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, arg);
798 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
799 unsigned int cmd, unsigned long arg)
805 return nvme_ioctl(bdev, mode, cmd, arg);
808 #define nvme_compat_ioctl NULL
811 static int nvme_open(struct block_device *bdev, fmode_t mode)
813 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
816 static void nvme_release(struct gendisk *disk, fmode_t mode)
818 struct nvme_ns *ns = disk->private_data;
820 module_put(ns->ctrl->ops->module);
824 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
826 /* some standard values */
828 geo->sectors = 1 << 5;
829 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
833 #ifdef CONFIG_BLK_DEV_INTEGRITY
834 static void nvme_init_integrity(struct nvme_ns *ns)
836 struct blk_integrity integrity;
838 memset(&integrity, 0, sizeof(integrity));
839 switch (ns->pi_type) {
840 case NVME_NS_DPS_PI_TYPE3:
841 integrity.profile = &t10_pi_type3_crc;
842 integrity.tag_size = sizeof(u16) + sizeof(u32);
843 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
845 case NVME_NS_DPS_PI_TYPE1:
846 case NVME_NS_DPS_PI_TYPE2:
847 integrity.profile = &t10_pi_type1_crc;
848 integrity.tag_size = sizeof(u16);
849 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
852 integrity.profile = NULL;
855 integrity.tuple_size = ns->ms;
856 blk_integrity_register(ns->disk, &integrity);
857 blk_queue_max_integrity_segments(ns->queue, 1);
860 static void nvme_init_integrity(struct nvme_ns *ns)
863 #endif /* CONFIG_BLK_DEV_INTEGRITY */
865 static void nvme_config_discard(struct nvme_ns *ns)
867 struct nvme_ctrl *ctrl = ns->ctrl;
868 u32 logical_block_size = queue_logical_block_size(ns->queue);
870 if (ctrl->quirks & NVME_QUIRK_DISCARD_ZEROES)
871 ns->queue->limits.discard_zeroes_data = 1;
873 ns->queue->limits.discard_zeroes_data = 0;
875 ns->queue->limits.discard_alignment = logical_block_size;
876 ns->queue->limits.discard_granularity = logical_block_size;
877 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
878 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
881 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
883 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
884 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
888 if ((*id)->ncap == 0) {
893 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
894 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
895 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
896 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
901 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
903 struct nvme_ns *ns = disk->private_data;
909 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
910 ns->lba_shift = id->lbaf[lbaf].ds;
911 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
912 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
915 * If identify namespace failed, use default 512 byte block size so
916 * block layer can use before failing read/write for 0 capacity.
918 if (ns->lba_shift == 0)
920 bs = 1 << ns->lba_shift;
921 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
922 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
923 id->dps & NVME_NS_DPS_PI_MASK : 0;
925 blk_mq_freeze_queue(disk->queue);
926 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
928 bs != queue_logical_block_size(disk->queue) ||
929 (ns->ms && ns->ext)))
930 blk_integrity_unregister(disk);
932 ns->pi_type = pi_type;
933 blk_queue_logical_block_size(ns->queue, bs);
935 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
936 nvme_init_integrity(ns);
937 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
938 set_capacity(disk, 0);
940 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
942 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
943 nvme_config_discard(ns);
944 blk_mq_unfreeze_queue(disk->queue);
947 static int nvme_revalidate_disk(struct gendisk *disk)
949 struct nvme_ns *ns = disk->private_data;
950 struct nvme_id_ns *id = NULL;
953 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
954 set_capacity(disk, 0);
958 ret = nvme_revalidate_ns(ns, &id);
962 __nvme_revalidate_disk(disk, id);
968 static char nvme_pr_type(enum pr_type type)
971 case PR_WRITE_EXCLUSIVE:
973 case PR_EXCLUSIVE_ACCESS:
975 case PR_WRITE_EXCLUSIVE_REG_ONLY:
977 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
979 case PR_WRITE_EXCLUSIVE_ALL_REGS:
981 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
988 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
989 u64 key, u64 sa_key, u8 op)
991 struct nvme_ns *ns = bdev->bd_disk->private_data;
992 struct nvme_command c;
993 u8 data[16] = { 0, };
995 put_unaligned_le64(key, &data[0]);
996 put_unaligned_le64(sa_key, &data[8]);
998 memset(&c, 0, sizeof(c));
999 c.common.opcode = op;
1000 c.common.nsid = cpu_to_le32(ns->ns_id);
