2 * NVM Express device driver
3 * Copyright (c) 2011-2014, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
15 #include <linux/blkdev.h>
16 #include <linux/blk-mq.h>
17 #include <linux/delay.h>
18 #include <linux/errno.h>
19 #include <linux/hdreg.h>
20 #include <linux/kernel.h>
21 #include <linux/module.h>
22 #include <linux/list_sort.h>
23 #include <linux/slab.h>
24 #include <linux/types.h>
26 #include <linux/ptrace.h>
27 #include <linux/nvme_ioctl.h>
28 #include <linux/t10-pi.h>
30 #include <asm/unaligned.h>
35 #define NVME_MINORS (1U << MINORBITS)
37 unsigned char admin_timeout = 60;
38 module_param(admin_timeout, byte, 0644);
39 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
40 EXPORT_SYMBOL_GPL(admin_timeout);
42 unsigned char nvme_io_timeout = 30;
43 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
44 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
45 EXPORT_SYMBOL_GPL(nvme_io_timeout);
47 unsigned char shutdown_timeout = 5;
48 module_param(shutdown_timeout, byte, 0644);
49 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
51 unsigned int nvme_max_retries = 5;
52 module_param_named(max_retries, nvme_max_retries, uint, 0644);
53 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
54 EXPORT_SYMBOL_GPL(nvme_max_retries);
56 static int nvme_char_major;
57 module_param(nvme_char_major, int, 0);
59 static LIST_HEAD(nvme_ctrl_list);
60 static DEFINE_SPINLOCK(dev_list_lock);
62 static struct class *nvme_class;
64 void nvme_cancel_request(struct request *req, void *data, bool reserved)
68 if (!blk_mq_request_started(req))
71 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
72 "Cancelling I/O %d", req->tag);
74 status = NVME_SC_ABORT_REQ;
75 if (blk_queue_dying(req->q))
76 status |= NVME_SC_DNR;
77 blk_mq_complete_request(req, status);
79 EXPORT_SYMBOL_GPL(nvme_cancel_request);
81 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
82 enum nvme_ctrl_state new_state)
84 enum nvme_ctrl_state old_state;
87 spin_lock_irq(&ctrl->lock);
89 old_state = ctrl->state;
94 case NVME_CTRL_RESETTING:
95 case NVME_CTRL_RECONNECTING:
102 case NVME_CTRL_RESETTING:
106 case NVME_CTRL_RECONNECTING:
113 case NVME_CTRL_RECONNECTING:
122 case NVME_CTRL_DELETING:
125 case NVME_CTRL_RESETTING:
126 case NVME_CTRL_RECONNECTING:
135 case NVME_CTRL_DELETING:
147 ctrl->state = new_state;
149 spin_unlock_irq(&ctrl->lock);
153 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
155 static void nvme_free_ns(struct kref *kref)
157 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
160 nvme_nvm_unregister(ns);
163 spin_lock(&dev_list_lock);
164 ns->disk->private_data = NULL;
165 spin_unlock(&dev_list_lock);
169 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
170 nvme_put_ctrl(ns->ctrl);
174 static void nvme_put_ns(struct nvme_ns *ns)
176 kref_put(&ns->kref, nvme_free_ns);
179 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
183 spin_lock(&dev_list_lock);
184 ns = disk->private_data;
186 if (!kref_get_unless_zero(&ns->kref))
188 if (!try_module_get(ns->ctrl->ops->module))
191 spin_unlock(&dev_list_lock);
196 kref_put(&ns->kref, nvme_free_ns);
198 spin_unlock(&dev_list_lock);
202 void nvme_requeue_req(struct request *req)
204 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
206 EXPORT_SYMBOL_GPL(nvme_requeue_req);
208 struct request *nvme_alloc_request(struct request_queue *q,
209 struct nvme_command *cmd, unsigned int flags, int qid)
213 if (qid == NVME_QID_ANY) {
214 req = blk_mq_alloc_request(q, nvme_is_write(cmd), flags);
216 req = blk_mq_alloc_request_hctx(q, nvme_is_write(cmd), flags,
222 req->cmd_type = REQ_TYPE_DRV_PRIV;
223 req->cmd_flags |= REQ_FAILFAST_DRIVER;
224 nvme_req(req)->cmd = cmd;
228 EXPORT_SYMBOL_GPL(nvme_alloc_request);
230 static inline void nvme_setup_flush(struct nvme_ns *ns,
231 struct nvme_command *cmnd)
233 memset(cmnd, 0, sizeof(*cmnd));
234 cmnd->common.opcode = nvme_cmd_flush;
235 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
238 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
239 struct nvme_command *cmnd)
241 struct nvme_dsm_range *range;
242 unsigned int nr_bytes = blk_rq_bytes(req);
244 range = kmalloc(sizeof(*range), GFP_ATOMIC);
246 return BLK_MQ_RQ_QUEUE_BUSY;
248 range->cattr = cpu_to_le32(0);
249 range->nlb = cpu_to_le32(nr_bytes >> ns->lba_shift);
250 range->slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
252 memset(cmnd, 0, sizeof(*cmnd));
253 cmnd->dsm.opcode = nvme_cmd_dsm;
254 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
256 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
258 req->special_vec.bv_page = virt_to_page(range);
259 req->special_vec.bv_offset = offset_in_page(range);
260 req->special_vec.bv_len = sizeof(*range);
261 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
263 return BLK_MQ_RQ_QUEUE_OK;
266 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
267 struct nvme_command *cmnd)
272 if (req->cmd_flags & REQ_FUA)
273 control |= NVME_RW_FUA;
274 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
275 control |= NVME_RW_LR;
277 if (req->cmd_flags & REQ_RAHEAD)
278 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
280 memset(cmnd, 0, sizeof(*cmnd));
281 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
282 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
283 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
284 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
287 switch (ns->pi_type) {
288 case NVME_NS_DPS_PI_TYPE3:
289 control |= NVME_RW_PRINFO_PRCHK_GUARD;
291 case NVME_NS_DPS_PI_TYPE1:
292 case NVME_NS_DPS_PI_TYPE2:
293 control |= NVME_RW_PRINFO_PRCHK_GUARD |
294 NVME_RW_PRINFO_PRCHK_REF;
295 cmnd->rw.reftag = cpu_to_le32(
296 nvme_block_nr(ns, blk_rq_pos(req)));
299 if (!blk_integrity_rq(req))
300 control |= NVME_RW_PRINFO_PRACT;
303 cmnd->rw.control = cpu_to_le16(control);
304 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
307 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
308 struct nvme_command *cmd)
