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 = ctrl->state;
87 spin_lock_irq(&ctrl->lock);
92 case NVME_CTRL_RESETTING:
93 case NVME_CTRL_RECONNECTING:
100 case NVME_CTRL_RESETTING:
104 case NVME_CTRL_RECONNECTING:
111 case NVME_CTRL_RECONNECTING:
120 case NVME_CTRL_DELETING:
123 case NVME_CTRL_RESETTING:
124 case NVME_CTRL_RECONNECTING:
133 case NVME_CTRL_DELETING:
143 spin_unlock_irq(&ctrl->lock);
146 ctrl->state = new_state;
150 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
152 static void nvme_free_ns(struct kref *kref)
154 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
156 if (ns->type == NVME_NS_LIGHTNVM)
157 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
159 spin_lock(&dev_list_lock);
160 ns->disk->private_data = NULL;
161 spin_unlock(&dev_list_lock);
164 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
165 nvme_put_ctrl(ns->ctrl);
169 static void nvme_put_ns(struct nvme_ns *ns)
171 kref_put(&ns->kref, nvme_free_ns);
174 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
178 spin_lock(&dev_list_lock);
179 ns = disk->private_data;
181 if (!kref_get_unless_zero(&ns->kref))
183 if (!try_module_get(ns->ctrl->ops->module))
186 spin_unlock(&dev_list_lock);
191 kref_put(&ns->kref, nvme_free_ns);
193 spin_unlock(&dev_list_lock);
197 void nvme_requeue_req(struct request *req)
201 blk_mq_requeue_request(req);
202 spin_lock_irqsave(req->q->queue_lock, flags);
203 if (!blk_queue_stopped(req->q))
204 blk_mq_kick_requeue_list(req->q);
205 spin_unlock_irqrestore(req->q->queue_lock, flags);
207 EXPORT_SYMBOL_GPL(nvme_requeue_req);
209 struct request *nvme_alloc_request(struct request_queue *q,
210 struct nvme_command *cmd, unsigned int flags, int qid)
214 if (qid == NVME_QID_ANY) {
215 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
217 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
223 req->cmd_type = REQ_TYPE_DRV_PRIV;
224 req->cmd_flags |= REQ_FAILFAST_DRIVER;
225 req->cmd = (unsigned char *)cmd;
226 req->cmd_len = sizeof(struct nvme_command);
230 EXPORT_SYMBOL_GPL(nvme_alloc_request);
232 static inline void nvme_setup_flush(struct nvme_ns *ns,
233 struct nvme_command *cmnd)
235 memset(cmnd, 0, sizeof(*cmnd));
236 cmnd->common.opcode = nvme_cmd_flush;
237 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
240 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
241 struct nvme_command *cmnd)
243 struct nvme_dsm_range *range;
246 unsigned int nr_bytes = blk_rq_bytes(req);
248 range = kmalloc(sizeof(*range), GFP_ATOMIC);
250 return BLK_MQ_RQ_QUEUE_BUSY;
252 range->cattr = cpu_to_le32(0);
253 range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
254 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
256 memset(cmnd, 0, sizeof(*cmnd));
257 cmnd->dsm.opcode = nvme_cmd_dsm;
258 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
260 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
262 req->completion_data = range;
263 page = virt_to_page(range);
264 offset = offset_in_page(range);
265 blk_add_request_payload(req, page, offset, sizeof(*range));
268 * we set __data_len back to the size of the area to be discarded
269 * on disk. This allows us to report completion on the full amount
270 * of blocks described by the request.
272 req->__data_len = nr_bytes;
277 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
278 struct nvme_command *cmnd)
283 if (req->cmd_flags & REQ_FUA)
284 control |= NVME_RW_FUA;
285 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
286 control |= NVME_RW_LR;
288 if (req->cmd_flags & REQ_RAHEAD)
289 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
291 memset(cmnd, 0, sizeof(*cmnd));
292 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
293 cmnd->rw.command_id = req->tag;
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)
324 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
325 memcpy(cmd, req->cmd, sizeof(*cmd));
326 else if (req_op(req) == REQ_OP_FLUSH)
327 nvme_setup_flush(ns, cmd);
328 else if (req_op(req) == REQ_OP_DISCARD)
329 ret = nvme_setup_discard(ns, req, cmd);
331 nvme_setup_rw(ns, req, cmd);
335 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
338 * Returns 0 on success. If the result is negative, it's a Linux error code;
339 * if the result is positive, it's an NVM Express status code
341 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
342 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
343 unsigned timeout, int qid, int at_head, int flags)
348 req = nvme_alloc_request(q, cmd, flags, qid);
352 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
355 if (buffer && bufflen) {
356 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
361 blk_execute_rq(req->q, NULL, req, at_head);
364 blk_mq_free_request(req);
367 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
369 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
370 void *buffer, unsigned bufflen)
372 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
375 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
377 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
378 void __user *ubuffer, unsigned bufflen,
379 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
380 u32 *result, unsigned timeout)
382 bool write = nvme_is_write(cmd);
383 struct nvme_completion cqe;
384 struct nvme_ns *ns = q->queuedata;
385 struct gendisk *disk = ns ? ns->disk : NULL;
387 struct bio *bio = NULL;
