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 int nvme_error_status(struct request *req)
75 switch (nvme_req(req)->status & 0x7ff) {
78 case NVME_SC_CAP_EXCEEDED:
84 * XXX: these errors are a nasty side-band protocol to
85 * drivers/md/dm-mpath.c:noretry_error() that aren't documented
88 case NVME_SC_CMD_SEQ_ERROR:
90 case NVME_SC_ONCS_NOT_SUPPORTED:
92 case NVME_SC_WRITE_FAULT:
93 case NVME_SC_READ_ERROR:
94 case NVME_SC_UNWRITTEN_BLOCK:
99 static inline bool nvme_req_needs_retry(struct request *req)
101 if (blk_noretry_request(req))
103 if (nvme_req(req)->status & NVME_SC_DNR)
105 if (jiffies - req->start_time >= req->timeout)
107 if (nvme_req(req)->retries >= nvme_max_retries)
112 void nvme_complete_rq(struct request *req)
114 if (unlikely(nvme_req(req)->status && nvme_req_needs_retry(req))) {
115 nvme_req(req)->retries++;
116 blk_mq_requeue_request(req, !blk_mq_queue_stopped(req->q));
120 blk_mq_end_request(req, nvme_error_status(req));
122 EXPORT_SYMBOL_GPL(nvme_complete_rq);
124 void nvme_cancel_request(struct request *req, void *data, bool reserved)
128 if (!blk_mq_request_started(req))
131 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
132 "Cancelling I/O %d", req->tag);
134 status = NVME_SC_ABORT_REQ;
135 if (blk_queue_dying(req->q))
136 status |= NVME_SC_DNR;
137 nvme_req(req)->status = status;
138 blk_mq_complete_request(req);
141 EXPORT_SYMBOL_GPL(nvme_cancel_request);
143 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
144 enum nvme_ctrl_state new_state)
146 enum nvme_ctrl_state old_state;
147 bool changed = false;
149 spin_lock_irq(&ctrl->lock);
151 old_state = ctrl->state;
156 case NVME_CTRL_RESETTING:
157 case NVME_CTRL_RECONNECTING:
164 case NVME_CTRL_RESETTING:
168 case NVME_CTRL_RECONNECTING:
175 case NVME_CTRL_RECONNECTING:
184 case NVME_CTRL_DELETING:
187 case NVME_CTRL_RESETTING:
188 case NVME_CTRL_RECONNECTING:
197 case NVME_CTRL_DELETING:
209 ctrl->state = new_state;
211 spin_unlock_irq(&ctrl->lock);
215 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
217 static void nvme_free_ns(struct kref *kref)
219 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
222 nvme_nvm_unregister(ns);
225 spin_lock(&dev_list_lock);
226 ns->disk->private_data = NULL;
227 spin_unlock(&dev_list_lock);
231 ida_simple_remove(&ns->ctrl->ns_ida, ns->instance);
232 nvme_put_ctrl(ns->ctrl);
236 static void nvme_put_ns(struct nvme_ns *ns)
238 kref_put(&ns->kref, nvme_free_ns);
241 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk)
245 spin_lock(&dev_list_lock);
246 ns = disk->private_data;
248 if (!kref_get_unless_zero(&ns->kref))
250 if (!try_module_get(ns->ctrl->ops->module))
253 spin_unlock(&dev_list_lock);
258 kref_put(&ns->kref, nvme_free_ns);
260 spin_unlock(&dev_list_lock);
264 struct request *nvme_alloc_request(struct request_queue *q,
265 struct nvme_command *cmd, unsigned int flags, int qid)
267 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
270 if (qid == NVME_QID_ANY) {
271 req = blk_mq_alloc_request(q, op, flags);
273 req = blk_mq_alloc_request_hctx(q, op, flags,
279 req->cmd_flags |= REQ_FAILFAST_DRIVER;
280 nvme_req(req)->cmd = cmd;
284 EXPORT_SYMBOL_GPL(nvme_alloc_request);
286 static inline void nvme_setup_flush(struct nvme_ns *ns,
287 struct nvme_command *cmnd)
289 memset(cmnd, 0, sizeof(*cmnd));
290 cmnd->common.opcode = nvme_cmd_flush;
291 cmnd->common.nsid = cpu_to_le32(ns->ns_id);
294 static inline int nvme_setup_discard(struct nvme_ns *ns, struct request *req,
295 struct nvme_command *cmnd)
297 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
298 struct nvme_dsm_range *range;
301 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
303 return BLK_MQ_RQ_QUEUE_BUSY;
305 __rq_for_each_bio(bio, req) {
306 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
307 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
309 range[n].cattr = cpu_to_le32(0);
310 range[n].nlb = cpu_to_le32(nlb);
311 range[n].slba = cpu_to_le64(slba);
315 if (WARN_ON_ONCE(n != segments)) {
317 return BLK_MQ_RQ_QUEUE_ERROR;
320 memset(cmnd, 0, sizeof(*cmnd));
321 cmnd->dsm.opcode = nvme_cmd_dsm;
322 cmnd->dsm.nsid = cpu_to_le32(ns->ns_id);
323 cmnd->dsm.nr = cpu_to_le32(segments - 1);
324 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
326 req->special_vec.bv_page = virt_to_page(range);
327 req->special_vec.bv_offset = offset_in_page(range);
328 req->special_vec.bv_len = sizeof(*range) * segments;
329 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
331 return BLK_MQ_RQ_QUEUE_OK;
334 static inline void nvme_setup_rw(struct nvme_ns *ns, struct request *req,
335 struct nvme_command *cmnd)
340 if (req->cmd_flags & REQ_FUA)
341 control |= NVME_RW_FUA;
342 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
343 control |= NVME_RW_LR;
345 if (req->cmd_flags & REQ_RAHEAD)
346 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
348 memset(cmnd, 0, sizeof(*cmnd));
349 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
350 cmnd->rw.nsid = cpu_to_le32(ns->ns_id);
351 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
352 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
355 switch (ns->pi_type) {
356 case NVME_NS_DPS_PI_TYPE3:
357 control |= NVME_RW_PRINFO_PRCHK_GUARD;
359 case NVME_NS_DPS_PI_TYPE1:
360 case NVME_NS_DPS_PI_TYPE2:
361 control |= NVME_RW_PRINFO_PRCHK_GUARD |
362 NVME_RW_PRINFO_PRCHK_REF;
363 cmnd->rw.reftag = cpu_to_le32(
364 nvme_block_nr(ns, blk_rq_pos(req)));
367 if (!blk_integrity_rq(req))
368 control |= NVME_RW_PRINFO_PRACT;
371 cmnd->rw.control = cpu_to_le16(control);
372 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
375 int nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
376 struct nvme_command *cmd)
378 int ret = BLK_MQ_RQ_QUEUE_OK;
380 if (!(req->rq_flags & RQF_DONTPREP)) {
381 nvme_req(req)->retries = 0;
382 nvme_req(req)->flags = 0;
383 req->rq_flags |= RQF_DONTPREP;
386 switch (req_op(req)) {
389 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
392 nvme_setup_flush(ns, cmd);
394 case REQ_OP_WRITE_ZEROES:
395 /* currently only aliased to deallocate for a few ctrls: */
397 ret = nvme_setup_discard(ns, req, cmd);
401 nvme_setup_rw(ns, req, cmd);
405 return BLK_MQ_RQ_QUEUE_ERROR;
408 cmd->common.command_id = req->tag;
411 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
414 * Returns 0 on success. If the result is negative, it's a Linux error code;
415 * if the result is positive, it's an NVM Express status code
417 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
418 union nvme_result *result, void *buffer, unsigned bufflen,
419 unsigned timeout, int qid, int at_head, int flags)
424 req = nvme_alloc_request(q, cmd, flags, qid);