310 int ret = BLK_MQ_RQ_QUEUE_OK;
312 if (req->cmd_type == REQ_TYPE_DRV_PRIV)
313 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
314 else if (req_op(req) == REQ_OP_FLUSH)
315 nvme_setup_flush(ns, cmd);
316 else if (req_op(req) == REQ_OP_DISCARD)
317 ret = nvme_setup_discard(ns, req, cmd);
319 nvme_setup_rw(ns, req, cmd);
321 cmd->common.command_id = req->tag;
325 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
328 * Returns 0 on success. If the result is negative, it's a Linux error code;
329 * if the result is positive, it's an NVM Express status code
331 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
332 union nvme_result *result, void *buffer, unsigned bufflen,
333 unsigned timeout, int qid, int at_head, int flags)
338 req = nvme_alloc_request(q, cmd, flags, qid);
342 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
344 if (buffer && bufflen) {
345 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
350 blk_execute_rq(req->q, NULL, req, at_head);
352 *result = nvme_req(req)->result;
355 blk_mq_free_request(req);
358 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
360 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
361 void *buffer, unsigned bufflen)
363 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
366 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
368 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
369 void __user *ubuffer, unsigned bufflen,
370 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
371 u32 *result, unsigned timeout)
373 bool write = nvme_is_write(cmd);
374 struct nvme_ns *ns = q->queuedata;
375 struct gendisk *disk = ns ? ns->disk : NULL;
377 struct bio *bio = NULL;
381 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
385 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
387 if (ubuffer && bufflen) {
388 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
396 bio->bi_bdev = bdget_disk(disk, 0);
402 if (meta_buffer && meta_len) {
403 struct bio_integrity_payload *bip;
405 meta = kmalloc(meta_len, GFP_KERNEL);
412 if (copy_from_user(meta, meta_buffer,
419 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
425 bip->bip_iter.bi_size = meta_len;
426 bip->bip_iter.bi_sector = meta_seed;
428 ret = bio_integrity_add_page(bio, virt_to_page(meta),
429 meta_len, offset_in_page(meta));
430 if (ret != meta_len) {
437 blk_execute_rq(req->q, disk, req, 0);
440 *result = le32_to_cpu(nvme_req(req)->result.u32);
441 if (meta && !ret && !write) {
442 if (copy_to_user(meta_buffer, meta, meta_len))
449 if (disk && bio->bi_bdev)
451 blk_rq_unmap_user(bio);
454 blk_mq_free_request(req);
458 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
459 void __user *ubuffer, unsigned bufflen, u32 *result,
462 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
466 static void nvme_keep_alive_end_io(struct request *rq, int error)
468 struct nvme_ctrl *ctrl = rq->end_io_data;
470 blk_mq_free_request(rq);
473 dev_err(ctrl->device,
474 "failed nvme_keep_alive_end_io error=%d\n", error);
478 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
481 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
483 struct nvme_command c;
486 memset(&c, 0, sizeof(c));
487 c.common.opcode = nvme_admin_keep_alive;
489 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
494 rq->timeout = ctrl->kato * HZ;
495 rq->end_io_data = ctrl;
497 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
502 static void nvme_keep_alive_work(struct work_struct *work)
504 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
505 struct nvme_ctrl, ka_work);
507 if (nvme_keep_alive(ctrl)) {
508 /* allocation failure, reset the controller */
509 dev_err(ctrl->device, "keep-alive failed\n");
510 ctrl->ops->reset_ctrl(ctrl);
515 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
517 if (unlikely(ctrl->kato == 0))
520 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
521 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
523 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
525 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
527 if (unlikely(ctrl->kato == 0))
530 cancel_delayed_work_sync(&ctrl->ka_work);
532 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
534 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
536 struct nvme_command c = { };
539 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
540 c.identify.opcode = nvme_admin_identify;
541 c.identify.cns = cpu_to_le32(NVME_ID_CNS_CTRL);
543 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
547 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
548 sizeof(struct nvme_id_ctrl));
554 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
556 struct nvme_command c = { };
558 c.identify.opcode = nvme_admin_identify;
559 c.identify.cns = cpu_to_le32(NVME_ID_CNS_NS_ACTIVE_LIST);
560 c.identify.nsid = cpu_to_le32(nsid);
561 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
564 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
565 struct nvme_id_ns **id)
567 struct nvme_command c = { };
570 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
571 c.identify.opcode = nvme_admin_identify,
572 c.identify.nsid = cpu_to_le32(nsid),
574 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
578 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
579 sizeof(struct nvme_id_ns));
585 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
586 void *buffer, size_t buflen, u32 *result)
588 struct nvme_command c;
589 union nvme_result res;
592 memset(&c, 0, sizeof(c));
593 c.features.opcode = nvme_admin_get_features;
594 c.features.nsid = cpu_to_le32(nsid);
595 c.features.fid = cpu_to_le32(fid);
597 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
599 if (ret >= 0 && result)
600 *result = le32_to_cpu(res.u32);
604 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
605 void *buffer, size_t buflen, u32 *result)
607 struct nvme_command c;
608 union nvme_result res;
611 memset(&c, 0, sizeof(c));
612 c.features.opcode = nvme_admin_set_features;