391 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
395 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
398 if (ubuffer && bufflen) {
399 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
407 bio->bi_bdev = bdget_disk(disk, 0);
413 if (meta_buffer && meta_len) {
414 struct bio_integrity_payload *bip;
416 meta = kmalloc(meta_len, GFP_KERNEL);
423 if (copy_from_user(meta, meta_buffer,
430 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
436 bip->bip_iter.bi_size = meta_len;
437 bip->bip_iter.bi_sector = meta_seed;
439 ret = bio_integrity_add_page(bio, virt_to_page(meta),
440 meta_len, offset_in_page(meta));
441 if (ret != meta_len) {
448 blk_execute_rq(req->q, disk, req, 0);
451 *result = le32_to_cpu(cqe.result);
452 if (meta && !ret && !write) {
453 if (copy_to_user(meta_buffer, meta, meta_len))
460 if (disk && bio->bi_bdev)
462 blk_rq_unmap_user(bio);
465 blk_mq_free_request(req);
469 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
470 void __user *ubuffer, unsigned bufflen, u32 *result,
473 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
477 static void nvme_keep_alive_end_io(struct request *rq, int error)
479 struct nvme_ctrl *ctrl = rq->end_io_data;
481 blk_mq_free_request(rq);
484 dev_err(ctrl->device,
485 "failed nvme_keep_alive_end_io error=%d\n", error);
489 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
492 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
494 struct nvme_command c;
497 memset(&c, 0, sizeof(c));
498 c.common.opcode = nvme_admin_keep_alive;
500 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
505 rq->timeout = ctrl->kato * HZ;
506 rq->end_io_data = ctrl;
508 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
513 static void nvme_keep_alive_work(struct work_struct *work)
515 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
516 struct nvme_ctrl, ka_work);
518 if (nvme_keep_alive(ctrl)) {
519 /* allocation failure, reset the controller */
520 dev_err(ctrl->device, "keep-alive failed\n");
521 ctrl->ops->reset_ctrl(ctrl);
526 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
528 if (unlikely(ctrl->kato == 0))
531 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
532 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
534 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
536 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
538 if (unlikely(ctrl->kato == 0))
541 cancel_delayed_work_sync(&ctrl->ka_work);
543 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
545 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
547 struct nvme_command c = { };
550 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
551 c.identify.opcode = nvme_admin_identify;
552 c.identify.cns = cpu_to_le32(1);
554 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
558 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
559 sizeof(struct nvme_id_ctrl));
565 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
567 struct nvme_command c = { };
569 c.identify.opcode = nvme_admin_identify;
570 c.identify.cns = cpu_to_le32(2);
571 c.identify.nsid = cpu_to_le32(nsid);
572 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
575 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
576 struct nvme_id_ns **id)
578 struct nvme_command c = { };
581 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
582 c.identify.opcode = nvme_admin_identify,
583 c.identify.nsid = cpu_to_le32(nsid),
585 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
589 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
590 sizeof(struct nvme_id_ns));
596 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
597 dma_addr_t dma_addr, u32 *result)
599 struct nvme_command c;
600 struct nvme_completion cqe;
603 memset(&c, 0, sizeof(c));
604 c.features.opcode = nvme_admin_get_features;
605 c.features.nsid = cpu_to_le32(nsid);
606 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
607 c.features.fid = cpu_to_le32(fid);
609 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
612 *result = le32_to_cpu(cqe.result);
616 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
617 dma_addr_t dma_addr, u32 *result)
619 struct nvme_command c;
620 struct nvme_completion cqe;
623 memset(&c, 0, sizeof(c));
624 c.features.opcode = nvme_admin_set_features;
625 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
626 c.features.fid = cpu_to_le32(fid);
627 c.features.dword11 = cpu_to_le32(dword11);
629 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
632 *result = le32_to_cpu(cqe.result);
636 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
638 struct nvme_command c = { };
641 c.common.opcode = nvme_admin_get_log_page,
642 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
643 c.common.cdw10[0] = cpu_to_le32(
644 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
647 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
651 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
652 sizeof(struct nvme_smart_log));
658 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
660 u32 q_count = (*count - 1) | ((*count - 1) << 16);
662 int status, nr_io_queues;
664 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
670 * Degraded controllers might return an error when setting the queue
671 * count. We still want to be able to bring them online and offer
672 * access to the admin queue, as that might be only way to fix them up.
