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)
204 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
206 EXPORT_SYMBOL_GPL(nvme_requeue_req);
208 struct request *nvme_alloc_request(struct request_queue *q,
209 struct nvme_command *cmd, unsigned int flags, int qid)
211 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
214 if (qid == NVME_QID_ANY) {
215 req = blk_mq_alloc_request(q, op, flags);
217 req = blk_mq_alloc_request_hctx(q, op, flags,
223 req->cmd_flags |= REQ_FAILFAST_DRIVER;
224 nvme_req(req)->cmd = cmd;
228 EXPORT_SYMBOL_GPL(nvme_alloc_request);
230 static inline void nvme_setup_flush(struct nvme_ns *ns,
231 struct nvme_command *cmnd)
233 memset(cmnd, 0, sizeof(*cmnd));
234 cmnd->common.opcode = nvme_cmd_flush;
235 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
238 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
239 struct nvme_command *cmnd)
241 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
242 struct nvme_dsm_range *range;
245 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
247 return BLK_MQ_RQ_QUEUE_BUSY;
249 __rq_for_each_bio(bio, req) {
250 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
251 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
253 range[n].cattr = cpu_to_le32(0);
254 range[n].nlb = cpu_to_le32(nlb);
255 range[n].slba = cpu_to_le64(slba);
259 if (WARN_ON_ONCE(n != segments)) {
261 return BLK_MQ_RQ_QUEUE_ERROR;
264 memset(cmnd, 0, sizeof(*cmnd));
265 cmnd->dsm.opcode = nvme_cmd_dsm;
266 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
267 cmnd->dsm.nr = segments - 1;
268 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
270 req->special_vec.bv_page = virt_to_page(range);
271 req->special_vec.bv_offset = offset_in_page(range);
272 req->special_vec.bv_len = sizeof(*range) * segments;
273 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
275 return BLK_MQ_RQ_QUEUE_OK;
278 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
279 struct nvme_command *cmnd)
284 if (req->cmd_flags & REQ_FUA)
285 control |= NVME_RW_FUA;
286 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
287 control |= NVME_RW_LR;
289 if (req->cmd_flags & REQ_RAHEAD)
290 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
292 memset(cmnd, 0, sizeof(*cmnd));
293 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
294 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
295 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
296 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
299 switch (ns->pi_type) {
300 case NVME_NS_DPS_PI_TYPE3:
301 control |= NVME_RW_PRINFO_PRCHK_GUARD;
303 case NVME_NS_DPS_PI_TYPE1:
304 case NVME_NS_DPS_PI_TYPE2:
305 control |= NVME_RW_PRINFO_PRCHK_GUARD |
306 NVME_RW_PRINFO_PRCHK_REF;
307 cmnd->rw.reftag = cpu_to_le32(
308 nvme_block_nr(ns, blk_rq_pos(req)));
311 if (!blk_integrity_rq(req))
312 control |= NVME_RW_PRINFO_PRACT;
315 cmnd->rw.control = cpu_to_le16(control);
316 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
319 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
320 struct nvme_command *cmd)
322 int ret = BLK_MQ_RQ_QUEUE_OK;
324 switch (req_op(req)) {
327 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
330 nvme_setup_flush(ns, cmd);
333 ret = nvme_setup_discard(ns, req, cmd);
337 nvme_setup_rw(ns, req, cmd);
341 return BLK_MQ_RQ_QUEUE_ERROR;
344 cmd->common.command_id = req->tag;
347 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
350 * Returns 0 on success. If the result is negative, it's a Linux error code;
351 * if the result is positive, it's an NVM Express status code
353 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
354 union nvme_result *result, void *buffer, unsigned bufflen,
355 unsigned timeout, int qid, int at_head, int flags)
360 req = nvme_alloc_request(q, cmd, flags, qid);
364 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
366 if (buffer && bufflen) {
367 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
372 blk_execute_rq(req->q, NULL, req, at_head);
374 *result = nvme_req(req)->result;
377 blk_mq_free_request(req);
380 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
382 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
383 void *buffer, unsigned bufflen)
385 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
388 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
390 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
391 void __user *ubuffer, unsigned bufflen,
392 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
393 u32 *result, unsigned timeout)
395 bool write = nvme_is_write(cmd);
396 struct nvme_ns *ns = q->queuedata;
397 struct gendisk *disk = ns ? ns->disk : NULL;
399 struct bio *bio = NULL;
403 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
407 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
409 if (ubuffer && bufflen) {
410 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
418 bio->bi_bdev = bdget_disk(disk, 0);
424 if (meta_buffer && meta_len) {
425 struct bio_integrity_payload *bip;
427 meta = kmalloc(meta_len, GFP_KERNEL);
