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>
30 #include <asm/unaligned.h>
32 #define CREATE_TRACE_POINTS
38 #define NVME_MINORS (1U << MINORBITS)
40 unsigned int admin_timeout = 60;
41 module_param(admin_timeout, uint, 0644);
42 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
43 EXPORT_SYMBOL_GPL(admin_timeout);
45 unsigned int nvme_io_timeout = 30;
46 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
47 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
48 EXPORT_SYMBOL_GPL(nvme_io_timeout);
50 static unsigned char shutdown_timeout = 5;
51 module_param(shutdown_timeout, byte, 0644);
52 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
54 static u8 nvme_max_retries = 5;
55 module_param_named(max_retries, nvme_max_retries, byte, 0644);
56 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
58 static unsigned long default_ps_max_latency_us = 100000;
59 module_param(default_ps_max_latency_us, ulong, 0644);
60 MODULE_PARM_DESC(default_ps_max_latency_us,
61 "max power saving latency for new devices; use PM QOS to change per device");
63 static bool force_apst;
64 module_param(force_apst, bool, 0644);
65 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
68 module_param(streams, bool, 0644);
69 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
72 * nvme_wq - hosts nvme related works that are not reset or delete
73 * nvme_reset_wq - hosts nvme reset works
74 * nvme_delete_wq - hosts nvme delete works
76 * nvme_wq will host works such are scan, aen handling, fw activation,
77 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
78 * runs reset works which also flush works hosted on nvme_wq for
79 * serialization purposes. nvme_delete_wq host controller deletion
80 * works which flush reset works for serialization.
82 struct workqueue_struct *nvme_wq;
83 EXPORT_SYMBOL_GPL(nvme_wq);
85 struct workqueue_struct *nvme_reset_wq;
86 EXPORT_SYMBOL_GPL(nvme_reset_wq);
88 struct workqueue_struct *nvme_delete_wq;
89 EXPORT_SYMBOL_GPL(nvme_delete_wq);
91 static DEFINE_IDA(nvme_subsystems_ida);
92 static LIST_HEAD(nvme_subsystems);
93 static DEFINE_MUTEX(nvme_subsystems_lock);
95 static DEFINE_IDA(nvme_instance_ida);
96 static dev_t nvme_chr_devt;
97 static struct class *nvme_class;
98 static struct class *nvme_subsys_class;
100 static void nvme_ns_remove(struct nvme_ns *ns);
101 static int nvme_revalidate_disk(struct gendisk *disk);
102 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
103 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
106 static void nvme_set_queue_dying(struct nvme_ns *ns)
109 * Revalidating a dead namespace sets capacity to 0. This will end
110 * buffered writers dirtying pages that can't be synced.
112 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
114 revalidate_disk(ns->disk);
115 blk_set_queue_dying(ns->queue);
116 /* Forcibly unquiesce queues to avoid blocking dispatch */
117 blk_mq_unquiesce_queue(ns->queue);
120 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
123 * Only new queue scan work when admin and IO queues are both alive
125 if (ctrl->state == NVME_CTRL_LIVE)
126 queue_work(nvme_wq, &ctrl->scan_work);
129 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
131 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
133 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
137 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
139 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
143 ret = nvme_reset_ctrl(ctrl);
145 flush_work(&ctrl->reset_work);
146 if (ctrl->state != NVME_CTRL_LIVE &&
147 ctrl->state != NVME_CTRL_ADMIN_ONLY)
153 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
155 static void nvme_delete_ctrl_work(struct work_struct *work)
157 struct nvme_ctrl *ctrl =
158 container_of(work, struct nvme_ctrl, delete_work);
160 dev_info(ctrl->device,
161 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
163 flush_work(&ctrl->reset_work);
164 nvme_stop_ctrl(ctrl);
165 nvme_remove_namespaces(ctrl);
166 ctrl->ops->delete_ctrl(ctrl);
167 nvme_uninit_ctrl(ctrl);
171 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
173 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
175 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
179 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
181 int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
186 * Keep a reference until the work is flushed since ->delete_ctrl
187 * can free the controller.
190 ret = nvme_delete_ctrl(ctrl);
192 flush_work(&ctrl->delete_work);
196 EXPORT_SYMBOL_GPL(nvme_delete_ctrl_sync);
198 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
200 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
203 static blk_status_t nvme_error_status(struct request *req)
205 switch (nvme_req(req)->status & 0x7ff) {
206 case NVME_SC_SUCCESS:
208 case NVME_SC_CAP_EXCEEDED:
209 return BLK_STS_NOSPC;
210 case NVME_SC_LBA_RANGE:
211 return BLK_STS_TARGET;
212 case NVME_SC_BAD_ATTRIBUTES:
213 case NVME_SC_ONCS_NOT_SUPPORTED:
214 case NVME_SC_INVALID_OPCODE:
215 case NVME_SC_INVALID_FIELD:
216 case NVME_SC_INVALID_NS:
217 return BLK_STS_NOTSUPP;
218 case NVME_SC_WRITE_FAULT:
219 case NVME_SC_READ_ERROR:
220 case NVME_SC_UNWRITTEN_BLOCK:
221 case NVME_SC_ACCESS_DENIED:
222 case NVME_SC_READ_ONLY:
223 case NVME_SC_COMPARE_FAILED:
224 return BLK_STS_MEDIUM;
225 case NVME_SC_GUARD_CHECK:
226 case NVME_SC_APPTAG_CHECK:
227 case NVME_SC_REFTAG_CHECK:
228 case NVME_SC_INVALID_PI:
229 return BLK_STS_PROTECTION;
230 case NVME_SC_RESERVATION_CONFLICT:
231 return BLK_STS_NEXUS;
233 return BLK_STS_IOERR;
237 static inline bool nvme_req_needs_retry(struct request *req)
239 if (blk_noretry_request(req))
241 if (nvme_req(req)->status & NVME_SC_DNR)
243 if (nvme_req(req)->retries >= nvme_max_retries)
248 void nvme_complete_rq(struct request *req)
250 blk_status_t status = nvme_error_status(req);
252 trace_nvme_complete_rq(req);
254 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
255 if ((req->cmd_flags & REQ_NVME_MPATH) &&
256 blk_path_error(status)) {
257 nvme_failover_req(req);
261 if (!blk_queue_dying(req->q)) {
262 nvme_req(req)->retries++;
263 blk_mq_requeue_request(req, true);
267 blk_mq_end_request(req, status);
269 EXPORT_SYMBOL_GPL(nvme_complete_rq);
271 void nvme_cancel_request(struct request *req, void *data, bool reserved)
273 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
274 "Cancelling I/O %d", req->tag);
276 nvme_req(req)->status = NVME_SC_ABORT_REQ;
277 blk_mq_complete_request(req);
280 EXPORT_SYMBOL_GPL(nvme_cancel_request);
282 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
283 enum nvme_ctrl_state new_state)
285 enum nvme_ctrl_state old_state;
287 bool changed = false;
289 spin_lock_irqsave(&ctrl->lock, flags);
291 old_state = ctrl->state;
293 case NVME_CTRL_ADMIN_ONLY:
295 case NVME_CTRL_CONNECTING:
305 case NVME_CTRL_RESETTING:
306 case NVME_CTRL_CONNECTING:
313 case NVME_CTRL_RESETTING:
317 case NVME_CTRL_ADMIN_ONLY:
324 case NVME_CTRL_CONNECTING:
327 case NVME_CTRL_RESETTING:
334 case NVME_CTRL_DELETING:
337 case NVME_CTRL_ADMIN_ONLY:
338 case NVME_CTRL_RESETTING:
339 case NVME_CTRL_CONNECTING:
348 case NVME_CTRL_DELETING:
360 ctrl->state = new_state;
362 spin_unlock_irqrestore(&ctrl->lock, flags);
363 if (changed && ctrl->state == NVME_CTRL_LIVE)
364 nvme_kick_requeue_lists(ctrl);
367 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
369 static void nvme_free_ns_head(struct kref *ref)
371 struct nvme_ns_head *head =
372 container_of(ref, struct nvme_ns_head, ref);
374 nvme_mpath_remove_disk(head);
375 ida_simple_remove(&head->subsys->ns_ida, head->instance);
376 list_del_init(&head->entry);
377 cleanup_srcu_struct_quiesced(&head->srcu);
378 nvme_put_subsystem(head->subsys);
382 static void nvme_put_ns_head(struct nvme_ns_head *head)
384 kref_put(&head->ref, nvme_free_ns_head);
387 static void nvme_free_ns(struct kref *kref)
389 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
392 nvme_nvm_unregister(ns);
395 nvme_put_ns_head(ns->head);
396 nvme_put_ctrl(ns->ctrl);
400 static void nvme_put_ns(struct nvme_ns *ns)
402 kref_put(&ns->kref, nvme_free_ns);
405 static inline void nvme_clear_nvme_request(struct request *req)
407 if (!(req->rq_flags & RQF_DONTPREP)) {
408 nvme_req(req)->retries = 0;
409 nvme_req(req)->flags = 0;
410 req->rq_flags |= RQF_DONTPREP;
414 struct request *nvme_alloc_request(struct request_queue *q,
415 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
417 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
420 if (qid == NVME_QID_ANY) {
421 req = blk_mq_alloc_request(q, op, flags);
423 req = blk_mq_alloc_request_hctx(q, op, flags,
429 req->cmd_flags |= REQ_FAILFAST_DRIVER;
430 nvme_clear_nvme_request(req);
431 nvme_req(req)->cmd = cmd;
435 EXPORT_SYMBOL_GPL(nvme_alloc_request);
437 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
439 struct nvme_command c;
441 memset(&c, 0, sizeof(c));
443 c.directive.opcode = nvme_admin_directive_send;
444 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
445 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
446 c.directive.dtype = NVME_DIR_IDENTIFY;
447 c.directive.tdtype = NVME_DIR_STREAMS;
448 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
450 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
453 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
455 return nvme_toggle_streams(ctrl, false);
458 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
460 return nvme_toggle_streams(ctrl, true);
463 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
464 struct streams_directive_params *s, u32 nsid)
466 struct nvme_command c;
468 memset(&c, 0, sizeof(c));
469 memset(s, 0, sizeof(*s));
471 c.directive.opcode = nvme_admin_directive_recv;
472 c.directive.nsid = cpu_to_le32(nsid);
473 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
474 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
475 c.directive.dtype = NVME_DIR_STREAMS;
477 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
480 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
482 struct streams_directive_params s;
485 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
490 ret = nvme_enable_streams(ctrl);
494 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
498 ctrl->nssa = le16_to_cpu(s.nssa);
499 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
500 dev_info(ctrl->device, "too few streams (%u) available\n",
502 nvme_disable_streams(ctrl);
506 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
507 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
512 * Check if 'req' has a write hint associated with it. If it does, assign
513 * a valid namespace stream to the write.