1001 c.common.cdw10[0] = cpu_to_le32(cdw10);
1003 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1006 static int nvme_pr_register(struct block_device *bdev, u64 old,
1007 u64 new, unsigned flags)
1011 if (flags & ~PR_FL_IGNORE_KEY)
1014 cdw10 = old ? 2 : 0;
1015 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1016 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1017 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1020 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1021 enum pr_type type, unsigned flags)
1025 if (flags & ~PR_FL_IGNORE_KEY)
1028 cdw10 = nvme_pr_type(type) << 8;
1029 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1030 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1033 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1034 enum pr_type type, bool abort)
1036 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1037 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1040 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1042 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1043 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1046 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1048 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1049 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1052 static const struct pr_ops nvme_pr_ops = {
1053 .pr_register = nvme_pr_register,
1054 .pr_reserve = nvme_pr_reserve,
1055 .pr_release = nvme_pr_release,
1056 .pr_preempt = nvme_pr_preempt,
1057 .pr_clear = nvme_pr_clear,
1060 #ifdef CONFIG_BLK_SED_OPAL
1061 int nvme_sec_submit(struct opal_dev *dev, u16 spsp, u8 secp,
1062 void *buffer, size_t len, bool send)
1064 struct nvme_command cmd;
1065 struct nvme_ctrl *ctrl = NULL;
1067 memset(&cmd, 0, sizeof(cmd));
1069 cmd.common.opcode = nvme_admin_security_send;
1071 cmd.common.opcode = nvme_admin_security_recv;
1072 ctrl = container_of(dev, struct nvme_ctrl, opal_dev);
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 /* Checking for ctrl->tagset is a trick to avoid sleeping on module
1139 * load, since we only need the quirk on reset_controller. Notice
1140 * that the HGST device needs this delay only in firmware activation
1141 * procedure; unfortunately we have no (easy) way to verify this.
1143 if ((ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY) && ctrl->tagset)
1144 msleep(NVME_QUIRK_DELAY_AMOUNT);
1146 return nvme_wait_ready(ctrl, cap, false);
1148 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1150 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1153 * Default to a 4K page size, with the intention to update this
1154 * path in the future to accomodate architectures with differing
1155 * kernel and IO page sizes.
1157 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1160 if (page_shift < dev_page_min) {
1161 dev_err(ctrl->device,
1162 "Minimum device page size %u too large for host (%u)\n",
1163 1 << dev_page_min, 1 << page_shift);
1167 ctrl->page_size = 1 << page_shift;
1169 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1170 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1171 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1172 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1173 ctrl->ctrl_config |= NVME_CC_ENABLE;
1175 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1178 return nvme_wait_ready(ctrl, cap, true);
1180 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1182 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1184 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1188 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1189 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1191 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1195 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1196 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1200 if (fatal_signal_pending(current))
1202 if (time_after(jiffies, timeout)) {
1203 dev_err(ctrl->device,
1204 "Device shutdown incomplete; abort shutdown\n");
1211 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1213 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1214 struct request_queue *q)
1218 if (ctrl->max_hw_sectors) {
1220 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1222 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1223 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1225 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1226 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1227 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1228 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1230 blk_queue_write_cache(q, vwc, vwc);
1234 * Initialize the cached copies of the Identify data and various controller
1235 * register in our nvme_ctrl structure. This should be called as soon as
1236 * the admin queue is fully up and running.
