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>
29 #include <linux/pm_qos.h>
31 #include <asm/unaligned.h>
36 #define NVME_MINORS (1U << MINORBITS)
38 unsigned char admin_timeout = 60;
39 module_param(admin_timeout, byte, 0644);
40 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
41 EXPORT_SYMBOL_GPL(admin_timeout);
43 unsigned char nvme_io_timeout = 30;
44 module_param_named(io_timeout, nvme_io_timeout, byte, 0644);
45 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
46 EXPORT_SYMBOL_GPL(nvme_io_timeout);
48 unsigned char shutdown_timeout = 5;
49 module_param(shutdown_timeout, byte, 0644);
50 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
52 static u8 nvme_max_retries = 5;
53 module_param_named(max_retries, nvme_max_retries, byte, 0644);
54 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
56 static int nvme_char_major;
57 module_param(nvme_char_major, int, 0);
59 static unsigned long default_ps_max_latency_us = 25000;
60 module_param(default_ps_max_latency_us, ulong, 0644);
61 MODULE_PARM_DESC(default_ps_max_latency_us,
62 "max power saving latency for new devices; use PM QOS to change per device");
64 static bool force_apst;
65 module_param(force_apst, bool, 0644);
66 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
68 static LIST_HEAD(nvme_ctrl_list);
69 static DEFINE_SPINLOCK(dev_list_lock);
71 static struct class *nvme_class;
73 static blk_status_t nvme_error_status(struct request *req)
75 switch (nvme_req(req)->status & 0x7ff) {
78 case NVME_SC_CAP_EXCEEDED:
80 case NVME_SC_ONCS_NOT_SUPPORTED:
81 return BLK_STS_NOTSUPP;
82 case NVME_SC_WRITE_FAULT:
83 case NVME_SC_READ_ERROR:
84 case NVME_SC_UNWRITTEN_BLOCK:
85 return BLK_STS_MEDIUM;
91 static inline bool nvme_req_needs_retry(struct request *req)
93 if (blk_noretry_request(req))
95 if (nvme_req(req)->status & NVME_SC_DNR)
97 if (jiffies - req->start_time >= req->timeout)
99 if (nvme_req(req)->retries >= nvme_max_retries)
104 void nvme_complete_rq(struct request *req)
106 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
107 nvme_req(req)->retries++;
108 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
112 blk_mq_end_request(req, nvme_error_status(req));
114 EXPORT_SYMBOL_GPL(nvme_complete_rq);
116 void nvme_cancel_request(struct request *req, void *data, bool reserved)
120 if (!blk_mq_request_started(req))
123 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
124 "Cancelling I/O %d", req->tag);
126 status = NVME_SC_ABORT_REQ;
127 if (blk_queue_dying(req->q))
128 status |= NVME_SC_DNR;
129 nvme_req(req)->status = status;
130 blk_mq_complete_request(req);
133 EXPORT_SYMBOL_GPL(nvme_cancel_request);
135 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
136 enum nvme_ctrl_state new_state)
138 enum nvme_ctrl_state old_state;
139 bool changed = false;
141 spin_lock_irq(&ctrl->lock);
143 old_state = ctrl->state;
148 case NVME_CTRL_RESETTING:
149 case NVME_CTRL_RECONNECTING:
156 case NVME_CTRL_RESETTING:
160 case NVME_CTRL_RECONNECTING:
167 case NVME_CTRL_RECONNECTING:
176 case NVME_CTRL_DELETING:
179 case NVME_CTRL_RESETTING:
180 case NVME_CTRL_RECONNECTING:
189 case NVME_CTRL_DELETING:
201 ctrl->state = new_state;
203 spin_unlock_irq(&ctrl->lock);
207 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
209 static void nvme_free_ns(struct kref *kref)
211 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
214 nvme_nvm_unregister(ns);
217 spin_lock(&dev_list_lock);
218 ns->disk->private_data = NULL;
219 spin_unlock(&dev_list_lock);
223 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
224 nvme_put_ctrl(ns->ctrl);
228 static void nvme_put_ns(struct nvme_ns *ns)
230 kref_put(&ns->kref, nvme_free_ns);
233 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
237 spin_lock(&dev_list_lock);
238 ns = disk->private_data;
240 if (!kref_get_unless_zero(&ns->kref))
242 if (!try_module_get(ns->ctrl->ops->module))
245 spin_unlock(&dev_list_lock);
250 kref_put(&ns->kref, nvme_free_ns);
252 spin_unlock(&dev_list_lock);
256 struct request *nvme_alloc_request(struct request_queue *q,
257 struct nvme_command *cmd, unsigned int flags, int qid)
259 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
262 if (qid == NVME_QID_ANY) {
263 req = blk_mq_alloc_request(q, op, flags);
265 req = blk_mq_alloc_request_hctx(q, op, flags,
271 req->cmd_flags |= REQ_FAILFAST_DRIVER;
272 nvme_req(req)->cmd = cmd;
276 EXPORT_SYMBOL_GPL(nvme_alloc_request);
278 static inline void nvme_setup_flush(struct nvme_ns *ns,
279 struct nvme_command *cmnd)
281 memset(cmnd, 0, sizeof(*cmnd));
282 cmnd->common.opcode = nvme_cmd_flush;
283 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
286 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
287 struct nvme_command *cmnd)
289 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
290 struct nvme_dsm_range *range;
293 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
295 return BLK_STS_RESOURCE;
297 __rq_for_each_bio(bio, req) {
298 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
299 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
301 range[n].cattr = cpu_to_le32(0);
302 range[n].nlb = cpu_to_le32(nlb);
303 range[n].slba = cpu_to_le64(slba);
307 if (WARN_ON_ONCE(n != segments)) {
309 return BLK_STS_IOERR;
312 memset(cmnd, 0, sizeof(*cmnd));
313 cmnd->dsm.opcode = nvme_cmd_dsm;
314 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
315 cmnd->dsm.nr = cpu_to_le32(segments - 1);
316 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
318 req->special_vec.bv_page = virt_to_page(range);
319 req->special_vec.bv_offset = offset_in_page(range);
320 req->special_vec.bv_len = sizeof(*range) * segments;
321 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
326 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
327 struct nvme_command *cmnd)
332 if (req->cmd_flags & REQ_FUA)
333 control |= NVME_RW_FUA;
334 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
335 control |= NVME_RW_LR;
337 if (req->cmd_flags & REQ_RAHEAD)
338 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
340 memset(cmnd, 0, sizeof(*cmnd));
341 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
342 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
343 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
344 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
347 switch (ns->pi_type) {
348 case NVME_NS_DPS_PI_TYPE3:
349 control |= NVME_RW_PRINFO_PRCHK_GUARD;
351 case NVME_NS_DPS_PI_TYPE1:
352 case NVME_NS_DPS_PI_TYPE2:
353 control |= NVME_RW_PRINFO_PRCHK_GUARD |
354 NVME_RW_PRINFO_PRCHK_REF;
355 cmnd->rw.reftag = cpu_to_le32(
356 nvme_block_nr(ns, blk_rq_pos(req)));
359 if (!blk_integrity_rq(req))
360 control |= NVME_RW_PRINFO_PRACT;
363 cmnd->rw.control = cpu_to_le16(control);
364 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
367 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
368 struct nvme_command *cmd)
370 blk_status_t ret = BLK_STS_OK;
372 if (!(req->rq_flags & RQF_DONTPREP)) {
373 nvme_req(req)->retries = 0;
374 nvme_req(req)->flags = 0;
375 req->rq_flags |= RQF_DONTPREP;
378 switch (req_op(req)) {
381 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
384 nvme_setup_flush(ns, cmd);
