ctrl
->vs
);
1081 dev_err(ctrl
->device
, "Reading VS failed (%d)\n", ret
);
1085 ret
= ctrl
->ops
->reg_read64(ctrl
, NVME_REG_CAP
, &cap
);
1087 dev_err(ctrl
->device
, "Reading CAP failed (%d)\n", ret
);
1090 page_shift
= NVME_CAP_MPSMIN(cap
) + 12;
1092 if (ctrl
->vs
>= NVME_VS(1, 1))
1093 ctrl
->subsystem
= NVME_CAP_NSSRC(cap
);
1095 ret
= nvme_identify_ctrl(ctrl
, &id
);
1097 dev_err(ctrl
->device
, "Identify Controller failed (%d)\n", ret
);
1101 ctrl
->vid
= le16_to_cpu(id
->vid
);
1102 ctrl
->oncs
= le16_to_cpup(&id
->oncs
);
1103 atomic_set(&ctrl
->abort_limit
, id
->acl
+ 1);
1104 ctrl
->vwc
= id
->vwc
;
1105 ctrl
->cntlid
= le16_to_cpup(&id
->cntlid
);
1106 memcpy(ctrl
->serial
, id
->sn
, sizeof(id
->sn
));
1107 memcpy(ctrl
->model
, id
->mn
, sizeof(id
->mn
));
1108 memcpy(ctrl
->firmware_rev
, id
->fr
, sizeof(id
->fr
));
1110 ctrl
->max_hw_sectors
= 1 << (id
->mdts
+ page_shift
- 9);
1112 ctrl
->max_hw_sectors
= UINT_MAX
;
1114 if ((ctrl
->quirks
& NVME_QUIRK_STRIPE_SIZE
) && id
->vs
[3]) {
1115 unsigned int max_hw_sectors
;
1117 ctrl
->stripe_size
= 1 << (id
->vs
[3] + page_shift
);
1118 max_hw_sectors
= ctrl
->stripe_size
>> (page_shift
- 9);
1119 if (ctrl
->max_hw_sectors
) {
1120 ctrl
->max_hw_sectors
= min(max_hw_sectors
,
1121 ctrl
->max_hw_sectors
);
1123 ctrl
->max_hw_sectors
= max_hw_sectors
;
1127 nvme_set_queue_limits(ctrl
, ctrl
->admin_q
);
1132 EXPORT_SYMBOL_GPL(nvme_init_identify
);
1134 static int nvme_dev_open(struct inode
*inode
, struct file
*file
)
1136 struct nvme_ctrl
*ctrl
;
1137 int instance
= iminor(inode
);
1140 spin_lock(&dev_list_lock
);
1141 list_for_each_entry(ctrl
, &nvme_ctrl_list
, node
) {
1142 if (ctrl
->instance
!= instance
)
1145 if (!ctrl
->admin_q
) {
1149 if (!kref_get_unless_zero(&ctrl
->kref
))
1151 file
->private_data
= ctrl
;
1155 spin_unlock(&dev_list_lock
);
1160 static int nvme_dev_release(struct inode
*inode
, struct file
*file
)
1162 nvme_put_ctrl(file
->private_data
);
1166 static int nvme_dev_user_cmd(struct nvme_ctrl
*ctrl
, void __user
*argp
)
1171 mutex_lock(&ctrl
->namespaces_mutex
);
1172 if (list_empty(&ctrl
->namespaces
)) {
1177 ns
= list_first_entry(&ctrl
->namespaces
, struct nvme_ns
, list
);
1178 if (ns
!= list_last_entry(&ctrl
->namespaces
, struct nvme_ns
, list
)) {
1179 dev_warn(ctrl
->device
,
1180 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1185 dev_warn(ctrl
->device
,
1186 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1187 kref_get(&ns
->kref
);
1188 mutex_unlock(&ctrl
->namespaces_mutex
);
1190 ret
= nvme_user_cmd(ctrl
, ns
, argp
);
1195 mutex_unlock(&ctrl
->namespaces_mutex
);
1199 static long nvme_dev_ioctl(struct file
*file
, unsigned int cmd
,
1202 struct nvme_ctrl
*ctrl
= file
->private_data
;
1203 void __user
*argp
= (void __user
*)arg
;
1206 case NVME_IOCTL_ADMIN_CMD
:
1207 return nvme_user_cmd(ctrl
, NULL
, argp
);
1208 case NVME_IOCTL_IO_CMD
:
1209 return nvme_dev_user_cmd(ctrl
, argp
);
1210 case NVME_IOCTL_RESET
:
1211 dev_warn(ctrl
->device
, "resetting controller\n");
1212 return ctrl
->ops
