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);
159 if (ns->type == NVME_NS_LIGHTNVM)
160 nvme_nvm_unregister(ns->queue, ns->disk->disk_name);
162 spin_lock(&dev_list_lock);
163 ns->disk->private_data = NULL;
164 spin_unlock(&dev_list_lock);
167 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
168 nvme_put_ctrl(ns->ctrl);
172 static void nvme_put_ns(struct nvme_ns *ns)
174 kref_put(&ns->kref, nvme_free_ns);
177 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
181 spin_lock(&dev_list_lock);
182 ns = disk->private_data;
184 if (!kref_get_unless_zero(&ns->kref))
186 if (!try_module_get(ns->ctrl->ops->module))
189 spin_unlock(&dev_list_lock);
194 kref_put(&ns->kref, nvme_free_ns);
196 spin_unlock(&dev_list_lock);
200 void nvme_requeue_req(struct request *req)
204 blk_mq_requeue_request(req);
205 spin_lock_irqsave(req->q->queue_lock, flags);
206 if (!blk_queue_stopped(req->q))
207 blk_mq_kick_requeue_list(req->q);
208 spin_unlock_irqrestore(req->q->queue_lock, flags);
210 EXPORT_SYMBOL_GPL(nvme_requeue_req);
212 struct request *nvme_alloc_request(struct request_queue *q,
213 struct nvme_command *cmd, unsigned int flags, int qid)
217 if (qid == NVME_QID_ANY) {
218 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
220 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
226 req->cmd_type = REQ_TYPE_DRV_PRIV;
227 req->cmd_flags |= REQ_FAILFAST_DRIVER;
228 req->cmd = (unsigned char *)cmd;
229 req->cmd_len = sizeof(struct nvme_command);
233 EXPORT_SYMBOL_GPL(nvme_alloc_request);
235 static inline void nvme_setup_flush(struct nvme_ns *ns,
236 struct nvme_command *cmnd)
238 memset(cmnd, 0, sizeof(*cmnd));
239 cmnd->common.opcode = nvme_cmd_flush;
240 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
243 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
244 struct nvme_command *cmnd)
246 struct nvme_dsm_range *range;
249 unsigned int nr_bytes = blk_rq_bytes(req);
251 range = kmalloc(sizeof(*range), GFP_ATOMIC);
253 return BLK_MQ_RQ_QUEUE_BUSY;
255 range->cattr = cpu_to_le32(0);
256 range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
257 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
259 memset(cmnd, 0, sizeof(*cmnd));
260 cmnd->dsm.opcode = nvme_cmd_dsm;
261 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
263 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
265 req->completion_data = range;
266 page = virt_to_page(range);
267 offset = offset_in_page(range);
268 blk_add_request_payload(req, page, offset, sizeof(*range));
271 * we set __data_len back to the size of the area to be discarded
272 * on disk. This allows us to report completion on the full amount
273 * of blocks described by the request.
275 req->__data_len = nr_bytes;
280 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
281 struct nvme_command *cmnd)
286 if (req->cmd_flags & REQ_FUA)
287 control |= NVME_RW_FUA;
288 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
289 control |= NVME_RW_LR;
291 if (req->cmd_flags & REQ_RAHEAD)
292 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
294 memset(cmnd, 0, sizeof(*cmnd));
295 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
296 cmnd->rw.command_id = req->tag;
297 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
298 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
299 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
302 switch (ns->pi_type) {
303 case NVME_NS_DPS_PI_TYPE3:
304 control |= NVME_RW_PRINFO_PRCHK_GUARD;
306 case NVME_NS_DPS_PI_TYPE1:
307 case NVME_NS_DPS_PI_TYPE2:
308 control |= NVME_RW_PRINFO_PRCHK_GUARD |
309 NVME_RW_PRINFO_PRCHK_REF;
310 cmnd->rw.reftag = cpu_to_le32(
311 nvme_block_nr(ns, blk_rq_pos(req)));
314 if (!blk_integrity_rq(req))
315 control |= NVME_RW_PRINFO_PRACT;
318 cmnd->rw.control = cpu_to_le16(control);
319 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
322 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
323 struct nvme_command *cmd)
327 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
328 memcpy(cmd, req->cmd, sizeof(*cmd));
329 else if (req_op(req) == REQ_OP_FLUSH)
330 nvme_setup_flush(ns, cmd);
331 else if (req_op(req) == REQ_OP_DISCARD)
332 ret = nvme_setup_discard(ns, req, cmd);
334 nvme_setup_rw(ns, req, cmd);
338 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
341 * Returns 0 on success. If the result is negative, it's a Linux error code;
342 * if the result is positive, it's an NVM Express status code
344 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
345 struct nvme_completion *cqe, void *buffer, unsigned bufflen,
346 unsigned timeout, int qid, int at_head, int flags)
351 req = nvme_alloc_request(q, cmd, flags, qid);
355 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
358 if (buffer && bufflen) {
359 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
364 blk_execute_rq(req->q, NULL, req, at_head);
367 blk_mq_free_request(req);
370 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
372 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
373 void *buffer, unsigned bufflen)
375 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
378 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
380 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
381 void __user *ubuffer, unsigned bufflen,
382 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
383 u32 *result, unsigned timeout)
385 bool write = nvme_is_write(cmd);
386 struct nvme_completion cqe;
387 struct nvme_ns *ns = q->queuedata;
388 struct gendisk *disk = ns ? ns->disk : NULL;