515 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
516 struct request *req, u16 *control,
519 enum rw_hint streamid = req->write_hint;
521 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
525 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
528 *control |= NVME_RW_DTYPE_STREAMS;
529 *dsmgmt |= streamid << 16;
532 if (streamid < ARRAY_SIZE(req->q->write_hints))
533 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
536 static inline void nvme_setup_flush(struct nvme_ns *ns,
537 struct nvme_command *cmnd)
539 memset(cmnd, 0, sizeof(*cmnd));
540 cmnd->common.opcode = nvme_cmd_flush;
541 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
544 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
545 struct nvme_command *cmnd)
547 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
548 struct nvme_dsm_range *range;
551 range = kmalloc_array(segments, sizeof(*range), GFP_ATOMIC);
553 return BLK_STS_RESOURCE;
555 __rq_for_each_bio(bio, req) {
556 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
557 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
560 range[n].cattr = cpu_to_le32(0);
561 range[n].nlb = cpu_to_le32(nlb);
562 range[n].slba = cpu_to_le64(slba);
567 if (WARN_ON_ONCE(n != segments)) {
569 return BLK_STS_IOERR;
572 memset(cmnd, 0, sizeof(*cmnd));
573 cmnd->dsm.opcode = nvme_cmd_dsm;
574 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
575 cmnd->dsm.nr = cpu_to_le32(segments - 1);
576 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
578 req->special_vec.bv_page = virt_to_page(range);
579 req->special_vec.bv_offset = offset_in_page(range);
580 req->special_vec.bv_len = sizeof(*range) * segments;
581 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
586 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
587 struct request *req, struct nvme_command *cmnd)
589 struct nvme_ctrl *ctrl = ns->ctrl;
593 if (req->cmd_flags & REQ_FUA)
594 control |= NVME_RW_FUA;
595 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
596 control |= NVME_RW_LR;
598 if (req->cmd_flags & REQ_RAHEAD)
599 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
601 memset(cmnd, 0, sizeof(*cmnd));
602 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
603 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
604 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
605 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
607 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
608 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
612 * If formated with metadata, the block layer always provides a
613 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
614 * we enable the PRACT bit for protection information or set the
615 * namespace capacity to zero to prevent any I/O.
617 if (!blk_integrity_rq(req)) {
618 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
619 return BLK_STS_NOTSUPP;
620 control |= NVME_RW_PRINFO_PRACT;
621 } else if (req_op(req) == REQ_OP_WRITE) {
622 t10_pi_prepare(req, ns->pi_type);
625 switch (ns->pi_type) {
626 case NVME_NS_DPS_PI_TYPE3:
627 control |= NVME_RW_PRINFO_PRCHK_GUARD;
629 case NVME_NS_DPS_PI_TYPE1:
630 case NVME_NS_DPS_PI_TYPE2:
631 control |= NVME_RW_PRINFO_PRCHK_GUARD |
632 NVME_RW_PRINFO_PRCHK_REF;
633 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
638 cmnd->rw.control = cpu_to_le16(control);
639 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
643 void nvme_cleanup_cmd(struct request *req)
645 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
646 nvme_req(req)->status == 0) {
647 struct nvme_ns *ns = req->rq_disk->private_data;
649 t10_pi_complete(req, ns->pi_type,
650 blk_rq_bytes(req) >> ns->lba_shift);
652 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
653 kfree(page_address(req->special_vec.bv_page) +
654 req->special_vec.bv_offset);
657 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
659 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
660 struct nvme_command *cmd)
662 blk_status_t ret = BLK_STS_OK;
664 nvme_clear_nvme_request(req);
666 switch (req_op(req)) {
669 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
672 nvme_setup_flush(ns, cmd);
674 case REQ_OP_WRITE_ZEROES:
675 /* currently only aliased to deallocate for a few ctrls: */
677 ret = nvme_setup_discard(ns, req, cmd);
681 ret = nvme_setup_rw(ns, req, cmd);
685 return BLK_STS_IOERR;
688 cmd->common.command_id = req->tag;
689 trace_nvme_setup_cmd(req, cmd);
692 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
695 * Returns 0 on success. If the result is negative, it's a Linux error code;
696 * if the result is positive, it's an NVM Express status code
698 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
699 union nvme_result *result, void *buffer, unsigned bufflen,
700 unsigned timeout, int qid, int at_head,
701 blk_mq_req_flags_t flags)
706 req = nvme_alloc_request(q, cmd, flags, qid);
710 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
712 if (buffer && bufflen) {
713 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
718 blk_execute_rq(req->q, NULL, req, at_head);
720 *result = nvme_req(req)->result;
721 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
724 ret = nvme_req(req)->status;
726 blk_mq_free_request(req);
729 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
731 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
732 void *buffer, unsigned bufflen)
734 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
737 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
739 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
740 unsigned len, u32 seed, bool write)
742 struct bio_integrity_payload *bip;
746 buf = kmalloc(len, GFP_KERNEL);
751 if (write && copy_from_user(buf, ubuf, len))
754 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
760 bip->bip_iter.bi_size = len;
761 bip->bip_iter.bi_sector = seed;
762 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
763 offset_in_page(buf));
773 static int nvme_submit_user_cmd(struct request_queue *q,
774 struct nvme_command *cmd, void __user *ubuffer,
775 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
776 u32 meta_seed, u32 *result, unsigned timeout)
778 bool write = nvme_is_write(cmd);
779 struct nvme_ns *ns = q->queuedata;
780 struct gendisk *disk = ns ? ns->disk : NULL;
782 struct bio *bio = NULL;
786 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
790 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
791 nvme_req(req)->flags |= NVME_REQ_USERCMD;
793 if (ubuffer && bufflen) {
794 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
800 if (disk && meta_buffer && meta_len) {
801 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
807 req->cmd_flags |= REQ_INTEGRITY;
811 blk_execute_rq(req->q, disk, req, 0);
812 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
815 ret = nvme_req(req)->status;
817 *result = le32_to_cpu(nvme_req(req)->result.u32);
818 if (meta && !ret && !write) {
819 if (copy_to_user(meta_buffer, meta, meta_len))
825 blk_rq_unmap_user(bio);
827 blk_mq_free_request(req);
831 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
833 struct nvme_ctrl *ctrl = rq->end_io_data;
835 bool startka = false;
837 blk_mq_free_request(rq);
840 dev_err(ctrl->device,
841 "failed nvme_keep_alive_end_io error=%d\n",
846 spin_lock_irqsave(&ctrl->lock, flags);
847 if (ctrl->state == NVME_CTRL_LIVE ||
848 ctrl->state == NVME_CTRL_CONNECTING)
850 spin_unlock_irqrestore(&ctrl->lock, flags);
852 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
855 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
859 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
864 rq->timeout = ctrl->kato * HZ;
865 rq->end_io_data = ctrl;
867 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
872 static void nvme_keep_alive_work(struct work_struct *work)
874 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
875 struct nvme_ctrl, ka_work);
877 if (nvme_keep_alive(ctrl)) {
878 /* allocation failure, reset the controller */
879 dev_err(ctrl->device, "keep-alive failed\n");
880 nvme_reset_ctrl(ctrl);
885 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
887 if (unlikely(ctrl->kato == 0))
890 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
893 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
895 if (unlikely(ctrl->kato == 0))
898 cancel_delayed_work_sync(&ctrl->ka_work);
900 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
902 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
904 struct nvme_command c = { };
907 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
908 c.identify.opcode = nvme_admin_identify;
909 c.identify.cns = NVME_ID_CNS_CTRL;
911 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
915 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
916 sizeof(struct nvme_id_ctrl));
922 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
923 struct nvme_ns_ids *ids)
925 struct nvme_command c = { };
931 c.identify.opcode = nvme_admin_identify;
932 c.identify.nsid = cpu_to_le32(nsid);
933 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
935 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
939 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
940 NVME_IDENTIFY_DATA_SIZE);
944 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
945 struct nvme_ns_id_desc *cur = data + pos;
951 case NVME_NIDT_EUI64:
952 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
953 dev_warn(ctrl->device,
954 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
958 len = NVME_NIDT_EUI64_LEN;
959 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
961 case NVME_NIDT_NGUID:
962 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
963 dev_warn(ctrl->device,
964 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
968 len = NVME_NIDT_NGUID_LEN;
969 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
972 if (cur->nidl != NVME_NIDT_UUID_LEN) {
973 dev_warn(ctrl->device,
974 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
978 len = NVME_NIDT_UUID_LEN;
979 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
982 /* Skip unnkown types */
994 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
996 struct nvme_command c = { };
998 c.identify.opcode = nvme_admin_identify;
999 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1000 c.identify.nsid = cpu_to_le32(nsid);
1001 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1002 NVME_IDENTIFY_DATA_SIZE);
1005 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1008 struct nvme_id_ns *id;
1009 struct nvme_command c = { };
1012 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1013 c.identify.opcode = nvme_admin_identify;
1014 c.identify.nsid = cpu_to_le32(nsid);
1015 c.identify.cns = NVME_ID_CNS_NS;
1017 id = kmalloc(sizeof(*id), GFP_KERNEL);
1021 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1023 dev_warn(ctrl->device, "Identify namespace failed\n");
1031 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1032 void *buffer, size_t buflen, u32 *result)
1034 struct nvme_command c;
1035 union nvme_result res;
1038 memset(&c, 0, sizeof(c));
1039 c.features.opcode = nvme_admin_set_features;
1040 c.features.fid = cpu_to_le32(fid);
1041 c.features.dword11 = cpu_to_le32(dword11);
1043 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1044 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1045 if (ret >= 0 && result)
1046 *result = le32_to_cpu(res.u32);
1050 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1052 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1054 int status, nr_io_queues;
1056 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1062 * Degraded controllers might return an error when setting the queue
1063 * count. We still want to be able to bring them online and offer
1064 * access to the admin queue, as that might be only way to fix them up.