613 c.features.fid = cpu_to_le32(fid);
614 c.features.dword11 = cpu_to_le32(dword11);
616 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
617 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
618 if (ret >= 0 && result)
619 *result = le32_to_cpu(res.u32);
623 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
625 struct nvme_command c = { };
628 c.common.opcode = nvme_admin_get_log_page,
629 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
630 c.common.cdw10[0] = cpu_to_le32(
631 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
634 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
638 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
639 sizeof(struct nvme_smart_log));
645 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
647 u32 q_count = (*count - 1) | ((*count - 1) << 16);
649 int status, nr_io_queues;
651 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
657 * Degraded controllers might return an error when setting the queue
658 * count. We still want to be able to bring them online and offer
659 * access to the admin queue, as that might be only way to fix them up.
662 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
665 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
666 *count = min(*count, nr_io_queues);
671 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
673 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
675 struct nvme_user_io io;
676 struct nvme_command c;
677 unsigned length, meta_len;
678 void __user *metadata;
680 if (copy_from_user(&io, uio, sizeof(io)))
688 case nvme_cmd_compare:
694 length = (io.nblocks + 1) << ns->lba_shift;
695 meta_len = (io.nblocks + 1) * ns->ms;
696 metadata = (void __user *)(uintptr_t)io.metadata;
701 } else if (meta_len) {
702 if ((io.metadata & 3) || !io.metadata)
706 memset(&c, 0, sizeof(c));
707 c.rw.opcode = io.opcode;
708 c.rw.flags = io.flags;
709 c.rw.nsid = cpu_to_le32(ns->ns_id);
710 c.rw.slba = cpu_to_le64(io.slba);
711 c.rw.length = cpu_to_le16(io.nblocks);
712 c.rw.control = cpu_to_le16(io.control);
713 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
714 c.rw.reftag = cpu_to_le32(io.reftag);
715 c.rw.apptag = cpu_to_le16(io.apptag);
716 c.rw.appmask = cpu_to_le16(io.appmask);
718 return __nvme_submit_user_cmd(ns->queue, &c,
719 (void __user *)(uintptr_t)io.addr, length,
720 metadata, meta_len, io.slba, NULL, 0);
723 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
724 struct nvme_passthru_cmd __user *ucmd)
726 struct nvme_passthru_cmd cmd;
727 struct nvme_command c;
728 unsigned timeout = 0;
731 if (!capable(CAP_SYS_ADMIN))
733 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
738 memset(&c, 0, sizeof(c));
739 c.common.opcode = cmd.opcode;
740 c.common.flags = cmd.flags;
741 c.common.nsid = cpu_to_le32(cmd.nsid);
742 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
743 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
744 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
745 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
746 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
747 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
748 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
749 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
752 timeout = msecs_to_jiffies(cmd.timeout_ms);
754 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
755 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
756 &cmd.result, timeout);
758 if (put_user(cmd.result, &ucmd->result))
765 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
766 unsigned int cmd, unsigned long arg)
768 struct nvme_ns *ns = bdev->bd_disk->private_data;
772 force_successful_syscall_return();
774 case NVME_IOCTL_ADMIN_CMD:
775 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
776 case NVME_IOCTL_IO_CMD:
777 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
778 case NVME_IOCTL_SUBMIT_IO:
779 return nvme_submit_io(ns, (void __user *)arg);
780 #ifdef CONFIG_BLK_DEV_NVME_SCSI
781 case SG_GET_VERSION_NUM:
782 return nvme_sg_get_version_num((void __user *)arg);
784 return nvme_sg_io(ns, (void __user *)arg);
789 return nvme_nvm_ioctl(ns, cmd, arg);
791 if (is_sed_ioctl(cmd))
792 return sed_ioctl(ns->ctrl->opal_dev, cmd,
793 (void __user *) arg);
799 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
800 unsigned int cmd, unsigned long arg)
806 return nvme_ioctl(bdev, mode, cmd, arg);
809 #define nvme_compat_ioctl NULL
812 static int nvme_open(struct block_device *bdev, fmode_t mode)
814 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
817 static void nvme_release(struct gendisk *disk, fmode_t mode)
819 struct nvme_ns *ns = disk->private_data;
821 module_put(ns->ctrl->ops->module);
825 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
827 /* some standard values */
829 geo->sectors = 1 << 5;
830 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
834 #ifdef CONFIG_BLK_DEV_INTEGRITY
835 static void nvme_init_integrity(struct nvme_ns *ns)
837 struct blk_integrity integrity;
839 memset(&integrity, 0, sizeof(integrity));
840 switch (ns->pi_type) {
841 case NVME_NS_DPS_PI_TYPE3:
842 integrity.profile = &t10_pi_type3_crc;
843 integrity.tag_size = sizeof(u16) + sizeof(u32);
844 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
846 case NVME_NS_DPS_PI_TYPE1:
847 case NVME_NS_DPS_PI_TYPE2:
848 integrity.profile = &t10_pi_type1_crc;
849 integrity.tag_size = sizeof(u16);
850 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
853 integrity.profile = NULL;
856 integrity.tuple_size = ns->ms;
857 blk_integrity_register(ns->disk, &integrity);
858 blk_queue_max_integrity_segments(ns->queue, 1);
861 static void nvme_init_integrity(struct nvme_ns *ns)
864 #endif /* CONFIG_BLK_DEV_INTEGRITY */
866 static void nvme_config_discard(struct nvme_ns *ns)
868 struct nvme_ctrl *ctrl = ns->ctrl;
869 u32 logical_block_size = queue_logical_block_size(ns->queue);
871 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