434 if (copy_from_user(meta, meta_buffer,
441 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
447 bip->bip_iter.bi_size = meta_len;
448 bip->bip_iter.bi_sector = meta_seed;
450 ret = bio_integrity_add_page(bio, virt_to_page(meta),
451 meta_len, offset_in_page(meta));
452 if (ret != meta_len) {
459 blk_execute_rq(req->q, disk, req, 0);
462 *result = le32_to_cpu(nvme_req(req)->result.u32);
463 if (meta && !ret && !write) {
464 if (copy_to_user(meta_buffer, meta, meta_len))
471 if (disk && bio->bi_bdev)
473 blk_rq_unmap_user(bio);
476 blk_mq_free_request(req);
480 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
481 void __user *ubuffer, unsigned bufflen, u32 *result,
484 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
488 static void nvme_keep_alive_end_io(struct request *rq, int error)
490 struct nvme_ctrl *ctrl = rq->end_io_data;
492 blk_mq_free_request(rq);
495 dev_err(ctrl->device,
496 "failed nvme_keep_alive_end_io error=%d\n", error);
500 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
503 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
505 struct nvme_command c;
508 memset(&c, 0, sizeof(c));
509 c.common.opcode = nvme_admin_keep_alive;
511 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
516 rq->timeout = ctrl->kato * HZ;
517 rq->end_io_data = ctrl;
519 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
524 static void nvme_keep_alive_work(struct work_struct *work)
526 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
527 struct nvme_ctrl, ka_work);
529 if (nvme_keep_alive(ctrl)) {
530 /* allocation failure, reset the controller */
531 dev_err(ctrl->device, "keep-alive failed\n");
532 ctrl->ops->reset_ctrl(ctrl);
537 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
539 if (unlikely(ctrl->kato == 0))
542 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
543 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
545 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
547 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
549 if (unlikely(ctrl->kato == 0))
552 cancel_delayed_work_sync(&ctrl->ka_work);
554 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
556 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
558 struct nvme_command c = { };
561 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
562 c.identify.opcode = nvme_admin_identify;
563 c.identify.cns = cpu_to_le32(NVME_ID_CNS_CTRL);
565 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
569 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
570 sizeof(struct nvme_id_ctrl));
576 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
578 struct nvme_command c = { };
580 c.identify.opcode = nvme_admin_identify;
581 c.identify.cns = cpu_to_le32(NVME_ID_CNS_NS_ACTIVE_LIST);
582 c.identify.nsid = cpu_to_le32(nsid);
583 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
586 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
587 struct nvme_id_ns **id)
589 struct nvme_command c = { };
592 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
593 c.identify.opcode = nvme_admin_identify;
594 c.identify.nsid = cpu_to_le32(nsid);
596 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
600 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
601 sizeof(struct nvme_id_ns));
607 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
608 void *buffer, size_t buflen, u32 *result)
610 struct nvme_command c;
611 union nvme_result res;
614 memset(&c, 0, sizeof(c));
615 c.features.opcode = nvme_admin_get_features;
616 c.features.nsid = cpu_to_le32(nsid);
617 c.features.fid = cpu_to_le32(fid);
619 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
621 if (ret >= 0 && result)
622 *result = le32_to_cpu(res.u32);
626 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
627 void *buffer, size_t buflen, u32 *result)
629 struct nvme_command c;
630 union nvme_result res;
633 memset(&c, 0, sizeof(c));
634 c.features.opcode = nvme_admin_set_features;
635 c.features.fid = cpu_to_le32(fid);
636 c.features.dword11 = cpu_to_le32(dword11);
638 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
639 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
640 if (ret >= 0 && result)
641 *result = le32_to_cpu(res.u32);
645 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
647 struct nvme_command c = { };
650 c.common.opcode = nvme_admin_get_log_page,
651 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
652 c.common.cdw10[0] = cpu_to_le32(
653 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
656 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
660 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
661 sizeof(struct nvme_smart_log));
667 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
669 u32 q_count = (*count - 1) | ((*count - 1) << 16);
671 int status, nr_io_queues;
673 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
679 * Degraded controllers might return an error when setting the queue
680 * count. We still want to be able to bring them online and offer
681 * access to the admin queue, as that might be only way to fix them up.