1238 int nvme_init_identify(struct nvme_ctrl *ctrl)
1240 struct nvme_id_ctrl *id;
1242 int ret, page_shift;
1245 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1247 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1251 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1253 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1256 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1258 if (ctrl->vs >= NVME_VS(1, 1, 0))
1259 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1261 ret = nvme_identify_ctrl(ctrl, &id);
1263 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1267 ctrl->vid = le16_to_cpu(id->vid);
1268 ctrl->oncs = le16_to_cpup(&id->oncs);
1269 atomic_set(&ctrl->abort_limit, id->acl + 1);
1270 ctrl->vwc = id->vwc;
1271 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1272 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1273 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1274 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1276 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1278 max_hw_sectors = UINT_MAX;
1279 ctrl->max_hw_sectors =
1280 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1282 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1283 ctrl->sgls = le32_to_cpu(id->sgls);
1284 ctrl->kas = le16_to_cpu(id->kas);
1286 if (ctrl->ops->is_fabrics) {
1287 ctrl->icdoff = le16_to_cpu(id->icdoff);
1288 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1289 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1290 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1293 * In fabrics we need to verify the cntlid matches the
1296 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1299 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1301 "keep-alive support is mandatory for fabrics\n");
1305 ctrl->cntlid = le16_to_cpu(id->cntlid);
1311 EXPORT_SYMBOL_GPL(nvme_init_identify);
1313 static int nvme_dev_open(struct inode *inode, struct file *file)
1315 struct nvme_ctrl *ctrl;
1316 int instance = iminor(inode);
1319 spin_lock(&dev_list_lock);
1320 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1321 if (ctrl->instance != instance)
1324 if (!ctrl->admin_q) {
1328 if (!kref_get_unless_zero(&ctrl->kref))
1330 file->private_data = ctrl;
1334 spin_unlock(&dev_list_lock);
1339 static int nvme_dev_release(struct inode *inode, struct file *file)
1341 nvme_put_ctrl(file->private_data);
1345 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1350 mutex_lock(&ctrl->namespaces_mutex);
1351 if (list_empty(&ctrl->namespaces)) {
1356 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1357 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1358 dev_warn(ctrl->device,
1359 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1364 dev_warn(ctrl->device,
1365 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1366 kref_get(&ns->kref);
1367 mutex_unlock(&ctrl->namespaces_mutex);
1369 ret = nvme_user_cmd(ctrl, ns, argp);
1374 mutex_unlock(&ctrl->namespaces_mutex);
1378 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1381 struct nvme_ctrl *ctrl = file->private_data;
1382 void __user *argp = (void __user *)arg;
1385 case NVME_IOCTL_ADMIN_CMD:
1386 return nvme_user_cmd(ctrl, NULL, argp);
1387 case NVME_IOCTL_IO_CMD:
1388 return nvme_dev_user_cmd(ctrl, argp);
1389 case NVME_IOCTL_RESET:
1390 dev_warn(ctrl->device, "resetting controller\n");
1391 return ctrl->ops->reset_ctrl(ctrl);
1392 case NVME_IOCTL_SUBSYS_RESET:
1393 return nvme_reset_subsystem(ctrl);
1394 case NVME_IOCTL_RESCAN:
1395 nvme_queue_scan(ctrl);
1402 static const struct file_operations nvme_dev_fops = {
1403 .owner = THIS_MODULE,
1404 .open = nvme_dev_open,
1405 .release = nvme_dev_release,
1406 .unlocked_ioctl = nvme_dev_ioctl,
1407 .compat_ioctl = nvme_dev_ioctl,
1410 static ssize_t nvme_sysfs_reset(struct device *dev,
1411 struct device_attribute *attr, const char *buf,
1414 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1417 ret = ctrl->ops->reset_ctrl(ctrl);
1422 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1424 static ssize_t nvme_sysfs_rescan(struct device *dev,
1425 struct device_attribute *attr, const char *buf,