872 ns->queue->limits.discard_zeroes_data = 1;
874 ns->queue->limits.discard_zeroes_data = 0;
876 ns->queue->limits.discard_alignment = logical_block_size;
877 ns->queue->limits.discard_granularity = logical_block_size;
878 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
879 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
882 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
884 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
885 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
889 if ((*id)->ncap == 0) {
894 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
895 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
896 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
897 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
902 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
904 struct nvme_ns *ns = disk->private_data;
910 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
911 ns->lba_shift = id->lbaf[lbaf].ds;
912 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
913 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
916 * If identify namespace failed, use default 512 byte block size so
917 * block layer can use before failing read/write for 0 capacity.
919 if (ns->lba_shift == 0)
921 bs = 1 << ns->lba_shift;
922 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
923 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
924 id->dps & NVME_NS_DPS_PI_MASK : 0;
926 blk_mq_freeze_queue(disk->queue);
927 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
929 bs != queue_logical_block_size(disk->queue) ||
930 (ns->ms && ns->ext)))
931 blk_integrity_unregister(disk);
933 ns->pi_type = pi_type;
934 blk_queue_logical_block_size(ns->queue, bs);
936 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
937 nvme_init_integrity(ns);
938 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
939 set_capacity(disk, 0);
941 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
943 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
944 nvme_config_discard(ns);
945 blk_mq_unfreeze_queue(disk->queue);
948 static int nvme_revalidate_disk(struct gendisk *disk)
950 struct nvme_ns *ns = disk->private_data;
951 struct nvme_id_ns *id = NULL;
954 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
955 set_capacity(disk, 0);
959 ret = nvme_revalidate_ns(ns, &id);
963 __nvme_revalidate_disk(disk, id);
969 static char nvme_pr_type(enum pr_type type)
972 case PR_WRITE_EXCLUSIVE:
974 case PR_EXCLUSIVE_ACCESS:
976 case PR_WRITE_EXCLUSIVE_REG_ONLY:
978 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
980 case PR_WRITE_EXCLUSIVE_ALL_REGS:
982 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
989 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
990 u64 key, u64 sa_key, u8 op)
992 struct nvme_ns *ns = bdev->bd_disk->private_data;
993 struct nvme_command c;
994 u8 data[16] = { 0, };
996 put_unaligned_le64(key, &data[0]);
997 put_unaligned_le64(sa_key, &data[8]);
999 memset(&c, 0, sizeof(c));
1000 c.common.opcode = op;
1001 c.common.nsid = cpu_to_le32(ns->ns_id);
1002 c.common.cdw10[0] = cpu_to_le32(cdw10);
1004 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1007 static int nvme_pr_register(struct block_device *bdev, u64 old,
1008 u64 new, unsigned flags)
1012 if (flags & ~PR_FL_IGNORE_KEY)
1015 cdw10 = old ? 2 : 0;
1016 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1017 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1018 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1021 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1022 enum pr_type type, unsigned flags)
1026 if (flags & ~PR_FL_IGNORE_KEY)
1029 cdw10 = nvme_pr_type(type) << 8;
1030 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1031 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1034 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1035 enum pr_type type, bool abort)
1037 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1038 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1041 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1043 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1044 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1047 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1049 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1050 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1053 static const struct pr_ops nvme_pr_ops = {
1054 .pr_register = nvme_pr_register,
1055 .pr_reserve = nvme_pr_reserve,
1056 .pr_release = nvme_pr_release,
1057 .pr_preempt = nvme_pr_preempt,
1058 .pr_clear = nvme_pr_clear,
1061 #ifdef CONFIG_BLK_SED_OPAL
1062 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1065 struct nvme_ctrl *ctrl = data;
1066 struct nvme_command cmd;
1068 memset(&cmd, 0, sizeof(cmd));
1070 cmd.common.opcode = nvme_admin_security_send;
1072 cmd.common.opcode = nvme_admin_security_recv;
1073 cmd.common.nsid = 0;
1074 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1075 cmd.common.cdw10[1] = cpu_to_le32(len);
1077 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1078 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1080 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1081 #endif /* CONFIG_BLK_SED_OPAL */
1083 static const struct block_device_operations nvme_fops = {
1084 .owner = THIS_MODULE,
1085 .ioctl = nvme_ioctl,
1086 .compat_ioctl = nvme_compat_ioctl,
1088 .release = nvme_release,
1089 .getgeo = nvme_getgeo,
1090 .revalidate_disk= nvme_revalidate_disk,
1091 .pr_ops = &nvme_pr_ops,
1094 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1096 unsigned long timeout =
1097 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1098 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1101 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1104 if ((csts & NVME_CSTS_RDY) == bit)
1108 if (fatal_signal_pending(current))
1110 if (time_after(jiffies, timeout)) {
1111 dev_err(ctrl->device,
1112 "Device not ready; aborting %s\n", enabled ?