684 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
687 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
688 *count = min(*count, nr_io_queues);
693 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
695 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
697 struct nvme_user_io io;
698 struct nvme_command c;
699 unsigned length, meta_len;
700 void __user *metadata;
702 if (copy_from_user(&io, uio, sizeof(io)))
710 case nvme_cmd_compare:
716 length = (io.nblocks + 1) << ns->lba_shift;
717 meta_len = (io.nblocks + 1) * ns->ms;
718 metadata = (void __user *)(uintptr_t)io.metadata;
723 } else if (meta_len) {
724 if ((io.metadata & 3) || !io.metadata)
728 memset(&c, 0, sizeof(c));
729 c.rw.opcode = io.opcode;
730 c.rw.flags = io.flags;
731 c.rw.nsid = cpu_to_le32(ns->ns_id);
732 c.rw.slba = cpu_to_le64(io.slba);
733 c.rw.length = cpu_to_le16(io.nblocks);
734 c.rw.control = cpu_to_le16(io.control);
735 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
736 c.rw.reftag = cpu_to_le32(io.reftag);
737 c.rw.apptag = cpu_to_le16(io.apptag);
738 c.rw.appmask = cpu_to_le16(io.appmask);
740 return __nvme_submit_user_cmd(ns->queue, &c,
741 (void __user *)(uintptr_t)io.addr, length,
742 metadata, meta_len, io.slba, NULL, 0);
745 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
746 struct nvme_passthru_cmd __user *ucmd)
748 struct nvme_passthru_cmd cmd;
749 struct nvme_command c;
750 unsigned timeout = 0;
753 if (!capable(CAP_SYS_ADMIN))
755 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
760 memset(&c, 0, sizeof(c));
761 c.common.opcode = cmd.opcode;
762 c.common.flags = cmd.flags;
763 c.common.nsid = cpu_to_le32(cmd.nsid);
764 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
765 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
766 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
767 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
768 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
769 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
770 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
771 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
774 timeout = msecs_to_jiffies(cmd.timeout_ms);
776 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
777 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
778 &cmd.result, timeout);
780 if (put_user(cmd.result, &ucmd->result))
787 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
788 unsigned int cmd, unsigned long arg)
790 struct nvme_ns *ns = bdev->bd_disk->private_data;
794 force_successful_syscall_return();
796 case NVME_IOCTL_ADMIN_CMD:
797 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
798 case NVME_IOCTL_IO_CMD:
799 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
800 case NVME_IOCTL_SUBMIT_IO:
801 return nvme_submit_io(ns, (void __user *)arg);
802 #ifdef CONFIG_BLK_DEV_NVME_SCSI
803 case SG_GET_VERSION_NUM:
804 return nvme_sg_get_version_num((void __user *)arg);
806 return nvme_sg_io(ns, (void __user *)arg);
811 return nvme_nvm_ioctl(ns, cmd, arg);
813 if (is_sed_ioctl(cmd))
814 return sed_ioctl(ns->ctrl->opal_dev, cmd,
815 (void __user *) arg);
821 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
822 unsigned int cmd, unsigned long arg)
828 return nvme_ioctl(bdev, mode, cmd, arg);
831 #define nvme_compat_ioctl NULL
834 static int nvme_open(struct block_device *bdev, fmode_t mode)
836 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
839 static void nvme_release(struct gendisk *disk, fmode_t mode)
841 struct nvme_ns *ns = disk->private_data;
843 module_put(ns->ctrl->ops->module);
847 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
849 /* some standard values */
851 geo->sectors = 1 << 5;
852 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
856 #ifdef CONFIG_BLK_DEV_INTEGRITY
857 static void nvme_init_integrity(struct nvme_ns *ns)
859 struct blk_integrity integrity;
861 memset(&integrity, 0, sizeof(integrity));
862 switch (ns->pi_type) {
863 case NVME_NS_DPS_PI_TYPE3:
864 integrity.profile = &t10_pi_type3_crc;
865 integrity.tag_size = sizeof(u16) + sizeof(u32);
866 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
868 case NVME_NS_DPS_PI_TYPE1:
869 case NVME_NS_DPS_PI_TYPE2:
870 integrity.profile = &t10_pi_type1_crc;
871 integrity.tag_size = sizeof(u16);
872 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
875 integrity.profile = NULL;
878 integrity.tuple_size = ns->ms;
879 blk_integrity_register(ns->disk, &integrity);
880 blk_queue_max_integrity_segments(ns->queue, 1);
883 static void nvme_init_integrity(struct nvme_ns *ns)
886 #endif /* CONFIG_BLK_DEV_INTEGRITY */
888 static void nvme_config_discard(struct nvme_ns *ns)
890 struct nvme_ctrl *ctrl = ns->ctrl;
891 u32 logical_block_size = queue_logical_block_size(ns->queue);
893 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
894 NVME_DSM_MAX_RANGES);
896 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
897 ns->queue->limits.discard_zeroes_data = 1;
899 ns->queue->limits.discard_zeroes_data = 0;
901 ns->queue->limits.discard_alignment = logical_block_size;
902 ns->queue->limits.discard_granularity = logical_block_size;
903 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
904 blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
905 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
908 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
910 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
911 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
915 if ((*id)->ncap == 0) {
920 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
921 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
922 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
923 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
928 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
930 struct nvme_ns *ns = disk->private_data;
936 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
937 ns->lba_shift = id->lbaf[lbaf].ds;
938 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
939 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
942 * If identify namespace failed, use default 512 byte block size so
943 * block layer can use before failing read/write for 0 capacity.