1067 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1070 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1071 *count = min(*count, nr_io_queues);
1076 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1078 #define NVME_AEN_SUPPORTED \
1079 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1081 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1083 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1086 if (!supported_aens)
1089 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1092 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1096 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1098 struct nvme_user_io io;
1099 struct nvme_command c;
1100 unsigned length, meta_len;
1101 void __user *metadata;
1103 if (copy_from_user(&io, uio, sizeof(io)))
1108 switch (io.opcode) {
1109 case nvme_cmd_write:
1111 case nvme_cmd_compare:
1117 length = (io.nblocks + 1) << ns->lba_shift;
1118 meta_len = (io.nblocks + 1) * ns->ms;
1119 metadata = (void __user *)(uintptr_t)io.metadata;
1124 } else if (meta_len) {
1125 if ((io.metadata & 3) || !io.metadata)
1129 memset(&c, 0, sizeof(c));
1130 c.rw.opcode = io.opcode;
1131 c.rw.flags = io.flags;
1132 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1133 c.rw.slba = cpu_to_le64(io.slba);
1134 c.rw.length = cpu_to_le16(io.nblocks);
1135 c.rw.control = cpu_to_le16(io.control);
1136 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1137 c.rw.reftag = cpu_to_le32(io.reftag);
1138 c.rw.apptag = cpu_to_le16(io.apptag);
1139 c.rw.appmask = cpu_to_le16(io.appmask);
1141 return nvme_submit_user_cmd(ns->queue, &c,
1142 (void __user *)(uintptr_t)io.addr, length,
1143 metadata, meta_len, io.slba, NULL, 0);
1146 static u32 nvme_known_admin_effects(u8 opcode)
1149 case nvme_admin_format_nvm:
1150 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1151 NVME_CMD_EFFECTS_CSE_MASK;
1152 case nvme_admin_sanitize_nvm:
1153 return NVME_CMD_EFFECTS_CSE_MASK;
1160 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1167 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1168 if (effects & ~NVME_CMD_EFFECTS_CSUPP)
1169 dev_warn(ctrl->device,
1170 "IO command:%02x has unhandled effects:%08x\n",
1176 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1178 effects = nvme_known_admin_effects(opcode);
1181 * For simplicity, IO to all namespaces is quiesced even if the command
1182 * effects say only one namespace is affected.
1184 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1185 mutex_lock(&ctrl->scan_lock);
1186 nvme_start_freeze(ctrl);
1187 nvme_wait_freeze(ctrl);
1192 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1196 down_read(&ctrl->namespaces_rwsem);
1197 list_for_each_entry(ns, &ctrl->namespaces, list)
1198 if (ns->disk && nvme_revalidate_disk(ns->disk))
1199 nvme_set_queue_dying(ns);
1200 up_read(&ctrl->namespaces_rwsem);
1202 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1205 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1208 * Revalidate LBA changes prior to unfreezing. This is necessary to
1209 * prevent memory corruption if a logical block size was changed by
1212 if (effects & NVME_CMD_EFFECTS_LBCC)
1213 nvme_update_formats(ctrl);
1214 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1215 nvme_unfreeze(ctrl);
1216 mutex_unlock(&ctrl->scan_lock);
1218 if (effects & NVME_CMD_EFFECTS_CCC)
1219 nvme_init_identify(ctrl);
1220 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1221 nvme_queue_scan(ctrl);
1224 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1225 struct nvme_passthru_cmd __user *ucmd)
1227 struct nvme_passthru_cmd cmd;
1228 struct nvme_command c;
1229 unsigned timeout = 0;
1233 if (!capable(CAP_SYS_ADMIN))
1235 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1240 memset(&c, 0, sizeof(c));
1241 c.common.opcode = cmd.opcode;
1242 c.common.flags = cmd.flags;
1243 c.common.nsid = cpu_to_le32(cmd.nsid);
1244 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1245 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1246 c.common.cdw10[0] = cpu_to_le32(cmd.cdw10);
1247 c.common.cdw10[1] = cpu_to_le32(cmd.cdw11);
1248 c.common.cdw10[2] = cpu_to_le32(cmd.cdw12);
1249 c.common.cdw10[3] = cpu_to_le32(cmd.cdw13);
1250 c.common.cdw10[4] = cpu_to_le32(cmd.cdw14);
1251 c.common.cdw10[5] = cpu_to_le32(cmd.cdw15);
1254 timeout = msecs_to_jiffies(cmd.timeout_ms);
1256 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1257 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1258 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1259 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1260 0, &cmd.result, timeout);
1261 nvme_passthru_end(ctrl, effects);
1264 if (put_user(cmd.result, &ucmd->result))
1272 * Issue ioctl requests on the first available path. Note that unlike normal
1273 * block layer requests we will not retry failed request on another controller.
1275 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1276 struct nvme_ns_head **head, int *srcu_idx)
1278 #ifdef CONFIG_NVME_MULTIPATH
1279 if (disk->fops == &nvme_ns_head_ops) {
1282 *head = disk->private_data;
1283 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1284 ns = nvme_find_path(*head);
1286 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1292 return disk->private_data;
1295 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1298 srcu_read_unlock(&head->srcu, idx);
1301 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1302 unsigned int cmd, unsigned long arg)
1304 struct nvme_ns_head *head = NULL;
1305 void __user *argp = (void __user *)arg;
1309 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1311 return -EWOULDBLOCK;
1314 * Handle ioctls that apply to the controller instead of the namespace
1315 * seperately and drop the ns SRCU reference early. This avoids a
1316 * deadlock when deleting namespaces using the passthrough interface.
1318 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1319 struct nvme_ctrl *ctrl = ns->ctrl;
1321 nvme_get_ctrl(ns->ctrl);
1322 nvme_put_ns_from_disk(head, srcu_idx);
1324 if (cmd == NVME_IOCTL_ADMIN_CMD)
1325 ret = nvme_user_cmd(ctrl, NULL, argp);
1327 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1329 nvme_put_ctrl(ctrl);
1335 force_successful_syscall_return();
1336 ret = ns->head->ns_id;
1338 case NVME_IOCTL_IO_CMD:
1339 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1341 case NVME_IOCTL_SUBMIT_IO:
1342 ret = nvme_submit_io(ns, argp);
1346 ret = nvme_nvm_ioctl(ns, cmd, arg);
1351 nvme_put_ns_from_disk(head, srcu_idx);
1355 static int nvme_open(struct block_device *bdev, fmode_t mode)
1357 struct nvme_ns *ns = bdev->bd_disk->private_data;
1359 #ifdef CONFIG_NVME_MULTIPATH
1360 /* should never be called due to GENHD_FL_HIDDEN */
1361 if (WARN_ON_ONCE(ns->head->disk))
1364 if (!kref_get_unless_zero(&ns->kref))
1366 if (!try_module_get(ns->ctrl->ops->module))
1377 static void nvme_release(struct gendisk *disk, fmode_t mode)
1379 struct nvme_ns *ns = disk->private_data;
1381 module_put(ns->ctrl->ops->module);
1385 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1387 /* some standard values */
1388 geo->heads = 1 << 6;
1389 geo->sectors = 1 << 5;
1390 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1394 #ifdef CONFIG_BLK_DEV_INTEGRITY
1395 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1397 struct blk_integrity integrity;
1399 memset(&integrity, 0, sizeof(integrity));
1401 case NVME_NS_DPS_PI_TYPE3:
1402 integrity.profile = &t10_pi_type3_crc;
1403 integrity.tag_size = sizeof(u16) + sizeof(u32);
1404 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1406 case NVME_NS_DPS_PI_TYPE1:
1407 case NVME_NS_DPS_PI_TYPE2:
1408 integrity.profile = &t10_pi_type1_crc;
1409 integrity.tag_size = sizeof(u16);
1410 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1413 integrity.profile = NULL;
1416 integrity.tuple_size = ms;
1417 blk_integrity_register(disk, &integrity);
1418 blk_queue_max_integrity_segments(disk->queue, 1);
1421 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1424 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1426 static void nvme_set_chunk_size(struct nvme_ns *ns)
1428 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1429 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1432 static void nvme_config_discard(struct nvme_ns *ns)
1434 struct nvme_ctrl *ctrl = ns->ctrl;
1435 struct request_queue *queue = ns->queue;
1436 u32 size = queue_logical_block_size(queue);
1438 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1439 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1443 if (ctrl->nr_streams && ns->sws && ns->sgs)
1444 size *= ns->sws * ns->sgs;
1446 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1447 NVME_DSM_MAX_RANGES);
1449 queue->limits.discard_alignment = 0;
1450 queue->limits.discard_granularity = size;
1452 /* If discard is already enabled, don't reset queue limits */
1453 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1456 blk_queue_max_discard_sectors(queue, UINT_MAX);
1457 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1459 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1460 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1463 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1464 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1466 memset(ids, 0, sizeof(*ids));
1468 if (ctrl->vs >= NVME_VS(1, 1, 0))
1469 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1470 if (ctrl->vs >= NVME_VS(1, 2, 0))
1471 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1472 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1473 /* Don't treat error as fatal we potentially
1474 * already have a NGUID or EUI-64
1476 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1477 dev_warn(ctrl->device,
1478 "%s: Identify Descriptors failed\n", __func__);
1482 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1484 return !uuid_is_null(&ids->uuid) ||
1485 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1486 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1489 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1491 return uuid_equal(&a->uuid, &b->uuid) &&
1492 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1493 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1496 static void nvme_update_disk_info(struct gendisk *disk,
1497 struct nvme_ns *ns, struct nvme_id_ns *id)
1499 sector_t capacity = le64_to_cpup(&id->nsze) << (ns->lba_shift - 9);
1500 unsigned short bs = 1 << ns->lba_shift;
1502 if (ns->lba_shift > PAGE_SHIFT) {
1503 /* unsupported block size, set capacity to 0 later */
1506 blk_mq_freeze_queue(disk->queue);
1507 blk_integrity_unregister(disk);
1509 blk_queue_logical_block_size(disk->queue, bs);
1510 blk_queue_physical_block_size(disk->queue, bs);
1511 blk_queue_io_min(disk->queue, bs);
1513 if (ns->ms && !ns->ext &&
1514 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1515 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1516 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1517 ns->lba_shift > PAGE_SHIFT)
1520 set_capacity(disk, capacity);
1521 nvme_config_discard(ns);
1523 if (id->nsattr & (1 << 0))
1524 set_disk_ro(disk, true);
1526 set_disk_ro(disk, false);
1528 blk_mq_unfreeze_queue(disk->queue);
1531 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1533 struct nvme_ns *ns = disk->private_data;
1536 * If identify namespace failed, use default 512 byte block size so
1537 * block layer can use before failing read/write for 0 capacity.