1113 "initialisation" : "reset");
1122 * If the device has been passed off to us in an enabled state, just clear
1123 * the enabled bit. The spec says we should set the 'shutdown notification
1124 * bits', but doing so may cause the device to complete commands to the
1125 * admin queue ... and we don't know what memory that might be pointing at!
1127 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1131 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1132 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1134 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1138 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1139 msleep(NVME_QUIRK_DELAY_AMOUNT);
1141 return nvme_wait_ready(ctrl, cap, false);
1143 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1145 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1148 * Default to a 4K page size, with the intention to update this
1149 * path in the future to accomodate architectures with differing
1150 * kernel and IO page sizes.
1152 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1155 if (page_shift < dev_page_min) {
1156 dev_err(ctrl->device,
1157 "Minimum device page size %u too large for host (%u)\n",
1158 1 << dev_page_min, 1 << page_shift);
1162 ctrl->page_size = 1 << page_shift;
1164 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1165 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1166 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1167 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1168 ctrl->ctrl_config |= NVME_CC_ENABLE;
1170 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1173 return nvme_wait_ready(ctrl, cap, true);
1175 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1177 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1179 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1183 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1184 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1186 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1190 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1191 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1195 if (fatal_signal_pending(current))
1197 if (time_after(jiffies, timeout)) {
1198 dev_err(ctrl->device,
1199 "Device shutdown incomplete; abort shutdown\n");
1206 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1208 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1209 struct request_queue *q)
1213 if (ctrl->max_hw_sectors) {
1215 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1217 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1218 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1220 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1221 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1222 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1223 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1225 blk_queue_write_cache(q, vwc, vwc);
1229 * Initialize the cached copies of the Identify data and various controller
1230 * register in our nvme_ctrl structure. This should be called as soon as
1231 * the admin queue is fully up and running.
1233 int nvme_init_identify(struct nvme_ctrl *ctrl)
1235 struct nvme_id_ctrl *id;
1237 int ret, page_shift;
1240 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1242 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1246 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1248 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1251 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1253 if (ctrl->vs >= NVME_VS(1, 1, 0))
1254 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1256 ret = nvme_identify_ctrl(ctrl, &id);
1258 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1262 ctrl->oacs = le16_to_cpu(id->oacs);
1263 ctrl->vid = le16_to_cpu(id->vid);
1264 ctrl->oncs = le16_to_cpup(&id->oncs);
1265 atomic_set(&ctrl->abort_limit, id->acl + 1);
1266 ctrl->vwc = id->vwc;
1267 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1268 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1269 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1270 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1272 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1274 max_hw_sectors = UINT_MAX;
1275 ctrl->max_hw_sectors =
1276 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1278 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1279 ctrl->sgls = le32_to_cpu(id->sgls);
1280 ctrl->kas = le16_to_cpu(id->kas);
1282 if (ctrl->ops->is_fabrics) {
1283 ctrl->icdoff = le16_to_cpu(id->icdoff);
1284 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1285 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1286 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1289 * In fabrics we need to verify the cntlid matches the
1292 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1295 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1297 "keep-alive support is mandatory for fabrics\n");
1301 ctrl->cntlid = le16_to_cpu(id->cntlid);
1307 EXPORT_SYMBOL_GPL(nvme_init_identify);
1309 static int nvme_dev_open(struct inode *inode, struct file *file)
1311 struct nvme_ctrl *ctrl;
1312 int instance = iminor(inode);
1315 spin_lock(&dev_list_lock);
1316 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1317 if (ctrl->instance != instance)
1320 if (!ctrl->admin_q) {
1324 if (!kref_get_unless_zero(&ctrl->kref))
1326 file->private_data = ctrl;
1330 spin_unlock(&dev_list_lock);
1335 static int nvme_dev_release(struct inode *inode, struct file *file)
1337 nvme_put_ctrl(file->private_data);
1341 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1346 mutex_lock(&ctrl->namespaces_mutex);
1347 if (list_empty(&ctrl->namespaces)) {
1352 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1353 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1354 dev_warn(ctrl->device,
1355 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1360 dev_warn(ctrl->device,
1361 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1362 kref_get(&ns->kref);