386 case REQ_OP_WRITE_ZEROES:
387 /* currently only aliased to deallocate for a few ctrls: */
389 ret = nvme_setup_discard(ns, req, cmd);
393 nvme_setup_rw(ns, req, cmd);
397 return BLK_STS_IOERR;
400 cmd->common.command_id = req->tag;
403 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
406 * Returns 0 on success. If the result is negative, it's a Linux error code;
407 * if the result is positive, it's an NVM Express status code
409 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
410 union nvme_result *result, void *buffer, unsigned bufflen,
411 unsigned timeout, int qid, int at_head, int flags)
416 req = nvme_alloc_request(q, cmd, flags, qid);
420 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
422 if (buffer && bufflen) {
423 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
428 blk_execute_rq(req->q, NULL, req, at_head);
430 *result = nvme_req(req)->result;
431 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
434 ret = nvme_req(req)->status;
436 blk_mq_free_request(req);
439 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
441 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
442 void *buffer, unsigned bufflen)
444 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
447 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
449 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
450 void __user *ubuffer, unsigned bufflen,
451 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
452 u32 *result, unsigned timeout)
454 bool write = nvme_is_write(cmd);
455 struct nvme_ns *ns = q->queuedata;
456 struct gendisk *disk = ns ? ns->disk : NULL;
458 struct bio *bio = NULL;
462 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
466 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
468 if (ubuffer && bufflen) {
469 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
477 bio->bi_bdev = bdget_disk(disk, 0);
483 if (meta_buffer && meta_len) {
484 struct bio_integrity_payload *bip;
486 meta = kmalloc(meta_len, GFP_KERNEL);
493 if (copy_from_user(meta, meta_buffer,
500 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
506 bip->bip_iter.bi_size = meta_len;
507 bip->bip_iter.bi_sector = meta_seed;
509 ret = bio_integrity_add_page(bio, virt_to_page(meta),
510 meta_len, offset_in_page(meta));
511 if (ret != meta_len) {
518 blk_execute_rq(req->q, disk, req, 0);
519 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
522 ret = nvme_req(req)->status;
524 *result = le32_to_cpu(nvme_req(req)->result.u32);
525 if (meta && !ret && !write) {
526 if (copy_to_user(meta_buffer, meta, meta_len))
533 if (disk && bio->bi_bdev)
535 blk_rq_unmap_user(bio);
538 blk_mq_free_request(req);
542 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
543 void __user *ubuffer, unsigned bufflen, u32 *result,
546 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
550 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
552 struct nvme_ctrl *ctrl = rq->end_io_data;
554 blk_mq_free_request(rq);
557 dev_err(ctrl->device,
558 "failed nvme_keep_alive_end_io error=%d\n",
563 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
566 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
568 struct nvme_command c;
571 memset(&c, 0, sizeof(c));
572 c.common.opcode = nvme_admin_keep_alive;
574 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
579 rq->timeout = ctrl->kato * HZ;
580 rq->end_io_data = ctrl;
582 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
587 static void nvme_keep_alive_work(struct work_struct *work)
589 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
590 struct nvme_ctrl, ka_work);
592 if (nvme_keep_alive(ctrl)) {
593 /* allocation failure, reset the controller */
594 dev_err(ctrl->device, "keep-alive failed\n");
595 ctrl->ops->reset_ctrl(ctrl);
600 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
602 if (unlikely(ctrl->kato == 0))
605 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
606 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
608 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
610 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
612 if (unlikely(ctrl->kato == 0))
615 cancel_delayed_work_sync(&ctrl->ka_work);
617 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
619 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
621 struct nvme_command c = { };
624 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
625 c.identify.opcode = nvme_admin_identify;
626 c.identify.cns = NVME_ID_CNS_CTRL;
628 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
632 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
633 sizeof(struct nvme_id_ctrl));
639 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
641 struct nvme_command c = { };
643 c.identify.opcode = nvme_admin_identify;
644 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
645 c.identify.nsid = cpu_to_le32(nsid);
646 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
649 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
650 struct nvme_id_ns **id)
652 struct nvme_command c = { };
655 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
656 c.identify.opcode = nvme_admin_identify;
657 c.identify.nsid = cpu_to_le32(nsid);
658 c.identify.cns = NVME_ID_CNS_NS;
660 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
664 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
665 sizeof(struct nvme_id_ns));
671 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
672 void *buffer, size_t buflen, u32 *result)
674 struct nvme_command c;
675 union nvme_result res;
678 memset(&c, 0, sizeof(c));
679 c.features.opcode = nvme_admin_get_features;
680 c.features.nsid = cpu_to_le32(nsid);
681 c.features.fid = cpu_to_le32(fid);
683 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
685 if (ret >= 0 && result)
686 *result = le32_to_cpu(res.u32);
690 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
691 void *buffer, size_t buflen, u32 *result)
693 struct nvme_command c;
694 union nvme_result res;
697 memset(&c, 0, sizeof(c));
698 c.features.opcode = nvme_admin_set_features;
699 c.features.fid = cpu_to_le32(fid);
700 c.features.dword11 = cpu_to_le32(dword11);
702 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
703 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
704 if (ret >= 0 && result)
705 *result = le32_to_cpu(res.u32);
709 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
711 struct nvme_command c = { };
714 c.common.opcode = nvme_admin_get_log_page,
715 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
716 c.common.cdw10[0] = cpu_to_le32(
717 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
720 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
724 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
725 sizeof(struct nvme_smart_log));
731 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
733 u32 q_count = (*count - 1) | ((*count - 1) << 16);
735 int status, nr_io_queues;
737 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
743 * Degraded controllers might return an error when setting the queue
744 * count. We still want to be able to bring them online and offer
745 * access to the admin queue, as that might be only way to fix them up.