675 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
678 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
679 *count = min(*count, nr_io_queues);
684 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
686 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
688 struct nvme_user_io io;
689 struct nvme_command c;
690 unsigned length, meta_len;
691 void __user *metadata;
693 if (copy_from_user(&io, uio, sizeof(io)))
701 case nvme_cmd_compare:
707 length = (io.nblocks + 1) << ns->lba_shift;
708 meta_len = (io.nblocks + 1) * ns->ms;
709 metadata = (void __user *)(uintptr_t)io.metadata;
714 } else if (meta_len) {
715 if ((io.metadata & 3) || !io.metadata)
719 memset(&c, 0, sizeof(c));
720 c.rw.opcode = io.opcode;
721 c.rw.flags = io.flags;
722 c.rw.nsid = cpu_to_le32(ns->ns_id);
723 c.rw.slba = cpu_to_le64(io.slba);
724 c.rw.length = cpu_to_le16(io.nblocks);
725 c.rw.control = cpu_to_le16(io.control);
726 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
727 c.rw.reftag = cpu_to_le32(io.reftag);
728 c.rw.apptag = cpu_to_le16(io.apptag);
729 c.rw.appmask = cpu_to_le16(io.appmask);
731 return __nvme_submit_user_cmd(ns->queue, &c,
732 (void __user *)(uintptr_t)io.addr, length,
733 metadata, meta_len, io.slba, NULL, 0);
736 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
737 struct nvme_passthru_cmd __user *ucmd)
739 struct nvme_passthru_cmd cmd;
740 struct nvme_command c;
741 unsigned timeout = 0;
744 if (!capable(CAP_SYS_ADMIN))
746 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
751 memset(&c, 0, sizeof(c));
752 c.common.opcode = cmd.opcode;
753 c.common.flags = cmd.flags;
754 c.common.nsid = cpu_to_le32(cmd.nsid);
755 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
756 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
757 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
758 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
759 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
760 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
761 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
762 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
765 timeout = msecs_to_jiffies(cmd.timeout_ms);
767 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
768 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
769 &cmd.result, timeout);
771 if (put_user(cmd.result, &ucmd->result))
778 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
779 unsigned int cmd, unsigned long arg)
781 struct nvme_ns *ns = bdev->bd_disk->private_data;
785 force_successful_syscall_return();
787 case NVME_IOCTL_ADMIN_CMD:
788 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
789 case NVME_IOCTL_IO_CMD:
790 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
791 case NVME_IOCTL_SUBMIT_IO:
792 return nvme_submit_io(ns, (void __user *)arg);
793 #ifdef CONFIG_BLK_DEV_NVME_SCSI
794 case SG_GET_VERSION_NUM:
795 return nvme_sg_get_version_num((void __user *)arg);
797 return nvme_sg_io(ns, (void __user *)arg);
805 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
806 unsigned int cmd, unsigned long arg)
812 return nvme_ioctl(bdev, mode, cmd, arg);
815 #define nvme_compat_ioctl NULL
818 static int nvme_open(struct block_device *bdev, fmode_t mode)
820 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
823 static void nvme_release(struct gendisk *disk, fmode_t mode)
825 struct nvme_ns *ns = disk->private_data;
827 module_put(ns->ctrl->ops->module);
831 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
833 /* some standard values */
835 geo->sectors = 1 << 5;
836 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
840 #ifdef CONFIG_BLK_DEV_INTEGRITY
841 static void nvme_init_integrity(struct nvme_ns *ns)
843 struct blk_integrity integrity;
845 memset(&integrity, 0, sizeof(integrity));
846 switch (ns->pi_type) {
847 case NVME_NS_DPS_PI_TYPE3:
848 integrity.profile = &t10_pi_type3_crc;
849 integrity.tag_size = sizeof(u16) + sizeof(u32);
850 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
852 case NVME_NS_DPS_PI_TYPE1:
853 case NVME_NS_DPS_PI_TYPE2:
854 integrity.profile = &t10_pi_type1_crc;
855 integrity.tag_size = sizeof(u16);
856 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
859 integrity.profile = NULL;
862 integrity.tuple_size = ns->ms;
863 blk_integrity_register(ns->disk, &integrity);
864 blk_queue_max_integrity_segments(ns->queue, 1);
867 static void nvme_init_integrity(struct nvme_ns *ns)
870 #endif /* CONFIG_BLK_DEV_INTEGRITY */
872 static void nvme_config_discard(struct nvme_ns *ns)
874 struct nvme_ctrl *ctrl = ns->ctrl;
875 u32 logical_block_size = queue_logical_block_size(ns->queue);
877 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
878 ns->queue->limits.discard_zeroes_data = 1;
880 ns->queue->limits.discard_zeroes_data = 0;
882 ns->queue->limits.discard_alignment = logical_block_size;
883 ns->queue->limits.discard_granularity = logical_block_size;
884 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
885 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
888 static int nvme_revalidate_disk(struct gendisk *disk)
890 struct nvme_ns *ns = disk->private_data;
891 struct nvme_id_ns *id;
896 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
897 set_capacity(disk, 0);
900 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
901 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
910 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
911 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
912 dev_warn(disk_to_dev(ns->disk),
913 "%s: LightNVM init failure\n", __func__);
917 ns->type = NVME_NS_LIGHTNVM;
920 if (ns->ctrl->vs >= NVME_VS(1, 1))
921 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
922 if (ns->ctrl->vs >= NVME_VS(1, 2))
923 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
926 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
927 ns->lba_shift = id->lbaf[lbaf].ds;
928 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
929 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
932 * If identify namespace failed, use default 512 byte block size so
933 * block layer can use before failing read/write for 0 capacity.