1363 mutex_unlock(&ctrl->namespaces_mutex);
1365 ret = nvme_user_cmd(ctrl, ns, argp);
1370 mutex_unlock(&ctrl->namespaces_mutex);
1374 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1377 struct nvme_ctrl *ctrl = file->private_data;
1378 void __user *argp = (void __user *)arg;
1381 case NVME_IOCTL_ADMIN_CMD:
1382 return nvme_user_cmd(ctrl, NULL, argp);
1383 case NVME_IOCTL_IO_CMD:
1384 return nvme_dev_user_cmd(ctrl, argp);
1385 case NVME_IOCTL_RESET:
1386 dev_warn(ctrl->device, "resetting controller\n");
1387 return ctrl->ops->reset_ctrl(ctrl);
1388 case NVME_IOCTL_SUBSYS_RESET:
1389 return nvme_reset_subsystem(ctrl);
1390 case NVME_IOCTL_RESCAN:
1391 nvme_queue_scan(ctrl);
1398 static const struct file_operations nvme_dev_fops = {
1399 .owner = THIS_MODULE,
1400 .open = nvme_dev_open,
1401 .release = nvme_dev_release,
1402 .unlocked_ioctl = nvme_dev_ioctl,
1403 .compat_ioctl = nvme_dev_ioctl,
1406 static ssize_t nvme_sysfs_reset(struct device *dev,
1407 struct device_attribute *attr, const char *buf,
1410 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1413 ret = ctrl->ops->reset_ctrl(ctrl);
1418 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1420 static ssize_t nvme_sysfs_rescan(struct device *dev,
1421 struct device_attribute *attr, const char *buf,
1424 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1426 nvme_queue_scan(ctrl);
1429 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1431 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1434 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1435 struct nvme_ctrl *ctrl = ns->ctrl;
1436 int serial_len = sizeof(ctrl->serial);
1437 int model_len = sizeof(ctrl->model);
1439 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1440 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1442 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1443 return sprintf(buf, "eui.%8phN\n", ns->eui);
1445 while (ctrl->serial[serial_len - 1] == ' ')
1447 while (ctrl->model[model_len - 1] == ' ')
1450 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1451 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1453 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1455 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1458 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1459 return sprintf(buf, "%pU\n", ns->uuid);
1461 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1463 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1466 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1467 return sprintf(buf, "%8phd\n", ns->eui);
1469 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1471 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1474 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1475 return sprintf(buf, "%d\n", ns->ns_id);
1477 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1479 static struct attribute *nvme_ns_attrs[] = {
1480 &dev_attr_wwid.attr,
1481 &dev_attr_uuid.attr,
1483 &dev_attr_nsid.attr,
1487 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1488 struct attribute *a, int n)
1490 struct device *dev = container_of(kobj, struct device, kobj);
1491 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1493 if (a == &dev_attr_uuid.attr) {
1494 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1497 if (a == &dev_attr_eui.attr) {
1498 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1504 static const struct attribute_group nvme_ns_attr_group = {
1505 .attrs = nvme_ns_attrs,
1506 .is_visible = nvme_ns_attrs_are_visible,
1509 #define nvme_show_str_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, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1516 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1518 #define nvme_show_int_function(field) \
1519 static ssize_t field##_show(struct device *dev, \
1520 struct device_attribute *attr, char *buf) \
1522 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1523 return sprintf(buf, "%d\n", ctrl->field); \
1525 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1527 nvme_show_str_function(model);
1528 nvme_show_str_function(serial);
1529 nvme_show_str_function(firmware_rev);
1530 nvme_show_int_function(cntlid);
1532 static ssize_t nvme_sysfs_delete(struct device *dev,
1533 struct device_attribute *attr, const char *buf,
1536 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1538 if (device_remove_file_self(dev, attr))
1539 ctrl->ops->delete_ctrl(ctrl);
1542 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1544 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1545 struct device_attribute *attr,
1548 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1550 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1552 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1554 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1555 struct device_attribute *attr,
1558 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1560 return snprintf(buf, PAGE_SIZE, "%s\n",
1561 ctrl->ops->get_subsysnqn(ctrl));
1563 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1565 static ssize_t nvme_sysfs_show_address(struct device *dev,
1566 struct device_attribute *attr,
1569 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1571 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1573 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1575 static struct attribute *nvme_dev_attrs[] = {
1576 &dev_attr_reset_controller.attr,
1577 &dev_attr_rescan_controller.attr,
1578 &dev_attr_model.attr,
1579 &dev_attr_serial.attr,
1580 &dev_attr_firmware_rev.attr,
1581 &dev_attr_cntlid.attr,
1582 &dev_attr_delete_controller.attr,
1583 &dev_attr_transport.attr,
1584 &dev_attr_subsysnqn.attr,
1585 &dev_attr_address.attr,
1589 #define CHECK_ATTR(ctrl, a, name) \
1590 if ((a) == &dev_attr_##name.attr && \
1591 !(ctrl)->ops->get_##name) \
1594 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1595 struct attribute *a, int n)
1597 struct device *dev = container_of(kobj, struct device, kobj);