748 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
751 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
752 *count = min(*count, nr_io_queues);
757 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
759 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
761 struct nvme_user_io io;
762 struct nvme_command c;
763 unsigned length, meta_len;
764 void __user *metadata;
766 if (copy_from_user(&io, uio, sizeof(io)))
774 case nvme_cmd_compare:
780 length = (io.nblocks + 1) << ns->lba_shift;
781 meta_len = (io.nblocks + 1) * ns->ms;
782 metadata = (void __user *)(uintptr_t)io.metadata;
787 } else if (meta_len) {
788 if ((io.metadata & 3) || !io.metadata)
792 memset(&c, 0, sizeof(c));
793 c.rw.opcode = io.opcode;
794 c.rw.flags = io.flags;
795 c.rw.nsid = cpu_to_le32(ns->ns_id);
796 c.rw.slba = cpu_to_le64(io.slba);
797 c.rw.length = cpu_to_le16(io.nblocks);
798 c.rw.control = cpu_to_le16(io.control);
799 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
800 c.rw.reftag = cpu_to_le32(io.reftag);
801 c.rw.apptag = cpu_to_le16(io.apptag);
802 c.rw.appmask = cpu_to_le16(io.appmask);
804 return __nvme_submit_user_cmd(ns->queue, &c,
805 (void __user *)(uintptr_t)io.addr, length,
806 metadata, meta_len, io.slba, NULL, 0);
809 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
810 struct nvme_passthru_cmd __user *ucmd)
812 struct nvme_passthru_cmd cmd;
813 struct nvme_command c;
814 unsigned timeout = 0;
817 if (!capable(CAP_SYS_ADMIN))
819 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
824 memset(&c, 0, sizeof(c));
825 c.common.opcode = cmd.opcode;
826 c.common.flags = cmd.flags;
827 c.common.nsid = cpu_to_le32(cmd.nsid);
828 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
829 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
830 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
831 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
832 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
833 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
834 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
835 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
838 timeout = msecs_to_jiffies(cmd.timeout_ms);
840 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
841 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
842 &cmd.result, timeout);
844 if (put_user(cmd.result, &ucmd->result))
851 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
852 unsigned int cmd, unsigned long arg)
854 struct nvme_ns *ns = bdev->bd_disk->private_data;
858 force_successful_syscall_return();
860 case NVME_IOCTL_ADMIN_CMD:
861 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
862 case NVME_IOCTL_IO_CMD:
863 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
864 case NVME_IOCTL_SUBMIT_IO:
865 return nvme_submit_io(ns, (void __user *)arg);
866 #ifdef CONFIG_BLK_DEV_NVME_SCSI
867 case SG_GET_VERSION_NUM:
868 return nvme_sg_get_version_num((void __user *)arg);
870 return nvme_sg_io(ns, (void __user *)arg);
875 return nvme_nvm_ioctl(ns, cmd, arg);
877 if (is_sed_ioctl(cmd))
878 return sed_ioctl(ns->ctrl->opal_dev, cmd,
879 (void __user *) arg);
885 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
886 unsigned int cmd, unsigned long arg)
892 return nvme_ioctl(bdev, mode, cmd, arg);
895 #define nvme_compat_ioctl NULL
898 static int nvme_open(struct block_device *bdev, fmode_t mode)
900 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
903 static void nvme_release(struct gendisk *disk, fmode_t mode)
905 struct nvme_ns *ns = disk->private_data;
907 module_put(ns->ctrl->ops->module);
911 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
913 /* some standard values */
915 geo->sectors = 1 << 5;
916 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
920 #ifdef CONFIG_BLK_DEV_INTEGRITY
921 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
924 struct nvme_ns *ns = disk->private_data;
928 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
929 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
931 /* PI implementation requires metadata equal t10 pi tuple size */
932 if (ns->ms == sizeof(struct t10_pi_tuple))
933 pi_type = id->dps & NVME_NS_DPS_PI_MASK;
935 if (blk_get_integrity(disk) &&
936 (ns->pi_type != pi_type || ns->ms != old_ms ||
937 bs != queue_logical_block_size(disk->queue) ||
938 (ns->ms && ns->ext)))
939 blk_integrity_unregister(disk);
941 ns->pi_type = pi_type;
944 static void nvme_init_integrity(struct nvme_ns *ns)
946 struct blk_integrity integrity;
948 memset(&integrity, 0, sizeof(integrity));
949 switch (ns->pi_type) {
950 case NVME_NS_DPS_PI_TYPE3:
951 integrity.profile = &t10_pi_type3_crc;
952 integrity.tag_size = sizeof(u16) + sizeof(u32);
953 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
955 case NVME_NS_DPS_PI_TYPE1:
956 case NVME_NS_DPS_PI_TYPE2:
957 integrity.profile = &t10_pi_type1_crc;
958 integrity.tag_size = sizeof(u16);
959 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
962 integrity.profile = NULL;
965 integrity.tuple_size = ns->ms;
966 blk_integrity_register(ns->disk, &integrity);
967 blk_queue_max_integrity_segments(ns->queue, 1);
970 static void nvme_prep_integrity(struct gendisk *disk, struct nvme_id_ns *id,
974 static void nvme_init_integrity(struct nvme_ns *ns)
977 #endif /* CONFIG_BLK_DEV_INTEGRITY */
979 static void nvme_config_discard(struct nvme_ns *ns)
981 struct nvme_ctrl *ctrl = ns->ctrl;
982 u32 logical_block_size = queue_logical_block_size(ns->queue);
984 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
985 NVME_DSM_MAX_RANGES);
987 ns->queue->limits.discard_alignment = logical_block_size;
988 ns->queue->limits.discard_granularity = logical_block_size;
989 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
990 blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
991 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
993 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
994 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
997 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
999 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
1000 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
1004 if ((*id)->ncap == 0) {
1009 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
1010 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
1011 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
1012 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
1017 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1019 struct nvme_ns *ns = disk->private_data;
1023 * If identify namespace failed, use default 512 byte block size so
1024 * block layer can use before failing read/write for 0 capacity.