935 if (ns->lba_shift == 0)
937 bs = 1 << ns->lba_shift;
938 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
939 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
940 id->dps & NVME_NS_DPS_PI_MASK : 0;
942 blk_mq_freeze_queue(disk->queue);
943 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
945 bs != queue_logical_block_size(disk->queue) ||
946 (ns->ms && ns->ext)))
947 blk_integrity_unregister(disk);
949 ns->pi_type = pi_type;
950 blk_queue_logical_block_size(ns->queue, bs);
952 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
953 nvme_init_integrity(ns);
954 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
955 set_capacity(disk, 0);
957 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
959 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
960 nvme_config_discard(ns);
961 blk_mq_unfreeze_queue(disk->queue);
967 static char nvme_pr_type(enum pr_type type)
970 case PR_WRITE_EXCLUSIVE:
972 case PR_EXCLUSIVE_ACCESS:
974 case PR_WRITE_EXCLUSIVE_REG_ONLY:
976 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
978 case PR_WRITE_EXCLUSIVE_ALL_REGS:
980 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
987 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
988 u64 key, u64 sa_key, u8 op)
990 struct nvme_ns *ns = bdev->bd_disk->private_data;
991 struct nvme_command c;
992 u8 data[16] = { 0, };
994 put_unaligned_le64(key, &data[0]);
995 put_unaligned_le64(sa_key, &data[8]);
997 memset(&c, 0, sizeof(c));
998 c.common.opcode = op;
999 c.common.nsid = cpu_to_le32(ns->ns_id);
1000 c.common.cdw10[0] = cpu_to_le32(cdw10);
1002 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1005 static int nvme_pr_register(struct block_device *bdev, u64 old,
1006 u64 new, unsigned flags)
1010 if (flags & ~PR_FL_IGNORE_KEY)
1013 cdw10 = old ? 2 : 0;
1014 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1015 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1016 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1019 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1020 enum pr_type type, unsigned flags)
1024 if (flags & ~PR_FL_IGNORE_KEY)
1027 cdw10 = nvme_pr_type(type) << 8;
1028 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1029 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1032 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1033 enum pr_type type, bool abort)
1035 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1036 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1039 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1041 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1042 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1045 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1047 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1048 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1051 static const struct pr_ops nvme_pr_ops = {
1052 .pr_register = nvme_pr_register,
1053 .pr_reserve = nvme_pr_reserve,
1054 .pr_release = nvme_pr_release,
1055 .pr_preempt = nvme_pr_preempt,
1056 .pr_clear = nvme_pr_clear,
1059 static const struct block_device_operations nvme_fops = {
1060 .owner = THIS_MODULE,
1061 .ioctl = nvme_ioctl,
1062 .compat_ioctl = nvme_compat_ioctl,
1064 .release = nvme_release,
1065 .getgeo = nvme_getgeo,
1066 .revalidate_disk= nvme_revalidate_disk,
1067 .pr_ops = &nvme_pr_ops,
1070 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1072 unsigned long timeout =
1073 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1074 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1077 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1078 if ((csts & NVME_CSTS_RDY) == bit)
1082 if (fatal_signal_pending(current))
1084 if (time_after(jiffies, timeout)) {
1085 dev_err(ctrl->device,
1086 "Device not ready; aborting %s\n", enabled ?
1087 "initialisation" : "reset");
1096 * If the device has been passed off to us in an enabled state, just clear
1097 * the enabled bit. The spec says we should set the 'shutdown notification
1098 * bits', but doing so may cause the device to complete commands to the
1099 * admin queue ... and we don't know what memory that might be pointing at!
1101 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1105 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1106 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1108 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1112 /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1113 * load, since we only need the quirk on reset_controller. Notice
1114 * that the HGST device needs this delay only in firmware activation
1115 * procedure; unfortunately we have no (easy) way to verify this.
1117 if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1118 msleep(NVME_QUIRK_DELAY_AMOUNT);
1120 return nvme_wait_ready(ctrl, cap, false);
1122 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1124 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1127 * Default to a 4K page size, with the intention to update this
1128 * path in the future to accomodate architectures with differing
1129 * kernel and IO page sizes.
1131 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1134 if (page_shift < dev_page_min) {
1135 dev_err(ctrl->device,
1136 "Minimum device page size %u too large for host (%u)\n",
1137 1 << dev_page_min, 1 << page_shift);
1141 ctrl->page_size = 1 << page_shift;
1143 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1144 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1145 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1146 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1147 ctrl->ctrl_config |= NVME_CC_ENABLE;
1149 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1152 return nvme_wait_ready(ctrl, cap, true);
1154 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1156 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1158 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1162 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1163 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1165 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1169 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1170 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1174 if (fatal_signal_pending(current))
1176 if (time_after(jiffies, timeout)) {
1177 dev_err(ctrl->device,
1178 "Device shutdown incomplete; abort shutdown\n");
1185 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1187 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1188 struct request_queue *q)
1192 if (ctrl->max_hw_sectors) {
1194 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1196 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1197 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1199 if (ctrl->stripe_size)
1200 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1201 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1202 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1204 blk_queue_write_cache(q, vwc, vwc);
1208 * Initialize the cached copies of the Identify data and various controller
1209 * register in our nvme_ctrl structure. This should be called as soon as
1210 * the admin queue is fully up and running.