390 struct bio *bio = NULL;
394 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
398 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
401 if (ubuffer && bufflen) {
402 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
410 bio->bi_bdev = bdget_disk(disk, 0);
416 if (meta_buffer && meta_len) {
417 struct bio_integrity_payload *bip;
419 meta = kmalloc(meta_len, GFP_KERNEL);
426 if (copy_from_user(meta, meta_buffer,
433 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
439 bip->bip_iter.bi_size = meta_len;
440 bip->bip_iter.bi_sector = meta_seed;
442 ret = bio_integrity_add_page(bio, virt_to_page(meta),
443 meta_len, offset_in_page(meta));
444 if (ret != meta_len) {
451 blk_execute_rq(req->q, disk, req, 0);
454 *result = le32_to_cpu(cqe.result);
455 if (meta && !ret && !write) {
456 if (copy_to_user(meta_buffer, meta, meta_len))
463 if (disk && bio->bi_bdev)
465 blk_rq_unmap_user(bio);
468 blk_mq_free_request(req);
472 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
473 void __user *ubuffer, unsigned bufflen, u32 *result,
476 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
480 static void nvme_keep_alive_end_io(struct request *rq, int error)
482 struct nvme_ctrl *ctrl = rq->end_io_data;
484 blk_mq_free_request(rq);
487 dev_err(ctrl->device,
488 "failed nvme_keep_alive_end_io error=%d\n", error);
492 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
495 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
497 struct nvme_command c;
500 memset(&c, 0, sizeof(c));
501 c.common.opcode = nvme_admin_keep_alive;
503 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
508 rq->timeout = ctrl->kato * HZ;
509 rq->end_io_data = ctrl;
511 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
516 static void nvme_keep_alive_work(struct work_struct *work)
518 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
519 struct nvme_ctrl, ka_work);
521 if (nvme_keep_alive(ctrl)) {
522 /* allocation failure, reset the controller */
523 dev_err(ctrl->device, "keep-alive failed\n");
524 ctrl->ops->reset_ctrl(ctrl);
529 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
531 if (unlikely(ctrl->kato == 0))
534 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
535 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
537 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
539 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
541 if (unlikely(ctrl->kato == 0))
544 cancel_delayed_work_sync(&ctrl->ka_work);
546 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
548 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
550 struct nvme_command c = { };
553 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
554 c.identify.opcode = nvme_admin_identify;
555 c.identify.cns = cpu_to_le32(1);
557 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
561 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
562 sizeof(struct nvme_id_ctrl));
568 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
570 struct nvme_command c = { };
572 c.identify.opcode = nvme_admin_identify;
573 c.identify.cns = cpu_to_le32(2);
574 c.identify.nsid = cpu_to_le32(nsid);
575 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
578 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
579 struct nvme_id_ns **id)
581 struct nvme_command c = { };
584 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
585 c.identify.opcode = nvme_admin_identify,
586 c.identify.nsid = cpu_to_le32(nsid),
588 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
592 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
593 sizeof(struct nvme_id_ns));
599 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
600 dma_addr_t dma_addr, u32 *result)
602 struct nvme_command c;
603 struct nvme_completion cqe;
606 memset(&c, 0, sizeof(c));
607 c.features.opcode = nvme_admin_get_features;
608 c.features.nsid = cpu_to_le32(nsid);
609 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
610 c.features.fid = cpu_to_le32(fid);
612 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
614 if (ret >= 0 && result)
615 *result = le32_to_cpu(cqe.result);
619 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
620 dma_addr_t dma_addr, u32 *result)
622 struct nvme_command c;
623 struct nvme_completion cqe;
626 memset(&c, 0, sizeof(c));
627 c.features.opcode = nvme_admin_set_features;
628 c.features.dptr.prp1 = cpu_to_le64(dma_addr);
629 c.features.fid = cpu_to_le32(fid);
630 c.features.dword11 = cpu_to_le32(dword11);
632 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &cqe, NULL, 0, 0,
634 if (ret >= 0 && result)
635 *result = le32_to_cpu(cqe.result);
639 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
641 struct nvme_command c = { };
644 c.common.opcode = nvme_admin_get_log_page,
645 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
646 c.common.cdw10[0] = cpu_to_le32(
647 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
650 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
654 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
655 sizeof(struct nvme_smart_log));
661 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
663 u32 q_count = (*count - 1) | ((*count - 1) << 16);
665 int status, nr_io_queues;
667 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, 0,
673 * Degraded controllers might return an error when setting the queue
674 * count. We still want to be able to bring them online and offer
675 * access to the admin queue, as that might be only way to fix them up.