1026 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1027 if (ns->lba_shift == 0)
1029 bs = 1 << ns->lba_shift;
1031 blk_mq_freeze_queue(disk->queue);
1033 if (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
1034 nvme_prep_integrity(disk, id, bs);
1035 blk_queue_logical_block_size(ns->queue, bs);
1036 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1037 nvme_init_integrity(ns);
1038 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1039 set_capacity(disk, 0);
1041 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1043 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1044 nvme_config_discard(ns);
1045 blk_mq_unfreeze_queue(disk->queue);
1048 static int nvme_revalidate_disk(struct gendisk *disk)
1050 struct nvme_ns *ns = disk->private_data;
1051 struct nvme_id_ns *id = NULL;
1054 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1055 set_capacity(disk, 0);
1059 ret = nvme_revalidate_ns(ns, &id);
1063 __nvme_revalidate_disk(disk, id);
1069 static char nvme_pr_type(enum pr_type type)
1072 case PR_WRITE_EXCLUSIVE:
1074 case PR_EXCLUSIVE_ACCESS:
1076 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1078 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1080 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1082 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1089 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1090 u64 key, u64 sa_key, u8 op)
1092 struct nvme_ns *ns = bdev->bd_disk->private_data;
1093 struct nvme_command c;
1094 u8 data[16] = { 0, };
1096 put_unaligned_le64(key, &data[0]);
1097 put_unaligned_le64(sa_key, &data[8]);
1099 memset(&c, 0, sizeof(c));
1100 c.common.opcode = op;
1101 c.common.nsid = cpu_to_le32(ns->ns_id);
1102 c.common.cdw10[0] = cpu_to_le32(cdw10);
1104 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1107 static int nvme_pr_register(struct block_device *bdev, u64 old,
1108 u64 new, unsigned flags)
1112 if (flags & ~PR_FL_IGNORE_KEY)
1115 cdw10 = old ? 2 : 0;
1116 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1117 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1118 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1121 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1122 enum pr_type type, unsigned flags)
1126 if (flags & ~PR_FL_IGNORE_KEY)
1129 cdw10 = nvme_pr_type(type) << 8;
1130 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1131 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1134 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1135 enum pr_type type, bool abort)
1137 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1138 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1141 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1143 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1144 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1147 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1149 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1150 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1153 static const struct pr_ops nvme_pr_ops = {
1154 .pr_register = nvme_pr_register,
1155 .pr_reserve = nvme_pr_reserve,
1156 .pr_release = nvme_pr_release,
1157 .pr_preempt = nvme_pr_preempt,
1158 .pr_clear = nvme_pr_clear,
1161 #ifdef CONFIG_BLK_SED_OPAL
1162 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1165 struct nvme_ctrl *ctrl = data;
1166 struct nvme_command cmd;
1168 memset(&cmd, 0, sizeof(cmd));
1170 cmd.common.opcode = nvme_admin_security_send;
1172 cmd.common.opcode = nvme_admin_security_recv;
1173 cmd.common.nsid = 0;
1174 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1175 cmd.common.cdw10[1] = cpu_to_le32(len);
1177 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1178 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1180 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1181 #endif /* CONFIG_BLK_SED_OPAL */
1183 static const struct block_device_operations nvme_fops = {
1184 .owner = THIS_MODULE,
1185 .ioctl = nvme_ioctl,
1186 .compat_ioctl = nvme_compat_ioctl,
1188 .release = nvme_release,
1189 .getgeo = nvme_getgeo,
1190 .revalidate_disk= nvme_revalidate_disk,
1191 .pr_ops = &nvme_pr_ops,
1194 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1196 unsigned long timeout =
1197 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1198 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1201 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1204 if ((csts & NVME_CSTS_RDY) == bit)
1208 if (fatal_signal_pending(current))
1210 if (time_after(jiffies, timeout)) {
1211 dev_err(ctrl->device,
1212 "Device not ready; aborting %s\n", enabled ?
1213 "initialisation" : "reset");
1222 * If the device has been passed off to us in an enabled state, just clear
1223 * the enabled bit. The spec says we should set the 'shutdown notification
1224 * bits', but doing so may cause the device to complete commands to the
1225 * admin queue ... and we don't know what memory that might be pointing at!
1227 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1231 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1232 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1234 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1238 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1239 msleep(NVME_QUIRK_DELAY_AMOUNT);
1241 return nvme_wait_ready(ctrl, cap, false);
1243 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1245 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1248 * Default to a 4K page size, with the intention to update this
1249 * path in the future to accomodate architectures with differing
1250 * kernel and IO page sizes.
1252 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1255 if (page_shift < dev_page_min) {
1256 dev_err(ctrl->device,
1257 "Minimum device page size %u too large for host (%u)\n",
1258 1 << dev_page_min, 1 << page_shift);
1262 ctrl->page_size = 1 << page_shift;
1264 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1265 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1266 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1267 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1268 ctrl->ctrl_config |= NVME_CC_ENABLE;
1270 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1273 return nvme_wait_ready(ctrl, cap, true);
1275 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1277 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1279 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1283 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1284 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1286 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1290 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1291 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1295 if (fatal_signal_pending(current))
1297 if (time_after(jiffies, timeout)) {
1298 dev_err(ctrl->device,
1299 "Device shutdown incomplete; abort shutdown\n");
1306 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1308 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1309 struct request_queue *q)
1313 if (ctrl->max_hw_sectors) {
1315 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1317 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1318 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1320 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1321 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1322 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1323 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1325 blk_queue_write_cache(q, vwc, vwc);
1328 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1331 * APST (Autonomous Power State Transition) lets us program a
1332 * table of power state transitions that the controller will
1333 * perform automatically. We configure it with a simple
1334 * heuristic: we are willing to spend at most 2% of the time
1335 * transitioning between power states. Therefore, when running
1336 * in any given state, we will enter the next lower-power
1337 * non-operational state after waiting 50 * (enlat + exlat)
1338 * microseconds, as long as that state's total latency is under
1339 * the requested maximum latency.
1341 * We will not autonomously enter any non-operational state for
1342 * which the total latency exceeds ps_max_latency_us. Users
1343 * can set ps_max_latency_us to zero to turn off APST.
1347 struct nvme_feat_auto_pst *table;
1353 * If APST isn't supported or if we haven't been initialized yet,
1354 * then don't do anything.