428 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
430 if (buffer && bufflen) {
431 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
436 blk_execute_rq(req->q, NULL, req, at_head);
438 *result = nvme_req(req)->result;
439 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
442 ret = nvme_req(req)->status;
444 blk_mq_free_request(req);
447 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
449 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
450 void *buffer, unsigned bufflen)
452 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
455 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
457 int __nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
458 void __user *ubuffer, unsigned bufflen,
459 void __user *meta_buffer, unsigned meta_len, u32 meta_seed,
460 u32 *result, unsigned timeout)
462 bool write = nvme_is_write(cmd);
463 struct nvme_ns *ns = q->queuedata;
464 struct gendisk *disk = ns ? ns->disk : NULL;
466 struct bio *bio = NULL;
470 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
474 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
476 if (ubuffer && bufflen) {
477 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
485 bio->bi_bdev = bdget_disk(disk, 0);
491 if (meta_buffer && meta_len) {
492 struct bio_integrity_payload *bip;
494 meta = kmalloc(meta_len, GFP_KERNEL);
501 if (copy_from_user(meta, meta_buffer,
508 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
514 bip->bip_iter.bi_size = meta_len;
515 bip->bip_iter.bi_sector = meta_seed;
517 ret = bio_integrity_add_page(bio, virt_to_page(meta),
518 meta_len, offset_in_page(meta));
519 if (ret != meta_len) {
526 blk_execute_rq(req->q, disk, req, 0);
527 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
530 ret = nvme_req(req)->status;
532 *result = le32_to_cpu(nvme_req(req)->result.u32);
533 if (meta && !ret && !write) {
534 if (copy_to_user(meta_buffer, meta, meta_len))
541 if (disk && bio->bi_bdev)
543 blk_rq_unmap_user(bio);
546 blk_mq_free_request(req);
550 int nvme_submit_user_cmd(struct request_queue *q, struct nvme_command *cmd,
551 void __user *ubuffer, unsigned bufflen, u32 *result,
554 return __nvme_submit_user_cmd(q, cmd, ubuffer, bufflen, NULL, 0, 0,
558 static void nvme_keep_alive_end_io(struct request *rq, int error)
560 struct nvme_ctrl *ctrl = rq->end_io_data;
562 blk_mq_free_request(rq);
565 dev_err(ctrl->device,
566 "failed nvme_keep_alive_end_io error=%d\n", error);
570 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
573 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
575 struct nvme_command c;
578 memset(&c, 0, sizeof(c));
579 c.common.opcode = nvme_admin_keep_alive;
581 rq = nvme_alloc_request(ctrl->admin_q, &c, BLK_MQ_REQ_RESERVED,
586 rq->timeout = ctrl->kato * HZ;
587 rq->end_io_data = ctrl;
589 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
594 static void nvme_keep_alive_work(struct work_struct *work)
596 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
597 struct nvme_ctrl, ka_work);
599 if (nvme_keep_alive(ctrl)) {
600 /* allocation failure, reset the controller */
601 dev_err(ctrl->device, "keep-alive failed\n");
602 ctrl->ops->reset_ctrl(ctrl);
607 void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
609 if (unlikely(ctrl->kato == 0))
612 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
613 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
615 EXPORT_SYMBOL_GPL(nvme_start_keep_alive);
617 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
619 if (unlikely(ctrl->kato == 0))
622 cancel_delayed_work_sync(&ctrl->ka_work);
624 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
626 int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
628 struct nvme_command c = { };
631 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
632 c.identify.opcode = nvme_admin_identify;
633 c.identify.cns = NVME_ID_CNS_CTRL;
635 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
639 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
640 sizeof(struct nvme_id_ctrl));
646 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
648 struct nvme_command c = { };
650 c.identify.opcode = nvme_admin_identify;
651 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
652 c.identify.nsid = cpu_to_le32(nsid);
653 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list, 0x1000);
656 int nvme_identify_ns(struct nvme_ctrl *dev, unsigned nsid,
657 struct nvme_id_ns **id)
659 struct nvme_command c = { };
662 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
663 c.identify.opcode = nvme_admin_identify;
664 c.identify.nsid = cpu_to_le32(nsid);
665 c.identify.cns = NVME_ID_CNS_NS;
667 *id = kmalloc(sizeof(struct nvme_id_ns), GFP_KERNEL);
671 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
672 sizeof(struct nvme_id_ns));
678 int nvme_get_features(struct nvme_ctrl *dev, unsigned fid, unsigned nsid,
679 void *buffer, size_t buflen, u32 *result)
681 struct nvme_command c;
682 union nvme_result res;
685 memset(&c, 0, sizeof(c));
686 c.features.opcode = nvme_admin_get_features;
687 c.features.nsid = cpu_to_le32(nsid);
688 c.features.fid = cpu_to_le32(fid);
690 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res, buffer, buflen, 0,
692 if (ret >= 0 && result)
693 *result = le32_to_cpu(res.u32);
697 int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
698 void *buffer, size_t buflen, u32 *result)
700 struct nvme_command c;
701 union nvme_result res;
704 memset(&c, 0, sizeof(c));
705 c.features.opcode = nvme_admin_set_features;
706 c.features.fid = cpu_to_le32(fid);
707 c.features.dword11 = cpu_to_le32(dword11);
709 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
710 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
711 if (ret >= 0 && result)
712 *result = le32_to_cpu(res.u32);
716 int nvme_get_log_page(struct nvme_ctrl *dev, struct nvme_smart_log **log)
718 struct nvme_command c = { };
721 c.common.opcode = nvme_admin_get_log_page,
722 c.common.nsid = cpu_to_le32(0xFFFFFFFF),
723 c.common.cdw10[0] = cpu_to_le32(
724 (((sizeof(struct nvme_smart_log) / 4) - 1) << 16) |
727 *log = kmalloc(sizeof(struct nvme_smart_log), GFP_KERNEL);
731 error = nvme_submit_sync_cmd(dev->admin_q, &c, *log,
732 sizeof(struct nvme_smart_log));
738 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
740 u32 q_count = (*count - 1) | ((*count - 1) << 16);
742 int status, nr_io_queues;
744 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
750 * Degraded controllers might return an error when setting the queue
751 * count. We still want to be able to bring them online and offer
752 * access to the admin queue, as that might be only way to fix them up.