1539 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1540 if (ns->lba_shift == 0)
1542 ns->noiob = le16_to_cpu(id->noiob);
1543 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1544 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1545 /* the PI implementation requires metadata equal t10 pi tuple size */
1546 if (ns->ms == sizeof(struct t10_pi_tuple))
1547 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1552 nvme_set_chunk_size(ns);
1553 nvme_update_disk_info(disk, ns, id);
1555 nvme_nvm_update_nvm_info(ns);
1556 #ifdef CONFIG_NVME_MULTIPATH
1557 if (ns->head->disk) {
1558 nvme_update_disk_info(ns->head->disk, ns, id);
1559 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1564 static int nvme_revalidate_disk(struct gendisk *disk)
1566 struct nvme_ns *ns = disk->private_data;
1567 struct nvme_ctrl *ctrl = ns->ctrl;
1568 struct nvme_id_ns *id;
1569 struct nvme_ns_ids ids;
1572 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1573 set_capacity(disk, 0);
1577 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1581 if (id->ncap == 0) {
1586 __nvme_revalidate_disk(disk, id);
1587 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1588 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1589 dev_err(ctrl->device,
1590 "identifiers changed for nsid %d\n", ns->head->ns_id);
1599 static char nvme_pr_type(enum pr_type type)
1602 case PR_WRITE_EXCLUSIVE:
1604 case PR_EXCLUSIVE_ACCESS:
1606 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1608 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1610 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1612 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1619 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1620 u64 key, u64 sa_key, u8 op)
1622 struct nvme_ns_head *head = NULL;
1624 struct nvme_command c;
1626 u8 data[16] = { 0, };
1628 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1630 return -EWOULDBLOCK;
1632 put_unaligned_le64(key, &data[0]);
1633 put_unaligned_le64(sa_key, &data[8]);
1635 memset(&c, 0, sizeof(c));
1636 c.common.opcode = op;
1637 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1638 c.common.cdw10[0] = cpu_to_le32(cdw10);
1640 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1641 nvme_put_ns_from_disk(head, srcu_idx);
1645 static int nvme_pr_register(struct block_device *bdev, u64 old,
1646 u64 new, unsigned flags)
1650 if (flags & ~PR_FL_IGNORE_KEY)
1653 cdw10 = old ? 2 : 0;
1654 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1655 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1656 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1659 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1660 enum pr_type type, unsigned flags)
1664 if (flags & ~PR_FL_IGNORE_KEY)
1667 cdw10 = nvme_pr_type(type) << 8;
1668 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1669 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1672 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1673 enum pr_type type, bool abort)
1675 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1676 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1679 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1681 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1682 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1685 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1687 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1688 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1691 static const struct pr_ops nvme_pr_ops = {
1692 .pr_register = nvme_pr_register,
1693 .pr_reserve = nvme_pr_reserve,
1694 .pr_release = nvme_pr_release,
1695 .pr_preempt = nvme_pr_preempt,
1696 .pr_clear = nvme_pr_clear,
1699 #ifdef CONFIG_BLK_SED_OPAL
1700 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1703 struct nvme_ctrl *ctrl = data;
1704 struct nvme_command cmd;
1706 memset(&cmd, 0, sizeof(cmd));
1708 cmd.common.opcode = nvme_admin_security_send;
1710 cmd.common.opcode = nvme_admin_security_recv;
1711 cmd.common.nsid = 0;
1712 cmd.common.cdw10[0] = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1713 cmd.common.cdw10[1] = cpu_to_le32(len);
1715 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1716 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0);
1718 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1719 #endif /* CONFIG_BLK_SED_OPAL */
1721 static const struct block_device_operations nvme_fops = {
1722 .owner = THIS_MODULE,
1723 .ioctl = nvme_ioctl,
1724 .compat_ioctl = nvme_ioctl,
1726 .release = nvme_release,
1727 .getgeo = nvme_getgeo,
1728 .revalidate_disk= nvme_revalidate_disk,
1729 .pr_ops = &nvme_pr_ops,
1732 #ifdef CONFIG_NVME_MULTIPATH
1733 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1735 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1737 if (!kref_get_unless_zero(&head->ref))
1742 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1744 nvme_put_ns_head(disk->private_data);
1747 const struct block_device_operations nvme_ns_head_ops = {
1748 .owner = THIS_MODULE,
1749 .open = nvme_ns_head_open,
1750 .release = nvme_ns_head_release,
1751 .ioctl = nvme_ioctl,
1752 .compat_ioctl = nvme_ioctl,
1753 .getgeo = nvme_getgeo,
1754 .pr_ops = &nvme_pr_ops,
1756 #endif /* CONFIG_NVME_MULTIPATH */
1758 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1760 unsigned long timeout =
1761 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1762 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1765 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1768 if ((csts & NVME_CSTS_RDY) == bit)
1772 if (fatal_signal_pending(current))
1774 if (time_after(jiffies, timeout)) {
1775 dev_err(ctrl->device,
1776 "Device not ready; aborting %s\n", enabled ?
1777 "initialisation" : "reset");
1786 * If the device has been passed off to us in an enabled state, just clear
1787 * the enabled bit. The spec says we should set the 'shutdown notification
1788 * bits', but doing so may cause the device to complete commands to the
1789 * admin queue ... and we don't know what memory that might be pointing at!
1791 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1795 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1796 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1798 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1802 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1803 msleep(NVME_QUIRK_DELAY_AMOUNT);
1805 return nvme_wait_ready(ctrl, cap, false);
1807 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1809 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1812 * Default to a 4K page size, with the intention to update this
1813 * path in the future to accomodate architectures with differing
1814 * kernel and IO page sizes.
1816 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1819 if (page_shift < dev_page_min) {
1820 dev_err(ctrl->device,
1821 "Minimum device page size %u too large for host (%u)\n",
1822 1 << dev_page_min, 1 << page_shift);
1826 ctrl->page_size = 1 << page_shift;
1828 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1829 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1830 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1831 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1832 ctrl->ctrl_config |= NVME_CC_ENABLE;
1834 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1837 return nvme_wait_ready(ctrl, cap, true);
1839 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1841 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1843 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1847 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1848 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1850 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1854 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1855 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1859 if (fatal_signal_pending(current))
1861 if (time_after(jiffies, timeout)) {
1862 dev_err(ctrl->device,
1863 "Device shutdown incomplete; abort shutdown\n");
1870 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1872 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1873 struct request_queue *q)
1877 if (ctrl->max_hw_sectors) {
1879 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1881 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1882 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1883 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1885 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1886 is_power_of_2(ctrl->max_hw_sectors))
1887 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1888 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1889 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1891 blk_queue_write_cache(q, vwc, vwc);
1894 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1899 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1902 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1903 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1906 dev_warn_once(ctrl->device,
1907 "could not set timestamp (%d)\n", ret);
1911 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
1914 * APST (Autonomous Power State Transition) lets us program a
1915 * table of power state transitions that the controller will
1916 * perform automatically. We configure it with a simple
1917 * heuristic: we are willing to spend at most 2% of the time
1918 * transitioning between power states. Therefore, when running
1919 * in any given state, we will enter the next lower-power
1920 * non-operational state after waiting 50 * (enlat + exlat)
1921 * microseconds, as long as that state's exit latency is under
1922 * the requested maximum latency.
1924 * We will not autonomously enter any non-operational state for
1925 * which the total latency exceeds ps_max_latency_us. Users
1926 * can set ps_max_latency_us to zero to turn off APST.
1930 struct nvme_feat_auto_pst *table;
1936 * If APST isn't supported or if we haven't been initialized yet,
1937 * then don't do anything.
1942 if (ctrl->npss > 31) {
1943 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
1947 table = kzalloc(sizeof(*table), GFP_KERNEL);
1951 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
1952 /* Turn off APST. */
1954 dev_dbg(ctrl->device, "APST disabled\n");
1956 __le64 target = cpu_to_le64(0);
1960 * Walk through all states from lowest- to highest-power.