1428 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1430 nvme_queue_scan(ctrl);
1433 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1435 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1438 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1439 struct nvme_ctrl *ctrl = ns->ctrl;
1440 int serial_len = sizeof(ctrl->serial);
1441 int model_len = sizeof(ctrl->model);
1443 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1444 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1446 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1447 return sprintf(buf, "eui.%8phN\n", ns->eui);
1449 while (ctrl->serial[serial_len - 1] == ' ')
1451 while (ctrl->model[model_len - 1] == ' ')
1454 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1455 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1457 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1459 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1462 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1463 return sprintf(buf, "%pU\n", ns->uuid);
1465 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1467 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1470 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1471 return sprintf(buf, "%8phd\n", ns->eui);
1473 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1475 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1478 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1479 return sprintf(buf, "%d\n", ns->ns_id);
1481 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1483 static struct attribute *nvme_ns_attrs[] = {
1484 &dev_attr_wwid.attr,
1485 &dev_attr_uuid.attr,
1487 &dev_attr_nsid.attr,
1491 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1492 struct attribute *a, int n)
1494 struct device *dev = container_of(kobj, struct device, kobj);
1495 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1497 if (a == &dev_attr_uuid.attr) {
1498 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1501 if (a == &dev_attr_eui.attr) {
1502 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1508 static const struct attribute_group nvme_ns_attr_group = {
1509 .attrs = nvme_ns_attrs,
1510 .is_visible = nvme_ns_attrs_are_visible,
1513 #define nvme_show_str_function(field) \
1514 static ssize_t field##_show(struct device *dev, \
1515 struct device_attribute *attr, char *buf) \
1517 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1518 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1520 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1522 #define nvme_show_int_function(field) \
1523 static ssize_t field##_show(struct device *dev, \
1524 struct device_attribute *attr, char *buf) \
1526 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1527 return sprintf(buf, "%d\n", ctrl->field); \
1529 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1531 nvme_show_str_function(model);
1532 nvme_show_str_function(serial);
1533 nvme_show_str_function(firmware_rev);
1534 nvme_show_int_function(cntlid);
1536 static ssize_t nvme_sysfs_delete(struct device *dev,
1537 struct device_attribute *attr, const char *buf,
1540 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1542 if (device_remove_file_self(dev, attr))
1543 ctrl->ops->delete_ctrl(ctrl);
1546 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1548 static ssize_t nvme_sysfs_show_transport(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", ctrl->ops->name);
1556 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1558 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1559 struct device_attribute *attr,
1562 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1564 return snprintf(buf, PAGE_SIZE, "%s\n",
1565 ctrl->ops->get_subsysnqn(ctrl));
1567 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1569 static ssize_t nvme_sysfs_show_address(struct device *dev,
1570 struct device_attribute *attr,
1573 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1575 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1577 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1579 static struct attribute *nvme_dev_attrs[] = {
1580 &dev_attr_reset_controller.attr,
1581 &dev_attr_rescan_controller.attr,