755 dev_err(ctrl->dev, "Could not set queue count (%d)\n", status);
758 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
759 *count = min(*count, nr_io_queues);
764 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
766 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
768 struct nvme_user_io io;
769 struct nvme_command c;
770 unsigned length, meta_len;
771 void __user *metadata;
773 if (copy_from_user(&io, uio, sizeof(io)))
781 case nvme_cmd_compare:
787 length = (io.nblocks + 1) << ns->lba_shift;
788 meta_len = (io.nblocks + 1) * ns->ms;
789 metadata = (void __user *)(uintptr_t)io.metadata;
794 } else if (meta_len) {
795 if ((io.metadata & 3) || !io.metadata)
799 memset(&c, 0, sizeof(c));
800 c.rw.opcode = io.opcode;
801 c.rw.flags = io.flags;
802 c.rw.nsid = cpu_to_le32(ns->ns_id);
803 c.rw.slba = cpu_to_le64(io.slba);
804 c.rw.length = cpu_to_le16(io.nblocks);
805 c.rw.control = cpu_to_le16(io.control);
806 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
807 c.rw.reftag = cpu_to_le32(io.reftag);
808 c.rw.apptag = cpu_to_le16(io.apptag);
809 c.rw.appmask = cpu_to_le16(io.appmask);
811 return __nvme_submit_user_cmd(ns->queue, &c,
812 (void __user *)(uintptr_t)io.addr, length,
813 metadata, meta_len, io.slba, NULL, 0);
816 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
817 struct nvme_passthru_cmd __user *ucmd)
819 struct nvme_passthru_cmd cmd;
820 struct nvme_command c;
821 unsigned timeout = 0;
824 if (!capable(CAP_SYS_ADMIN))
826 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
831 memset(&c, 0, sizeof(c));
832 c.common.opcode = cmd.opcode;
833 c.common.flags = cmd.flags;
834 c.common.nsid = cpu_to_le32(cmd.nsid);
835 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
836 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
837 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
838 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
839 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
840 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
841 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
842 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
845 timeout = msecs_to_jiffies(cmd.timeout_ms);
847 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
848 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
849 &cmd.result, timeout);
851 if (put_user(cmd.result, &ucmd->result))
858 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
859 unsigned int cmd, unsigned long arg)
861 struct nvme_ns *ns = bdev->bd_disk->private_data;
865 force_successful_syscall_return();
867 case NVME_IOCTL_ADMIN_CMD:
868 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
869 case NVME_IOCTL_IO_CMD:
870 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
871 case NVME_IOCTL_SUBMIT_IO:
872 return nvme_submit_io(ns, (void __user *)arg);
873 #ifdef CONFIG_BLK_DEV_NVME_SCSI
874 case SG_GET_VERSION_NUM:
875 return nvme_sg_get_version_num((void __user *)arg);
877 return nvme_sg_io(ns, (void __user *)arg);
882 return nvme_nvm_ioctl(ns, cmd, arg);
884 if (is_sed_ioctl(cmd))
885 return sed_ioctl(ns->ctrl->opal_dev, cmd,
886 (void __user *) arg);
892 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
893 unsigned int cmd, unsigned long arg)
899 return nvme_ioctl(bdev, mode, cmd, arg);
902 #define nvme_compat_ioctl NULL
905 static int nvme_open(struct block_device *bdev, fmode_t mode)
907 return nvme_get_ns_from_disk(bdev->bd_disk) ? 0 : -ENXIO;
910 static void nvme_release(struct gendisk *disk, fmode_t mode)
912 struct nvme_ns *ns = disk->private_data;
914 module_put(ns->ctrl->ops->module);
918 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
920 /* some standard values */
922 geo->sectors = 1 << 5;
923 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
927 #ifdef CONFIG_BLK_DEV_INTEGRITY
928 static void nvme_init_integrity(struct nvme_ns *ns)
930 struct blk_integrity integrity;
932 memset(&integrity, 0, sizeof(integrity));
933 switch (ns->pi_type) {
934 case NVME_NS_DPS_PI_TYPE3:
935 integrity.profile = &t10_pi_type3_crc;
936 integrity.tag_size = sizeof(u16) + sizeof(u32);
937 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
939 case NVME_NS_DPS_PI_TYPE1:
940 case NVME_NS_DPS_PI_TYPE2:
941 integrity.profile = &t10_pi_type1_crc;
942 integrity.tag_size = sizeof(u16);
943 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
946 integrity.profile = NULL;
949 integrity.tuple_size = ns->ms;
950 blk_integrity_register(ns->disk, &integrity);
951 blk_queue_max_integrity_segments(ns->queue, 1);
954 static void nvme_init_integrity(struct nvme_ns *ns)
957 #endif /* CONFIG_BLK_DEV_INTEGRITY */
959 static void nvme_config_discard(struct nvme_ns *ns)
961 struct nvme_ctrl *ctrl = ns->ctrl;
962 u32 logical_block_size = queue_logical_block_size(ns->queue);
964 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
965 NVME_DSM_MAX_RANGES);
967 ns->queue->limits.discard_alignment = logical_block_size;
968 ns->queue->limits.discard_granularity = logical_block_size;
969 blk_queue_max_discard_sectors(ns->queue, UINT_MAX);
970 blk_queue_max_discard_segments(ns->queue, NVME_DSM_MAX_RANGES);
971 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, ns->queue);
973 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
974 blk_queue_max_write_zeroes_sectors(ns->queue, UINT_MAX);
977 static int nvme_revalidate_ns(struct nvme_ns *ns, struct nvme_id_ns **id)
979 if (nvme_identify_ns(ns->ctrl, ns->ns_id, id)) {
980 dev_warn(ns->ctrl->dev, "%s: Identify failure\n", __func__);
984 if ((*id)->ncap == 0) {
989 if (ns->ctrl->vs >= NVME_VS(1, 1, 0))
990 memcpy(ns->eui, (*id)->eui64, sizeof(ns->eui));
991 if (ns->ctrl->vs >= NVME_VS(1, 2, 0))
992 memcpy(ns->uuid, (*id)->nguid, sizeof(ns->uuid));
997 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
999 struct nvme_ns *ns = disk->private_data;
1005 lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1006 ns->lba_shift = id->lbaf[lbaf].ds;
1007 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1008 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1011 * If identify namespace failed, use default 512 byte block size so
1012 * block layer can use before failing read/write for 0 capacity.
1014 if (ns->lba_shift == 0)
1016 bs = 1 << ns->lba_shift;
1017 /* XXX: PI implementation requires metadata equal t10 pi tuple size */
1018 pi_type = ns->ms == sizeof(struct t10_pi_tuple) ?