->reset_ctrl(ctrl
);
1213 case NVME_IOCTL_SUBSYS_RESET
:
1214 return nvme_reset_subsystem(ctrl
);
1220 static const struct file_operations nvme_dev_fops
= {
1221 .owner
= THIS_MODULE
,
1222 .open
= nvme_dev_open
,
1223 .release
= nvme_dev_release
,
1224 .unlocked_ioctl
= nvme_dev_ioctl
,
1225 .compat_ioctl
= nvme_dev_ioctl
,
1228 static ssize_t
nvme_sysfs_reset(struct device
*dev
,
1229 struct device_attribute
*attr
, const char *buf
,
1232 struct nvme_ctrl
*ctrl
= dev_get_drvdata(dev
);
1235 ret
= ctrl
->ops
->reset_ctrl(ctrl
);
1240 static DEVICE_ATTR(reset_controller
, S_IWUSR
, NULL
, nvme_sysfs_reset
);
1242 static ssize_t
wwid_show(struct device
*dev
, struct device_attribute
*attr
,
1245 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1246 struct nvme_ctrl
*ctrl
= ns
->ctrl
;
1247 int serial_len
= sizeof(ctrl
->serial
);
1248 int model_len
= sizeof(ctrl
->model
);
1250 if (memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1251 return sprintf(buf
, "eui.%16phN\n", ns
->uuid
);
1253 if (memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1254 return sprintf(buf
, "eui.%8phN\n", ns
->eui
);
1256 while (ctrl
->serial
[serial_len
- 1] == ' ')
1258 while (ctrl
->model
[model_len
- 1] == ' ')
1261 return sprintf(buf
, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl
->vid
,
1262 serial_len
, ctrl
->serial
, model_len
, ctrl
->model
, ns
->ns_id
);
1264 static DEVICE_ATTR(wwid
, S_IRUGO
, wwid_show
, NULL
);
1266 static ssize_t
uuid_show(struct device
*dev
, struct device_attribute
*attr
,
1269 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1270 return sprintf(buf
, "%pU\n", ns
->uuid
);
1272 static DEVICE_ATTR(uuid
, S_IRUGO
, uuid_show
, NULL
);
1274 static ssize_t
eui_show(struct device
*dev
, struct device_attribute
*attr
,
1277 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1278 return sprintf(buf
, "%8phd\n", ns
->eui
);
1280 static DEVICE_ATTR(eui
, S_IRUGO
, eui_show
, NULL
);
1282 static ssize_t
nsid_show(struct device
*dev
, struct device_attribute
*attr
,
1285 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1286 return sprintf(buf
, "%d\n", ns
->ns_id
);
1288 static DEVICE_ATTR(nsid
, S_IRUGO
, nsid_show
, NULL
);
1290 static struct attribute
*nvme_ns_attrs
[] = {
1291 &dev_attr_wwid
.attr
,
1292 &dev_attr_uuid
.attr
,
1294 &dev_attr_nsid
.attr
,
1298 static umode_t
nvme_attrs_are_visible(struct kobject
*kobj
,
1299 struct attribute
*a
, int n
)
1301 struct device
*dev
= container_of(kobj
, struct device
, kobj
);
1302 struct nvme_ns
*ns
= dev_to_disk(dev
)->private_data
;
1304 if (a
== &dev_attr_uuid
.attr
) {
1305 if (!memchr_inv(ns
->uuid
, 0, sizeof(ns
->uuid
)))
1308 if (a
== &dev_attr_eui
.attr
) {
1309 if (!memchr_inv(ns
->eui
, 0, sizeof(ns
->eui
)))
1315 static const struct attribute_group nvme_ns_attr_group
= {
1316 .attrs
= nvme_ns_attrs
,
1317 .is_visible
= nvme_attrs_are_visible
,
1320 #define nvme_show_str_function(field) \
1321 static ssize_t field##_show(struct device *dev, \
1322 struct device_attribute *attr, char *buf) \