1212 int nvme_init_identify(struct nvme_ctrl *ctrl)
1214 struct nvme_id_ctrl *id;
1216 int ret, page_shift;
1219 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1221 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1225 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1227 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1230 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1232 if (ctrl->vs >= NVME_VS(1, 1))
1233 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1235 ret = nvme_identify_ctrl(ctrl, &id);
1237 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1241 ctrl->vid = le16_to_cpu(id->vid);
1242 ctrl->oncs = le16_to_cpup(&id->oncs);
1243 atomic_set(&ctrl->abort_limit, id->acl + 1);
1244 ctrl->vwc = id->vwc;
1245 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1246 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1247 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1248 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1250 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1252 max_hw_sectors = UINT_MAX;
1253 ctrl->max_hw_sectors =
1254 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1256 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1257 unsigned int max_hw_sectors;
1259 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1260 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1261 if (ctrl->max_hw_sectors) {
1262 ctrl->max_hw_sectors = min(max_hw_sectors,
1263 ctrl->max_hw_sectors);
1265 ctrl->max_hw_sectors = max_hw_sectors;
1269 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1270 ctrl->sgls = le32_to_cpu(id->sgls);
1271 ctrl->kas = le16_to_cpu(id->kas);
1273 if (ctrl->ops->is_fabrics) {
1274 ctrl->icdoff = le16_to_cpu(id->icdoff);
1275 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1276 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1277 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1280 * In fabrics we need to verify the cntlid matches the
1283 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1286 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1288 "keep-alive support is mandatory for fabrics\n");
1292 ctrl->cntlid = le16_to_cpu(id->cntlid);
1298 EXPORT_SYMBOL_GPL(nvme_init_identify);
1300 static int nvme_dev_open(struct inode *inode, struct file *file)
1302 struct nvme_ctrl *ctrl;
1303 int instance = iminor(inode);
1306 spin_lock(&dev_list_lock);
1307 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1308 if (ctrl->instance != instance)
1311 if (!ctrl->admin_q) {
1315 if (!kref_get_unless_zero(&ctrl->kref))
1317 file->private_data = ctrl;
1321 spin_unlock(&dev_list_lock);
1326 static int nvme_dev_release(struct inode *inode, struct file *file)
1328 nvme_put_ctrl(file->private_data);
1332 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1337 mutex_lock(&ctrl->namespaces_mutex);
1338 if (list_empty(&ctrl->namespaces)) {
1343 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1344 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1345 dev_warn(ctrl->device,
1346 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1351 dev_warn(ctrl->device,
1352 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1353 kref_get(&ns->kref);
1354 mutex_unlock(&ctrl->namespaces_mutex);
1356 ret = nvme_user_cmd(ctrl, ns, argp);
1361 mutex_unlock(&ctrl->namespaces_mutex);
1365 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1368 struct nvme_ctrl *ctrl = file->private_data;
1369 void __user *argp = (void __user *)arg;
1372 case NVME_IOCTL_ADMIN_CMD:
1373 return nvme_user_cmd(ctrl, NULL, argp);
1374 case NVME_IOCTL_IO_CMD:
1375 return nvme_dev_user_cmd(ctrl, argp);
1376 case NVME_IOCTL_RESET:
1377 dev_warn(ctrl->device, "resetting controller\n");
1378 return ctrl->ops->reset_ctrl(ctrl);
1379 case NVME_IOCTL_SUBSYS_RESET:
1380 return nvme_reset_subsystem(ctrl);
1381 case NVME_IOCTL_RESCAN:
1382 nvme_queue_scan(ctrl);
1389 static const struct file_operations nvme_dev_fops = {
1390 .owner = THIS_MODULE,
1391 .open = nvme_dev_open,
1392 .release = nvme_dev_release,
1393 .unlocked_ioctl = nvme_dev_ioctl,
1394 .compat_ioctl = nvme_dev_ioctl,
1397 static ssize_t nvme_sysfs_reset(struct device *dev,
1398 struct device_attribute *attr, const char *buf,
1401 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1404 ret = ctrl->ops->reset_ctrl(ctrl);
1409 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1411 static ssize_t nvme_sysfs_rescan(struct device *dev,
1412 struct device_attribute *attr, const char *buf,
1415 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1417 nvme_queue_scan(ctrl);
1420 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1422 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1425 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1426 struct nvme_ctrl *ctrl = ns->ctrl;
1427 int serial_len = sizeof(ctrl->serial);
1428 int model_len = sizeof(ctrl->model);