1019 id->dps & NVME_NS_DPS_PI_MASK : 0;
1021 blk_mq_freeze_queue(disk->queue);
1022 if (blk_get_integrity(disk) && (ns->pi_type != pi_type ||
1024 bs != queue_logical_block_size(disk->queue) ||
1025 (ns->ms && ns->ext)))
1026 blk_integrity_unregister(disk);
1028 ns->pi_type = pi_type;
1029 blk_queue_logical_block_size(ns->queue, bs);
1031 if (ns->ms && !blk_get_integrity(disk) && !ns->ext)
1032 nvme_init_integrity(ns);
1033 if (ns->ms && !(ns->ms == 8 && ns->pi_type) && !blk_get_integrity(disk))
1034 set_capacity(disk, 0);
1036 set_capacity(disk, le64_to_cpup(&id->nsze) << (ns->lba_shift - 9));
1038 if (ns->ctrl->oncs & NVME_CTRL_ONCS_DSM)
1039 nvme_config_discard(ns);
1040 blk_mq_unfreeze_queue(disk->queue);
1043 static int nvme_revalidate_disk(struct gendisk *disk)
1045 struct nvme_ns *ns = disk->private_data;
1046 struct nvme_id_ns *id = NULL;
1049 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1050 set_capacity(disk, 0);
1054 ret = nvme_revalidate_ns(ns, &id);
1058 __nvme_revalidate_disk(disk, id);
1064 static char nvme_pr_type(enum pr_type type)
1067 case PR_WRITE_EXCLUSIVE:
1069 case PR_EXCLUSIVE_ACCESS:
1071 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1073 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1075 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1077 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1084 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1085 u64 key, u64 sa_key, u8 op)
1087 struct nvme_ns *ns = bdev->bd_disk->private_data;
1088 struct nvme_command c;
1089 u8 data[16] = { 0, };
1091 put_unaligned_le64(key, &data[0]);
1092 put_unaligned_le64(sa_key, &data[8]);
1094 memset(&c, 0, sizeof(c));
1095 c.common.opcode = op;
1096 c.common.nsid = cpu_to_le32(ns->ns_id);
1097 c.common.cdw10[0] = cpu_to_le32(cdw10);
1099 return nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1102 static int nvme_pr_register(struct block_device *bdev, u64 old,
1103 u64 new, unsigned flags)
1107 if (flags & ~PR_FL_IGNORE_KEY)
1110 cdw10 = old ? 2 : 0;
1111 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1112 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1113 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1116 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1117 enum pr_type type, unsigned flags)
1121 if (flags & ~PR_FL_IGNORE_KEY)
1124 cdw10 = nvme_pr_type(type) << 8;
1125 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1126 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1129 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1130 enum pr_type type, bool abort)
1132 u32 cdw10 = nvme_pr_type(type) << 8 | abort ? 2 : 1;
1133 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1136 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1138 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1139 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1142 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1144 u32 cdw10 = nvme_pr_type(type) << 8 | key ? 1 << 3 : 0;
1145 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1148 static const struct pr_ops nvme_pr_ops = {
1149 .pr_register = nvme_pr_register,
1150 .pr_reserve = nvme_pr_reserve,
1151 .pr_release = nvme_pr_release,
1152 .pr_preempt = nvme_pr_preempt,
1153 .pr_clear = nvme_pr_clear,
1156 #ifdef CONFIG_BLK_SED_OPAL
1157 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1160 struct nvme_ctrl *ctrl = data;
1161 struct nvme_command cmd;
1163 memset(&cmd, 0, sizeof(cmd));
1165 cmd.common.opcode = nvme_admin_security_send;
1167 cmd.common.opcode = nvme_admin_security_recv;
1168 cmd.common.nsid = 0;
1169 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1170 cmd.common.cdw10[1] = cpu_to_le32(len);
1172 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1173 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1175 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1176 #endif /* CONFIG_BLK_SED_OPAL */
1178 static const struct block_device_operations nvme_fops = {
1179 .owner = THIS_MODULE,
1180 .ioctl = nvme_ioctl,
1181 .compat_ioctl = nvme_compat_ioctl,
1183 .release = nvme_release,
1184 .getgeo = nvme_getgeo,
1185 .revalidate_disk= nvme_revalidate_disk,
1186 .pr_ops = &nvme_pr_ops,
1189 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1191 unsigned long timeout =
1192 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1193 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1196 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1199 if ((csts & NVME_CSTS_RDY) == bit)
1203 if (fatal_signal_pending(current))
1205 if (time_after(jiffies, timeout)) {
1206 dev_err(ctrl->device,
1207 "Device not ready; aborting %s\n", enabled ?
1208 "initialisation" : "reset");
1217 * If the device has been passed off to us in an enabled state, just clear
1218 * the enabled bit. The spec says we should set the 'shutdown notification
1219 * bits', but doing so may cause the device to complete commands to the
1220 * admin queue ... and we don't know what memory that might be pointing at!
1222 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1226 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1227 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1229 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1233 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1234 msleep(NVME_QUIRK_DELAY_AMOUNT);
1236 return nvme_wait_ready(ctrl, cap, false);
1238 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1240 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1243 * Default to a 4K page size, with the intention to update this
1244 * path in the future to accomodate architectures with differing
1245 * kernel and IO page sizes.
1247 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1250 if (page_shift < dev_page_min) {
1251 dev_err(ctrl->device,
1252 "Minimum device page size %u too large for host (%u)\n",
1253 1 << dev_page_min, 1 << page_shift);
1257 ctrl->page_size = 1 << page_shift;
1259 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1260 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1261 ctrl->ctrl_config |= NVME_CC_ARB_RR | NVME_CC_SHN_NONE;
1262 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1263 ctrl->ctrl_config |= NVME_CC_ENABLE;
1265 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1268 return nvme_wait_ready(ctrl, cap, true);
1270 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1272 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1274 unsigned long timeout = SHUTDOWN_TIMEOUT + jiffies;
1278 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1279 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1281 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1285 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1286 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1290 if (fatal_signal_pending(current))
1292 if (time_after(jiffies, timeout)) {
1293 dev_err(ctrl->device,
1294 "Device shutdown incomplete; abort shutdown\n");
1301 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1303 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1304 struct request_queue *q)
1308 if (ctrl->max_hw_sectors) {
1310 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1312 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1313 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1315 if (ctrl->quirks & NVME_QUIRK_STRIPE_SIZE)
1316 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1317 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1318 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1320 blk_queue_write_cache(q, vwc, vwc);
1323 static void nvme_configure_apst(struct nvme_ctrl *ctrl)
1326 * APST (Autonomous Power State Transition) lets us program a
1327 * table of power state transitions that the controller will
1328 * perform automatically. We configure it with a simple
1329 * heuristic: we are willing to spend at most 2% of the time
1330 * transitioning between power states. Therefore, when running
1331 * in any given state, we will enter the next lower-power
1332 * non-operational state after waiting 50 * (enlat + exlat)
1333 * microseconds, as long as that state's total latency is under
1334 * the requested maximum latency.
1336 * We will not autonomously enter any non-operational state for
1337 * which the total latency exceeds ps_max_latency_us. Users
1338 * can set ps_max_latency_us to zero to turn off APST.
1342 struct nvme_feat_auto_pst *table;
1348 * If APST isn't supported or if we haven't been initialized yet,
1349 * then don't do anything.
1354 if (ctrl->npss > 31) {
1355 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1359 table = kzalloc(sizeof(*table), GFP_KERNEL);
1363 if (ctrl->ps_max_latency_us == 0) {
1364 /* Turn off APST. */
1366 dev_dbg(ctrl->device, "APST disabled\n");
1368 __le64 target = cpu_to_le64(0);
1372 * Walk through all states from lowest- to highest-power.
1373 * According to the spec, lower-numbered states use more
1374 * power. NPSS, despite the name, is the index of the
1375 * lowest-power state, not the number of states.
1377 for (state = (int)ctrl->npss; state >= 0; state--) {
1378 u64 total_latency_us, transition_ms;
1381 table->entries[state] = target;
1384 * Don't allow transitions to the deepest state
1385 * if it's quirked off.
1387 if (state == ctrl->npss &&
1388 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1392 * Is this state a useful non-operational state for
1393 * higher-power states to autonomously transition to?