1324 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1325 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1327 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1329 #define nvme_show_int_function(field) \
1330 static ssize_t field##_show(struct device *dev, \
1331 struct device_attribute *attr, char *buf) \
1333 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1334 return sprintf(buf, "%d\n", ctrl->field); \
1336 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1338 nvme_show_str_function(model
);
1339 nvme_show_str_function(serial
);
1340 nvme_show_str_function(firmware_rev
);
1341 nvme_show_int_function(cntlid
);
1343 static struct attribute
*nvme_dev_attrs
[] = {
1344 &dev_attr_reset_controller
.attr
,
1345 &dev_attr_model
.attr
,
1346 &dev_attr_serial
.attr
,
1347 &dev_attr_firmware_rev
.attr
,
1348 &dev_attr_cntlid
.attr
,
1352 static struct attribute_group nvme_dev_attrs_group
= {
1353 .attrs
= nvme_dev_attrs
,
1356 static const struct attribute_group
*nvme_dev_attr_groups
[] = {
1357 &nvme_dev_attrs_group
,
1361 static int ns_cmp(void *priv
, struct list_head
*a
, struct list_head
*b
)
1363 struct nvme_ns
*nsa
= container_of(a
, struct nvme_ns
, list
);
1364 struct nvme_ns
*nsb
= container_of(b
, struct nvme_ns
, list
);
1366 return nsa
->ns_id
- nsb
->ns_id
;
1369 static struct nvme_ns
*nvme_find_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1373 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1375 list_for_each_entry(ns
, &ctrl
->namespaces
, list
) {
1376 if (ns
->ns_id
== nsid
)
1378 if (ns
->ns_id
> nsid
)
1384 static void nvme_alloc_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1387 struct gendisk
*disk
;
1388 int node
= dev_to_node(ctrl
->dev
);
1390 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1392 ns
= kzalloc_node(sizeof(*ns
), GFP_KERNEL
, node
);
1396 ns
->instance
= ida_simple_get(&ctrl
->ns_ida
, 1, 0, GFP_KERNEL
);
1397 if (ns
->instance
< 0)
1400 ns
->queue
= blk_mq_init_queue(ctrl
->tagset
);
1401 if (IS_ERR(ns
->queue
))
1402 goto out_release_instance
;
1403 queue_flag_set_unlocked(QUEUE_FLAG_NONROT
, ns
->queue
);
1404 ns
->queue
->queuedata
= ns
;
1407 disk
= alloc_disk_node(0, node
);
1409 goto out_free_queue
;
1411 kref_init(&ns
->kref
);
1414 ns
->lba_shift
= 9; /* set to a default value for 512 until disk is validated */
1417 blk_queue_logical_block_size(ns
->queue
, 1 << ns
->lba_shift
);
1418 nvme_set_queue_limits(ctrl
, ns
->queue
);
1420 disk
->major
= nvme_major
;
1421 disk
->first_minor
= 0;
1422 disk
->fops
= &nvme_fops
;
1423 disk
->private_data
= ns
;
1424 disk
->queue
= ns
->queue
;
1425 disk
->driverfs_dev
= ctrl
->device
;
1426 disk
->flags
= GENHD_FL_EXT_DEVT
;
1427 sprintf(disk
->disk_name
, "nvme%dn%d", ctrl
->instance
, ns
->instance
);
1429 if (nvme_revalidate_disk(ns
->disk
))
1432 list_add_tail_rcu(&ns
->list
, &ctrl
->namespaces
);
1433 kref_get(&ctrl
->kref
);
1434 if (ns
->type
== NVME_NS_LIGHTNVM
)
1438 if (sysfs_create_group(&disk_to_dev(ns
->disk
)->kobj
,
1439 &nvme_ns_attr_group
))
1440 pr_warn("%s: failed to create sysfs group for identification\n",