678 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
681 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
682 *count = min(*count, nr_io_queues);
687 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
689 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
691 struct nvme_user_io io;
692 struct nvme_command c;
693 unsigned length, meta_len;
694 void __user *metadata;
696 if (copy_from_user(&io, uio, sizeof(io)))
704 case nvme_cmd_compare:
710 length = (io.nblocks + 1) << ns->lba_shift;
711 meta_len = (io.nblocks + 1) * ns->ms;
712 metadata = (void __user *)(uintptr_t)io.metadata;
717 } else if (meta_len) {
718 if ((io.metadata & 3) || !io.metadata)
722 memset(&c, 0, sizeof(c));
723 c.rw.opcode = io.opcode;
724 c.rw.flags = io.flags;
725 c.rw.nsid = cpu_to_le32(ns->ns_id);
726 c.rw.slba = cpu_to_le64(io.slba);
727 c.rw.length = cpu_to_le16(io.nblocks);
728 c.rw.control = cpu_to_le16(io.control);
729 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
730 c.rw.reftag = cpu_to_le32(io.reftag);
731 c.rw.apptag = cpu_to_le16(io.apptag);
732 c.rw.appmask = cpu_to_le16(io.appmask);
734 return __nvme_submit_user_cmd(ns->queue, &c,
735 (void __user *)(uintptr_t)io.addr, length,
736 metadata, meta_len, io.slba, NULL, 0);
739 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
740 struct nvme_passthru_cmd __user *ucmd)
742 struct nvme_passthru_cmd cmd;
743 struct nvme_command c;
744 unsigned timeout = 0;
747 if (!capable(CAP_SYS_ADMIN))
749 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
754 memset(&c, 0, sizeof(c));
755 c.common.opcode = cmd.opcode;
756 c.common.flags = cmd.flags;
757 c.common.nsid = cpu_to_le32(cmd.nsid);
758 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
759 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
760 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
761 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
762 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
763 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
764 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
765 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
768 timeout = msecs_to_jiffies(cmd.timeout_ms);
770 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
771 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
772 &cmd.result, timeout);
774 if (put_user(cmd.result, &ucmd->result))
781 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
782 unsigned int cmd, unsigned long arg)
784 struct nvme_ns *ns = bdev->bd_disk->private_data;
788 force_successful_syscall_return();
790 case NVME_IOCTL_ADMIN_CMD:
791 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
792 case NVME_IOCTL_IO_CMD:
793 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
794 case NVME_IOCTL_SUBMIT_IO:
795 return nvme_submit_io(ns, (void __user *)arg);
796 #ifdef CONFIG_BLK_DEV_NVME_SCSI
797 case SG_GET_VERSION_NUM:
798 return nvme_sg_get_version_num((void __user *)arg);
800 return nvme_sg_io(ns, (void __user *)arg);
808 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
809 unsigned int cmd, unsigned long arg)
815 return nvme_ioctl(bdev, mode, cmd, arg);
818 #define nvme_compat_ioctl NULL
821 static int nvme_open(struct block_device *bdev, fmode_t mode)
823 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
826 static void nvme_release(struct gendisk *disk, fmode_t mode)
828 struct nvme_ns *ns = disk->private_data;
830 module_put(ns->ctrl->ops->module);
834 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
836 /* some standard values */
838 geo->sectors = 1 << 5;
839 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
843 #ifdef CONFIG_BLK_DEV_INTEGRITY
844 static void nvme_init_integrity(struct nvme_ns *ns)
846 struct blk_integrity integrity;
848 memset(&integrity, 0, sizeof(integrity));
849 switch (ns->pi_type) {
850 case NVME_NS_DPS_PI_TYPE3:
851 integrity.profile = &t10_pi_type3_crc;
852 integrity.tag_size = sizeof(u16) + sizeof(u32);
853 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
855 case NVME_NS_DPS_PI_TYPE1:
856 case NVME_NS_DPS_PI_TYPE2:
857 integrity.profile = &t10_pi_type1_crc;
858 integrity.tag_size = sizeof(u16);
859 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
862 integrity.profile = NULL;
865 integrity.tuple_size = ns->ms;
866 blk_integrity_register(ns->disk, &integrity);
867 blk_queue_max_integrity_segments(ns->queue, 1);
870 static void nvme_init_integrity(struct nvme_ns *ns)
873 #endif /* CONFIG_BLK_DEV_INTEGRITY */
875 static void nvme_config_discard(struct nvme_ns *ns)
877 struct nvme_ctrl *ctrl = ns->ctrl;
878 u32 logical_block_size = queue_logical_block_size(ns->queue);
880 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
881 ns->queue->limits.discard_zeroes_data = 1;
883 ns->queue->limits.discard_zeroes_data = 0;
885 ns->queue->limits.discard_alignment = logical_block_size;
886 ns->queue->limits.discard_granularity = logical_block_size;
887 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
888 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
891 static int nvme_revalidate_disk(struct gendisk *disk)
893 struct nvme_ns *ns = disk->private_data;
894 struct nvme_id_ns *id;
899 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
900 set_capacity(disk, 0);
903 if (nvme_identify_ns(ns->ctrl, ns->ns_id, &id)) {
904 dev_warn(disk_to_dev(ns->disk), "%s: Identify failure\n",
913 if (nvme_nvm_ns_supported(ns, id) && ns->type != NVME_NS_LIGHTNVM) {
914 if (nvme_nvm_register(ns->queue, disk->disk_name)) {
915 dev_warn(disk_to_dev(ns->disk),
916 "%s: LightNVM init failure\n", __func__);
920 ns->type = NVME_NS_LIGHTNVM;
923 if (ns->ctrl->vs >= NVME_VS(1, 1))
924 memcpy(ns->eui, id->eui64, sizeof(ns->eui));
925 if (ns->ctrl->vs >= NVME_VS(1, 2))
926 memcpy(ns->uuid, id->nguid, sizeof(ns->uuid));
929 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
930 ns->lba_shift = id->lbaf[lbaf].ds;
931 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
932 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
935 * If identify namespace failed, use default 512 byte block size so
936 * block layer can use before failing read/write for 0 capacity.