945 if (ns->lba_shift == 0)
947 bs = 1 << ns->lba_shift;
948 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
949 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
950 id->dps & NVME_NS_DPS_PI_MASK : 0;
952 blk_mq_freeze_queue(disk->queue);
953 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
955 bs != queue_logical_block_size(disk->queue) ||
956 (ns->ms && ns->ext)))
957 blk_integrity_unregister(disk);
959 ns->pi_type = pi_type;
960 blk_queue_logical_block_size(ns->queue, bs);
962 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
963 nvme_init_integrity(ns);
964 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
965 set_capacity(disk, 0);
967 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
969 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
970 nvme_config_discard(ns);
971 blk_mq_unfreeze_queue(disk->queue);
974 static int nvme_revalidate_disk(struct gendisk *disk)
976 struct nvme_ns *ns = disk->private_data;
977 struct nvme_id_ns *id = NULL;
980 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
981 set_capacity(disk, 0);
985 ret = nvme_revalidate_ns(ns, &id);
989 __nvme_revalidate_disk(disk, id);
995 static char nvme_pr_type(enum pr_type type)
998 case PR_WRITE_EXCLUSIVE:
1000 case PR_EXCLUSIVE_ACCESS:
1002 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1004 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1006 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1008 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1015 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1016 u64 key, u64 sa_key, u8 op)
1018 struct nvme_ns *ns = bdev->bd_disk->private_data;
1019 struct nvme_command c;
1020 u8 data[16] = { 0, };
1022 put_unaligned_le64(key, &data[0]);
1023 put_unaligned_le64(sa_key, &data[8]);
1025 memset(&c, 0, sizeof(c));
1026 c.common.opcode = op;
1027 c.common.nsid = cpu_to_le32(ns->ns_id);
1028 c.common.cdw10[0] = cpu_to_le32(cdw10);
1030 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1033 static int nvme_pr_register(struct block_device *bdev, u64 old,
1034 u64 new, unsigned flags)
1038 if (flags & ~PR_FL_IGNORE_KEY)
1041 cdw10 = old ? 2 : 0;
1042 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1043 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1044 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1047 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1048 enum pr_type type, unsigned flags)
1052 if (flags & ~PR_FL_IGNORE_KEY)
1055 cdw10 = nvme_pr_type(type) << 8;
1056 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1057 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1060 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1061 enum pr_type type, bool abort)
1063 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1064 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1067 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1069 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1070 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1073 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1075 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1076 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1079 static const struct pr_ops nvme_pr_ops = {
1080 .pr_register = nvme_pr_register,
1081 .pr_reserve = nvme_pr_reserve,
1082 .pr_release = nvme_pr_release,
1083 .pr_preempt = nvme_pr_preempt,
1084 .pr_clear = nvme_pr_clear,
1087 #ifdef CONFIG_BLK_SED_OPAL
1088 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1091 struct nvme_ctrl *ctrl = data;
1092 struct nvme_command cmd;
1094 memset(&cmd, 0, sizeof(cmd));
1096 cmd.common.opcode = nvme_admin_security_send;
1098 cmd.common.opcode = nvme_admin_security_recv;
1099 cmd.common.nsid = 0;
1100 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1101 cmd.common.cdw10[1] = cpu_to_le32(len);
1103 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1104 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1106 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1107 #endif /* CONFIG_BLK_SED_OPAL */
1109 static const struct block_device_operations nvme_fops = {
1110 .owner = THIS_MODULE,
1111 .ioctl = nvme_ioctl,
1112 .compat_ioctl = nvme_compat_ioctl,
1114 .release = nvme_release,
1115 .getgeo = nvme_getgeo,
1116 .revalidate_disk= nvme_revalidate_disk,
1117 .pr_ops = &nvme_pr_ops,
1120 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1122 unsigned long timeout =
1123 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1124 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1127 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1130 if ((csts & NVME_CSTS_RDY) == bit)
1134 if (fatal_signal_pending(current))
1136 if (time_after(jiffies, timeout)) {
1137 dev_err(ctrl->device,
1138 "Device not ready; aborting %s\n", enabled ?
1139 "initialisation" : "reset");
1148 * If the device has been passed off to us in an enabled state, just clear
1149 * the enabled bit. The spec says we should set the 'shutdown notification
1150 * bits', but doing so may cause the device to complete commands to the
1151 * admin queue ... and we don't know what memory that might be pointing at!
1153 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1157 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1158 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1160 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1164 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1165 msleep(NVME_QUIRK_DELAY_AMOUNT);
1167 return nvme_wait_ready(ctrl, cap, false);
1169 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1171 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1174 * Default to a 4K page size, with the intention to update this
1175 * path in the future to accomodate architectures with differing
1176 * kernel and IO page sizes.
1178 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1181 if (page_shift < dev_page_min) {
1182 dev_err(ctrl->device,
1183 "Minimum device page size %u too large for host (%u)\n",
1184 1 << dev_page_min, 1 << page_shift);
1188 ctrl->page_size = 1 << page_shift;
1190 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1191 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1192 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1193 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1194 ctrl->ctrl_config |= NVME_CC_ENABLE;
1196 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1199 return nvme_wait_ready(ctrl, cap, true);
1201 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1203 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1205 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1209 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1210 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1212 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1216 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1217 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1221 if (fatal_signal_pending(current))
1223 if (time_after(jiffies, timeout)) {
1224 dev_err(ctrl->device,
1225 "Device shutdown incomplete; abort shutdown\n");
1232 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1234 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1235 struct request_queue *q)
1239 if (ctrl->max_hw_sectors) {
1241 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1243 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1244 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1246 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1247 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1248 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1249 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1251 blk_queue_write_cache(q, vwc, vwc);
1255 * Initialize the cached copies of the Identify data and various controller
1256 * register in our nvme_ctrl structure. This should be called as soon as
1257 * the admin queue is fully up and running.