1441 ns
->disk
->disk_name
);
1446 blk_cleanup_queue(ns
->queue
);
1447 out_release_instance
:
1448 ida_simple_remove(&ctrl
->ns_ida
, ns
->instance
);
1453 static void nvme_ns_remove(struct nvme_ns
*ns
)
1455 lockdep_assert_held(&ns
->ctrl
->namespaces_mutex
);
1457 if (test_and_set_bit(NVME_NS_REMOVING
, &ns
->flags
))
1460 if (ns
->disk
->flags
& GENHD_FL_UP
) {
1461 if (blk_get_integrity(ns
->disk
))
1462 blk_integrity_unregister(ns
->disk
);
1463 sysfs_remove_group(&disk_to_dev(ns
->disk
)->kobj
,
1464 &nvme_ns_attr_group
);
1465 del_gendisk(ns
->disk
);
1466 blk_mq_abort_requeue_list(ns
->queue
);
1467 blk_cleanup_queue(ns
->queue
);
1469 list_del_init(&ns
->list
);
1474 static void nvme_validate_ns(struct nvme_ctrl
*ctrl
, unsigned nsid
)
1478 ns
= nvme_find_ns(ctrl
, nsid
);
1480 if (revalidate_disk(ns
->disk
))
1483 nvme_alloc_ns(ctrl
, nsid
);
1486 static int nvme_scan_ns_list(struct nvme_ctrl
*ctrl
, unsigned nn
)
1490 unsigned i
, j
, nsid
, prev
= 0, num_lists
= DIV_ROUND_UP(nn
, 1024);
1493 ns_list
= kzalloc(0x1000, GFP_KERNEL
);
1497 for (i
= 0; i
< num_lists
; i
++) {
1498 ret
= nvme_identify_ns_list(ctrl
, prev
, ns_list
);
1502 for (j
= 0; j
< min(nn
, 1024U); j
++) {
1503 nsid
= le32_to_cpu(ns_list
[j
]);
1507 nvme_validate_ns(ctrl
, nsid
);
1509 while (++prev
< nsid
) {
1510 ns
= nvme_find_ns(ctrl
, prev
);
1522 static void nvme_scan_ns_sequential(struct nvme_ctrl
*ctrl
, unsigned nn
)
1524 struct nvme_ns
*ns
, *next
;
1527 lockdep_assert_held(&ctrl
->namespaces_mutex
);
1529 for (i
= 1; i
<= nn
; i
++)
1530 nvme_validate_ns(ctrl
, i
);
1532 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
) {
1538 static void nvme_scan_work(struct work_struct
*work
)
1540 struct nvme_ctrl
*ctrl
=
1541 container_of(work
, struct nvme_ctrl
, scan_work
);
1542 struct nvme_id_ctrl
*id
;
1545 if (ctrl
->state
!= NVME_CTRL_LIVE
)
1548 if (nvme_identify_ctrl(ctrl
, &id
))
1551 mutex_lock(&ctrl
->namespaces_mutex
);
1552 nn
= le32_to_cpu(id
->nn
);
1553 if (ctrl
->vs
>= NVME_VS(1, 1) &&
1554 !(ctrl
->quirks
& NVME_QUIRK_IDENTIFY_CNS
)) {
1555 if (!nvme_scan_ns_list(ctrl
, nn
))
1558 nvme_scan_ns_sequential(ctrl
, nn
);
1560 list_sort(NULL
, &ctrl
->namespaces
, ns_cmp
);
1561 mutex_unlock(&ctrl
->namespaces_mutex
);
1564 if (ctrl
->ops
->post_scan
)
1565 ctrl
->ops
->post_scan(ctrl
);
1568 void nvme_queue_scan(struct nvme_ctrl
*ctrl
)
1571 * Do not queue new scan work when a controller is reset during
1574 if (ctrl
->state
== NVME_CTRL_LIVE
)
1575 schedule_work(&ctrl
->scan_work
);
1577 EXPORT_SYMBOL_GPL(nvme_queue_scan
);
1579 void nvme_remove_namespaces(struct nvme_ctrl
*ctrl
)
1581 struct nvme_ns
*ns
, *next
;
1583 mutex_lock(&ctrl
->namespaces_mutex
);
1584 list_for_each_entry_safe(ns
, next
, &ctrl
->namespaces
, list
)
1586 mutex_unlock(&ctrl
->namespaces_mutex
);
1588 EXPORT_SYMBOL_GPL(nvme_remove_namespaces
);
1590 static void nvme_async_event_work(struct work_struct
*work
)
1592 struct nvme_ctrl
*ctrl
=
1593 container_of(work
, struct nvme_ctrl
, async_event_work
);
1595 spin_lock_irq(&ctrl