1430 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1431 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1433 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1434 return sprintf(buf, "eui.%8phN\n", ns->eui);
1436 while (ctrl->serial[serial_len - 1] == ' ')
1438 while (ctrl->model[model_len - 1] == ' ')
1441 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1442 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1444 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1446 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1449 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1450 return sprintf(buf, "%pU\n", ns->uuid);
1452 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1454 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1457 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1458 return sprintf(buf, "%8phd\n", ns->eui);
1460 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1462 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1465 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1466 return sprintf(buf, "%d\n", ns->ns_id);
1468 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1470 static struct attribute *nvme_ns_attrs[] = {
1471 &dev_attr_wwid.attr,
1472 &dev_attr_uuid.attr,
1474 &dev_attr_nsid.attr,
1478 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1479 struct attribute *a, int n)
1481 struct device *dev = container_of(kobj, struct device, kobj);
1482 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1484 if (a == &dev_attr_uuid.attr) {
1485 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1488 if (a == &dev_attr_eui.attr) {
1489 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1495 static const struct attribute_group nvme_ns_attr_group = {
1496 .attrs = nvme_ns_attrs,
1497 .is_visible = nvme_ns_attrs_are_visible,
1500 #define nvme_show_str_function(field) \
1501 static ssize_t field##_show(struct device *dev, \
1502 struct device_attribute *attr, char *buf) \
1504 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1505 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1507 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1509 #define nvme_show_int_function(field) \
1510 static ssize_t field##_show(struct device *dev, \
1511 struct device_attribute *attr, char *buf) \
1513 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1514 return sprintf(buf, "%d\n", ctrl->field); \
1516 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1518 nvme_show_str_function(model);
1519 nvme_show_str_function(serial);
1520 nvme_show_str_function(firmware_rev);
1521 nvme_show_int_function(cntlid);
1523 static ssize_t nvme_sysfs_delete(struct device *dev,
1524 struct device_attribute *attr, const char *buf,
1527 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1529 if (device_remove_file_self(dev, attr))
1530 ctrl->ops->delete_ctrl(ctrl);
1533 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1535 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1536 struct device_attribute *attr,
1539 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1541 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1543 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1545 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1546 struct device_attribute *attr,
1549 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1551 return snprintf(buf, PAGE_SIZE, "%s\n",
1552 ctrl->ops->get_subsysnqn(ctrl));
1554 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1556 static ssize_t nvme_sysfs_show_address(struct device *dev,
1557 struct device_attribute *attr,
1560 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1562 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1564 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1566 static struct attribute *nvme_dev_attrs[] = {
1567 &dev_attr_reset_controller.attr,
1568 &dev_attr_rescan_controller.attr,
1569 &dev_attr_model.attr,
1570 &dev_attr_serial.attr,
1571 &dev_attr_firmware_rev.attr,
1572 &dev_attr_cntlid.attr,
1573 &dev_attr_delete_controller.attr,
1574 &dev_attr_transport.attr,
1575 &dev_attr_subsysnqn.attr,
1576 &dev_attr_address.attr,
1580 #define CHECK_ATTR(ctrl, a, name) \
1581 if ((a) == &dev_attr_##name.attr && \
1582 !(ctrl)->ops->get_##name) \
1585 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1586 struct attribute *a, int n)
1588 struct device *dev = container_of(kobj, struct device, kobj);
1589 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1591 if (a == &dev_attr_delete_controller.attr) {
1592 if (!ctrl->ops->delete_ctrl)
1596 CHECK_ATTR(ctrl, a, subsysnqn);
1597 CHECK_ATTR(ctrl, a, address);
1602 static struct attribute_group nvme_dev_attrs_group = {
1603 .attrs = nvme_dev_attrs,
1604 .is_visible = nvme_dev_attrs_are_visible,
1607 static const struct attribute_group *nvme_dev_attr_groups[] = {
1608 &nvme_dev_attrs_group,
1612 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1614 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1615 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1617 return nsa->ns_id - nsb->ns_id;