1259 int nvme_init_identify(struct nvme_ctrl *ctrl)
1261 struct nvme_id_ctrl *id;
1263 int ret, page_shift;
1266 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1268 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1272 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1274 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1277 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1279 if (ctrl->vs >= NVME_VS(1, 1, 0))
1280 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1282 ret = nvme_identify_ctrl(ctrl, &id);
1284 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1288 ctrl->oacs = le16_to_cpu(id->oacs);
1289 ctrl->vid = le16_to_cpu(id->vid);
1290 ctrl->oncs = le16_to_cpup(&id->oncs);
1291 atomic_set(&ctrl->abort_limit, id->acl + 1);
1292 ctrl->vwc = id->vwc;
1293 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1294 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1295 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1296 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1298 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1300 max_hw_sectors = UINT_MAX;
1301 ctrl->max_hw_sectors =
1302 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1304 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1305 ctrl->sgls = le32_to_cpu(id->sgls);
1306 ctrl->kas = le16_to_cpu(id->kas);
1308 if (ctrl->ops->is_fabrics) {
1309 ctrl->icdoff = le16_to_cpu(id->icdoff);
1310 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1311 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1312 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1315 * In fabrics we need to verify the cntlid matches the
1318 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1321 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1323 "keep-alive support is mandatory for fabrics\n");
1327 ctrl->cntlid = le16_to_cpu(id->cntlid);
1333 EXPORT_SYMBOL_GPL(nvme_init_identify);
1335 static int nvme_dev_open(struct inode *inode, struct file *file)
1337 struct nvme_ctrl *ctrl;
1338 int instance = iminor(inode);
1341 spin_lock(&dev_list_lock);
1342 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1343 if (ctrl->instance != instance)
1346 if (!ctrl->admin_q) {
1350 if (!kref_get_unless_zero(&ctrl->kref))
1352 file->private_data = ctrl;
1356 spin_unlock(&dev_list_lock);
1361 static int nvme_dev_release(struct inode *inode, struct file *file)
1363 nvme_put_ctrl(file->private_data);
1367 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1372 mutex_lock(&ctrl->namespaces_mutex);
1373 if (list_empty(&ctrl->namespaces)) {
1378 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1379 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1380 dev_warn(ctrl->device,
1381 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1386 dev_warn(ctrl->device,
1387 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1388 kref_get(&ns->kref);
1389 mutex_unlock(&ctrl->namespaces_mutex);
1391 ret = nvme_user_cmd(ctrl, ns, argp);
1396 mutex_unlock(&ctrl->namespaces_mutex);
1400 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1403 struct nvme_ctrl *ctrl = file->private_data;
1404 void __user *argp = (void __user *)arg;
1407 case NVME_IOCTL_ADMIN_CMD:
1408 return nvme_user_cmd(ctrl, NULL, argp);
1409 case NVME_IOCTL_IO_CMD:
1410 return nvme_dev_user_cmd(ctrl, argp);
1411 case NVME_IOCTL_RESET:
1412 dev_warn(ctrl->device, "resetting controller\n");
1413 return ctrl->ops->reset_ctrl(ctrl);
1414 case NVME_IOCTL_SUBSYS_RESET:
1415 return nvme_reset_subsystem(ctrl);
1416 case NVME_IOCTL_RESCAN:
1417 nvme_queue_scan(ctrl);
1424 static const struct file_operations nvme_dev_fops = {
1425 .owner = THIS_MODULE,
1426 .open = nvme_dev_open,
1427 .release = nvme_dev_release,
1428 .unlocked_ioctl = nvme_dev_ioctl,
1429 .compat_ioctl = nvme_dev_ioctl,
1432 static ssize_t nvme_sysfs_reset(struct device *dev,
1433 struct device_attribute *attr, const char *buf,
1436 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1439 ret = ctrl->ops->reset_ctrl(ctrl);
1444 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1446 static ssize_t nvme_sysfs_rescan(struct device *dev,
1447 struct device_attribute *attr, const char *buf,
1450 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1452 nvme_queue_scan(ctrl);
1455 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1457 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1460 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1461 struct nvme_ctrl *ctrl = ns->ctrl;
1462 int serial_len = sizeof(ctrl->serial);
1463 int model_len = sizeof(ctrl->model);
1465 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1466 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1468 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1469 return sprintf(buf, "eui.%8phN\n", ns->eui);
1471 while (ctrl->serial[serial_len - 1] == ' ')
1473 while (ctrl->model[model_len - 1] == ' ')
1476 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1477 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1479 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1481 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1484 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1485 return sprintf(buf, "%pU\n", ns->uuid);