938 if (ns->lba_shift == 0)
940 bs = 1 << ns->lba_shift;
941 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
942 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
943 id->dps & NVME_NS_DPS_PI_MASK : 0;
945 blk_mq_freeze_queue(disk->queue);
946 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
948 bs != queue_logical_block_size(disk->queue) ||
949 (ns->ms && ns->ext)))
950 blk_integrity_unregister(disk);
952 ns->pi_type = pi_type;
953 blk_queue_logical_block_size(ns->queue, bs);
955 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
956 nvme_init_integrity(ns);
957 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
958 set_capacity(disk, 0);
960 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
962 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
963 nvme_config_discard(ns);
964 blk_mq_unfreeze_queue(disk->queue);
970 static char nvme_pr_type(enum pr_type type)
973 case PR_WRITE_EXCLUSIVE:
975 case PR_EXCLUSIVE_ACCESS:
977 case PR_WRITE_EXCLUSIVE_REG_ONLY:
979 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
981 case PR_WRITE_EXCLUSIVE_ALL_REGS:
983 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
990 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
991 u64 key, u64 sa_key, u8 op)
993 struct nvme_ns *ns = bdev->bd_disk->private_data;
994 struct nvme_command c;
995 u8 data[16] = { 0, };
997 put_unaligned_le64(key, &data[0]);
998 put_unaligned_le64(sa_key, &data[8]);
1000 memset(&c, 0, sizeof(c));
1001 c.common.opcode = op;
1002 c.common.nsid = cpu_to_le32(ns->ns_id);
1003 c.common.cdw10[0] = cpu_to_le32(cdw10);
1005 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1008 static int nvme_pr_register(struct block_device *bdev, u64 old,
1009 u64 new, unsigned flags)
1013 if (flags & ~PR_FL_IGNORE_KEY)
1016 cdw10 = old ? 2 : 0;
1017 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1018 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1019 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1022 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1023 enum pr_type type, unsigned flags)
1027 if (flags & ~PR_FL_IGNORE_KEY)
1030 cdw10 = nvme_pr_type(type) << 8;
1031 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1032 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1035 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1036 enum pr_type type, bool abort)
1038 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1039 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1042 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1044 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1045 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1048 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1050 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1051 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1054 static const struct pr_ops nvme_pr_ops = {
1055 .pr_register = nvme_pr_register,
1056 .pr_reserve = nvme_pr_reserve,
1057 .pr_release = nvme_pr_release,
1058 .pr_preempt = nvme_pr_preempt,
1059 .pr_clear = nvme_pr_clear,
1062 static const struct block_device_operations nvme_fops = {
1063 .owner = THIS_MODULE,
1064 .ioctl = nvme_ioctl,
1065 .compat_ioctl = nvme_compat_ioctl,
1067 .release = nvme_release,
1068 .getgeo = nvme_getgeo,
1069 .revalidate_disk= nvme_revalidate_disk,
1070 .pr_ops = &nvme_pr_ops,
1073 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1075 unsigned long timeout =
1076 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1077 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1080 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1081 if ((csts & NVME_CSTS_RDY) == bit)
1085 if (fatal_signal_pending(current))
1087 if (time_after(jiffies, timeout)) {
1088 dev_err(ctrl->device,
1089 "Device not ready; aborting %s\n", enabled ?
1090 "initialisation" : "reset");
1099 * If the device has been passed off to us in an enabled state, just clear
1100 * the enabled bit. The spec says we should set the 'shutdown notification
1101 * bits', but doing so may cause the device to complete commands to the
1102 * admin queue ... and we don't know what memory that might be pointing at!
1104 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1108 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1109 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1111 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1115 /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1116 * load, since we only need the quirk on reset_controller. Notice
1117 * that the HGST device needs this delay only in firmware activation
1118 * procedure; unfortunately we have no (easy) way to verify this.
1120 if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1121 msleep(NVME_QUIRK_DELAY_AMOUNT);
1123 return nvme_wait_ready(ctrl, cap, false);
1125 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1127 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1130 * Default to a 4K page size, with the intention to update this
1131 * path in the future to accomodate architectures with differing
1132 * kernel and IO page sizes.
1134 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1137 if (page_shift < dev_page_min) {
1138 dev_err(ctrl->device,
1139 "Minimum device page size %u too large for host (%u)\n",
1140 1 << dev_page_min, 1 << page_shift);
1144 ctrl->page_size = 1 << page_shift;
1146 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1147 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1148 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1149 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1150 ctrl->ctrl_config |= NVME_CC_ENABLE;
1152 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1155 return nvme_wait_ready(ctrl, cap, true);
1157 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1159 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1161 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1165 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1166 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1168 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1172 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1173 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1177 if (fatal_signal_pending(current))
1179 if (time_after(jiffies, timeout)) {
1180 dev_err(ctrl->device,
1181 "Device shutdown incomplete; abort shutdown\n");
1188 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1190 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1191 struct request_queue *q)
1195 if (ctrl->max_hw_sectors) {
1197 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1199 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1200 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1202 if (ctrl->stripe_size)
1203 blk_queue_chunk_sectors(q, ctrl->stripe_size >> 9);
1204 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1205 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1207 blk_queue_write_cache(q, vwc, vwc);
1211 * Initialize the cached copies of the Identify data and various controller
1212 * register in our nvme_ctrl structure. This should be called as soon as
1213 * the admin queue is fully up and running.