1359 if (ctrl->npss > 31) {
1360 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1364 table = kzalloc(sizeof(*table), GFP_KERNEL);
1368 if (ctrl->ps_max_latency_us == 0) {
1369 /* Turn off APST. */
1371 dev_dbg(ctrl->device, "APST disabled\n");
1373 __le64 target = cpu_to_le64(0);
1377 * Walk through all states from lowest- to highest-power.
1378 * According to the spec, lower-numbered states use more
1379 * power. NPSS, despite the name, is the index of the
1380 * lowest-power state, not the number of states.
1382 for (state = (int)ctrl->npss; state >= 0; state--) {
1383 u64 total_latency_us, transition_ms;
1386 table->entries[state] = target;
1389 * Don't allow transitions to the deepest state
1390 * if it's quirked off.
1392 if (state == ctrl->npss &&
1393 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1397 * Is this state a useful non-operational state for
1398 * higher-power states to autonomously transition to?
1400 if (!(ctrl->psd[state].flags &
1401 NVME_PS_FLAGS_NON_OP_STATE))
1405 (u64)le32_to_cpu(ctrl->psd[state].entry_lat) +
1406 + le32_to_cpu(ctrl->psd[state].exit_lat);
1407 if (total_latency_us > ctrl->ps_max_latency_us)
1411 * This state is good. Use it as the APST idle
1412 * target for higher power states.
1414 transition_ms = total_latency_us + 19;
1415 do_div(transition_ms, 20);
1416 if (transition_ms > (1 << 24) - 1)
1417 transition_ms = (1 << 24) - 1;
1419 target = cpu_to_le64((state << 3) |
1420 (transition_ms << 8));
1425 if (total_latency_us > max_lat_us)
1426 max_lat_us = total_latency_us;
1432 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1434 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1435 max_ps, max_lat_us, (int)sizeof(*table), table);
1439 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1440 table, sizeof(*table), NULL);
1442 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1447 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1449 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1453 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1454 case PM_QOS_LATENCY_ANY:
1462 if (ctrl->ps_max_latency_us != latency) {
1463 ctrl->ps_max_latency_us = latency;
1464 nvme_configure_apst(ctrl);
1468 struct nvme_core_quirk_entry {
1470 * NVMe model and firmware strings are padded with spaces. For
1471 * simplicity, strings in the quirk table are padded with NULLs
1477 unsigned long quirks;
1480 static const struct nvme_core_quirk_entry core_quirks[] = {
1483 * This Toshiba device seems to die using any APST states. See:
1484 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1487 .mn = "THNSF5256GPUK TOSHIBA",
1488 .quirks = NVME_QUIRK_NO_APST,
1492 /* match is null-terminated but idstr is space-padded. */
1493 static bool string_matches(const char *idstr, const char *match, size_t len)
1500 matchlen = strlen(match);
1501 WARN_ON_ONCE(matchlen > len);
1503 if (memcmp(idstr, match, matchlen))
1506 for (; matchlen < len; matchlen++)
1507 if (idstr[matchlen] != ' ')
1513 static bool quirk_matches(const struct nvme_id_ctrl *id,
1514 const struct nvme_core_quirk_entry *q)
1516 return q->vid == le16_to_cpu(id->vid) &&
1517 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1518 string_matches(id->fr, q->fr, sizeof(id->fr));
1522 * Initialize the cached copies of the Identify data and various controller
1523 * register in our nvme_ctrl structure. This should be called as soon as
1524 * the admin queue is fully up and running.
1526 int nvme_init_identify(struct nvme_ctrl *ctrl)
1528 struct nvme_id_ctrl *id;
1530 int ret, page_shift;
1534 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1536 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1540 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1542 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1545 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1547 if (ctrl->vs >= NVME_VS(1, 1, 0))
1548 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1550 ret = nvme_identify_ctrl(ctrl, &id);
1552 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1556 if (!ctrl->identified) {
1558 * Check for quirks. Quirk can depend on firmware version,
1559 * so, in principle, the set of quirks present can change
1560 * across a reset. As a possible future enhancement, we
1561 * could re-scan for quirks every time we reinitialize
1562 * the device, but we'd have to make sure that the driver
1563 * behaves intelligently if the quirks change.
1568 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1569 if (quirk_matches(id, &core_quirks[i]))
1570 ctrl->quirks |= core_quirks[i].quirks;
1574 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1575 dev_warn(ctrl->dev, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1576 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1579 ctrl->oacs = le16_to_cpu(id->oacs);
1580 ctrl->vid = le16_to_cpu(id->vid);
1581 ctrl->oncs = le16_to_cpup(&id->oncs);
1582 atomic_set(&ctrl->abort_limit, id->acl + 1);
1583 ctrl->vwc = id->vwc;
1584 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1585 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1586 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1587 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1589 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1591 max_hw_sectors = UINT_MAX;
1592 ctrl->max_hw_sectors =
1593 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1595 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1596 ctrl->sgls = le32_to_cpu(id->sgls);
1597 ctrl->kas = le16_to_cpu(id->kas);
1599 ctrl->npss = id->npss;
1600 prev_apsta = ctrl->apsta;
1601 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1602 if (force_apst && id->apsta) {
1603 dev_warn(ctrl->dev, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1609 ctrl->apsta = id->apsta;
1611 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1613 if (ctrl->ops->flags & NVME_F_FABRICS) {
1614 ctrl->icdoff = le16_to_cpu(id->icdoff);
1615 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1616 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1617 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1620 * In fabrics we need to verify the cntlid matches the
1623 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1626 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1628 "keep-alive support is mandatory for fabrics\n");
1632 ctrl->cntlid = le16_to_cpu(id->cntlid);
1637 if (ctrl->apsta && !prev_apsta)
1638 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1639 else if (!ctrl->apsta && prev_apsta)
1640 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1642 nvme_configure_apst(ctrl);
1644 ctrl->identified = true;
1648 EXPORT_SYMBOL_GPL(nvme_init_identify);
1650 static int nvme_dev_open(struct inode *inode, struct file *file)
1652 struct nvme_ctrl *ctrl;