->lock
);
1596 while (ctrl
->event_limit
> 0) {
1597 int aer_idx
= --ctrl
->event_limit
;
1599 spin_unlock_irq(&ctrl
->lock
);
1600 ctrl
->ops
->submit_async_event(ctrl
, aer_idx
);
1601 spin_lock_irq(&ctrl
->lock
);
1603 spin_unlock_irq(&ctrl
->lock
);
1606 void nvme_complete_async_event(struct nvme_ctrl
*ctrl
,
1607 struct nvme_completion
*cqe
)
1609 u16 status
= le16_to_cpu(cqe
->status
) >> 1;
1610 u32 result
= le32_to_cpu(cqe
->result
);
1612 if (status
== NVME_SC_SUCCESS
|| status
== NVME_SC_ABORT_REQ
) {
1613 ++ctrl
->event_limit
;
1614 schedule_work(&ctrl
->async_event_work
);
1617 if (status
!= NVME_SC_SUCCESS
)
1620 switch (result
& 0xff07) {
1621 case NVME_AER_NOTICE_NS_CHANGED
:
1622 dev_info(ctrl
->device
, "rescanning\n");
1623 nvme_queue_scan(ctrl
);
1626 dev_warn(ctrl
->device
, "async event result %08x\n", result
);
1629 EXPORT_SYMBOL_GPL(nvme_complete_async_event
);
1631 void nvme_queue_async_events(struct nvme_ctrl
*ctrl
)
1633 ctrl
->event_limit
= NVME_NR_AERS
;
1634 schedule_work(&ctrl
->async_event_work
);
1636 EXPORT_SYMBOL_GPL(nvme_queue_async_events
);
1638 static DEFINE_IDA(nvme_instance_ida
);
1640 static int nvme_set_instance(struct nvme_ctrl
*ctrl
)
1642 int instance
, error
;
1645 if (!ida_pre_get(&nvme_instance_ida
, GFP_KERNEL
))
1648 spin_lock(&dev_list_lock
);
1649 error
= ida_get_new(&nvme_instance_ida
, &instance
);
1650 spin_unlock(&dev_list_lock
);
1651 } while (error
== -EAGAIN
);
1656 ctrl
->instance
= instance
;
1660 static void nvme_release_instance(struct nvme_ctrl
*ctrl
)
1662 spin_lock(&dev_list_lock
);
1663 ida_remove(&nvme_instance_ida
, ctrl
->instance
);
1664 spin_unlock(&dev_list_lock
);
1667 void nvme_uninit_ctrl(struct nvme_ctrl
*ctrl
)
1669 flush_work(&ctrl
->async_event_work
);
1670 flush_work(&ctrl
->scan_work
);
1671 nvme_remove_namespaces(ctrl
);
1673 device_destroy(nvme_class
, MKDEV(nvme_char_major
, ctrl
->instance
));
1675 spin_lock(&dev_list_lock
);
1676 list_del(&ctrl
->node
);
1677 spin_unlock(&dev_list_lock
);
1679 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl
);
1681 static void nvme_free_ctrl(struct kref
*kref
)
1683 struct nvme_ctrl
*ctrl
= container_of(kref
, struct nvme_ctrl
, kref
);
1685 put_device(ctrl
->device
);
1686 nvme_release_instance(ctrl
);
1687 ida_destroy(&ctrl
->ns_ida
);
1689 ctrl
->ops
->free_ctrl(ctrl
);
1692 void nvme_put_ctrl(struct nvme_ctrl
*ctrl
)
1694 kref_put(&ctrl
->kref
, nvme_free_ctrl
);
1696 EXPORT_SYMBOL_GPL(nvme_put_ctrl
);
1699 * Initialize a NVMe controller structures. This needs to be called during
1700 * earliest initialization so that we have the initialized structured around
1703 int nvme_init_ctrl(struct nvme_ctrl
*ctrl
, struct device
*dev
,
1704 const struct nvme_ctrl_ops
*ops
, unsigned long quirks
)
1708 ctrl
->state
= NVME_CTRL_NEW
;
1709 spin_lock_init(&ctrl
->lock
);
1710 INIT_LIST_HEAD(&ctrl
->namespaces
);
1711 mutex_init(&ctrl
->namespaces_mutex
);
1712 kref_init(&ctrl
->kref
);
1715 ctrl
->quirks
= quirks
;
1716 INIT_WORK(&ctrl
->scan_work
, nvme_scan_work
);
1717 INIT_WORK(&ctrl