1582 &dev_attr_model.attr,
1583 &dev_attr_serial.attr,
1584 &dev_attr_firmware_rev.attr,
1585 &dev_attr_cntlid.attr,
1586 &dev_attr_delete_controller.attr,
1587 &dev_attr_transport.attr,
1588 &dev_attr_subsysnqn.attr,
1589 &dev_attr_address.attr,
1593 #define CHECK_ATTR(ctrl, a, name) \
1594 if ((a) == &dev_attr_##name.attr && \
1595 !(ctrl)->ops->get_##name) \
1598 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1599 struct attribute *a, int n)
1601 struct device *dev = container_of(kobj, struct device, kobj);
1602 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1604 if (a == &dev_attr_delete_controller.attr) {
1605 if (!ctrl->ops->delete_ctrl)
1609 CHECK_ATTR(ctrl, a, subsysnqn);
1610 CHECK_ATTR(ctrl, a, address);
1615 static struct attribute_group nvme_dev_attrs_group = {
1616 .attrs = nvme_dev_attrs,
1617 .is_visible = nvme_dev_attrs_are_visible,
1620 static const struct attribute_group *nvme_dev_attr_groups[] = {
1621 &nvme_dev_attrs_group,
1625 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1627 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1628 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1630 return nsa->ns_id - nsb->ns_id;
1633 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1635 struct nvme_ns *ns, *ret = NULL;
1637 mutex_lock(&ctrl->namespaces_mutex);
1638 list_for_each_entry(ns, &ctrl->namespaces, list) {
1639 if (ns->ns_id == nsid) {
1640 kref_get(&ns->kref);
1644 if (ns->ns_id > nsid)
1647 mutex_unlock(&ctrl->namespaces_mutex);
1651 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1654 struct gendisk *disk;
1655 struct nvme_id_ns *id;
1656 char disk_name[DISK_NAME_LEN];
1657 int node = dev_to_node(ctrl->dev);
1659 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
1663 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
1664 if (ns->instance < 0)
1667 ns->queue = blk_mq_init_queue(ctrl->tagset);
1668 if (IS_ERR(ns->queue))
1669 goto out_release_instance;
1670 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
1671 ns->queue->queuedata = ns;
1674 kref_init(&ns->kref);
1676 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
1678 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
1679 nvme_set_queue_limits(ctrl, ns->queue);
1681 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
1683 if (nvme_revalidate_ns(ns, &id))
1684 goto out_free_queue;
1686 if (nvme_nvm_ns_supported(ns, id) &&
1687 nvme_nvm_register(ns, disk_name, node)) {
1688 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
1692 disk = alloc_disk_node(0, node);
1696 disk->fops = &nvme_fops;
1697 disk->private_data = ns;
1698 disk->queue = ns->queue;
1699 disk->flags = GENHD_FL_EXT_DEVT;
1700 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
1703 __nvme_revalidate_disk(disk, id);
1705 mutex_lock(&ctrl->namespaces_mutex);
1706 list_add_tail(&ns->list, &ctrl->namespaces);
1707 mutex_unlock(&ctrl->namespaces_mutex);
1709 kref_get(&ctrl->kref);
1713 device_add_disk(ctrl->device, ns->disk);
1714 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
1715 &nvme_ns_attr_group))
1716 pr_warn("%s: failed to create sysfs group for identification\n",
1717 ns->disk->disk_name);
1718 if (ns->ndev && nvme_nvm_register_sysfs(ns))
1719 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
1720 ns->disk->disk_name);
1725 blk_cleanup_queue(ns->queue);
1726 out_release_instance:
1727 ida_simple_remove(&ctrl->ns_ida, ns->instance);
1732 static void nvme_ns_remove(struct nvme_ns *ns)
1734 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
1737 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
1738 if (blk_get_integrity(ns->disk))
1739 blk_integrity_unregister(ns->disk);
1740 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
1741 &nvme_ns_attr_group);
1743 nvme_nvm_unregister_sysfs(ns);
1744 del_gendisk(ns->disk);
1745 blk_mq_abort_requeue_list(ns->queue);
1746 blk_cleanup_queue(ns->queue);
1749 mutex_lock(&ns->ctrl->namespaces_mutex);
1750 list_del_init(&ns->list);
1751 mutex_unlock(&ns->ctrl->namespaces_mutex);
1756 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1760 ns = nvme_find_get_ns(ctrl, nsid);
1762 if (ns->disk && revalidate_disk(ns->disk))
1766 nvme_alloc_ns(ctrl, nsid);