1653 int instance = iminor(inode);
1656 spin_lock(&dev_list_lock);
1657 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1658 if (ctrl->instance != instance)
1661 if (!ctrl->admin_q) {
1665 if (!kref_get_unless_zero(&ctrl->kref))
1667 file->private_data = ctrl;
1671 spin_unlock(&dev_list_lock);
1676 static int nvme_dev_release(struct inode *inode, struct file *file)
1678 nvme_put_ctrl(file->private_data);
1682 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1687 mutex_lock(&ctrl->namespaces_mutex);
1688 if (list_empty(&ctrl->namespaces)) {
1693 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1694 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1695 dev_warn(ctrl->device,
1696 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1701 dev_warn(ctrl->device,
1702 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1703 kref_get(&ns->kref);
1704 mutex_unlock(&ctrl->namespaces_mutex);
1706 ret = nvme_user_cmd(ctrl, ns, argp);
1711 mutex_unlock(&ctrl->namespaces_mutex);
1715 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1718 struct nvme_ctrl *ctrl = file->private_data;
1719 void __user *argp = (void __user *)arg;
1722 case NVME_IOCTL_ADMIN_CMD:
1723 return nvme_user_cmd(ctrl, NULL, argp);
1724 case NVME_IOCTL_IO_CMD:
1725 return nvme_dev_user_cmd(ctrl, argp);
1726 case NVME_IOCTL_RESET:
1727 dev_warn(ctrl->device, "resetting controller\n");
1728 return ctrl->ops->reset_ctrl(ctrl);
1729 case NVME_IOCTL_SUBSYS_RESET:
1730 return nvme_reset_subsystem(ctrl);
1731 case NVME_IOCTL_RESCAN:
1732 nvme_queue_scan(ctrl);
1739 static const struct file_operations nvme_dev_fops = {
1740 .owner = THIS_MODULE,
1741 .open = nvme_dev_open,
1742 .release = nvme_dev_release,
1743 .unlocked_ioctl = nvme_dev_ioctl,
1744 .compat_ioctl = nvme_dev_ioctl,
1747 static ssize_t nvme_sysfs_reset(struct device *dev,
1748 struct device_attribute *attr, const char *buf,
1751 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1754 ret = ctrl->ops->reset_ctrl(ctrl);
1759 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1761 static ssize_t nvme_sysfs_rescan(struct device *dev,
1762 struct device_attribute *attr, const char *buf,
1765 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1767 nvme_queue_scan(ctrl);
1770 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1772 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1775 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1776 struct nvme_ctrl *ctrl = ns->ctrl;
1777 int serial_len = sizeof(ctrl->serial);
1778 int model_len = sizeof(ctrl->model);
1780 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1781 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1783 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1784 return sprintf(buf, "eui.%8phN\n", ns->eui);
1786 while (ctrl->serial[serial_len - 1] == ' ')
1788 while (ctrl->model[model_len - 1] == ' ')
1791 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1792 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1794 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1796 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1799 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1800 return sprintf(buf, "%pU\n", ns->uuid);
1802 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1804 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1807 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1808 return sprintf(buf, "%8phd\n", ns->eui);
1810 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1812 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1815 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1816 return sprintf(buf, "%d\n", ns->ns_id);
1818 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1820 static struct attribute *nvme_ns_attrs[] = {
1821 &dev_attr_wwid.attr,
1822 &dev_attr_uuid.attr,
1824 &dev_attr_nsid.attr,
1828 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1829 struct attribute *a, int n)
1831 struct device *dev = container_of(kobj, struct device, kobj);
1832 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1834 if (a == &dev_attr_uuid.attr) {
1835 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1838 if (a == &dev_attr_eui.attr) {
1839 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1845 static const struct attribute_group nvme_ns_attr_group = {
1846 .attrs = nvme_ns_attrs,
1847 .is_visible = nvme_ns_attrs_are_visible,
1850 #define nvme_show_str_function(field) \
1851 static ssize_t field##_show(struct device *dev, \
1852 struct device_attribute *attr, char *buf) \
1854 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1855 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1857 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1859 #define nvme_show_int_function(field) \
1860 static ssize_t field##_show(struct device *dev, \
1861 struct device_attribute *attr, char *buf) \
1863 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1864 return sprintf(buf, "%d\n", ctrl->field); \
1866 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1868 nvme_show_str_function(model);
1869 nvme_show_str_function(serial);
1870 nvme_show_str_function(firmware_rev);
1871 nvme_show_int_function(cntlid);
1873 static ssize_t nvme_sysfs_delete(struct device *dev,
1874 struct device_attribute *attr, const char *buf,
1877 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1879 if (device_remove_file_self(dev, attr))
1880 ctrl->ops->delete_ctrl(ctrl);
1883 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1885 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1886 struct device_attribute *attr,
1889 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1891 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1893 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1895 static ssize_t nvme_sysfs_show_state(struct device *dev,
1896 struct device_attribute *attr,
1899 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1900 static const char *const state_name[] = {
1901 [NVME_CTRL_NEW] = "new",
1902 [NVME_CTRL_LIVE] = "live",
1903 [NVME_CTRL_RESETTING] = "resetting",
1904 [NVME_CTRL_RECONNECTING]= "reconnecting",
1905 [NVME_CTRL_DELETING] = "deleting",
1906 [NVME_CTRL_DEAD] = "dead",
1909 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1910 state_name[ctrl->state])
1911 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1913 return sprintf(buf, "unknown state\n");
1916 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1918 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1919 struct device_attribute *attr,
1922 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1924 return snprintf(buf, PAGE_SIZE, "%s\n",
1925 ctrl->ops->get_subsysnqn(ctrl));
1927 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1929 static ssize_t nvme_sysfs_show_address(struct device *dev,
1930 struct device_attribute *attr,