1395 if (!(ctrl->psd[state].flags &
1396 NVME_PS_FLAGS_NON_OP_STATE))
1400 (u64)le32_to_cpu(ctrl->psd[state].entry_lat) +
1401 + le32_to_cpu(ctrl->psd[state].exit_lat);
1402 if (total_latency_us > ctrl->ps_max_latency_us)
1406 * This state is good. Use it as the APST idle
1407 * target for higher power states.
1409 transition_ms = total_latency_us + 19;
1410 do_div(transition_ms, 20);
1411 if (transition_ms > (1 << 24) - 1)
1412 transition_ms = (1 << 24) - 1;
1414 target = cpu_to_le64((state << 3) |
1415 (transition_ms << 8));
1420 if (total_latency_us > max_lat_us)
1421 max_lat_us = total_latency_us;
1427 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
1429 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
1430 max_ps, max_lat_us, (int)sizeof(*table), table);
1434 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
1435 table, sizeof(*table), NULL);
1437 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
1442 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
1444 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1448 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
1449 case PM_QOS_LATENCY_ANY:
1457 if (ctrl->ps_max_latency_us != latency) {
1458 ctrl->ps_max_latency_us = latency;
1459 nvme_configure_apst(ctrl);
1463 struct nvme_core_quirk_entry {
1465 * NVMe model and firmware strings are padded with spaces. For
1466 * simplicity, strings in the quirk table are padded with NULLs
1472 unsigned long quirks;
1475 static const struct nvme_core_quirk_entry core_quirks[] = {
1478 * This Toshiba device seems to die using any APST states. See:
1479 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
1482 .mn = "THNSF5256GPUK TOSHIBA",
1483 .quirks = NVME_QUIRK_NO_APST,
1487 /* match is null-terminated but idstr is space-padded. */
1488 static bool string_matches(const char *idstr, const char *match, size_t len)
1495 matchlen = strlen(match);
1496 WARN_ON_ONCE(matchlen > len);
1498 if (memcmp(idstr, match, matchlen))
1501 for (; matchlen < len; matchlen++)
1502 if (idstr[matchlen] != ' ')
1508 static bool quirk_matches(const struct nvme_id_ctrl *id,
1509 const struct nvme_core_quirk_entry *q)
1511 return q->vid == le16_to_cpu(id->vid) &&
1512 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
1513 string_matches(id->fr, q->fr, sizeof(id->fr));
1517 * Initialize the cached copies of the Identify data and various controller
1518 * register in our nvme_ctrl structure. This should be called as soon as
1519 * the admin queue is fully up and running.
1521 int nvme_init_identify(struct nvme_ctrl *ctrl)
1523 struct nvme_id_ctrl *id;
1525 int ret, page_shift;
1529 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
1531 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
1535 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
1537 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1540 page_shift = NVME_CAP_MPSMIN(cap) + 12;
1542 if (ctrl->vs >= NVME_VS(1, 1, 0))
1543 ctrl->subsystem = NVME_CAP_NSSRC(cap);
1545 ret = nvme_identify_ctrl(ctrl, &id);
1547 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
1551 if (!ctrl->identified) {
1553 * Check for quirks. Quirk can depend on firmware version,
1554 * so, in principle, the set of quirks present can change
1555 * across a reset. As a possible future enhancement, we
1556 * could re-scan for quirks every time we reinitialize
1557 * the device, but we'd have to make sure that the driver
1558 * behaves intelligently if the quirks change.
1563 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
1564 if (quirk_matches(id, &core_quirks[i]))
1565 ctrl->quirks |= core_quirks[i].quirks;
1569 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
1570 dev_warn(ctrl->dev, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
1571 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
1574 ctrl->oacs = le16_to_cpu(id->oacs);
1575 ctrl->vid = le16_to_cpu(id->vid);
1576 ctrl->oncs = le16_to_cpup(&id->oncs);
1577 atomic_set(&ctrl->abort_limit, id->acl + 1);
1578 ctrl->vwc = id->vwc;
1579 ctrl->cntlid = le16_to_cpup(&id->cntlid);
1580 memcpy(ctrl->serial, id->sn, sizeof(id->sn));
1581 memcpy(ctrl->model, id->mn, sizeof(id->mn));
1582 memcpy(ctrl->firmware_rev, id->fr, sizeof(id->fr));
1584 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
1586 max_hw_sectors = UINT_MAX;
1587 ctrl->max_hw_sectors =
1588 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
1590 nvme_set_queue_limits(ctrl, ctrl->admin_q);
1591 ctrl->sgls = le32_to_cpu(id->sgls);
1592 ctrl->kas = le16_to_cpu(id->kas);
1594 ctrl->npss = id->npss;
1595 prev_apsta = ctrl->apsta;
1596 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
1597 if (force_apst && id->apsta) {
1598 dev_warn(ctrl->dev, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
1604 ctrl->apsta = id->apsta;
1606 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
1608 if (ctrl->ops->flags & NVME_F_FABRICS) {
1609 ctrl->icdoff = le16_to_cpu(id->icdoff);
1610 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
1611 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
1612 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
1615 * In fabrics we need to verify the cntlid matches the
1618 if (ctrl->cntlid != le16_to_cpu(id->cntlid))
1621 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
1623 "keep-alive support is mandatory for fabrics\n");
1627 ctrl->cntlid = le16_to_cpu(id->cntlid);
1632 if (ctrl->apsta && !prev_apsta)
1633 dev_pm_qos_expose_latency_tolerance(ctrl->device);
1634 else if (!ctrl->apsta && prev_apsta)
1635 dev_pm_qos_hide_latency_tolerance(ctrl->device);
1637 nvme_configure_apst(ctrl);
1639 ctrl->identified = true;
1643 EXPORT_SYMBOL_GPL(nvme_init_identify);
1645 static int nvme_dev_open(struct inode *inode, struct file *file)
1647 struct nvme_ctrl *ctrl;
1648 int instance = iminor(inode);
1651 spin_lock(&dev_list_lock);
1652 list_for_each_entry(ctrl, &nvme_ctrl_list, node) {
1653 if (ctrl->instance != instance)
1656 if (!ctrl->admin_q) {
1660 if (!kref_get_unless_zero(&ctrl->kref))
1662 file->private_data = ctrl;
1666 spin_unlock(&dev_list_lock);
1671 static int nvme_dev_release(struct inode *inode, struct file *file)
1673 nvme_put_ctrl(file->private_data);
1677 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
1682 mutex_lock(&ctrl->namespaces_mutex);
1683 if (list_empty(&ctrl->namespaces)) {
1688 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
1689 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
1690 dev_warn(ctrl->device,