1769 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
1772 struct nvme_ns *ns, *next;
1774 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
1775 if (ns->ns_id > nsid)
1780 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
1784 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
1787 ns_list = kzalloc(0x1000, GFP_KERNEL);
1791 for (i = 0; i < num_lists; i++) {
1792 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
1796 for (j = 0; j < min(nn, 1024U); j++) {
1797 nsid = le32_to_cpu(ns_list[j]);
1801 nvme_validate_ns(ctrl, nsid);
1803 while (++prev < nsid) {
1804 ns = nvme_find_get_ns(ctrl, prev);
1814 nvme_remove_invalid_namespaces(ctrl, prev);
1820 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
1824 for (i = 1; i <= nn; i++)
1825 nvme_validate_ns(ctrl, i);
1827 nvme_remove_invalid_namespaces(ctrl, nn);
1830 static void nvme_scan_work(struct work_struct *work)
1832 struct nvme_ctrl *ctrl =
1833 container_of(work, struct nvme_ctrl, scan_work);
1834 struct nvme_id_ctrl *id;
1837 if (ctrl->state != NVME_CTRL_LIVE)
1840 if (nvme_identify_ctrl(ctrl, &id))
1843 nn = le32_to_cpu(id->nn);
1844 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1845 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
1846 if (!nvme_scan_ns_list(ctrl, nn))
1849 nvme_scan_ns_sequential(ctrl, nn);
1851 mutex_lock(&ctrl->namespaces_mutex);
1852 list_sort(NULL, &ctrl->namespaces, ns_cmp);
1853 mutex_unlock(&ctrl->namespaces_mutex);
1857 void nvme_queue_scan(struct nvme_ctrl *ctrl)
1860 * Do not queue new scan work when a controller is reset during
1863 if (ctrl->state == NVME_CTRL_LIVE)
1864 schedule_work(&ctrl->scan_work);
1866 EXPORT_SYMBOL_GPL(nvme_queue_scan);
1869 * This function iterates the namespace list unlocked to allow recovery from
1870 * controller failure. It is up to the caller to ensure the namespace list is
1871 * not modified by scan work while this function is executing.
1873 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
1875 struct nvme_ns *ns, *next;
1878 * The dead states indicates the controller was not gracefully
1879 * disconnected. In that case, we won't be able to flush any data while
1880 * removing the namespaces' disks; fail all the queues now to avoid
1881 * potentially having to clean up the failed sync later.
1883 if (ctrl->state == NVME_CTRL_DEAD)
1884 nvme_kill_queues(ctrl);
1886 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
1889 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
1891 static void nvme_async_event_work(struct work_struct *work)
1893 struct nvme_ctrl *ctrl =
1894 container_of(work, struct nvme_ctrl, async_event_work);
1896 spin_lock_irq(&ctrl->lock);
1897 while (ctrl->event_limit > 0) {
1898 int aer_idx = --ctrl->event_limit;
1900 spin_unlock_irq(&ctrl->lock);
1901 ctrl->ops->submit_async_event(ctrl, aer_idx);
1902 spin_lock_irq(&ctrl->lock);
1904 spin_unlock_irq(&ctrl->lock);
1907 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
1908 union nvme_result *res)
1910 u32 result = le32_to_cpu(res->u32);
1913 switch (le16_to_cpu(status) >> 1) {
1914 case NVME_SC_SUCCESS:
1917 case NVME_SC_ABORT_REQ:
1918 ++ctrl->event_limit;
1919 schedule_work(&ctrl->async_event_work);
1928 switch (result & 0xff07) {
1929 case NVME_AER_NOTICE_NS_CHANGED:
1930 dev_info(ctrl->device, "rescanning\n");
1931 nvme_queue_scan(ctrl);
1934 dev_warn(ctrl->device, "async event result %08x\n", result);
1937 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
1939 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
1941 ctrl->event_limit = NVME_NR_AERS;
1942 schedule_work(&ctrl->async_event_work);
1944 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
1946 static DEFINE_IDA(nvme_instance_ida);
1948 static int nvme_set_instance(struct nvme_ctrl *ctrl)
1950 int instance, error;
1953 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
1956 spin_lock(&dev_list_lock);
1957 error = ida_get_new(&nvme_instance_ida, &instance);
1958 spin_unlock(&dev_list_lock);
1959 } while (error == -EAGAIN);
1964 ctrl->instance = instance;
1968 static void nvme_release_instance(struct nvme_ctrl *ctrl)
1970 spin_lock(&dev_list_lock);