->async_event_work
, nvme_async_event_work
);
1719 ret
= nvme_set_instance(ctrl
);
1723 ctrl
->device
= device_create_with_groups(nvme_class
, ctrl
->dev
,
1724 MKDEV(nvme_char_major
, ctrl
->instance
),
1725 ctrl
, nvme_dev_attr_groups
,
1726 "nvme%d", ctrl
->instance
);
1727 if (IS_ERR(ctrl
->device
)) {
1728 ret
= PTR_ERR(ctrl
->device
);
1729 goto out_release_instance
;
1731 get_device(ctrl
->device
);
1732 ida_init(&ctrl
->ns_ida
);
1734 spin_lock(&dev_list_lock
);
1735 list_add_tail(&ctrl
->node
, &nvme_ctrl_list
);
1736 spin_unlock(&dev_list_lock
);
1739 out_release_instance
:
1740 nvme_release_instance(ctrl
);
1744 EXPORT_SYMBOL_GPL(nvme_init_ctrl
);
1747 * nvme_kill_queues(): Ends all namespace queues
1748 * @ctrl: the dead controller that needs to end
1750 * Call this function when the driver determines it is unable to get the
1751 * controller in a state capable of servicing IO.
1753 void nvme_kill_queues(struct nvme_ctrl
*ctrl
)
1758 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1759 if (!kref_get_unless_zero(&ns
->kref
))
1763 * Revalidating a dead namespace sets capacity to 0. This will
1764 * end buffered writers dirtying pages that can't be synced.
1766 if (!test_and_set_bit(NVME_NS_DEAD
, &ns
->flags
))
1767 revalidate_disk(ns
->disk
);
1769 blk_set_queue_dying(ns
->queue
);
1770 blk_mq_abort_requeue_list(ns
->queue
);
1771 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
1777 EXPORT_SYMBOL_GPL(nvme_kill_queues
);
1779 void nvme_stop_queues(struct nvme_ctrl
*ctrl
)
1784 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1785 spin_lock_irq(ns
->queue
->queue_lock
);
1786 queue_flag_set(QUEUE_FLAG_STOPPED
, ns
->queue
);
1787 spin_unlock_irq(ns
->queue
->queue_lock
);
1789 blk_mq_cancel_requeue_work(ns
->queue
);
1790 blk_mq_stop_hw_queues(ns
->queue
);
1794 EXPORT_SYMBOL_GPL(nvme_stop_queues
);
1796 void nvme_start_queues(struct nvme_ctrl
*ctrl
)
1801 list_for_each_entry_rcu(ns
, &ctrl
->namespaces
, list
) {
1802 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED
, ns
->queue
);
1803 blk_mq_start_stopped_hw_queues(ns
->queue
, true);
1804 blk_mq_kick_requeue_list(ns
->queue
);
1808 EXPORT_SYMBOL_GPL(nvme_start_queues
);
1810 int __init
nvme_core_init(void)
1814 result
= register_blkdev(nvme_major
, "nvme");
1817 else if (result
> 0)
1818 nvme_major
= result
;
1820 result
= __register_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme",
1823 goto unregister_blkdev
;
1824 else if (result
> 0)
1825 nvme_char_major
= result
;
1827 nvme_class
= class_create(THIS_MODULE
, "nvme");
1828 if (IS_ERR(nvme_class
)) {
1829 result
= PTR_ERR(nvme_class
);
1830 goto unregister_chrdev
;
1836 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
1838 unregister_blkdev(nvme_major
, "nvme");
1842 void nvme_core_exit(void)
1844 class_destroy(nvme_class
);
1845 __unregister_chrdev(nvme_char_major
, 0, NVME_MINORS
, "nvme");
1846 unregister_blkdev(nvme_major
, "nvme");
1849 MODULE_LICENSE("GPL");
1850 MODULE_VERSION("1.0");
1851 module_init(nvme_core_init
);
1852 module_exit(nvme_core_exit
);