1691 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
1696 dev_warn(ctrl->device,
1697 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
1698 kref_get(&ns->kref);
1699 mutex_unlock(&ctrl->namespaces_mutex);
1701 ret = nvme_user_cmd(ctrl, ns, argp);
1706 mutex_unlock(&ctrl->namespaces_mutex);
1710 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
1713 struct nvme_ctrl *ctrl = file->private_data;
1714 void __user *argp = (void __user *)arg;
1717 case NVME_IOCTL_ADMIN_CMD:
1718 return nvme_user_cmd(ctrl, NULL, argp);
1719 case NVME_IOCTL_IO_CMD:
1720 return nvme_dev_user_cmd(ctrl, argp);
1721 case NVME_IOCTL_RESET:
1722 dev_warn(ctrl->device, "resetting controller\n");
1723 return ctrl->ops->reset_ctrl(ctrl);
1724 case NVME_IOCTL_SUBSYS_RESET:
1725 return nvme_reset_subsystem(ctrl);
1726 case NVME_IOCTL_RESCAN:
1727 nvme_queue_scan(ctrl);
1734 static const struct file_operations nvme_dev_fops = {
1735 .owner = THIS_MODULE,
1736 .open = nvme_dev_open,
1737 .release = nvme_dev_release,
1738 .unlocked_ioctl = nvme_dev_ioctl,
1739 .compat_ioctl = nvme_dev_ioctl,
1742 static ssize_t nvme_sysfs_reset(struct device *dev,
1743 struct device_attribute *attr, const char *buf,
1746 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1749 ret = ctrl->ops->reset_ctrl(ctrl);
1754 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
1756 static ssize_t nvme_sysfs_rescan(struct device *dev,
1757 struct device_attribute *attr, const char *buf,
1760 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1762 nvme_queue_scan(ctrl);
1765 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
1767 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
1770 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1771 struct nvme_ctrl *ctrl = ns->ctrl;
1772 int serial_len = sizeof(ctrl->serial);
1773 int model_len = sizeof(ctrl->model);
1775 if (memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1776 return sprintf(buf, "eui.%16phN\n", ns->uuid);
1778 if (memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1779 return sprintf(buf, "eui.%8phN\n", ns->eui);
1781 while (ctrl->serial[serial_len - 1] == ' ')
1783 while (ctrl->model[model_len - 1] == ' ')
1786 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", ctrl->vid,
1787 serial_len, ctrl->serial, model_len, ctrl->model, ns->ns_id);
1789 static DEVICE_ATTR(wwid, S_IRUGO, wwid_show, NULL);
1791 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
1794 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1795 return sprintf(buf, "%pU\n", ns->uuid);
1797 static DEVICE_ATTR(uuid, S_IRUGO, uuid_show, NULL);
1799 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
1802 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1803 return sprintf(buf, "%8phd\n", ns->eui);
1805 static DEVICE_ATTR(eui, S_IRUGO, eui_show, NULL);
1807 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
1810 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1811 return sprintf(buf, "%d\n", ns->ns_id);
1813 static DEVICE_ATTR(nsid, S_IRUGO, nsid_show, NULL);
1815 static struct attribute *nvme_ns_attrs[] = {
1816 &dev_attr_wwid.attr,
1817 &dev_attr_uuid.attr,
1819 &dev_attr_nsid.attr,
1823 static umode_t nvme_ns_attrs_are_visible(struct kobject *kobj,
1824 struct attribute *a, int n)
1826 struct device *dev = container_of(kobj, struct device, kobj);
1827 struct nvme_ns *ns = nvme_get_ns_from_dev(dev);
1829 if (a == &dev_attr_uuid.attr) {
1830 if (!memchr_inv(ns->uuid, 0, sizeof(ns->uuid)))
1833 if (a == &dev_attr_eui.attr) {
1834 if (!memchr_inv(ns->eui, 0, sizeof(ns->eui)))
1840 static const struct attribute_group nvme_ns_attr_group = {
1841 .attrs = nvme_ns_attrs,
1842 .is_visible = nvme_ns_attrs_are_visible,
1845 #define nvme_show_str_function(field) \
1846 static ssize_t field##_show(struct device *dev, \
1847 struct device_attribute *attr, char *buf) \
1849 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1850 return sprintf(buf, "%.*s\n", (int)sizeof(ctrl->field), ctrl->field); \
1852 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1854 #define nvme_show_int_function(field) \
1855 static ssize_t field##_show(struct device *dev, \
1856 struct device_attribute *attr, char *buf) \
1858 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
1859 return sprintf(buf, "%d\n", ctrl->field); \
1861 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
1863 nvme_show_str_function(model);
1864 nvme_show_str_function(serial);
1865 nvme_show_str_function(firmware_rev);
1866 nvme_show_int_function(cntlid);
1868 static ssize_t nvme_sysfs_delete(struct device *dev,
1869 struct device_attribute *attr, const char *buf,
1872 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1874 if (device_remove_file_self(dev, attr))
1875 ctrl->ops->delete_ctrl(ctrl);
1878 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
1880 static ssize_t nvme_sysfs_show_transport(struct device *dev,
1881 struct device_attribute *attr,
1884 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1886 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
1888 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
1890 static ssize_t nvme_sysfs_show_state(struct device *dev,
1891 struct device_attribute *attr,
1894 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1895 static const char *const state_name[] = {
1896 [NVME_CTRL_NEW] = "new",
1897 [NVME_CTRL_LIVE] = "live",
1898 [NVME_CTRL_RESETTING] = "resetting",
1899 [NVME_CTRL_RECONNECTING]= "reconnecting",
1900 [NVME_CTRL_DELETING] = "deleting",
1901 [NVME_CTRL_DEAD] = "dead",
1904 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
1905 state_name[ctrl->state])
1906 return sprintf(buf, "%s\n", state_name[ctrl->state]);
1908 return sprintf(buf, "unknown state\n");
1911 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
1913 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
1914 struct device_attribute *attr,
1917 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1919 return snprintf(buf, PAGE_SIZE, "%s\n",
1920 ctrl->ops->get_subsysnqn(ctrl));
1922 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
1924 static ssize_t nvme_sysfs_show_address(struct device *dev,
1925 struct device_attribute *attr,
1928 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1930 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
1932 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
1934 static struct attribute *nvme_dev_attrs[] = {
1935 &dev_attr_reset_controller.attr,
1936 &dev_attr_rescan_controller.attr,
1937 &dev_attr_model.attr,