1598 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1600 if (a == &dev_attr_delete_controller.attr) {
1601 if (!ctrl->ops->delete_ctrl)
1605 CHECK_ATTR(ctrl, a, subsysnqn);
1606 CHECK_ATTR(ctrl, a, address);
1611 static struct attribute_group nvme_dev_attrs_group = {
1612 .attrs = nvme_dev_attrs,
1613 .is_visible = nvme_dev_attrs_are_visible,
1616 static const struct attribute_group *nvme_dev_attr_groups[] = {
1617 &nvme_dev_attrs_group,
1621 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1623 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1624 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1626 return nsa->ns_id - nsb->ns_id;
1629 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1631 struct nvme_ns *ns, *ret = NULL;
1633 mutex_lock(&ctrl->namespaces_mutex);
1634 list_for_each_entry(ns, &ctrl->namespaces, list) {
1635 if (ns->ns_id == nsid) {
1636 kref_get(&ns->kref);
1640 if (ns->ns_id > nsid)
1643 mutex_unlock(&ctrl->namespaces_mutex);
1647 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1650 struct gendisk *disk;
1651 struct nvme_id_ns *id;
1652 char disk_name[DISK_NAME_LEN];
1653 int node = dev_to_node(ctrl->dev);
1655 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1659 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1660 if (ns->instance < 0)
1663 ns->queue = blk_mq_init_queue(ctrl->tagset);
1664 if (IS_ERR(ns->queue))
1665 goto out_release_instance;
1666 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1667 ns->queue->queuedata = ns;
1670 kref_init(&ns->kref);
1672 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1674 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1675 nvme_set_queue_limits(ctrl, ns->queue);
1677 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1679 if (nvme_revalidate_ns(ns, &id))
1680 goto out_free_queue;
1682 if (nvme_nvm_ns_supported(ns, id) &&
1683 nvme_nvm_register(ns, disk_name, node)) {
1684 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1688 disk = alloc_disk_node(0, node);
1692 disk->fops = &nvme_fops;
1693 disk->private_data = ns;
1694 disk->queue = ns->queue;
1695 disk->flags = GENHD_FL_EXT_DEVT;
1696 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1699 __nvme_revalidate_disk(disk, id);
1701 mutex_lock(&ctrl->namespaces_mutex);
1702 list_add_tail(&ns->list, &ctrl->namespaces);
1703 mutex_unlock(&ctrl->namespaces_mutex);
1705 kref_get(&ctrl->kref);
1709 device_add_disk(ctrl->device, ns->disk);
1710 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1711 &nvme_ns_attr_group))
1712 pr_warn("%s: failed to create sysfs group for identification\n",
1713 ns->disk->disk_name);
1714 if (ns->ndev && nvme_nvm_register_sysfs(ns))
1715 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
1716 ns->disk->disk_name);
1721 blk_cleanup_queue(ns->queue);
1722 out_release_instance:
1723 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1728 static void nvme_ns_remove(struct nvme_ns *ns)
1730 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1733 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1734 if (blk_get_integrity(ns->disk))
1735 blk_integrity_unregister(ns->disk);
1736 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1737 &nvme_ns_attr_group);
1739 nvme_nvm_unregister_sysfs(ns);
1740 del_gendisk(ns->disk);
1741 blk_mq_abort_requeue_list(ns->queue);
1742 blk_cleanup_queue(ns->queue);
1745 mutex_lock(&ns->ctrl->namespaces_mutex);
1746 list_del_init(&ns->list);
1747 mutex_unlock(&ns->ctrl->namespaces_mutex);
1752 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1756 ns = nvme_find_get_ns(ctrl, nsid);
1758 if (ns->disk && revalidate_disk(ns->disk))
1762 nvme_alloc_ns(ctrl, nsid);
1765 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1768 struct nvme_ns *ns, *next;
1770 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1771 if (ns->ns_id > nsid)
1776 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1780 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1783 ns_list = kzalloc(0x1000, GFP_KERNEL);
1787 for (i = 0; i < num_lists; i++) {
1788 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1792 for (j = 0; j < min(nn, 1024U); j++) {
1793 nsid = le32_to_cpu(ns_list[j]);
1797 nvme_validate_ns(ctrl, nsid);
1799 while (++prev < nsid) {
1800 ns = nvme_find_get_ns(ctrl, prev);
1810 nvme_remove_invalid_namespaces(ctrl, prev);
1816 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1820 for (i = 1; i <= nn; i++)
1821 nvme_validate_ns(ctrl, i);
1823 nvme_remove_invalid_namespaces(ctrl, nn);
1826 static void nvme_scan_work(struct work_struct *work)
1828 struct nvme_ctrl *ctrl =
1829 container_of(work, struct nvme_ctrl, scan_work);
1830 struct nvme_id_ctrl *id;
1833 if (ctrl->state != NVME_CTRL_LIVE)
1836 if (nvme_identify_ctrl(ctrl, &id))
1839 nn = le32_to_cpu(id->nn);
1840 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1841 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1842 if (!nvme_scan_ns_list(ctrl, nn))
1845 nvme_scan_ns_sequential(ctrl, nn);
1847 mutex_lock(&ctrl->namespaces_mutex);
1848 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1849 mutex_unlock(&ctrl->namespaces_mutex);
1853 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1856 * Do not queue new scan work when a controller is reset during
1859 if (ctrl->state == NVME_CTRL_LIVE)
1860 schedule_work(&ctrl->scan_work);
1862 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1865 * This function iterates the namespace list unlocked to allow recovery from
1866 * controller failure. It is up to the caller to ensure the namespace list is
1867 * not modified by scan work while this function is executing.
1869 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1871 struct nvme_ns *ns, *next;
1874 * The dead states indicates the controller was not gracefully
1875 * disconnected. In that case, we won't be able to flush any data while
1876 * removing the namespaces' disks; fail all the queues now to avoid
1877 * potentially having to clean up the failed sync later.