1933 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1935 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1937 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1939 static struct attribute *nvme_dev_attrs[] = {
1940 &dev_attr_reset_controller.attr,
1941 &dev_attr_rescan_controller.attr,
1942 &dev_attr_model.attr,
1943 &dev_attr_serial.attr,
1944 &dev_attr_firmware_rev.attr,
1945 &dev_attr_cntlid.attr,
1946 &dev_attr_delete_controller.attr,
1947 &dev_attr_transport.attr,
1948 &dev_attr_subsysnqn.attr,
1949 &dev_attr_address.attr,
1950 &dev_attr_state.attr,
1954 #define CHECK_ATTR(ctrl, a, name) \
1955 if ((a) == &dev_attr_##name.attr && \
1956 !(ctrl)->ops->get_##name) \
1959 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1960 struct attribute *a, int n)
1962 struct device *dev = container_of(kobj, struct device, kobj);
1963 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1965 if (a == &dev_attr_delete_controller.attr) {
1966 if (!ctrl->ops->delete_ctrl)
1970 CHECK_ATTR(ctrl, a, subsysnqn);
1971 CHECK_ATTR(ctrl, a, address);
1976 static struct attribute_group nvme_dev_attrs_group = {
1977 .attrs = nvme_dev_attrs,
1978 .is_visible = nvme_dev_attrs_are_visible,
1981 static const struct attribute_group *nvme_dev_attr_groups[] = {
1982 &nvme_dev_attrs_group,
1986 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1988 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1989 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1991 return nsa->ns_id - nsb->ns_id;
1994 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1996 struct nvme_ns *ns, *ret = NULL;
1998 mutex_lock(&ctrl->namespaces_mutex);
1999 list_for_each_entry(ns, &ctrl->namespaces, list) {
2000 if (ns->ns_id == nsid) {
2001 kref_get(&ns->kref);
2005 if (ns->ns_id > nsid)
2008 mutex_unlock(&ctrl->namespaces_mutex);
2012 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2015 struct gendisk *disk;
2016 struct nvme_id_ns *id;
2017 char disk_name[DISK_NAME_LEN];
2018 int node = dev_to_node(ctrl->dev);
2020 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2024 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2025 if (ns->instance < 0)
2028 ns->queue = blk_mq_init_queue(ctrl->tagset);
2029 if (IS_ERR(ns->queue))
2030 goto out_release_instance;
2031 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2032 ns->queue->queuedata = ns;
2035 kref_init(&ns->kref);
2037 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2039 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2040 nvme_set_queue_limits(ctrl, ns->queue);
2042 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2044 if (nvme_revalidate_ns(ns, &id))
2045 goto out_free_queue;
2047 if (nvme_nvm_ns_supported(ns, id) &&
2048 nvme_nvm_register(ns, disk_name, node)) {
2049 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
2053 disk = alloc_disk_node(0, node);
2057 disk->fops = &nvme_fops;
2058 disk->private_data = ns;
2059 disk->queue = ns->queue;
2060 disk->flags = GENHD_FL_EXT_DEVT;
2061 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2064 __nvme_revalidate_disk(disk, id);
2066 mutex_lock(&ctrl->namespaces_mutex);
2067 list_add_tail(&ns->list, &ctrl->namespaces);
2068 mutex_unlock(&ctrl->namespaces_mutex);
2070 kref_get(&ctrl->kref);
2074 device_add_disk(ctrl->device, ns->disk);
2075 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2076 &nvme_ns_attr_group))
2077 pr_warn("%s: failed to create sysfs group for identification\n",
2078 ns->disk->disk_name);
2079 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2080 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2081 ns->disk->disk_name);
2086 blk_cleanup_queue(ns->queue);
2087 out_release_instance:
2088 ida_simple_remove(&ctrl->ns_ida, ns->instance);
2093 static void nvme_ns_remove(struct nvme_ns *ns)
2095 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2098 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2099 if (blk_get_integrity(ns->disk))
2100 blk_integrity_unregister(ns->disk);
2101 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2102 &nvme_ns_attr_group);
2104 nvme_nvm_unregister_sysfs(ns);
2105 del_gendisk(ns->disk);
2106 blk_cleanup_queue(ns->queue);
2109 mutex_lock(&ns->ctrl->namespaces_mutex);
2110 list_del_init(&ns->list);
2111 mutex_unlock(&ns->ctrl->namespaces_mutex);
2116 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2120 ns = nvme_find_get_ns(ctrl, nsid);
2122 if (ns->disk && revalidate_disk(ns->disk))
2126 nvme_alloc_ns(ctrl, nsid);
2129 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2132 struct nvme_ns *ns, *next;
2134 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2135 if (ns->ns_id > nsid)
2140 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2144 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2147 ns_list = kzalloc(0x1000, GFP_KERNEL);
2151 for (i = 0; i < num_lists; i++) {
2152 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2156 for (j = 0; j < min(nn, 1024U); j++) {
2157 nsid = le32_to_cpu(ns_list[j]);
2161 nvme_validate_ns(ctrl, nsid);
2163 while (++prev < nsid) {
2164 ns = nvme_find_get_ns(ctrl, prev);
2174 nvme_remove_invalid_namespaces(ctrl, prev);
2180 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2184 for (i = 1; i <= nn; i++)
2185 nvme_validate_ns(ctrl, i);
2187 nvme_remove_invalid_namespaces(ctrl, nn);
2190 static void nvme_scan_work(struct work_struct *work)
2192 struct nvme_ctrl *ctrl =
2193 container_of(work, struct nvme_ctrl, scan_work);
2194 struct nvme_id_ctrl *id;
2197 if (ctrl->state != NVME_CTRL_LIVE)
2200 if (nvme_identify_ctrl(ctrl, &id))
2203 nn = le32_to_cpu(id->nn);
2204 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2205 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2206 if (!nvme_scan_ns_list(ctrl, nn))
2209 nvme_scan_ns_sequential(ctrl, nn);
2211 mutex_lock(&ctrl->namespaces_mutex);
2212 list_sort(NULL, &ctrl->namespaces, ns_cmp);
2213 mutex_unlock(&ctrl->namespaces_mutex);
2217 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2220 * Do not queue new scan work when a controller is reset during
2223 if (ctrl->state == NVME_CTRL_LIVE)
2224 schedule_work(&ctrl->scan_work);
2226 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2229 * This function iterates the namespace list unlocked to allow recovery from
2230 * controller failure. It is up to the caller to ensure the namespace list is
2231 * not modified by scan work while this function is executing.
2233 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2235 struct nvme_ns *ns, *next;
2238 * The dead states indicates the controller was not gracefully
2239 * disconnected. In that case, we won't be able to flush any data while
2240 * removing the namespaces' disks; fail all the queues now to avoid
2241 * potentially having to clean up the failed sync later.