1938 &dev_attr_serial.attr,
1939 &dev_attr_firmware_rev.attr,
1940 &dev_attr_cntlid.attr,
1941 &dev_attr_delete_controller.attr,
1942 &dev_attr_transport.attr,
1943 &dev_attr_subsysnqn.attr,
1944 &dev_attr_address.attr,
1945 &dev_attr_state.attr,
1949 #define CHECK_ATTR(ctrl, a, name) \
1950 if ((a) == &dev_attr_##name.attr && \
1951 !(ctrl)->ops->get_##name) \
1954 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
1955 struct attribute *a, int n)
1957 struct device *dev = container_of(kobj, struct device, kobj);
1958 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
1960 if (a == &dev_attr_delete_controller.attr) {
1961 if (!ctrl->ops->delete_ctrl)
1965 CHECK_ATTR(ctrl, a, subsysnqn);
1966 CHECK_ATTR(ctrl, a, address);
1971 static struct attribute_group nvme_dev_attrs_group = {
1972 .attrs = nvme_dev_attrs,
1973 .is_visible = nvme_dev_attrs_are_visible,
1976 static const struct attribute_group *nvme_dev_attr_groups[] = {
1977 &nvme_dev_attrs_group,
1981 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
1983 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
1984 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
1986 return nsa->ns_id - nsb->ns_id;
1989 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
1991 struct nvme_ns *ns, *ret = NULL;
1993 mutex_lock(&ctrl->namespaces_mutex);
1994 list_for_each_entry(ns, &ctrl->namespaces, list) {
1995 if (ns->ns_id == nsid) {
1996 kref_get(&ns->kref);
2000 if (ns->ns_id > nsid)
2003 mutex_unlock(&ctrl->namespaces_mutex);
2007 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2010 struct gendisk *disk;
2011 struct nvme_id_ns *id;
2012 char disk_name[DISK_NAME_LEN];
2013 int node = dev_to_node(ctrl->dev);
2015 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
2019 ns->instance = ida_simple_get(&ctrl->ns_ida, 1, 0, GFP_KERNEL);
2020 if (ns->instance < 0)
2023 ns->queue = blk_mq_init_queue(ctrl->tagset);
2024 if (IS_ERR(ns->queue))
2025 goto out_release_instance;
2026 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, ns->queue);
2027 ns->queue->queuedata = ns;
2030 kref_init(&ns->kref);
2032 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
2034 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
2035 nvme_set_queue_limits(ctrl, ns->queue);
2037 sprintf(disk_name, "nvme%dn%d", ctrl->instance, ns->instance);
2039 if (nvme_revalidate_ns(ns, &id))
2040 goto out_free_queue;
2042 if (nvme_nvm_ns_supported(ns, id) &&
2043 nvme_nvm_register(ns, disk_name, node)) {
2044 dev_warn(ctrl->dev, "%s: LightNVM init failure\n", __func__);
2048 disk = alloc_disk_node(0, node);
2052 disk->fops = &nvme_fops;
2053 disk->private_data = ns;
2054 disk->queue = ns->queue;
2055 disk->flags = GENHD_FL_EXT_DEVT;
2056 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
2059 __nvme_revalidate_disk(disk, id);
2061 mutex_lock(&ctrl->namespaces_mutex);
2062 list_add_tail(&ns->list, &ctrl->namespaces);
2063 mutex_unlock(&ctrl->namespaces_mutex);
2065 kref_get(&ctrl->kref);
2069 device_add_disk(ctrl->device, ns->disk);
2070 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
2071 &nvme_ns_attr_group))
2072 pr_warn("%s: failed to create sysfs group for identification\n",
2073 ns->disk->disk_name);
2074 if (ns->ndev && nvme_nvm_register_sysfs(ns))
2075 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
2076 ns->disk->disk_name);
2081 blk_cleanup_queue(ns->queue);
2082 out_release_instance:
2083 ida_simple_remove(&ctrl->ns_ida, ns->instance);
2088 static void nvme_ns_remove(struct nvme_ns *ns)
2090 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
2093 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
2094 if (blk_get_integrity(ns->disk))
2095 blk_integrity_unregister(ns->disk);
2096 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
2097 &nvme_ns_attr_group);
2099 nvme_nvm_unregister_sysfs(ns);
2100 del_gendisk(ns->disk);
2101 blk_cleanup_queue(ns->queue);
2104 mutex_lock(&ns->ctrl->namespaces_mutex);
2105 list_del_init(&ns->list);
2106 mutex_unlock(&ns->ctrl->namespaces_mutex);
2111 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
2115 ns = nvme_find_get_ns(ctrl, nsid);
2117 if (ns->disk && revalidate_disk(ns->disk))
2121 nvme_alloc_ns(ctrl, nsid);
2124 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
2127 struct nvme_ns *ns, *next;
2129 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
2130 if (ns->ns_id > nsid)
2135 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
2139 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
2142 ns_list = kzalloc(0x1000, GFP_KERNEL);
2146 for (i = 0; i < num_lists; i++) {
2147 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
2151 for (j = 0; j < min(nn, 1024U); j++) {
2152 nsid = le32_to_cpu(ns_list[j]);
2156 nvme_validate_ns(ctrl, nsid);
2158 while (++prev < nsid) {
2159 ns = nvme_find_get_ns(ctrl, prev);
2169 nvme_remove_invalid_namespaces(ctrl, prev);
2175 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
2179 for (i = 1; i <= nn; i++)
2180 nvme_validate_ns(ctrl, i);
2182 nvme_remove_invalid_namespaces(ctrl, nn);
2185 static void nvme_scan_work(struct work_struct *work)
2187 struct nvme_ctrl *ctrl =
2188 container_of(work, struct nvme_ctrl, scan_work);
2189 struct nvme_id_ctrl *id;
2192 if (ctrl->state != NVME_CTRL_LIVE)
2195 if (nvme_identify_ctrl(ctrl, &id))
2198 nn = le32_to_cpu(id->nn);
2199 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
2200 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
2201 if (!nvme_scan_ns_list(ctrl, nn))
2204 nvme_scan_ns_sequential(ctrl, nn);
2206 mutex_lock(&ctrl->namespaces_mutex);
2207 list_sort(NULL, &ctrl->namespaces, ns_cmp);
2208 mutex_unlock(&ctrl->namespaces_mutex);
2212 void nvme_queue_scan(struct nvme_ctrl *ctrl)
2215 * Do not queue new scan work when a controller is reset during
2218 if (ctrl->state == NVME_CTRL_LIVE)
2219 schedule_work(&ctrl->scan_work);
2221 EXPORT_SYMBOL_GPL(nvme_queue_scan);
2224 * This function iterates the namespace list unlocked to allow recovery from
2225 * controller failure. It is up to the caller to ensure the namespace list is
2226 * not modified by scan work while this function is executing.
2228 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
2230 struct nvme_ns *ns, *next;
2233 * The dead states indicates the controller was not gracefully
2234 * disconnected. In that case, we won't be able to flush any data while
2235 * removing the namespaces' disks; fail all the queues now to avoid
2236 * potentially having to clean up the failed sync later.