1487 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1489 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1492 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1493 return sprintf(buf, "%8phd\n", ns->eui);
1495 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1497 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1500 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1501 return sprintf(buf, "%d\n", ns->ns_id);
1503 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1505 static struct attribute *nvme_ns_attrs[] = {
1506 &dev_attr_wwid.attr,
1507 &dev_attr_uuid.attr,
1509 &dev_attr_nsid.attr,
1513 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1514 struct attribute *a, int n)
1516 struct device *dev = container_of(kobj, struct device, kobj);
1517 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1519 if (a == &dev_attr_uuid.attr) {
1520 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1523 if (a == &dev_attr_eui.attr) {
1524 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1530 static const struct attribute_group nvme_ns_attr_group = {
1531 .attrs = nvme_ns_attrs,
1532 .is_visible = nvme_ns_attrs_are_visible,
1535 #define nvme_show_str_function(field) \
1536 static ssize_t field##_show(struct device *dev, \
1537 struct device_attribute *attr, char *buf) \
1539 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1540 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1542 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1544 #define nvme_show_int_function(field) \
1545 static ssize_t field##_show(struct device *dev, \
1546 struct device_attribute *attr, char *buf) \
1548 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1549 return sprintf(buf, "%d\n", ctrl->field); \
1551 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1553 nvme_show_str_function(model);
1554 nvme_show_str_function(serial);
1555 nvme_show_str_function(firmware_rev);
1556 nvme_show_int_function(cntlid);
1558 static ssize_t nvme_sysfs_delete(struct device *dev,
1559 struct device_attribute *attr, const char *buf,
1562 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1564 if (device_remove_file_self(dev, attr))
1565 ctrl->ops->delete_ctrl(ctrl);
1568 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1570 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1571 struct device_attribute *attr,
1574 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1576 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1578 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1580 static ssize_t nvme_sysfs_show_state(struct device *dev,
1581 struct device_attribute *attr,
1584 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1585 static const char *const state_name[] = {
1586 [NVME_CTRL_NEW] = "new",
1587 [NVME_CTRL_LIVE] = "live",
1588 [NVME_CTRL_RESETTING] = "resetting",
1589 [NVME_CTRL_RECONNECTING]= "reconnecting",
1590 [NVME_CTRL_DELETING] = "deleting",
1591 [NVME_CTRL_DEAD] = "dead",
1594 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1595 state_name[ctrl->state])
1596 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1598 return sprintf(buf, "unknown state\n");
1601 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1603 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1604 struct device_attribute *attr,
1607 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1609 return snprintf(buf, PAGE_SIZE, "%s\n",
1610 ctrl->ops->get_subsysnqn(ctrl));
1612 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1614 static ssize_t nvme_sysfs_show_address(struct device *dev,
1615 struct device_attribute *attr,
1618 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1620 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1622 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1624 static struct attribute *nvme_dev_attrs[] = {
1625 &dev_attr_reset_controller.attr,
1626 &dev_attr_rescan_controller.attr,
1627 &dev_attr_model.attr,
1628 &dev_attr_serial.attr,
1629 &dev_attr_firmware_rev.attr,
1630 &dev_attr_cntlid.attr,
1631 &dev_attr_delete_controller.attr,
1632 &dev_attr_transport.attr,
1633 &dev_attr_subsysnqn.attr,
1634 &dev_attr_address.attr,
1635 &dev_attr_state.attr,
1639 #define CHECK_ATTR(ctrl, a, name) \
1640 if ((a) == &dev_attr_##name.attr && \
1641 !(ctrl)->ops->get_##name) \
1644 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1645 struct attribute *a, int n)
1647 struct device *dev = container_of(kobj, struct device, kobj);
1648 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1650 if (a == &dev_attr_delete_controller.attr) {
1651 if (!ctrl->ops->delete_ctrl)
1655 CHECK_ATTR(ctrl, a, subsysnqn);
1656 CHECK_ATTR(ctrl, a, address);
1661 static struct attribute_group nvme_dev_attrs_group = {
1662 .attrs = nvme_dev_attrs,
1663 .is_visible = nvme_dev_attrs_are_visible,
1666 static const struct attribute_group *nvme_dev_attr_groups[] = {
1667 &nvme_dev_attrs_group,
1671 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1673 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1674 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1676 return nsa->ns_id - nsb->ns_id;
1679 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1681 struct nvme_ns *ns, *ret = NULL;