1620 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1622 struct nvme_ns *ns, *ret = NULL;
1624 mutex_lock(&ctrl->namespaces_mutex);
1625 list_for_each_entry(ns, &ctrl->namespaces, list) {
1626 if (ns->ns_id == nsid) {
1627 kref_get(&ns->kref);
1631 if (ns->ns_id > nsid)
1634 mutex_unlock(&ctrl->namespaces_mutex);
1638 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1641 struct gendisk *disk;
1642 int node = dev_to_node(ctrl->dev);
1644 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1648 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1649 if (ns->instance < 0)
1652 ns->queue = blk_mq_init_queue(ctrl->tagset);
1653 if (IS_ERR(ns->queue))
1654 goto out_release_instance;
1655 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1656 ns->queue->queuedata = ns;
1659 disk = alloc_disk_node(0, node);
1661 goto out_free_queue;
1663 kref_init(&ns->kref);
1666 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1669 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1670 nvme_set_queue_limits(ctrl, ns->queue);
1672 disk->fops = &nvme_fops;
1673 disk->private_data = ns;
1674 disk->queue = ns->queue;
1675 disk->flags = GENHD_FL_EXT_DEVT;
1676 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1678 if (nvme_revalidate_disk(ns->disk))
1681 mutex_lock(&ctrl->namespaces_mutex);
1682 list_add_tail(&ns->list, &ctrl->namespaces);
1683 mutex_unlock(&ctrl->namespaces_mutex);
1685 kref_get(&ctrl->kref);
1686 if (ns->type == NVME_NS_LIGHTNVM)
1689 device_add_disk(ctrl->device, ns->disk);
1690 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1691 &nvme_ns_attr_group))
1692 pr_warn("%s: failed to create sysfs group for identification\n",
1693 ns->disk->disk_name);
1698 blk_cleanup_queue(ns->queue);
1699 out_release_instance:
1700 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1705 static void nvme_ns_remove(struct nvme_ns *ns)
1707 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1710 if (ns->disk->flags & GENHD_FL_UP) {
1711 if (blk_get_integrity(ns->disk))
1712 blk_integrity_unregister(ns->disk);
1713 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1714 &nvme_ns_attr_group);
1715 del_gendisk(ns->disk);
1716 blk_mq_abort_requeue_list(ns->queue);
1717 blk_cleanup_queue(ns->queue);
1720 mutex_lock(&ns->ctrl->namespaces_mutex);
1721 list_del_init(&ns->list);
1722 mutex_unlock(&ns->ctrl->namespaces_mutex);
1727 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1731 ns = nvme_find_get_ns(ctrl, nsid);
1733 if (revalidate_disk(ns->disk))
1737 nvme_alloc_ns(ctrl, nsid);
1740 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1743 struct nvme_ns *ns, *next;
1745 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1746 if (ns->ns_id > nsid)
1751 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1755 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1758 ns_list = kzalloc(0x1000, GFP_KERNEL);
1762 for (i = 0; i < num_lists; i++) {
1763 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1767 for (j = 0; j < min(nn, 1024U); j++) {
1768 nsid = le32_to_cpu(ns_list[j]);
1772 nvme_validate_ns(ctrl, nsid);
1774 while (++prev < nsid) {
1775 ns = nvme_find_get_ns(ctrl, prev);
1785 nvme_remove_invalid_namespaces(ctrl, prev);
1791 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1795 for (i = 1; i <= nn; i++)
1796 nvme_validate_ns(ctrl, i);
1798 nvme_remove_invalid_namespaces(ctrl, nn);
1801 static void nvme_scan_work(struct work_struct *work)
1803 struct nvme_ctrl *ctrl =
1804 container_of(work, struct nvme_ctrl, scan_work);
1805 struct nvme_id_ctrl *id;
1808 if (ctrl->state != NVME_CTRL_LIVE)
1811 if (nvme_identify_ctrl(ctrl, &id))
1814 nn = le32_to_cpu(id->nn);
1815 if (ctrl->vs >= NVME_VS(1, 1) &&
1816 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1817 if (!nvme_scan_ns_list(ctrl, nn))
1820 nvme_scan_ns_sequential(ctrl, nn);
1822 mutex_lock(&ctrl->namespaces_mutex);
1823 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1824 mutex_unlock(&ctrl->namespaces_mutex);
1827 if (ctrl->ops->post_scan)
1828 ctrl->ops->post_scan(ctrl);
1831 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1834 * Do not queue new scan work when a controller is reset during
1837 if (ctrl->state == NVME_CTRL_LIVE)
1838 schedule_work(&ctrl->scan_work);
1840 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1843 * This function iterates the namespace list unlocked to allow recovery from
1844 * controller failure. It is up to the caller to ensure the namespace list is
1845 * not modified by scan work while this function is executing.
1847 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1849 struct nvme_ns *ns, *next;
1852 * The dead states indicates the controller was not gracefully
1853 * disconnected. In that case, we won't be able to flush any data while
1854 * removing the namespaces' disks; fail all the queues now to avoid
1855 * potentially having to clean up the failed sync later.