2238 if (ctrl->state == NVME_CTRL_DEAD)
2239 nvme_kill_queues(ctrl);
2241 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list)
2244 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
2246 static void nvme_async_event_work(struct work_struct *work)
2248 struct nvme_ctrl *ctrl =
2249 container_of(work, struct nvme_ctrl, async_event_work);
2251 spin_lock_irq(&ctrl->lock);
2252 while (ctrl->event_limit > 0) {
2253 int aer_idx = --ctrl->event_limit;
2255 spin_unlock_irq(&ctrl->lock);
2256 ctrl->ops->submit_async_event(ctrl, aer_idx);
2257 spin_lock_irq(&ctrl->lock);
2259 spin_unlock_irq(&ctrl->lock);
2262 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
2263 union nvme_result *res)
2265 u32 result = le32_to_cpu(res->u32);
2268 switch (le16_to_cpu(status) >> 1) {
2269 case NVME_SC_SUCCESS:
2272 case NVME_SC_ABORT_REQ:
2273 ++ctrl->event_limit;
2274 schedule_work(&ctrl->async_event_work);
2283 switch (result & 0xff07) {
2284 case NVME_AER_NOTICE_NS_CHANGED:
2285 dev_info(ctrl->device, "rescanning\n");
2286 nvme_queue_scan(ctrl);
2289 dev_warn(ctrl->device, "async event result %08x\n", result);
2292 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
2294 void nvme_queue_async_events(struct nvme_ctrl *ctrl)
2296 ctrl->event_limit = NVME_NR_AERS;
2297 schedule_work(&ctrl->async_event_work);
2299 EXPORT_SYMBOL_GPL(nvme_queue_async_events);
2301 static DEFINE_IDA(nvme_instance_ida);
2303 static int nvme_set_instance(struct nvme_ctrl *ctrl)
2305 int instance, error;
2308 if (!ida_pre_get(&nvme_instance_ida, GFP_KERNEL))
2311 spin_lock(&dev_list_lock);
2312 error = ida_get_new(&nvme_instance_ida, &instance);
2313 spin_unlock(&dev_list_lock);
2314 } while (error == -EAGAIN);
2319 ctrl->instance = instance;
2323 static void nvme_release_instance(struct nvme_ctrl *ctrl)
2325 spin_lock(&dev_list_lock);
2326 ida_remove(&nvme_instance_ida, ctrl->instance);
2327 spin_unlock(&dev_list_lock);
2330 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
2332 flush_work(&ctrl->async_event_work);
2333 flush_work(&ctrl->scan_work);
2334 nvme_remove_namespaces(ctrl);
2336 device_destroy(nvme_class, MKDEV(nvme_char_major, ctrl->instance));
2338 spin_lock(&dev_list_lock);
2339 list_del(&ctrl->node);
2340 spin_unlock(&dev_list_lock);
2342 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
2344 static void nvme_free_ctrl(struct kref *kref)
2346 struct nvme_ctrl *ctrl = container_of(kref, struct nvme_ctrl, kref);
2348 put_device(ctrl->device);
2349 nvme_release_instance(ctrl);
2350 ida_destroy(&ctrl->ns_ida);
2352 ctrl->ops->free_ctrl(ctrl);
2355 void nvme_put_ctrl(struct nvme_ctrl *ctrl)
2357 kref_put(&ctrl->kref, nvme_free_ctrl);
2359 EXPORT_SYMBOL_GPL(nvme_put_ctrl);
2362 * Initialize a NVMe controller structures. This needs to be called during
2363 * earliest initialization so that we have the initialized structured around
2366 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
2367 const struct nvme_ctrl_ops *ops, unsigned long quirks)
2371 ctrl->state = NVME_CTRL_NEW;
2372 spin_lock_init(&ctrl->lock);
2373 INIT_LIST_HEAD(&ctrl->namespaces);
2374 mutex_init(&ctrl->namespaces_mutex);
2375 kref_init(&ctrl->kref);
2378 ctrl->quirks = quirks;
2379 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
2380 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
2382 ret = nvme_set_instance(ctrl);
2386 ctrl->device = device_create_with_groups(nvme_class, ctrl->dev,
2387 MKDEV(nvme_char_major, ctrl->instance),
2388 ctrl, nvme_dev_attr_groups,
2389 "nvme%d", ctrl->instance);
2390 if (IS_ERR(ctrl->device)) {
2391 ret = PTR_ERR(ctrl->device);
2392 goto out_release_instance;
2394 get_device(ctrl->device);
2395 ida_init(&ctrl->ns_ida);
2397 spin_lock(&dev_list_lock);
2398 list_add_tail(&ctrl->node, &nvme_ctrl_list);
2399 spin_unlock(&dev_list_lock);
2402 * Initialize latency tolerance controls. The sysfs files won't
2403 * be visible to userspace unless the device actually supports APST.
2405 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
2406 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
2407 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
2410 out_release_instance:
2411 nvme_release_instance(ctrl);
2415 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
2418 * nvme_kill_queues(): Ends all namespace queues
2419 * @ctrl: the dead controller that needs to end
2421 * Call this function when the driver determines it is unable to get the
2422 * controller in a state capable of servicing IO.
2424 void nvme_kill_queues(struct nvme_ctrl *ctrl)
2428 mutex_lock(&ctrl->namespaces_mutex);
2429 list_for_each_entry(ns, &ctrl->namespaces, list) {
2431 * Revalidating a dead namespace sets capacity to 0. This will
2432 * end buffered writers dirtying pages that can't be synced.
2434 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
2436 revalidate_disk(ns->disk);
2437 blk_set_queue_dying(ns->queue);
2440 * Forcibly start all queues to avoid having stuck requests.
2441 * Note that we must ensure the queues are not stopped
2442 * when the final removal happens.
2444 blk_mq_start_hw_queues(ns->queue);
2446 /* draining requests in requeue list */
2447 blk_mq_kick_requeue_list(ns->queue);
2449 mutex_unlock(&ctrl->namespaces_mutex);
2451 EXPORT_SYMBOL_GPL(nvme_kill_queues);
2453 void nvme_unfreeze(struct nvme_ctrl *ctrl)
2457 mutex_lock(&ctrl->namespaces_mutex);
2458 list_for_each_entry(ns, &ctrl->namespaces, list)
2459 blk_mq_unfreeze_queue(ns->queue);
2460 mutex_unlock(&ctrl->namespaces_mutex);
2462 EXPORT_SYMBOL_GPL(nvme_unfreeze);
2464 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
2468 mutex_lock(&ctrl->namespaces_mutex);
2469 list_for_each_entry(ns, &ctrl->namespaces, list) {
2470 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
2474 mutex_unlock(&ctrl->namespaces_mutex);
2476 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
2478 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
2482 mutex_lock(&ctrl->namespaces_mutex);
2483 list_for_each_entry(ns, &ctrl->namespaces, list)
2484 blk_mq_freeze_queue_wait(ns->queue);
2485 mutex_unlock(&ctrl->namespaces_mutex);
2487 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
2489 void nvme_start_freeze(struct nvme_ctrl *ctrl)
2493 mutex_lock(&ctrl->namespaces_mutex);
2494 list_for_each_entry(ns, &ctrl->namespaces, list)
2495 blk_freeze_queue_start(ns->queue);
2496 mutex_unlock(&ctrl->namespaces_mutex);
2498 EXPORT_SYMBOL_GPL(nvme_start_freeze);
2500 void nvme_stop_queues(struct nvme_ctrl *ctrl)
2504 mutex_lock(&ctrl->namespaces_mutex);
2505 list_for_each_entry(ns, &ctrl->namespaces, list)
2506 blk_mq_quiesce_queue(ns->queue);
2507 mutex_unlock(&ctrl->namespaces_mutex);
2509 EXPORT_SYMBOL_GPL(nvme_stop_queues);
2511 void nvme_start_queues(struct nvme_ctrl *ctrl)
2515 mutex_lock(&ctrl->namespaces_mutex);
2516 list_for_each_entry(ns, &ctrl->namespaces, list) {
2517 blk_mq_start_stopped_hw_queues(ns->queue, true);
2518 blk_mq_kick_requeue_list(ns->queue);
2520 mutex_unlock(&ctrl->namespaces_mutex);
2522 EXPORT_SYMBOL_GPL(nvme_start_queues);
2524 int __init nvme_core_init(void)
2528 result = __register_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme",
2532 else if (result > 0)
2533 nvme_char_major = result;
2535 nvme_class = class_create(THIS_MODULE, "nvme");
2536 if (IS_ERR(nvme_class)) {
2537 result = PTR_ERR(nvme_class);
2538 goto unregister_chrdev;
2544 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2548 void nvme_core_exit(void)
2550 class_destroy(nvme_class);
2551 __unregister_chrdev(nvme_char_major, 0, NVME_MINORS, "nvme");
2554 MODULE_LICENSE("GPL");
2555 MODULE_VERSION("1.0");
2556 module_init(nvme_core_init);
2557 module_exit(nvme_core_exit);