1971 ida_remove(&nvme_instance_ida, ctrl->instance);
1972 spin_unlock(&dev_list_lock);
1975 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
1977 flush_work(&ctrl->async_event_work);
1978 flush_work(&ctrl->scan_work);
1979 nvme_remove_namespaces(ctrl);
1981 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
1983 spin_lock(&dev_list_lock);
1984 list_del(&ctrl->node);
1985 spin_unlock(&dev_list_lock);
1987 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
1989 static void nvme_free_ctrl(struct kref *kref)
1991 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
1993 put_device(ctrl->device);
1994 nvme_release_instance(ctrl);
1995 ida_destroy(&ctrl->ns_ida);
1997 ctrl->ops->free_ctrl(ctrl);
2000 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2002 kref_put(&ctrl->kref, nvme_free_ctrl);
2004 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2007 * Initialize a NVMe controller structures. This needs to be called during
2008 * earliest initialization so that we have the initialized structured around
2011 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2012 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2016 ctrl->state = NVME_CTRL_NEW;
2017 spin_lock_init(&ctrl->lock);
2018 INIT_LIST_HEAD(&ctrl->namespaces);
2019 mutex_init(&ctrl->namespaces_mutex);
2020 kref_init(&ctrl->kref);
2023 ctrl->quirks = quirks;
2024 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2025 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2027 ret = nvme_set_instance(ctrl);
2031 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2032 MKDEV(nvme_char_major, ctrl->instance),
2033 ctrl, nvme_dev_attr_groups,
2034 "nvme%d", ctrl->instance);
2035 if (IS_ERR(ctrl->device)) {
2036 ret = PTR_ERR(ctrl->device);
2037 goto out_release_instance;
2039 get_device(ctrl->device);
2040 ida_init(&ctrl->ns_ida);
2042 spin_lock(&dev_list_lock);
2043 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2044 spin_unlock(&dev_list_lock);
2047 out_release_instance:
2048 nvme_release_instance(ctrl);
2052 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2055 * nvme_kill_queues(): Ends all namespace queues
2056 * @ctrl: the dead controller that needs to end
2058 * Call this function when the driver determines it is unable to get the
2059 * controller in a state capable of servicing IO.
2061 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2065 mutex_lock(&ctrl->namespaces_mutex);
2066 list_for_each_entry(ns, &ctrl->namespaces, list) {
2068 * Revalidating a dead namespace sets capacity to 0. This will
2069 * end buffered writers dirtying pages that can't be synced.
2071 if (ns->disk && !test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2072 revalidate_disk(ns->disk);
2074 blk_set_queue_dying(ns->queue);
2075 blk_mq_abort_requeue_list(ns->queue);
2076 blk_mq_start_stopped_hw_queues(ns->queue, true);
2078 mutex_unlock(&ctrl->namespaces_mutex);
2080 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2082 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2086 mutex_lock(&ctrl->namespaces_mutex);
2087 list_for_each_entry(ns, &ctrl->namespaces, list)
2088 blk_mq_quiesce_queue(ns->queue);
2089 mutex_unlock(&ctrl->namespaces_mutex);
2091 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2093 void nvme_start_queues(struct nvme_ctrl *ctrl)
2097 mutex_lock(&ctrl->namespaces_mutex);
2098 list_for_each_entry(ns, &ctrl->namespaces, list) {
2099 blk_mq_start_stopped_hw_queues(ns->queue, true);
2100 blk_mq_kick_requeue_list(ns->queue);
2102 mutex_unlock(&ctrl->namespaces_mutex);
2104 EXPORT_SYMBOL_GPL(nvme_start_queues);
2106 int __init nvme_core_init(void)
2110 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2114 else if (result > 0)
2115 nvme_char_major = result;
2117 nvme_class = class_create(THIS_MODULE, "nvme");
2118 if (IS_ERR(nvme_class)) {
2119 result = PTR_ERR(nvme_class);
2120 goto unregister_chrdev;
2126 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2130 void nvme_core_exit(void)
2132 class_destroy(nvme_class);
2133 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2136 MODULE_LICENSE("GPL");
2137 MODULE_VERSION("1.0");
2138 module_init(nvme_core_init);
2139 module_exit(nvme_core_exit);