1857 if (ctrl->state == NVME_CTRL_DEAD)
1858 nvme_kill_queues(ctrl);
1860 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1863 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1865 static void nvme_async_event_work(struct work_struct *work)
1867 struct nvme_ctrl *ctrl =
1868 container_of(work, struct nvme_ctrl, async_event_work);
1870 spin_lock_irq(&ctrl->lock);
1871 while (ctrl->event_limit > 0) {
1872 int aer_idx = --ctrl->event_limit;
1874 spin_unlock_irq(&ctrl->lock);
1875 ctrl->ops->submit_async_event(ctrl, aer_idx);
1876 spin_lock_irq(&ctrl->lock);
1878 spin_unlock_irq(&ctrl->lock);
1881 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1882 struct nvme_completion *cqe)
1884 u16 status = le16_to_cpu(cqe->status) >> 1;
1885 u32 result = le32_to_cpu(cqe->result);
1887 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1888 ++ctrl->event_limit;
1889 schedule_work(&ctrl->async_event_work);
1892 if (status != NVME_SC_SUCCESS)
1895 switch (result & 0xff07) {
1896 case NVME_AER_NOTICE_NS_CHANGED:
1897 dev_info(ctrl->device, "rescanning\n");
1898 nvme_queue_scan(ctrl);
1901 dev_warn(ctrl->device, "async event result %08x\n", result);
1904 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1906 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1908 ctrl->event_limit = NVME_NR_AERS;
1909 schedule_work(&ctrl->async_event_work);
1911 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1913 static DEFINE_IDA(nvme_instance_ida);
1915 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1917 int instance, error;
1920 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1923 spin_lock(&dev_list_lock);
1924 error = ida_get_new(&nvme_instance_ida, &instance);
1925 spin_unlock(&dev_list_lock);
1926 } while (error == -EAGAIN);
1931 ctrl->instance = instance;
1935 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1937 spin_lock(&dev_list_lock);
1938 ida_remove(&nvme_instance_ida, ctrl->instance);
1939 spin_unlock(&dev_list_lock);
1942 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1944 flush_work(&ctrl->async_event_work);
1945 flush_work(&ctrl->scan_work);
1946 nvme_remove_namespaces(ctrl);
1948 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1950 spin_lock(&dev_list_lock);
1951 list_del(&ctrl->node);
1952 spin_unlock(&dev_list_lock);
1954 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1956 static void nvme_free_ctrl(struct kref *kref)
1958 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1960 put_device(ctrl->device);
1961 nvme_release_instance(ctrl);
1962 ida_destroy(&ctrl->ns_ida);
1964 ctrl->ops->free_ctrl(ctrl);
1967 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1969 kref_put(&ctrl->kref, nvme_free_ctrl);
1971 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1974 * Initialize a NVMe controller structures. This needs to be called during
1975 * earliest initialization so that we have the initialized structured around
1978 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1979 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1983 ctrl->state = NVME_CTRL_NEW;
1984 spin_lock_init(&ctrl->lock);
1985 INIT_LIST_HEAD(&ctrl->namespaces);
1986 mutex_init(&ctrl->namespaces_mutex);
1987 kref_init(&ctrl->kref);
1990 ctrl->quirks = quirks;
1991 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1992 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1994 ret = nvme_set_instance(ctrl);
1998 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
1999 MKDEV(nvme_char_major, ctrl->instance),
2000 ctrl, nvme_dev_attr_groups,
2001 "nvme%d", ctrl->instance);
2002 if (IS_ERR(ctrl->device)) {
2003 ret = PTR_ERR(ctrl->device);
2004 goto out_release_instance;
2006 get_device(ctrl->device);
2007 ida_init(&ctrl->ns_ida);
2009 spin_lock(&dev_list_lock);
2010 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2011 spin_unlock(&dev_list_lock);
2014 out_release_instance:
2015 nvme_release_instance(ctrl);
2019 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2022 * nvme_kill_queues(): Ends all namespace queues
2023 * @ctrl: the dead controller that needs to end
2025 * Call this function when the driver determines it is unable to get the
2026 * controller in a state capable of servicing IO.
2028 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2032 mutex_lock(&ctrl->namespaces_mutex);
2033 list_for_each_entry(ns, &ctrl->namespaces, list) {
2035 * Revalidating a dead namespace sets capacity to 0. This will
2036 * end buffered writers dirtying pages that can't be synced.
2038 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2039 revalidate_disk(ns->disk);
2041 blk_set_queue_dying(ns->queue);
2042 blk_mq_abort_requeue_list(ns->queue);
2043 blk_mq_start_stopped_hw_queues(ns->queue, true);
2045 mutex_unlock(&ctrl->namespaces_mutex);
2047 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2049 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2053 mutex_lock(&ctrl->namespaces_mutex);
2054 list_for_each_entry(ns, &ctrl->namespaces, list) {
2055 spin_lock_irq(ns->queue->queue_lock);
2056 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2057 spin_unlock_irq(ns->queue->queue_lock);
2059 blk_mq_cancel_requeue_work(ns->queue);
2060 blk_mq_stop_hw_queues(ns->queue);
2062 mutex_unlock(&ctrl->namespaces_mutex);
2064 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2066 void nvme_start_queues(struct nvme_ctrl *ctrl)
2070 mutex_lock(&ctrl->namespaces_mutex);
2071 list_for_each_entry(ns, &ctrl->namespaces, list) {
2072 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2073 blk_mq_start_stopped_hw_queues(ns->queue, true);
2074 blk_mq_kick_requeue_list(ns->queue);
2076 mutex_unlock(&ctrl->namespaces_mutex);
2078 EXPORT_SYMBOL_GPL(nvme_start_queues);
2080 int __init nvme_core_init(void)
2084 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2088 else if (result > 0)
2089 nvme_char_major = result;
2091 nvme_class = class_create(THIS_MODULE, "nvme");
2092 if (IS_ERR(nvme_class)) {
2093 result = PTR_ERR(nvme_class);
2094 goto unregister_chrdev;
2100 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2104 void nvme_core_exit(void)
2106 class_destroy(nvme_class);
2107 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2110 MODULE_LICENSE("GPL");
2111 MODULE_VERSION("1.0");
2112 module_init(nvme_core_init);
2113 module_exit(nvme_core_exit);