1879 if (ctrl->state == NVME_CTRL_DEAD)
1880 nvme_kill_queues(ctrl);
1882 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1885 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1887 static void nvme_async_event_work(struct work_struct *work)
1889 struct nvme_ctrl *ctrl =
1890 container_of(work, struct nvme_ctrl, async_event_work);
1892 spin_lock_irq(&ctrl->lock);
1893 while (ctrl->event_limit > 0) {
1894 int aer_idx = --ctrl->event_limit;
1896 spin_unlock_irq(&ctrl->lock);
1897 ctrl->ops->submit_async_event(ctrl, aer_idx);
1898 spin_lock_irq(&ctrl->lock);
1900 spin_unlock_irq(&ctrl->lock);
1903 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
1904 union nvme_result *res)
1906 u32 result = le32_to_cpu(res->u32);
1909 switch (le16_to_cpu(status) >> 1) {
1910 case NVME_SC_SUCCESS:
1913 case NVME_SC_ABORT_REQ:
1914 ++ctrl->event_limit;
1915 schedule_work(&ctrl->async_event_work);
1924 switch (result & 0xff07) {
1925 case NVME_AER_NOTICE_NS_CHANGED:
1926 dev_info(ctrl->device, "rescanning\n");
1927 nvme_queue_scan(ctrl);
1930 dev_warn(ctrl->device, "async event result %08x\n", result);
1933 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1935 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1937 ctrl->event_limit = NVME_NR_AERS;
1938 schedule_work(&ctrl->async_event_work);
1940 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1942 static DEFINE_IDA(nvme_instance_ida);
1944 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1946 int instance, error;
1949 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1952 spin_lock(&dev_list_lock);
1953 error = ida_get_new(&nvme_instance_ida, &instance);
1954 spin_unlock(&dev_list_lock);
1955 } while (error == -EAGAIN);
1960 ctrl->instance = instance;
1964 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1966 spin_lock(&dev_list_lock);
1967 ida_remove(&nvme_instance_ida, ctrl->instance);
1968 spin_unlock(&dev_list_lock);
1971 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1973 flush_work(&ctrl->async_event_work);
1974 flush_work(&ctrl->scan_work);
1975 nvme_remove_namespaces(ctrl);
1977 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1979 spin_lock(&dev_list_lock);
1980 list_del(&ctrl->node);
1981 spin_unlock(&dev_list_lock);
1983 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1985 static void nvme_free_ctrl(struct kref *kref)
1987 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1989 put_device(ctrl->device);
1990 nvme_release_instance(ctrl);
1991 ida_destroy(&ctrl->ns_ida);
1993 ctrl->ops->free_ctrl(ctrl);
1996 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
1998 kref_put(&ctrl->kref, nvme_free_ctrl);
2000 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2003 * Initialize a NVMe controller structures. This needs to be called during
2004 * earliest initialization so that we have the initialized structured around
2007 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2008 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2012 ctrl->state = NVME_CTRL_NEW;
2013 spin_lock_init(&ctrl->lock);
2014 INIT_LIST_HEAD(&ctrl->namespaces);
2015 mutex_init(&ctrl->namespaces_mutex);
2016 kref_init(&ctrl->kref);
2019 ctrl->quirks = quirks;
2020 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2021 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2023 ret = nvme_set_instance(ctrl);
2027 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2028 MKDEV(nvme_char_major, ctrl->instance),
2029 ctrl, nvme_dev_attr_groups,
2030 "nvme%d", ctrl->instance);
2031 if (IS_ERR(ctrl->device)) {
2032 ret = PTR_ERR(ctrl->device);
2033 goto out_release_instance;
2035 get_device(ctrl->device);
2036 ida_init(&ctrl->ns_ida);
2038 spin_lock(&dev_list_lock);
2039 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2040 spin_unlock(&dev_list_lock);
2043 out_release_instance:
2044 nvme_release_instance(ctrl);
2048 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2051 * nvme_kill_queues(): Ends all namespace queues
2052 * @ctrl: the dead controller that needs to end
2054 * Call this function when the driver determines it is unable to get the
2055 * controller in a state capable of servicing IO.
2057 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2061 mutex_lock(&ctrl->namespaces_mutex);
2062 list_for_each_entry(ns, &ctrl->namespaces, list) {
2064 * Revalidating a dead namespace sets capacity to 0. This will
2065 * end buffered writers dirtying pages that can't be synced.
2067 if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2068 revalidate_disk(ns->disk);
2070 blk_set_queue_dying(ns->queue);
2071 blk_mq_abort_requeue_list(ns->queue);
2072 blk_mq_start_stopped_hw_queues(ns->queue, true);
2074 mutex_unlock(&ctrl->namespaces_mutex);
2076 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2078 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2082 mutex_lock(&ctrl->namespaces_mutex);
2083 list_for_each_entry(ns, &ctrl->namespaces, list)
2084 blk_mq_quiesce_queue(ns->queue);
2085 mutex_unlock(&ctrl->namespaces_mutex);
2087 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2089 void nvme_start_queues(struct nvme_ctrl *ctrl)
2093 mutex_lock(&ctrl->namespaces_mutex);
2094 list_for_each_entry(ns, &ctrl->namespaces, list) {
2095 blk_mq_start_stopped_hw_queues(ns->queue, true);
2096 blk_mq_kick_requeue_list(ns->queue);
2098 mutex_unlock(&ctrl->namespaces_mutex);
2100 EXPORT_SYMBOL_GPL(nvme_start_queues);
2102 int __init nvme_core_init(void)
2106 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2110 else if (result > 0)
2111 nvme_char_major = result;
2113 nvme_class = class_create(THIS_MODULE, "nvme");
2114 if (IS_ERR(nvme_class)) {
2115 result = PTR_ERR(nvme_class);
2116 goto unregister_chrdev;
2122 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2126 void nvme_core_exit(void)
2128 class_destroy(nvme_class);
2129 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2132 MODULE_LICENSE("GPL");
2133 MODULE_VERSION("1.0");
2134 module_init(nvme_core_init);
2135 module_exit(nvme_core_exit);