1961 * According to the spec, lower-numbered states use more
1962 * power. NPSS, despite the name, is the index of the
1963 * lowest-power state, not the number of states.
1965 for (state = (int)ctrl->npss; state >= 0; state--) {
1966 u64 total_latency_us, exit_latency_us, transition_ms;
1969 table->entries[state] = target;
1972 * Don't allow transitions to the deepest state
1973 * if it's quirked off.
1975 if (state == ctrl->npss &&
1976 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
1980 * Is this state a useful non-operational state for
1981 * higher-power states to autonomously transition to?
1983 if (!(ctrl->psd[state].flags &
1984 NVME_PS_FLAGS_NON_OP_STATE))
1988 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
1989 if (exit_latency_us > ctrl->ps_max_latency_us)
1994 le32_to_cpu(ctrl->psd[state].entry_lat);
1997 * This state is good. Use it as the APST idle
1998 * target for higher power states.
2000 transition_ms = total_latency_us + 19;
2001 do_div(transition_ms, 20);
2002 if (transition_ms > (1 << 24) - 1)
2003 transition_ms = (1 << 24) - 1;
2005 target = cpu_to_le64((state << 3) |
2006 (transition_ms << 8));
2011 if (total_latency_us > max_lat_us)
2012 max_lat_us = total_latency_us;
2018 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2020 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2021 max_ps, max_lat_us, (int)sizeof(*table), table);
2025 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2026 table, sizeof(*table), NULL);
2028 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2034 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2036 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2040 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2041 case PM_QOS_LATENCY_ANY:
2049 if (ctrl->ps_max_latency_us != latency) {
2050 ctrl->ps_max_latency_us = latency;
2051 nvme_configure_apst(ctrl);
2055 struct nvme_core_quirk_entry {
2057 * NVMe model and firmware strings are padded with spaces. For
2058 * simplicity, strings in the quirk table are padded with NULLs
2064 unsigned long quirks;
2067 static const struct nvme_core_quirk_entry core_quirks[] = {
2070 * This Toshiba device seems to die using any APST states. See:
2071 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2074 .mn = "THNSF5256GPUK TOSHIBA",
2075 .quirks = NVME_QUIRK_NO_APST,
2079 /* match is null-terminated but idstr is space-padded. */
2080 static bool string_matches(const char *idstr, const char *match, size_t len)
2087 matchlen = strlen(match);
2088 WARN_ON_ONCE(matchlen > len);
2090 if (memcmp(idstr, match, matchlen))
2093 for (; matchlen < len; matchlen++)
2094 if (idstr[matchlen] != ' ')
2100 static bool quirk_matches(const struct nvme_id_ctrl *id,
2101 const struct nvme_core_quirk_entry *q)
2103 return q->vid == le16_to_cpu(id->vid) &&
2104 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2105 string_matches(id->fr, q->fr, sizeof(id->fr));
2108 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2109 struct nvme_id_ctrl *id)
2114 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2115 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2116 strncpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2120 if (ctrl->vs >= NVME_VS(1, 2, 1))
2121 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2123 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2124 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2125 "nqn.2014.08.org.nvmexpress:%04x%04x",
2126 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2127 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2128 off += sizeof(id->sn);
2129 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2130 off += sizeof(id->mn);
2131 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2134 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2136 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2140 static void nvme_release_subsystem(struct device *dev)
2142 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2145 static void nvme_destroy_subsystem(struct kref *ref)
2147 struct nvme_subsystem *subsys =
2148 container_of(ref, struct nvme_subsystem, ref);
2150 mutex_lock(&nvme_subsystems_lock);
2151 list_del(&subsys->entry);
2152 mutex_unlock(&nvme_subsystems_lock);
2154 ida_destroy(&subsys->ns_ida);
2155 device_del(&subsys->dev);
2156 put_device(&subsys->dev);
2159 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2161 kref_put(&subsys->ref, nvme_destroy_subsystem);
2164 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2166 struct nvme_subsystem *subsys;
2168 lockdep_assert_held(&nvme_subsystems_lock);
2170 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2171 if (strcmp(subsys->subnqn, subsysnqn))
2173 if (!kref_get_unless_zero(&subsys->ref))
2181 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2182 struct device_attribute subsys_attr_##_name = \
2183 __ATTR(_name, _mode, _show, NULL)
2185 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2186 struct device_attribute *attr,
2189 struct nvme_subsystem *subsys =
2190 container_of(dev, struct nvme_subsystem, dev);
2192 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2194 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2196 #define nvme_subsys_show_str_function(field) \
2197 static ssize_t subsys_##field##_show(struct device *dev, \
2198 struct device_attribute *attr, char *buf) \
2200 struct nvme_subsystem *subsys = \
2201 container_of(dev, struct nvme_subsystem, dev); \
2202 return sprintf(buf, "%.*s\n", \
2203 (int)sizeof(subsys->field), subsys->field); \
2205 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2207 nvme_subsys_show_str_function(model);
2208 nvme_subsys_show_str_function(serial);
2209 nvme_subsys_show_str_function(firmware_rev);
2211 static struct attribute *nvme_subsys_attrs[] = {
2212 &subsys_attr_model.attr,
2213 &subsys_attr_serial.attr,
2214 &subsys_attr_firmware_rev.attr,
2215 &subsys_attr_subsysnqn.attr,
2219 static struct attribute_group nvme_subsys_attrs_group = {
2220 .attrs = nvme_subsys_attrs,
2223 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2224 &nvme_subsys_attrs_group,
2228 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2231 struct nvme_ctrl *ctrl;
2233 mutex_lock(&subsys->lock);
2234 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2235 if (ctrl->state != NVME_CTRL_DELETING &&
2236 ctrl->state != NVME_CTRL_DEAD)
2239 mutex_unlock(&subsys->lock);
2244 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2246 struct nvme_subsystem *subsys, *found;
2249 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2252 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2257 subsys->instance = ret;
2258 mutex_init(&subsys->lock);
2259 kref_init(&subsys->ref);
2260 INIT_LIST_HEAD(&subsys->ctrls);
2261 INIT_LIST_HEAD(&subsys->nsheads);
2262 nvme_init_subnqn(subsys, ctrl, id);
2263 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2264 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2265 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2266 subsys->vendor_id = le16_to_cpu(id->vid);
2267 subsys->cmic = id->cmic;
2269 subsys->dev.class = nvme_subsys_class;
2270 subsys->dev.release = nvme_release_subsystem;
2271 subsys->dev.groups = nvme_subsys_attrs_groups;
2272 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2273 device_initialize(&subsys->dev);
2275 mutex_lock(&nvme_subsystems_lock);
2276 found = __nvme_find_get_subsystem(subsys->subnqn);
2279 * Verify that the subsystem actually supports multiple
2280 * controllers, else bail out.
2282 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2283 nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2284 dev_err(ctrl->device,
2285 "ignoring ctrl due to duplicate subnqn (%s).\n",
2287 nvme_put_subsystem(found);
2292 __nvme_release_subsystem(subsys);
2295 ret = device_add(&subsys->dev);
2297 dev_err(ctrl->device,
2298 "failed to register subsystem device.\n");
2301 ida_init(&subsys->ns_ida);
2302 list_add_tail(&subsys->entry, &nvme_subsystems);
2305 ctrl->subsys = subsys;
2306 mutex_unlock(&nvme_subsystems_lock);
2308 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2309 dev_name(ctrl->device))) {
2310 dev_err(ctrl->device,
2311 "failed to create sysfs link from subsystem.\n");
2312 /* the transport driver will eventually put the subsystem */
2316 mutex_lock(&subsys->lock);
2317 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2318 mutex_unlock(&subsys->lock);
2323 mutex_unlock(&nvme_subsystems_lock);
2324 put_device(&subsys->dev);
2328 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2329 void *log, size_t size, u64 offset)
2331 struct nvme_command c = { };
2332 unsigned long dwlen = size / 4 - 1;
2334 c.get_log_page.opcode = nvme_admin_get_log_page;
2335 c.get_log_page.nsid = cpu_to_le32(nsid);
2336 c.get_log_page.lid = log_page;
2337 c.get_log_page.lsp = lsp;
2338 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2339 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2340 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2341 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2343 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2346 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2351 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2356 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2357 ctrl->effects, sizeof(*ctrl->effects), 0);
2359 kfree(ctrl->effects);
2360 ctrl->effects = NULL;
2366 * Initialize the cached copies of the Identify data and various controller
2367 * register in our nvme_ctrl structure. This should be called as soon as
2368 * the admin queue is fully up and running.
2370 int nvme_init_identify(struct nvme_ctrl *ctrl)
2372 struct nvme_id_ctrl *id;
2374 int ret, page_shift;
2376 bool prev_apst_enabled;
2378 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2380 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2384 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2386 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2389 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2391 if (ctrl->vs >= NVME_VS(1, 1, 0))
2392 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2394 ret = nvme_identify_ctrl(ctrl, &id);
2396 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2400 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2401 ret = nvme_get_effects_log(ctrl);
2406 if (!ctrl->identified) {
2409 ret = nvme_init_subsystem(ctrl, id);
2414 * Check for quirks. Quirk can depend on firmware version,
2415 * so, in principle, the set of quirks present can change
2416 * across a reset. As a possible future enhancement, we
2417 * could re-scan for quirks every time we reinitialize
2418 * the device, but we'd have to make sure that the driver
2419 * behaves intelligently if the quirks change.