1215 int nvme_init_identify(struct nvme_ctrl *ctrl)
1217 struct nvme_id_ctrl *id;
1219 int ret, page_shift;
1222 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1224 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1228 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1230 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1233 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1235 if (ctrl->vs >= NVME_VS(1, 1))
1236 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1238 ret = nvme_identify_ctrl(ctrl, &id);
1240 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1244 ctrl->vid = le16_to_cpu(id->vid);
1245 ctrl->oncs = le16_to_cpup(&id->oncs);
1246 atomic_set(&ctrl->abort_limit, id->acl + 1);
1247 ctrl->vwc = id->vwc;
1248 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1249 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1250 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1251 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1253 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1255 max_hw_sectors = UINT_MAX;
1256 ctrl->max_hw_sectors =
1257 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1259 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) && id->vs[3]) {
1260 unsigned int max_hw_sectors;
1262 ctrl->stripe_size = 1 << (id->vs[3] + page_shift);
1263 max_hw_sectors = ctrl->stripe_size >> (page_shift - 9);
1264 if (ctrl->max_hw_sectors) {
1265 ctrl->max_hw_sectors = min(max_hw_sectors,
1266 ctrl->max_hw_sectors);
1268 ctrl->max_hw_sectors = max_hw_sectors;
1272 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1273 ctrl->sgls = le32_to_cpu(id->sgls);
1274 ctrl->kas = le16_to_cpu(id->kas);
1276 if (ctrl->ops->is_fabrics) {
1277 ctrl->icdoff = le16_to_cpu(id->icdoff);
1278 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1279 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1280 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1283 * In fabrics we need to verify the cntlid matches the
1286 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1289 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1291 "keep-alive support is mandatory for fabrics\n");
1295 ctrl->cntlid = le16_to_cpu(id->cntlid);
1301 EXPORT_SYMBOL_GPL(nvme_init_identify);
1303 static int nvme_dev_open(struct inode *inode, struct file *file)
1305 struct nvme_ctrl *ctrl;
1306 int instance = iminor(inode);
1309 spin_lock(&dev_list_lock);
1310 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1311 if (ctrl->instance != instance)
1314 if (!ctrl->admin_q) {
1318 if (!kref_get_unless_zero(&ctrl->kref))
1320 file->private_data = ctrl;
1324 spin_unlock(&dev_list_lock);
1329 static int nvme_dev_release(struct inode *inode, struct file *file)
1331 nvme_put_ctrl(file->private_data);
1335 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1340 mutex_lock(&ctrl->namespaces_mutex);
1341 if (list_empty(&ctrl->namespaces)) {
1346 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1347 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1348 dev_warn(ctrl->device,
1349 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1354 dev_warn(ctrl->device,
1355 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1356 kref_get(&ns->kref);
1357 mutex_unlock(&ctrl->namespaces_mutex);
1359 ret = nvme_user_cmd(ctrl, ns, argp);
1364 mutex_unlock(&ctrl->namespaces_mutex);
1368 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1371 struct nvme_ctrl *ctrl = file->private_data;
1372 void __user *argp = (void __user *)arg;
1375 case NVME_IOCTL_ADMIN_CMD:
1376 return nvme_user_cmd(ctrl, NULL, argp);
1377 case NVME_IOCTL_IO_CMD:
1378 return nvme_dev_user_cmd(ctrl, argp);
1379 case NVME_IOCTL_RESET:
1380 dev_warn(ctrl->device, "resetting controller\n");
1381 return ctrl->ops->reset_ctrl(ctrl);
1382 case NVME_IOCTL_SUBSYS_RESET:
1383 return nvme_reset_subsystem(ctrl);
1384 case NVME_IOCTL_RESCAN:
1385 nvme_queue_scan(ctrl);
1392 static const struct file_operations nvme_dev_fops = {
1393 .owner = THIS_MODULE,
1394 .open = nvme_dev_open,
1395 .release = nvme_dev_release,
1396 .unlocked_ioctl = nvme_dev_ioctl,
1397 .compat_ioctl = nvme_dev_ioctl,
1400 static ssize_t nvme_sysfs_reset(struct device *dev,
1401 struct device_attribute *attr, const char *buf,
1404 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1407 ret = ctrl->ops->reset_ctrl(ctrl);
1412 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1414 static ssize_t nvme_sysfs_rescan(struct device *dev,
1415 struct device_attribute *attr, const char *buf,
1418 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1420 nvme_queue_scan(ctrl);
1423 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1425 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1428 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1429 struct nvme_ctrl *ctrl = ns->ctrl;
1430 int serial_len = sizeof(ctrl->serial);
1431 int model_len = sizeof(ctrl->model);