2243 if (ctrl->state == NVME_CTRL_DEAD)
2244 nvme_kill_queues(ctrl);
2246 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2249 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2251 static void nvme_async_event_work(struct work_struct *work)
2253 struct nvme_ctrl *ctrl =
2254 container_of(work, struct nvme_ctrl, async_event_work);
2256 spin_lock_irq(&ctrl->lock);
2257 while (ctrl->event_limit > 0) {
2258 int aer_idx = --ctrl->event_limit;
2260 spin_unlock_irq(&ctrl->lock);
2261 ctrl->ops->submit_async_event(ctrl, aer_idx);
2262 spin_lock_irq(&ctrl->lock);
2264 spin_unlock_irq(&ctrl->lock);
2267 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2268 union nvme_result *res)
2270 u32 result = le32_to_cpu(res->u32);
2273 switch (le16_to_cpu(status) >> 1) {
2274 case NVME_SC_SUCCESS:
2277 case NVME_SC_ABORT_REQ:
2278 ++ctrl->event_limit;
2279 schedule_work(&ctrl->async_event_work);
2288 switch (result & 0xff07) {
2289 case NVME_AER_NOTICE_NS_CHANGED:
2290 dev_info(ctrl->device, "rescanning\n");
2291 nvme_queue_scan(ctrl);
2294 dev_warn(ctrl->device, "async event result %08x\n", result);
2297 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2299 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2301 ctrl->event_limit = NVME_NR_AERS;
2302 schedule_work(&ctrl->async_event_work);
2304 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2306 static DEFINE_IDA(nvme_instance_ida);
2308 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2310 int instance, error;
2313 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2316 spin_lock(&dev_list_lock);
2317 error = ida_get_new(&nvme_instance_ida, &instance);
2318 spin_unlock(&dev_list_lock);
2319 } while (error == -EAGAIN);
2324 ctrl->instance = instance;
2328 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2330 spin_lock(&dev_list_lock);
2331 ida_remove(&nvme_instance_ida, ctrl->instance);
2332 spin_unlock(&dev_list_lock);
2335 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2337 flush_work(&ctrl->async_event_work);
2338 flush_work(&ctrl->scan_work);
2339 nvme_remove_namespaces(ctrl);
2341 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2343 spin_lock(&dev_list_lock);
2344 list_del(&ctrl->node);
2345 spin_unlock(&dev_list_lock);
2347 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2349 static void nvme_free_ctrl(struct kref *kref)
2351 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2353 put_device(ctrl->device);
2354 nvme_release_instance(ctrl);
2355 ida_destroy(&ctrl->ns_ida);
2357 ctrl->ops->free_ctrl(ctrl);
2360 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2362 kref_put(&ctrl->kref, nvme_free_ctrl);
2364 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2367 * Initialize a NVMe controller structures. This needs to be called during
2368 * earliest initialization so that we have the initialized structured around
2371 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2372 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2376 ctrl->state = NVME_CTRL_NEW;
2377 spin_lock_init(&ctrl->lock);
2378 INIT_LIST_HEAD(&ctrl->namespaces);
2379 mutex_init(&ctrl->namespaces_mutex);
2380 kref_init(&ctrl->kref);
2383 ctrl->quirks = quirks;
2384 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2385 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2387 ret = nvme_set_instance(ctrl);
2391 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2392 MKDEV(nvme_char_major, ctrl->instance),
2393 ctrl, nvme_dev_attr_groups,
2394 "nvme%d", ctrl->instance);
2395 if (IS_ERR(ctrl->device)) {
2396 ret = PTR_ERR(ctrl->device);
2397 goto out_release_instance;
2399 get_device(ctrl->device);
2400 ida_init(&ctrl->ns_ida);
2402 spin_lock(&dev_list_lock);
2403 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2404 spin_unlock(&dev_list_lock);
2407 * Initialize latency tolerance controls. The sysfs files won't
2408 * be visible to userspace unless the device actually supports APST.
2410 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2411 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2412 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2415 out_release_instance:
2416 nvme_release_instance(ctrl);
2420 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2423 * nvme_kill_queues(): Ends all namespace queues
2424 * @ctrl: the dead controller that needs to end
2426 * Call this function when the driver determines it is unable to get the
2427 * controller in a state capable of servicing IO.
2429 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2433 mutex_lock(&ctrl->namespaces_mutex);
2434 list_for_each_entry(ns, &ctrl->namespaces, list) {
2436 * Revalidating a dead namespace sets capacity to 0. This will
2437 * end buffered writers dirtying pages that can't be synced.
2439 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2441 revalidate_disk(ns->disk);
2442 blk_set_queue_dying(ns->queue);
2445 * Forcibly start all queues to avoid having stuck requests.
2446 * Note that we must ensure the queues are not stopped
2447 * when the final removal happens.
2449 blk_mq_start_hw_queues(ns->queue);
2451 /* draining requests in requeue list */
2452 blk_mq_kick_requeue_list(ns->queue);
2454 mutex_unlock(&ctrl->namespaces_mutex);
2456 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2458 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2462 mutex_lock(&ctrl->namespaces_mutex);
2463 list_for_each_entry(ns, &ctrl->namespaces, list)
2464 blk_mq_unfreeze_queue(ns->queue);
2465 mutex_unlock(&ctrl->namespaces_mutex);
2467 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2469 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2473 mutex_lock(&ctrl->namespaces_mutex);
2474 list_for_each_entry(ns, &ctrl->namespaces, list) {
2475 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2479 mutex_unlock(&ctrl->namespaces_mutex);
2481 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2483 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2487 mutex_lock(&ctrl->namespaces_mutex);
2488 list_for_each_entry(ns, &ctrl->namespaces, list)
2489 blk_mq_freeze_queue_wait(ns->queue);
2490 mutex_unlock(&ctrl->namespaces_mutex);
2492 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2494 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2498 mutex_lock(&ctrl->namespaces_mutex);
2499 list_for_each_entry(ns, &ctrl->namespaces, list)
2500 blk_freeze_queue_start(ns->queue);
2501 mutex_unlock(&ctrl->namespaces_mutex);
2503 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2505 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2509 mutex_lock(&ctrl->namespaces_mutex);
2510 list_for_each_entry(ns, &ctrl->namespaces, list)
2511 blk_mq_quiesce_queue(ns->queue);
2512 mutex_unlock(&ctrl->namespaces_mutex);
2514 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2516 void nvme_start_queues(struct nvme_ctrl *ctrl)
2520 mutex_lock(&ctrl->namespaces_mutex);
2521 list_for_each_entry(ns, &ctrl->namespaces, list) {
2522 blk_mq_start_stopped_hw_queues(ns->queue, true);
2523 blk_mq_kick_requeue_list(ns->queue);
2525 mutex_unlock(&ctrl->namespaces_mutex);
2527 EXPORT_SYMBOL_GPL(nvme_start_queues);
2529 int __init nvme_core_init(void)
2533 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2537 else if (result > 0)
2538 nvme_char_major = result;
2540 nvme_class = class_create(THIS_MODULE, "nvme");
2541 if (IS_ERR(nvme_class)) {
2542 result = PTR_ERR(nvme_class);
2543 goto unregister_chrdev;
2549 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2553 void nvme_core_exit(void)
2555 class_destroy(nvme_class);
2556 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2559 MODULE_LICENSE("GPL");
2560 MODULE_VERSION("1.0");
2561 module_init(nvme_core_init);
2562 module_exit(nvme_core_exit);