2421 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2422 if (quirk_matches(id, &core_quirks[i]))
2423 ctrl->quirks |= core_quirks[i].quirks;
2427 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2428 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2429 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2432 ctrl->oacs = le16_to_cpu(id->oacs);
2433 ctrl->oncs = le16_to_cpup(&id->oncs);
2434 ctrl->oaes = le32_to_cpu(id->oaes);
2435 atomic_set(&ctrl->abort_limit, id->acl + 1);
2436 ctrl->vwc = id->vwc;
2437 ctrl->cntlid = le16_to_cpup(&id->cntlid);
2439 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2441 max_hw_sectors = UINT_MAX;
2442 ctrl->max_hw_sectors =
2443 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2445 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2446 ctrl->sgls = le32_to_cpu(id->sgls);
2447 ctrl->kas = le16_to_cpu(id->kas);
2448 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2452 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2454 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2455 shutdown_timeout, 60);
2457 if (ctrl->shutdown_timeout != shutdown_timeout)
2458 dev_info(ctrl->device,
2459 "Shutdown timeout set to %u seconds\n",
2460 ctrl->shutdown_timeout);
2462 ctrl->shutdown_timeout = shutdown_timeout;
2464 ctrl->npss = id->npss;
2465 ctrl->apsta = id->apsta;
2466 prev_apst_enabled = ctrl->apst_enabled;
2467 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2468 if (force_apst && id->apsta) {
2469 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2470 ctrl->apst_enabled = true;
2472 ctrl->apst_enabled = false;
2475 ctrl->apst_enabled = id->apsta;
2477 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2479 if (ctrl->ops->flags & NVME_F_FABRICS) {
2480 ctrl->icdoff = le16_to_cpu(id->icdoff);
2481 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2482 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2483 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2486 * In fabrics we need to verify the cntlid matches the
2489 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2494 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2495 dev_err(ctrl->device,
2496 "keep-alive support is mandatory for fabrics\n");
2501 ctrl->cntlid = le16_to_cpu(id->cntlid);
2502 ctrl->hmpre = le32_to_cpu(id->hmpre);
2503 ctrl->hmmin = le32_to_cpu(id->hmmin);
2504 ctrl->hmminds = le32_to_cpu(id->hmminds);
2505 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2508 ret = nvme_mpath_init(ctrl, id);
2514 if (ctrl->apst_enabled && !prev_apst_enabled)
2515 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2516 else if (!ctrl->apst_enabled && prev_apst_enabled)
2517 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2519 ret = nvme_configure_apst(ctrl);
2523 ret = nvme_configure_timestamp(ctrl);
2527 ret = nvme_configure_directives(ctrl);
2531 ctrl->identified = true;
2539 EXPORT_SYMBOL_GPL(nvme_init_identify);
2541 static int nvme_dev_open(struct inode *inode, struct file *file)
2543 struct nvme_ctrl *ctrl =
2544 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2546 switch (ctrl->state) {
2547 case NVME_CTRL_LIVE:
2548 case NVME_CTRL_ADMIN_ONLY:
2551 return -EWOULDBLOCK;
2554 file->private_data = ctrl;
2558 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2563 down_read(&ctrl->namespaces_rwsem);
2564 if (list_empty(&ctrl->namespaces)) {
2569 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2570 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2571 dev_warn(ctrl->device,
2572 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2577 dev_warn(ctrl->device,
2578 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2579 kref_get(&ns->kref);
2580 up_read(&ctrl->namespaces_rwsem);
2582 ret = nvme_user_cmd(ctrl, ns, argp);
2587 up_read(&ctrl->namespaces_rwsem);
2591 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2594 struct nvme_ctrl *ctrl = file->private_data;
2595 void __user *argp = (void __user *)arg;
2598 case NVME_IOCTL_ADMIN_CMD:
2599 return nvme_user_cmd(ctrl, NULL, argp);
2600 case NVME_IOCTL_IO_CMD:
2601 return nvme_dev_user_cmd(ctrl, argp);
2602 case NVME_IOCTL_RESET:
2603 dev_warn(ctrl->device, "resetting controller\n");
2604 return nvme_reset_ctrl_sync(ctrl);
2605 case NVME_IOCTL_SUBSYS_RESET:
2606 return nvme_reset_subsystem(ctrl);
2607 case NVME_IOCTL_RESCAN:
2608 nvme_queue_scan(ctrl);
2615 static const struct file_operations nvme_dev_fops = {
2616 .owner = THIS_MODULE,
2617 .open = nvme_dev_open,
2618 .unlocked_ioctl = nvme_dev_ioctl,
2619 .compat_ioctl = nvme_dev_ioctl,
2622 static ssize_t nvme_sysfs_reset(struct device *dev,
2623 struct device_attribute *attr, const char *buf,
2626 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2629 ret = nvme_reset_ctrl_sync(ctrl);
2634 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2636 static ssize_t nvme_sysfs_rescan(struct device *dev,
2637 struct device_attribute *attr, const char *buf,
2640 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2642 nvme_queue_scan(ctrl);
2645 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2647 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2649 struct gendisk *disk = dev_to_disk(dev);
2651 if (disk->fops == &nvme_fops)
2652 return nvme_get_ns_from_dev(dev)->head;
2654 return disk->private_data;
2657 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2660 struct nvme_ns_head *head = dev_to_ns_head(dev);
2661 struct nvme_ns_ids *ids = &head->ids;
2662 struct nvme_subsystem *subsys = head->subsys;
2663 int serial_len = sizeof(subsys->serial);
2664 int model_len = sizeof(subsys->model);
2666 if (!uuid_is_null(&ids->uuid))
2667 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2669 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2670 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2672 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2673 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2675 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2676 subsys->serial[serial_len - 1] == '\0'))
2678 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2679 subsys->model[model_len - 1] == '\0'))
2682 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2683 serial_len, subsys->serial, model_len, subsys->model,
2686 static DEVICE_ATTR_RO(wwid);
2688 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2691 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2693 static DEVICE_ATTR_RO(nguid);
2695 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2698 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2700 /* For backward compatibility expose the NGUID to userspace if
2701 * we have no UUID set
2703 if (uuid_is_null(&ids->uuid)) {
2704 printk_ratelimited(KERN_WARNING
2705 "No UUID available providing old NGUID\n");
2706 return sprintf(buf, "%pU\n", ids->nguid);
2708 return sprintf(buf, "%pU\n", &ids->uuid);
2710 static DEVICE_ATTR_RO(uuid);
2712 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2715 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2717 static DEVICE_ATTR_RO(eui);
2719 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2722 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2724 static DEVICE_ATTR_RO(nsid);
2726 static struct attribute *nvme_ns_id_attrs[] = {
2727 &dev_attr_wwid.attr,
2728 &dev_attr_uuid.attr,
2729 &dev_attr_nguid.attr,
2731 &dev_attr_nsid.attr,
2732 #ifdef CONFIG_NVME_MULTIPATH
2733 &dev_attr_ana_grpid.attr,
2734 &dev_attr_ana_state.attr,
2739 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2740 struct attribute *a, int n)
2742 struct device *dev = container_of(kobj, struct device, kobj);
2743 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2745 if (a == &dev_attr_uuid.attr) {
2746 if (uuid_is_null(&ids->uuid) &&
2747 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2750 if (a == &dev_attr_nguid.attr) {
2751 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2754 if (a == &dev_attr_eui.attr) {
2755 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2758 #ifdef CONFIG_NVME_MULTIPATH
2759 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2760 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2762 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2769 const struct attribute_group nvme_ns_id_attr_group = {
2770 .attrs = nvme_ns_id_attrs,
2771 .is_visible = nvme_ns_id_attrs_are_visible,
2774 #define nvme_show_str_function(field) \
2775 static ssize_t field##_show(struct device *dev, \
2776 struct device_attribute *attr, char *buf) \
2778 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2779 return sprintf(buf, "%.*s\n", \
2780 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2782 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2784 nvme_show_str_function(model);
2785 nvme_show_str_function(serial);
2786 nvme_show_str_function(firmware_rev);
2788 #define nvme_show_int_function(field) \
2789 static ssize_t field##_show(struct device *dev, \
2790 struct device_attribute *attr, char *buf) \
2792 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2793 return sprintf(buf, "%d\n", ctrl->field); \
2795 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2797 nvme_show_int_function(cntlid);
2799 static ssize_t nvme_sysfs_delete(struct device *dev,
2800 struct device_attribute *attr, const char *buf,
2803 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2805 if (device_remove_file_self(dev, attr))
2806 nvme_delete_ctrl_sync(ctrl);
2809 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2811 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2812 struct device_attribute *attr,
2815 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2817 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2819 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2821 static ssize_t nvme_sysfs_show_state(struct device *dev,
2822 struct device_attribute *attr,
2825 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2826 static const char *const state_name[] = {
2827 [NVME_CTRL_NEW] = "new",
2828 [NVME_CTRL_LIVE] = "live",
2829 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2830 [NVME_CTRL_RESETTING] = "resetting",
2831 [NVME_CTRL_CONNECTING] = "connecting",
2832 [NVME_CTRL_DELETING] = "deleting",
2833 [NVME_CTRL_DEAD] = "dead",
2836 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2837 state_name[ctrl->state])
2838 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2840 return sprintf(buf, "unknown state\n");
2843 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2845 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2846 struct device_attribute *attr,