1683 mutex_lock(&ctrl->namespaces_mutex);
1684 list_for_each_entry(ns, &ctrl->namespaces, list) {
1685 if (ns->ns_id == nsid) {
1686 kref_get(&ns->kref);
1690 if (ns->ns_id > nsid)
1693 mutex_unlock(&ctrl->namespaces_mutex);
1697 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1700 struct gendisk *disk;
1701 struct nvme_id_ns *id;
1702 char disk_name[DISK_NAME_LEN];
1703 int node = dev_to_node(ctrl->dev);
1705 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1709 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1710 if (ns->instance < 0)
1713 ns->queue = blk_mq_init_queue(ctrl->tagset);
1714 if (IS_ERR(ns->queue))
1715 goto out_release_instance;
1716 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1717 ns->queue->queuedata = ns;
1720 kref_init(&ns->kref);
1722 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1724 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1725 nvme_set_queue_limits(ctrl, ns->queue);
1727 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1729 if (nvme_revalidate_ns(ns, &id))
1730 goto out_free_queue;
1732 if (nvme_nvm_ns_supported(ns, id) &&
1733 nvme_nvm_register(ns, disk_name, node)) {
1734 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1738 disk = alloc_disk_node(0, node);
1742 disk->fops = &nvme_fops;
1743 disk->private_data = ns;
1744 disk->queue = ns->queue;
1745 disk->flags = GENHD_FL_EXT_DEVT;
1746 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1749 __nvme_revalidate_disk(disk, id);
1751 mutex_lock(&ctrl->namespaces_mutex);
1752 list_add_tail(&ns->list, &ctrl->namespaces);
1753 mutex_unlock(&ctrl->namespaces_mutex);
1755 kref_get(&ctrl->kref);
1759 device_add_disk(ctrl->device, ns->disk);
1760 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1761 &nvme_ns_attr_group))
1762 pr_warn("%s: failed to create sysfs group for identification\n",
1763 ns->disk->disk_name);
1764 if (ns->ndev && nvme_nvm_register_sysfs(ns))
1765 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
1766 ns->disk->disk_name);
1771 blk_cleanup_queue(ns->queue);
1772 out_release_instance:
1773 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1778 static void nvme_ns_remove(struct nvme_ns *ns)
1780 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1783 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1784 if (blk_get_integrity(ns->disk))
1785 blk_integrity_unregister(ns->disk);
1786 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1787 &nvme_ns_attr_group);
1789 nvme_nvm_unregister_sysfs(ns);
1790 del_gendisk(ns->disk);
1791 blk_mq_abort_requeue_list(ns->queue);
1792 blk_cleanup_queue(ns->queue);
1795 mutex_lock(&ns->ctrl->namespaces_mutex);
1796 list_del_init(&ns->list);
1797 mutex_unlock(&ns->ctrl->namespaces_mutex);
1802 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1806 ns = nvme_find_get_ns(ctrl, nsid);
1808 if (ns->disk && revalidate_disk(ns->disk))
1812 nvme_alloc_ns(ctrl, nsid);
1815 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1818 struct nvme_ns *ns, *next;
1820 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1821 if (ns->ns_id > nsid)
1826 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1830 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1833 ns_list = kzalloc(0x1000, GFP_KERNEL);
1837 for (i = 0; i < num_lists; i++) {
1838 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1842 for (j = 0; j < min(nn, 1024U); j++) {
1843 nsid = le32_to_cpu(ns_list[j]);
1847 nvme_validate_ns(ctrl, nsid);
1849 while (++prev < nsid) {
1850 ns = nvme_find_get_ns(ctrl, prev);
1860 nvme_remove_invalid_namespaces(ctrl, prev);
1866 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1870 for (i = 1; i <= nn; i++)
1871 nvme_validate_ns(ctrl, i);
1873 nvme_remove_invalid_namespaces(ctrl, nn);
1876 static void nvme_scan_work(struct work_struct *work)
1878 struct nvme_ctrl *ctrl =
1879 container_of(work, struct nvme_ctrl, scan_work);
1880 struct nvme_id_ctrl *id;
1883 if (ctrl->state != NVME_CTRL_LIVE)
1886 if (nvme_identify_ctrl(ctrl, &id))
1889 nn = le32_to_cpu(id->nn);
1890 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1891 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1892 if (!nvme_scan_ns_list(ctrl, nn))
1895 nvme_scan_ns_sequential(ctrl, nn);
1897 mutex_lock(&ctrl->namespaces_mutex);
1898 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1899 mutex_unlock(&ctrl->namespaces_mutex);
1903 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1906 * Do not queue new scan work when a controller is reset during
1909 if (ctrl->state == NVME_CTRL_LIVE)
1910 schedule_work(&ctrl->scan_work);
1912 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1915 * This function iterates the namespace list unlocked to allow recovery from
1916 * controller failure. It is up to the caller to ensure the namespace list is
1917 * not modified by scan work while this function is executing.
1919 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1921 struct nvme_ns *ns, *next;
1924 * The dead states indicates the controller was not gracefully
1925 * disconnected. In that case, we won't be able to flush any data while
1926 * removing the namespaces' disks; fail all the queues now to avoid
1927 * potentially having to clean up the failed sync later.