1433 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1434 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1436 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1437 return sprintf(buf, "eui.%8phN\n", ns->eui);
1439 while (ctrl->serial[serial_len - 1] == ' ')
1441 while (ctrl->model[model_len - 1] == ' ')
1444 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1445 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1447 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1449 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1452 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1453 return sprintf(buf, "%pU\n", ns->uuid);
1455 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1457 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1460 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1461 return sprintf(buf, "%8phd\n", ns->eui);
1463 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1465 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1468 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1469 return sprintf(buf, "%d\n", ns->ns_id);
1471 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1473 static struct attribute *nvme_ns_attrs[] = {
1474 &dev_attr_wwid.attr,
1475 &dev_attr_uuid.attr,
1477 &dev_attr_nsid.attr,
1481 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1482 struct attribute *a, int n)
1484 struct device *dev = container_of(kobj, struct device, kobj);
1485 struct nvme_ns *ns = dev_to_disk(dev)->private_data;
1487 if (a == &dev_attr_uuid.attr) {
1488 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1491 if (a == &dev_attr_eui.attr) {
1492 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1498 static const struct attribute_group nvme_ns_attr_group = {
1499 .attrs = nvme_ns_attrs,
1500 .is_visible = nvme_ns_attrs_are_visible,
1503 #define nvme_show_str_function(field) \
1504 static ssize_t field##_show(struct device *dev, \
1505 struct device_attribute *attr, char *buf) \
1507 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1508 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1510 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1512 #define nvme_show_int_function(field) \
1513 static ssize_t field##_show(struct device *dev, \
1514 struct device_attribute *attr, char *buf) \
1516 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1517 return sprintf(buf, "%d\n", ctrl->field); \
1519 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1521 nvme_show_str_function(model);
1522 nvme_show_str_function(serial);
1523 nvme_show_str_function(firmware_rev);
1524 nvme_show_int_function(cntlid);
1526 static ssize_t nvme_sysfs_delete(struct device *dev,
1527 struct device_attribute *attr, const char *buf,
1530 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1532 if (device_remove_file_self(dev, attr))
1533 ctrl->ops->delete_ctrl(ctrl);
1536 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1538 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1539 struct device_attribute *attr,
1542 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1544 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1546 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1548 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1549 struct device_attribute *attr,
1552 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1554 return snprintf(buf, PAGE_SIZE, "%s\n",
1555 ctrl->ops->get_subsysnqn(ctrl));
1557 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1559 static ssize_t nvme_sysfs_show_address(struct device *dev,
1560 struct device_attribute *attr,
1563 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1565 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1567 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1569 static struct attribute *nvme_dev_attrs[] = {
1570 &dev_attr_reset_controller.attr,
1571 &dev_attr_rescan_controller.attr,
1572 &dev_attr_model.attr,
1573 &dev_attr_serial.attr,
1574 &dev_attr_firmware_rev.attr,
1575 &dev_attr_cntlid.attr,
1576 &dev_attr_delete_controller.attr,
1577 &dev_attr_transport.attr,
1578 &dev_attr_subsysnqn.attr,
1579 &dev_attr_address.attr,
1583 #define CHECK_ATTR(ctrl, a, name) \
1584 if ((a) == &dev_attr_##name.attr && \
1585 !(ctrl)->ops->get_##name) \
1588 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1589 struct attribute *a, int n)
1591 struct device *dev = container_of(kobj, struct device, kobj);
1592 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1594 if (a == &dev_attr_delete_controller.attr) {
1595 if (!ctrl->ops->delete_ctrl)
1599 CHECK_ATTR(ctrl, a, subsysnqn);
1600 CHECK_ATTR(ctrl, a, address);
1605 static struct attribute_group nvme_dev_attrs_group = {
1606 .attrs = nvme_dev_attrs,
1607 .is_visible = nvme_dev_attrs_are_visible,
1610 static const struct attribute_group *nvme_dev_attr_groups[] = {
1611 &nvme_dev_attrs_group,
1615 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1617 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1618 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1620 return nsa->ns_id - nsb->ns_id;