2849 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2851 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2853 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2855 static ssize_t nvme_sysfs_show_address(struct device *dev,
2856 struct device_attribute *attr,
2859 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2861 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
2863 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
2865 static struct attribute *nvme_dev_attrs[] = {
2866 &dev_attr_reset_controller.attr,
2867 &dev_attr_rescan_controller.attr,
2868 &dev_attr_model.attr,
2869 &dev_attr_serial.attr,
2870 &dev_attr_firmware_rev.attr,
2871 &dev_attr_cntlid.attr,
2872 &dev_attr_delete_controller.attr,
2873 &dev_attr_transport.attr,
2874 &dev_attr_subsysnqn.attr,
2875 &dev_attr_address.attr,
2876 &dev_attr_state.attr,
2880 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
2881 struct attribute *a, int n)
2883 struct device *dev = container_of(kobj, struct device, kobj);
2884 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2886 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
2888 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
2894 static struct attribute_group nvme_dev_attrs_group = {
2895 .attrs = nvme_dev_attrs,
2896 .is_visible = nvme_dev_attrs_are_visible,
2899 static const struct attribute_group *nvme_dev_attr_groups[] = {
2900 &nvme_dev_attrs_group,
2904 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
2907 struct nvme_ns_head *h;
2909 lockdep_assert_held(&subsys->lock);
2911 list_for_each_entry(h, &subsys->nsheads, entry) {
2912 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
2919 static int __nvme_check_ids(struct nvme_subsystem *subsys,
2920 struct nvme_ns_head *new)
2922 struct nvme_ns_head *h;
2924 lockdep_assert_held(&subsys->lock);
2926 list_for_each_entry(h, &subsys->nsheads, entry) {
2927 if (nvme_ns_ids_valid(&new->ids) &&
2928 !list_empty(&h->list) &&
2929 nvme_ns_ids_equal(&new->ids, &h->ids))
2936 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
2937 unsigned nsid, struct nvme_id_ns *id)
2939 struct nvme_ns_head *head;
2942 head = kzalloc(sizeof(*head), GFP_KERNEL);
2945 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
2948 head->instance = ret;
2949 INIT_LIST_HEAD(&head->list);
2950 ret = init_srcu_struct(&head->srcu);
2952 goto out_ida_remove;
2953 head->subsys = ctrl->subsys;
2955 kref_init(&head->ref);
2957 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
2959 ret = __nvme_check_ids(ctrl->subsys, head);
2961 dev_err(ctrl->device,
2962 "duplicate IDs for nsid %d\n", nsid);
2963 goto out_cleanup_srcu;
2966 ret = nvme_mpath_alloc_disk(ctrl, head);
2968 goto out_cleanup_srcu;
2970 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
2972 kref_get(&ctrl->subsys->ref);
2976 cleanup_srcu_struct(&head->srcu);
2978 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
2982 return ERR_PTR(ret);
2985 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
2986 struct nvme_id_ns *id)
2988 struct nvme_ctrl *ctrl = ns->ctrl;
2989 bool is_shared = id->nmic & (1 << 0);
2990 struct nvme_ns_head *head = NULL;
2993 mutex_lock(&ctrl->subsys->lock);
2995 head = __nvme_find_ns_head(ctrl->subsys, nsid);
2997 head = nvme_alloc_ns_head(ctrl, nsid, id);
2999 ret = PTR_ERR(head);
3003 struct nvme_ns_ids ids;
3005 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3006 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3007 dev_err(ctrl->device,
3008 "IDs don't match for shared namespace %d\n",
3015 list_add_tail(&ns->siblings, &head->list);
3019 mutex_unlock(&ctrl->subsys->lock);
3023 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3025 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3026 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3028 return nsa->head->ns_id - nsb->head->ns_id;
3031 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3033 struct nvme_ns *ns, *ret = NULL;
3035 down_read(&ctrl->namespaces_rwsem);
3036 list_for_each_entry(ns, &ctrl->namespaces, list) {
3037 if (ns->head->ns_id == nsid) {
3038 if (!kref_get_unless_zero(&ns->kref))
3043 if (ns->head->ns_id > nsid)
3046 up_read(&ctrl->namespaces_rwsem);
3050 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3052 struct streams_directive_params s;
3055 if (!ctrl->nr_streams)
3058 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3062 ns->sws = le32_to_cpu(s.sws);
3063 ns->sgs = le16_to_cpu(s.sgs);
3066 unsigned int bs = 1 << ns->lba_shift;
3068 blk_queue_io_min(ns->queue, bs * ns->sws);
3070 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3076 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3079 struct gendisk *disk;
3080 struct nvme_id_ns *id;
3081 char disk_name[DISK_NAME_LEN];
3082 int node = dev_to_node(ctrl->dev), flags = GENHD_FL_EXT_DEVT;
3084 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3088 ns->queue = blk_mq_init_queue(ctrl->tagset);
3089 if (IS_ERR(ns->queue))
3091 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3092 ns->queue->queuedata = ns;
3095 kref_init(&ns->kref);
3096 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3098 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3099 nvme_set_queue_limits(ctrl, ns->queue);
3101 id = nvme_identify_ns(ctrl, nsid);
3103 goto out_free_queue;
3108 if (nvme_init_ns_head(ns, nsid, id))
3110 nvme_setup_streams_ns(ctrl, ns);
3111 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3113 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3114 if (nvme_nvm_register(ns, disk_name, node)) {
3115 dev_warn(ctrl->device, "LightNVM init failure\n");
3120 disk = alloc_disk_node(0, node);
3124 disk->fops = &nvme_fops;
3125 disk->private_data = ns;
3126 disk->queue = ns->queue;
3127 disk->flags = flags;
3128 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3131 __nvme_revalidate_disk(disk, id);
3133 down_write(&ctrl->namespaces_rwsem);
3134 list_add_tail(&ns->list, &ctrl->namespaces);
3135 up_write(&ctrl->namespaces_rwsem);
3137 nvme_get_ctrl(ctrl);
3139 device_add_disk(ctrl->device, ns->disk);
3140 if (sysfs_create_group(&disk_to_dev(ns->disk)->kobj,
3141 &nvme_ns_id_attr_group))
3142 pr_warn("%s: failed to create sysfs group for identification\n",
3143 ns->disk->disk_name);
3144 if (ns->ndev && nvme_nvm_register_sysfs(ns))
3145 pr_warn("%s: failed to register lightnvm sysfs group for identification\n",
3146 ns->disk->disk_name);
3148 nvme_mpath_add_disk(ns, id);
3149 nvme_fault_inject_init(ns);
3154 mutex_lock(&ctrl->subsys->lock);
3155 list_del_rcu(&ns->siblings);
3156 mutex_unlock(&ctrl->subsys->lock);
3160 blk_cleanup_queue(ns->queue);
3165 static void nvme_ns_remove(struct nvme_ns *ns)
3167 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3170 nvme_fault_inject_fini(ns);
3172 mutex_lock(&ns->ctrl->subsys->lock);
3173 list_del_rcu(&ns->siblings);
3174 mutex_unlock(&ns->ctrl->subsys->lock);
3175 synchronize_rcu(); /* guarantee not available in head->list */
3176 nvme_mpath_clear_current_path(ns);
3177 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3179 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3180 sysfs_remove_group(&disk_to_dev(ns->disk)->kobj,
3181 &nvme_ns_id_attr_group);
3183 nvme_nvm_unregister_sysfs(ns);
3184 del_gendisk(ns->disk);
3185 blk_cleanup_queue(ns->queue);
3186 if (blk_get_integrity(ns->disk))
3187 blk_integrity_unregister(ns->disk);
3190 down_write(&ns->ctrl->namespaces_rwsem);
3191 list_del_init(&ns->list);
3192 up_write(&ns->ctrl->namespaces_rwsem);
3194 nvme_mpath_check_last_path(ns);
3198 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3202 ns = nvme_find_get_ns(ctrl, nsid);
3204 if (ns->disk && revalidate_disk(ns->disk))
3208 nvme_alloc_ns(ctrl, nsid);
3211 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3214 struct nvme_ns *ns, *next;
3217 down_write(&ctrl->namespaces_rwsem);
3218 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3219 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3220 list_move_tail(&ns->list, &rm_list);
3222 up_write(&ctrl->namespaces_rwsem);
3224 list_for_each_entry_safe(ns, next, &rm_list, list)
3229 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3233 unsigned i, j, nsid, prev = 0;
3234 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3237 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3241 for (i = 0; i < num_lists; i++) {
3242 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3246 for (j = 0; j < min(nn, 1024U); j++) {
3247 nsid = le32_to_cpu(ns_list[j]);
3251 nvme_validate_ns(ctrl, nsid);
3253 while (++prev < nsid) {
3254 ns = nvme_find_get_ns(ctrl, prev);
3264 nvme_remove_invalid_namespaces(ctrl, prev);
3270 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3274 for (i = 1; i <= nn; i++)
3275 nvme_validate_ns(ctrl, i);
3277 nvme_remove_invalid_namespaces(ctrl, nn);
3280 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3282 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3286 log = kzalloc(log_size, GFP_KERNEL);
3291 * We need to read the log to clear the AEN, but we don't want to rely
3292 * on it for the changed namespace information as userspace could have
3293 * raced with us in reading the log page, which could cause us to miss
3296 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3299 dev_warn(ctrl->device,
3300 "reading changed ns log failed: %d\n", error);
3305 static void nvme_scan_work(struct work_struct *work)
3307 struct nvme_ctrl *ctrl =
3308 container_of(work, struct nvme_ctrl, scan_work);
3309 struct nvme_id_ctrl *id;
3312 if (ctrl->state != NVME_CTRL_LIVE)
3315 WARN_ON_ONCE(!ctrl->tagset);
3317 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3318 dev_info(ctrl->device, "rescanning namespaces.\n");
3319 nvme_clear_changed_ns_log(ctrl);
3322 if (nvme_identify_ctrl(ctrl, &id))
3325 mutex_lock(&ctrl->scan_lock);
3326 nn = le32_to_cpu(id->nn);
3327 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3328 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3329 if (!nvme_scan_ns_list(ctrl, nn))
3332 nvme_scan_ns_sequential(ctrl, nn);
3334 mutex_unlock(&ctrl->scan_lock);
3336 down_write(&ctrl->namespaces_rwsem);
3337 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3338 up_write(&ctrl->namespaces_rwsem);
3342 * This function iterates the namespace list unlocked to allow recovery from
3343 * controller failure. It is up to the caller to ensure the namespace list is
3344 * not modified by scan work while this function is executing.