1929 if (ctrl->state == NVME_CTRL_DEAD)
1930 nvme_kill_queues(ctrl);
1932 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1935 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1937 static void nvme_async_event_work(struct work_struct *work)
1939 struct nvme_ctrl *ctrl =
1940 container_of(work, struct nvme_ctrl, async_event_work);
1942 spin_lock_irq(&ctrl->lock);
1943 while (ctrl->event_limit > 0) {
1944 int aer_idx = --ctrl->event_limit;
1946 spin_unlock_irq(&ctrl->lock);
1947 ctrl->ops->submit_async_event(ctrl, aer_idx);
1948 spin_lock_irq(&ctrl->lock);
1950 spin_unlock_irq(&ctrl->lock);
1953 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
1954 union nvme_result *res)
1956 u32 result = le32_to_cpu(res->u32);
1959 switch (le16_to_cpu(status) >> 1) {
1960 case NVME_SC_SUCCESS:
1963 case NVME_SC_ABORT_REQ:
1964 ++ctrl->event_limit;
1965 schedule_work(&ctrl->async_event_work);
1974 switch (result & 0xff07) {
1975 case NVME_AER_NOTICE_NS_CHANGED:
1976 dev_info(ctrl->device, "rescanning\n");
1977 nvme_queue_scan(ctrl);
1980 dev_warn(ctrl->device, "async event result %08x\n", result);
1983 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1985 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1987 ctrl->event_limit = NVME_NR_AERS;
1988 schedule_work(&ctrl->async_event_work);
1990 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1992 static DEFINE_IDA(nvme_instance_ida);
1994 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1996 int instance, error;
1999 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2002 spin_lock(&dev_list_lock);
2003 error = ida_get_new(&nvme_instance_ida, &instance);
2004 spin_unlock(&dev_list_lock);
2005 } while (error == -EAGAIN);
2010 ctrl->instance = instance;
2014 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2016 spin_lock(&dev_list_lock);
2017 ida_remove(&nvme_instance_ida, ctrl->instance);
2018 spin_unlock(&dev_list_lock);
2021 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2023 flush_work(&ctrl->async_event_work);
2024 flush_work(&ctrl->scan_work);
2025 nvme_remove_namespaces(ctrl);
2027 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2029 spin_lock(&dev_list_lock);
2030 list_del(&ctrl->node);
2031 spin_unlock(&dev_list_lock);
2033 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2035 static void nvme_free_ctrl(struct kref *kref)
2037 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2039 put_device(ctrl->device);
2040 nvme_release_instance(ctrl);
2041 ida_destroy(&ctrl->ns_ida);
2043 ctrl->ops->free_ctrl(ctrl);
2046 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2048 kref_put(&ctrl->kref, nvme_free_ctrl);
2050 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2053 * Initialize a NVMe controller structures. This needs to be called during
2054 * earliest initialization so that we have the initialized structured around
2057 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2058 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2062 ctrl->state = NVME_CTRL_NEW;
2063 spin_lock_init(&ctrl->lock);
2064 INIT_LIST_HEAD(&ctrl->namespaces);
2065 mutex_init(&ctrl->namespaces_mutex);
2066 kref_init(&ctrl->kref);
2069 ctrl->quirks = quirks;
2070 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2071 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2073 ret = nvme_set_instance(ctrl);
2077 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2078 MKDEV(nvme_char_major, ctrl->instance),
2079 ctrl, nvme_dev_attr_groups,
2080 "nvme%d", ctrl->instance);
2081 if (IS_ERR(ctrl->device)) {
2082 ret = PTR_ERR(ctrl->device);
2083 goto out_release_instance;
2085 get_device(ctrl->device);
2086 ida_init(&ctrl->ns_ida);
2088 spin_lock(&dev_list_lock);
2089 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2090 spin_unlock(&dev_list_lock);
2093 out_release_instance:
2094 nvme_release_instance(ctrl);
2098 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2101 * nvme_kill_queues(): Ends all namespace queues
2102 * @ctrl: the dead controller that needs to end
2104 * Call this function when the driver determines it is unable to get the
2105 * controller in a state capable of servicing IO.
2107 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2111 mutex_lock(&ctrl->namespaces_mutex);
2112 list_for_each_entry(ns, &ctrl->namespaces, list) {
2114 * Revalidating a dead namespace sets capacity to 0. This will
2115 * end buffered writers dirtying pages that can't be synced.
2117 if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2118 revalidate_disk(ns->disk);
2120 blk_set_queue_dying(ns->queue);
2121 blk_mq_abort_requeue_list(ns->queue);
2122 blk_mq_start_stopped_hw_queues(ns->queue, true);
2124 mutex_unlock(&ctrl->namespaces_mutex);
2126 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2128 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2132 mutex_lock(&ctrl->namespaces_mutex);
2133 list_for_each_entry(ns, &ctrl->namespaces, list)
2134 blk_mq_quiesce_queue(ns->queue);
2135 mutex_unlock(&ctrl->namespaces_mutex);
2137 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2139 void nvme_start_queues(struct nvme_ctrl *ctrl)
2143 mutex_lock(&ctrl->namespaces_mutex);
2144 list_for_each_entry(ns, &ctrl->namespaces, list) {
2145 blk_mq_start_stopped_hw_queues(ns->queue, true);
2146 blk_mq_kick_requeue_list(ns->queue);
2148 mutex_unlock(&ctrl->namespaces_mutex);
2150 EXPORT_SYMBOL_GPL(nvme_start_queues);
2152 int __init nvme_core_init(void)
2156 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2160 else if (result > 0)
2161 nvme_char_major = result;
2163 nvme_class = class_create(THIS_MODULE, "nvme");
2164 if (IS_ERR(nvme_class)) {
2165 result = PTR_ERR(nvme_class);
2166 goto unregister_chrdev;
2172 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2176 void nvme_core_exit(void)
2178 class_destroy(nvme_class);
2179 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2182 MODULE_LICENSE("GPL");
2183 MODULE_VERSION("1.0");
2184 module_init(nvme_core_init);
2185 module_exit(nvme_core_exit);