1623 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1625 struct nvme_ns *ns, *ret = NULL;
1627 mutex_lock(&ctrl->namespaces_mutex);
1628 list_for_each_entry(ns, &ctrl->namespaces, list) {
1629 if (ns->ns_id == nsid) {
1630 kref_get(&ns->kref);
1634 if (ns->ns_id > nsid)
1637 mutex_unlock(&ctrl->namespaces_mutex);
1641 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1644 struct gendisk *disk;
1645 int node = dev_to_node(ctrl->dev);
1647 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1651 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1652 if (ns->instance < 0)
1655 ns->queue = blk_mq_init_queue(ctrl->tagset);
1656 if (IS_ERR(ns->queue))
1657 goto out_release_instance;
1658 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1659 ns->queue->queuedata = ns;
1662 disk = alloc_disk_node(0, node);
1664 goto out_free_queue;
1666 kref_init(&ns->kref);
1669 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1672 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1673 nvme_set_queue_limits(ctrl, ns->queue);
1675 disk->fops = &nvme_fops;
1676 disk->private_data = ns;
1677 disk->queue = ns->queue;
1678 disk->flags = GENHD_FL_EXT_DEVT;
1679 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1681 if (nvme_revalidate_disk(ns->disk))
1684 mutex_lock(&ctrl->namespaces_mutex);
1685 list_add_tail(&ns->list, &ctrl->namespaces);
1686 mutex_unlock(&ctrl->namespaces_mutex);
1688 kref_get(&ctrl->kref);
1689 if (ns->type == NVME_NS_LIGHTNVM)
1692 device_add_disk(ctrl->device, ns->disk);
1693 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1694 &nvme_ns_attr_group))
1695 pr_warn("%s: failed to create sysfs group for identification\n",
1696 ns->disk->disk_name);
1701 blk_cleanup_queue(ns->queue);
1702 out_release_instance:
1703 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1708 static void nvme_ns_remove(struct nvme_ns *ns)
1710 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1713 if (ns->disk->flags & GENHD_FL_UP) {
1714 if (blk_get_integrity(ns->disk))
1715 blk_integrity_unregister(ns->disk);
1716 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1717 &nvme_ns_attr_group);
1718 del_gendisk(ns->disk);
1719 blk_mq_abort_requeue_list(ns->queue);
1720 blk_cleanup_queue(ns->queue);
1723 mutex_lock(&ns->ctrl->namespaces_mutex);
1724 list_del_init(&ns->list);
1725 mutex_unlock(&ns->ctrl->namespaces_mutex);
1730 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1734 ns = nvme_find_get_ns(ctrl, nsid);
1736 if (revalidate_disk(ns->disk))
1740 nvme_alloc_ns(ctrl, nsid);
1743 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1746 struct nvme_ns *ns, *next;
1748 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1749 if (ns->ns_id > nsid)
1754 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1758 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1761 ns_list = kzalloc(0x1000, GFP_KERNEL);
1765 for (i = 0; i < num_lists; i++) {
1766 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1770 for (j = 0; j < min(nn, 1024U); j++) {
1771 nsid = le32_to_cpu(ns_list[j]);
1775 nvme_validate_ns(ctrl, nsid);
1777 while (++prev < nsid) {
1778 ns = nvme_find_get_ns(ctrl, prev);
1788 nvme_remove_invalid_namespaces(ctrl, prev);
1794 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1798 for (i = 1; i <= nn; i++)
1799 nvme_validate_ns(ctrl, i);
1801 nvme_remove_invalid_namespaces(ctrl, nn);
1804 static void nvme_scan_work(struct work_struct *work)
1806 struct nvme_ctrl *ctrl =
1807 container_of(work, struct nvme_ctrl, scan_work);
1808 struct nvme_id_ctrl *id;
1811 if (ctrl->state != NVME_CTRL_LIVE)
1814 if (nvme_identify_ctrl(ctrl, &id))
1817 nn = le32_to_cpu(id->nn);
1818 if (ctrl->vs >= NVME_VS(1, 1) &&
1819 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1820 if (!nvme_scan_ns_list(ctrl, nn))
1823 nvme_scan_ns_sequential(ctrl, nn);
1825 mutex_lock(&ctrl->namespaces_mutex);
1826 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1827 mutex_unlock(&ctrl->namespaces_mutex);
1830 if (ctrl->ops->post_scan)
1831 ctrl->ops->post_scan(ctrl);
1834 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1837 * Do not queue new scan work when a controller is reset during
1840 if (ctrl->state == NVME_CTRL_LIVE)
1841 schedule_work(&ctrl->scan_work);
1843 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1846 * This function iterates the namespace list unlocked to allow recovery from
1847 * controller failure. It is up to the caller to ensure the namespace list is
1848 * not modified by scan work while this function is executing.
1850 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1852 struct nvme_ns *ns, *next;
1855 * The dead states indicates the controller was not gracefully
1856 * disconnected. In that case, we won't be able to flush any data while
1857 * removing the namespaces' disks; fail all the queues now to avoid
1858 * potentially having to clean up the failed sync later.