3346 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3348 struct nvme_ns *ns, *next;
3351 /* prevent racing with ns scanning */
3352 flush_work(&ctrl->scan_work);
3355 * The dead states indicates the controller was not gracefully
3356 * disconnected. In that case, we won't be able to flush any data while
3357 * removing the namespaces' disks; fail all the queues now to avoid
3358 * potentially having to clean up the failed sync later.
3360 if (ctrl->state == NVME_CTRL_DEAD)
3361 nvme_kill_queues(ctrl);
3363 down_write(&ctrl->namespaces_rwsem);
3364 list_splice_init(&ctrl->namespaces, &ns_list);
3365 up_write(&ctrl->namespaces_rwsem);
3367 list_for_each_entry_safe(ns, next, &ns_list, list)
3370 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3372 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3374 char *envp[2] = { NULL, NULL };
3375 u32 aen_result = ctrl->aen_result;
3377 ctrl->aen_result = 0;
3381 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3384 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3388 static void nvme_async_event_work(struct work_struct *work)
3390 struct nvme_ctrl *ctrl =
3391 container_of(work, struct nvme_ctrl, async_event_work);
3393 nvme_aen_uevent(ctrl);
3394 ctrl->ops->submit_async_event(ctrl);
3397 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3402 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3408 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3411 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3413 struct nvme_fw_slot_info_log *log;
3415 log = kmalloc(sizeof(*log), GFP_KERNEL);
3419 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3421 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3425 static void nvme_fw_act_work(struct work_struct *work)
3427 struct nvme_ctrl *ctrl = container_of(work,
3428 struct nvme_ctrl, fw_act_work);
3429 unsigned long fw_act_timeout;
3432 fw_act_timeout = jiffies +
3433 msecs_to_jiffies(ctrl->mtfa * 100);
3435 fw_act_timeout = jiffies +
3436 msecs_to_jiffies(admin_timeout * 1000);
3438 nvme_stop_queues(ctrl);
3439 while (nvme_ctrl_pp_status(ctrl)) {
3440 if (time_after(jiffies, fw_act_timeout)) {
3441 dev_warn(ctrl->device,
3442 "Fw activation timeout, reset controller\n");
3443 nvme_reset_ctrl(ctrl);
3449 if (ctrl->state != NVME_CTRL_LIVE)
3452 nvme_start_queues(ctrl);
3453 /* read FW slot information to clear the AER */
3454 nvme_get_fw_slot_info(ctrl);
3457 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3459 switch ((result & 0xff00) >> 8) {
3460 case NVME_AER_NOTICE_NS_CHANGED:
3461 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3462 nvme_queue_scan(ctrl);
3464 case NVME_AER_NOTICE_FW_ACT_STARTING:
3465 queue_work(nvme_wq, &ctrl->fw_act_work);
3467 #ifdef CONFIG_NVME_MULTIPATH
3468 case NVME_AER_NOTICE_ANA:
3469 if (!ctrl->ana_log_buf)
3471 queue_work(nvme_wq, &ctrl->ana_work);
3475 dev_warn(ctrl->device, "async event result %08x\n", result);
3479 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3480 volatile union nvme_result *res)
3482 u32 result = le32_to_cpu(res->u32);
3484 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3487 switch (result & 0x7) {
3488 case NVME_AER_NOTICE:
3489 nvme_handle_aen_notice(ctrl, result);
3491 case NVME_AER_ERROR:
3492 case NVME_AER_SMART:
3495 ctrl->aen_result = result;
3500 queue_work(nvme_wq, &ctrl->async_event_work);
3502 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3504 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3506 nvme_mpath_stop(ctrl);
3507 nvme_stop_keep_alive(ctrl);
3508 flush_work(&ctrl->async_event_work);
3509 cancel_work_sync(&ctrl->fw_act_work);
3510 if (ctrl->ops->stop_ctrl)
3511 ctrl->ops->stop_ctrl(ctrl);
3513 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3515 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3518 nvme_start_keep_alive(ctrl);
3520 if (ctrl->queue_count > 1) {
3521 nvme_queue_scan(ctrl);
3522 nvme_enable_aen(ctrl);
3523 queue_work(nvme_wq, &ctrl->async_event_work);
3524 nvme_start_queues(ctrl);
3527 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3529 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3531 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3532 cdev_device_del(&ctrl->cdev, ctrl->device);
3534 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3536 static void nvme_free_ctrl(struct device *dev)
3538 struct nvme_ctrl *ctrl =
3539 container_of(dev, struct nvme_ctrl, ctrl_device);
3540 struct nvme_subsystem *subsys = ctrl->subsys;
3542 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3543 kfree(ctrl->effects);
3544 nvme_mpath_uninit(ctrl);
3547 mutex_lock(&subsys->lock);
3548 list_del(&ctrl->subsys_entry);
3549 mutex_unlock(&subsys->lock);
3550 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3553 ctrl->ops->free_ctrl(ctrl);
3556 nvme_put_subsystem(subsys);
3560 * Initialize a NVMe controller structures. This needs to be called during
3561 * earliest initialization so that we have the initialized structured around
3564 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3565 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3569 ctrl->state = NVME_CTRL_NEW;
3570 spin_lock_init(&ctrl->lock);
3571 mutex_init(&ctrl->scan_lock);
3572 INIT_LIST_HEAD(&ctrl->namespaces);
3573 init_rwsem(&ctrl->namespaces_rwsem);
3576 ctrl->quirks = quirks;
3577 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3578 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3579 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3580 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3582 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3583 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3584 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3586 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3589 ctrl->instance = ret;
3591 device_initialize(&ctrl->ctrl_device);
3592 ctrl->device = &ctrl->ctrl_device;
3593 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3594 ctrl->device->class = nvme_class;
3595 ctrl->device->parent = ctrl->dev;
3596 ctrl->device->groups = nvme_dev_attr_groups;
3597 ctrl->device->release = nvme_free_ctrl;
3598 dev_set_drvdata(ctrl->device, ctrl);
3599 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3601 goto out_release_instance;
3603 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3604 ctrl->cdev.owner = ops->module;
3605 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3610 * Initialize latency tolerance controls. The sysfs files won't
3611 * be visible to userspace unless the device actually supports APST.
3613 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3614 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3615 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3619 kfree_const(dev->kobj.name);
3620 out_release_instance:
3621 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3625 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3628 * nvme_kill_queues(): Ends all namespace queues
3629 * @ctrl: the dead controller that needs to end
3631 * Call this function when the driver determines it is unable to get the
3632 * controller in a state capable of servicing IO.
3634 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3638 down_read(&ctrl->namespaces_rwsem);
3640 /* Forcibly unquiesce queues to avoid blocking dispatch */
3642 blk_mq_unquiesce_queue(ctrl->admin_q);
3644 list_for_each_entry(ns, &ctrl->namespaces, list)
3645 nvme_set_queue_dying(ns);
3647 up_read(&ctrl->namespaces_rwsem);
3649 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3651 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3655 down_read(&ctrl->namespaces_rwsem);
3656 list_for_each_entry(ns, &ctrl->namespaces, list)
3657 blk_mq_unfreeze_queue(ns->queue);
3658 up_read(&ctrl->namespaces_rwsem);
3660 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3662 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3666 down_read(&ctrl->namespaces_rwsem);
3667 list_for_each_entry(ns, &ctrl->namespaces, list) {
3668 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3672 up_read(&ctrl->namespaces_rwsem);
3674 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3676 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3680 down_read(&ctrl->namespaces_rwsem);
3681 list_for_each_entry(ns, &ctrl->namespaces, list)
3682 blk_mq_freeze_queue_wait(ns->queue);
3683 up_read(&ctrl->namespaces_rwsem);
3685 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3687 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3691 down_read(&ctrl->namespaces_rwsem);
3692 list_for_each_entry(ns, &ctrl->namespaces, list)
3693 blk_freeze_queue_start(ns->queue);
3694 up_read(&ctrl->namespaces_rwsem);
3696 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3698 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3702 down_read(&ctrl->namespaces_rwsem);
3703 list_for_each_entry(ns, &ctrl->namespaces, list)
3704 blk_mq_quiesce_queue(ns->queue);
3705 up_read(&ctrl->namespaces_rwsem);
3707 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3709 void nvme_start_queues(struct nvme_ctrl *ctrl)
3713 down_read(&ctrl->namespaces_rwsem);
3714 list_for_each_entry(ns, &ctrl->namespaces, list)
3715 blk_mq_unquiesce_queue(ns->queue);
3716 up_read(&ctrl->namespaces_rwsem);
3718 EXPORT_SYMBOL_GPL(nvme_start_queues);
3720 int __init nvme_core_init(void)
3722 int result = -ENOMEM;
3724 nvme_wq = alloc_workqueue("nvme-wq",
3725 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3729 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3730 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3734 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3735 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3736 if (!nvme_delete_wq)
3737 goto destroy_reset_wq;
3739 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3741 goto destroy_delete_wq;
3743 nvme_class = class_create(THIS_MODULE, "nvme");
3744 if (IS_ERR(nvme_class)) {
3745 result = PTR_ERR(nvme_class);
3746 goto unregister_chrdev;
3749 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3750 if (IS_ERR(nvme_subsys_class)) {
3751 result = PTR_ERR(nvme_subsys_class);
3757 class_destroy(nvme_class);
3759 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3761 destroy_workqueue(nvme_delete_wq);
3763 destroy_workqueue(nvme_reset_wq);
3765 destroy_workqueue(nvme_wq);
3770 void nvme_core_exit(void)
3772 ida_destroy(&nvme_subsystems_ida);
3773 class_destroy(nvme_subsys_class);
3774 class_destroy(nvme_class);
3775 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3776 destroy_workqueue(nvme_delete_wq);
3777 destroy_workqueue(nvme_reset_wq);
3778 destroy_workqueue(nvme_wq);
3781 MODULE_LICENSE("GPL");
3782 MODULE_VERSION("1.0");
3783 module_init(nvme_core_init);
3784 module_exit(nvme_core_exit);