1860 if (ctrl->state == NVME_CTRL_DEAD)
1861 nvme_kill_queues(ctrl);
1863 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1866 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1868 static void nvme_async_event_work(struct work_struct *work)
1870 struct nvme_ctrl *ctrl =
1871 container_of(work, struct nvme_ctrl, async_event_work);
1873 spin_lock_irq(&ctrl->lock);
1874 while (ctrl->event_limit > 0) {
1875 int aer_idx = --ctrl->event_limit;
1877 spin_unlock_irq(&ctrl->lock);
1878 ctrl->ops->submit_async_event(ctrl, aer_idx);
1879 spin_lock_irq(&ctrl->lock);
1881 spin_unlock_irq(&ctrl->lock);
1884 void nvme_complete_async_event(struct nvme_ctrl *ctrl,
1885 struct nvme_completion *cqe)
1887 u16 status = le16_to_cpu(cqe->status) >> 1;
1888 u32 result = le32_to_cpu(cqe->result);
1890 if (status == NVME_SC_SUCCESS || status == NVME_SC_ABORT_REQ) {
1891 ++ctrl->event_limit;
1892 schedule_work(&ctrl->async_event_work);
1895 if (status != NVME_SC_SUCCESS)
1898 switch (result & 0xff07) {
1899 case NVME_AER_NOTICE_NS_CHANGED:
1900 dev_info(ctrl->device, "rescanning\n");
1901 nvme_queue_scan(ctrl);
1904 dev_warn(ctrl->device, "async event result %08x\n", result);
1907 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1909 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1911 ctrl->event_limit = NVME_NR_AERS;
1912 schedule_work(&ctrl->async_event_work);
1914 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1916 static DEFINE_IDA(nvme_instance_ida);
1918 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1920 int instance, error;
1923 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1926 spin_lock(&dev_list_lock);
1927 error = ida_get_new(&nvme_instance_ida, &instance);
1928 spin_unlock(&dev_list_lock);
1929 } while (error == -EAGAIN);
1934 ctrl->instance = instance;
1938 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1940 spin_lock(&dev_list_lock);
1941 ida_remove(&nvme_instance_ida, ctrl->instance);
1942 spin_unlock(&dev_list_lock);
1945 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1947 flush_work(&ctrl->async_event_work);
1948 flush_work(&ctrl->scan_work);
1949 nvme_remove_namespaces(ctrl);
1951 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1953 spin_lock(&dev_list_lock);
1954 list_del(&ctrl->node);
1955 spin_unlock(&dev_list_lock);
1957 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1959 static void nvme_free_ctrl(struct kref *kref)
1961 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1963 put_device(ctrl->device);
1964 nvme_release_instance(ctrl);
1965 ida_destroy(&ctrl->ns_ida);
1967 ctrl->ops->free_ctrl(ctrl);
1970 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1972 kref_put(&ctrl->kref, nvme_free_ctrl);
1974 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
1977 * Initialize a NVMe controller structures. This needs to be called during
1978 * earliest initialization so that we have the initialized structured around
1981 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
1982 const struct nvme_ctrl_ops *ops, unsigned long quirks)
1986 ctrl->state = NVME_CTRL_NEW;
1987 spin_lock_init(&ctrl->lock);
1988 INIT_LIST_HEAD(&ctrl->namespaces);
1989 mutex_init(&ctrl->namespaces_mutex);
1990 kref_init(&ctrl->kref);
1993 ctrl->quirks = quirks;
1994 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
1995 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
1997 ret = nvme_set_instance(ctrl);
2001 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2002 MKDEV(nvme_char_major, ctrl->instance),
2003 ctrl, nvme_dev_attr_groups,
2004 "nvme%d", ctrl->instance);
2005 if (IS_ERR(ctrl->device)) {
2006 ret = PTR_ERR(ctrl->device);
2007 goto out_release_instance;
2009 get_device(ctrl->device);
2010 ida_init(&ctrl->ns_ida);
2012 spin_lock(&dev_list_lock);
2013 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2014 spin_unlock(&dev_list_lock);
2017 out_release_instance:
2018 nvme_release_instance(ctrl);
2022 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2025 * nvme_kill_queues(): Ends all namespace queues
2026 * @ctrl: the dead controller that needs to end
2028 * Call this function when the driver determines it is unable to get the
2029 * controller in a state capable of servicing IO.
2031 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2035 mutex_lock(&ctrl->namespaces_mutex);
2036 list_for_each_entry(ns, &ctrl->namespaces, list) {
2038 * Revalidating a dead namespace sets capacity to 0. This will
2039 * end buffered writers dirtying pages that can't be synced.
2041 if (!test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2042 revalidate_disk(ns->disk);
2044 blk_set_queue_dying(ns->queue);
2045 blk_mq_abort_requeue_list(ns->queue);
2046 blk_mq_start_stopped_hw_queues(ns->queue, true);
2048 mutex_unlock(&ctrl->namespaces_mutex);
2050 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2052 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2056 mutex_lock(&ctrl->namespaces_mutex);
2057 list_for_each_entry(ns, &ctrl->namespaces, list) {
2058 spin_lock_irq(ns->queue->queue_lock);
2059 queue_flag_set(QUEUE_FLAG_STOPPED, ns->queue);
2060 spin_unlock_irq(ns->queue->queue_lock);
2062 blk_mq_cancel_requeue_work(ns->queue);
2063 blk_mq_stop_hw_queues(ns->queue);
2065 mutex_unlock(&ctrl->namespaces_mutex);
2067 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2069 void nvme_start_queues(struct nvme_ctrl *ctrl)
2073 mutex_lock(&ctrl->namespaces_mutex);
2074 list_for_each_entry(ns, &ctrl->namespaces, list) {
2075 queue_flag_clear_unlocked(QUEUE_FLAG_STOPPED, ns->queue);
2076 blk_mq_start_stopped_hw_queues(ns->queue, true);
2077 blk_mq_kick_requeue_list(ns->queue);
2079 mutex_unlock(&ctrl->namespaces_mutex);
2081 EXPORT_SYMBOL_GPL(nvme_start_queues);
2083 int __init nvme_core_init(void)
2087 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2091 else if (result > 0)
2092 nvme_char_major = result;
2094 nvme_class = class_create(THIS_MODULE, "nvme");
2095 if (IS_ERR(nvme_class)) {
2096 result = PTR_ERR(nvme_class);
2097 goto unregister_chrdev;
2103 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2107 void nvme_core_exit(void)
2109 class_destroy(nvme_class);
2110 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2113 MODULE_LICENSE("GPL");
2114 MODULE_VERSION("1.0");
2115 module_init(nvme_core_init);
2116 module_exit(nvme_core_exit);