1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/list_sort.h>
15 #include <linux/slab.h>
16 #include <linux/types.h>
18 #include <linux/ptrace.h>
19 #include <linux/nvme_ioctl.h>
20 #include <linux/t10-pi.h>
21 #include <linux/pm_qos.h>
22 #include <asm/unaligned.h>
24 #define CREATE_TRACE_POINTS
30 #define NVME_MINORS (1U << MINORBITS)
32 unsigned int admin_timeout = 60;
33 module_param(admin_timeout, uint, 0644);
34 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
35 EXPORT_SYMBOL_GPL(admin_timeout);
37 unsigned int nvme_io_timeout = 30;
38 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
39 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
40 EXPORT_SYMBOL_GPL(nvme_io_timeout);
42 static unsigned char shutdown_timeout = 5;
43 module_param(shutdown_timeout, byte, 0644);
44 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
46 static u8 nvme_max_retries = 5;
47 module_param_named(max_retries, nvme_max_retries, byte, 0644);
48 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
50 static unsigned long default_ps_max_latency_us = 100000;
51 module_param(default_ps_max_latency_us, ulong, 0644);
52 MODULE_PARM_DESC(default_ps_max_latency_us,
53 "max power saving latency for new devices; use PM QOS to change per device");
55 static bool force_apst;
56 module_param(force_apst, bool, 0644);
57 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
60 module_param(streams, bool, 0644);
61 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
64 * nvme_wq - hosts nvme related works that are not reset or delete
65 * nvme_reset_wq - hosts nvme reset works
66 * nvme_delete_wq - hosts nvme delete works
68 * nvme_wq will host works such are scan, aen handling, fw activation,
69 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
70 * runs reset works which also flush works hosted on nvme_wq for
71 * serialization purposes. nvme_delete_wq host controller deletion
72 * works which flush reset works for serialization.
74 struct workqueue_struct *nvme_wq;
75 EXPORT_SYMBOL_GPL(nvme_wq);
77 struct workqueue_struct *nvme_reset_wq;
78 EXPORT_SYMBOL_GPL(nvme_reset_wq);
80 struct workqueue_struct *nvme_delete_wq;
81 EXPORT_SYMBOL_GPL(nvme_delete_wq);
83 static DEFINE_IDA(nvme_subsystems_ida);
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 revalidate_disk(ns->disk);
106 blk_set_queue_dying(ns->queue);
107 /* Forcibly unquiesce queues to avoid blocking dispatch */
108 blk_mq_unquiesce_queue(ns->queue);
111 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
114 * Only new queue scan work when admin and IO queues are both alive
116 if (ctrl->state == NVME_CTRL_LIVE)
117 queue_work(nvme_wq, &ctrl->scan_work);
120 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
122 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
124 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
128 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
130 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
134 ret = nvme_reset_ctrl(ctrl);
136 flush_work(&ctrl->reset_work);
137 if (ctrl->state != NVME_CTRL_LIVE &&
138 ctrl->state != NVME_CTRL_ADMIN_ONLY)
144 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
146 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
148 dev_info(ctrl->device,
149 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
151 flush_work(&ctrl->reset_work);
152 nvme_stop_ctrl(ctrl);
153 nvme_remove_namespaces(ctrl);
154 ctrl->ops->delete_ctrl(ctrl);
155 nvme_uninit_ctrl(ctrl);
159 static void nvme_delete_ctrl_work(struct work_struct *work)
161 struct nvme_ctrl *ctrl =
162 container_of(work, struct nvme_ctrl, delete_work);
164 nvme_do_delete_ctrl(ctrl);
167 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
169 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
171 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
175 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
177 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
182 * Keep a reference until nvme_do_delete_ctrl() complete,
183 * since ->delete_ctrl can free the controller.
186 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
189 nvme_do_delete_ctrl(ctrl);
194 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
196 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
199 static blk_status_t nvme_error_status(struct request *req)
201 switch (nvme_req(req)->status & 0x7ff) {
202 case NVME_SC_SUCCESS:
204 case NVME_SC_CAP_EXCEEDED:
205 return BLK_STS_NOSPC;
206 case NVME_SC_LBA_RANGE:
207 return BLK_STS_TARGET;
208 case NVME_SC_BAD_ATTRIBUTES:
209 case NVME_SC_ONCS_NOT_SUPPORTED:
210 case NVME_SC_INVALID_OPCODE:
211 case NVME_SC_INVALID_FIELD:
212 case NVME_SC_INVALID_NS:
213 return BLK_STS_NOTSUPP;
214 case NVME_SC_WRITE_FAULT:
215 case NVME_SC_READ_ERROR:
216 case NVME_SC_UNWRITTEN_BLOCK:
217 case NVME_SC_ACCESS_DENIED:
218 case NVME_SC_READ_ONLY:
219 case NVME_SC_COMPARE_FAILED:
220 return BLK_STS_MEDIUM;
221 case NVME_SC_GUARD_CHECK:
222 case NVME_SC_APPTAG_CHECK:
223 case NVME_SC_REFTAG_CHECK:
224 case NVME_SC_INVALID_PI:
225 return BLK_STS_PROTECTION;
226 case NVME_SC_RESERVATION_CONFLICT:
227 return BLK_STS_NEXUS;
229 return BLK_STS_IOERR;
233 static inline bool nvme_req_needs_retry(struct request *req)
235 if (blk_noretry_request(req))
237 if (nvme_req(req)->status & NVME_SC_DNR)
239 if (nvme_req(req)->retries >= nvme_max_retries)
244 static void nvme_retry_req(struct request *req)
246 struct nvme_ns *ns = req->q->queuedata;
247 unsigned long delay = 0;
250 /* The mask and shift result must be <= 3 */
251 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
253 delay = ns->ctrl->crdt[crd - 1] * 100;
255 nvme_req(req)->retries++;
256 blk_mq_requeue_request(req, false);
257 blk_mq_delay_kick_requeue_list(req->q, delay);
260 void nvme_complete_rq(struct request *req)
262 blk_status_t status = nvme_error_status(req);
264 trace_nvme_complete_rq(req);
266 if (nvme_req(req)->ctrl->kas)
267 nvme_req(req)->ctrl->comp_seen = true;
269 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
270 if ((req->cmd_flags & REQ_NVME_MPATH) &&
271 blk_path_error(status)) {
272 nvme_failover_req(req);
276 if (!blk_queue_dying(req->q)) {
281 blk_mq_end_request(req, status);
283 EXPORT_SYMBOL_GPL(nvme_complete_rq);
285 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
287 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
288 "Cancelling I/O %d", req->tag);
290 nvme_req(req)->status = NVME_SC_ABORT_REQ;
291 blk_mq_complete_request_sync(req);
294 EXPORT_SYMBOL_GPL(nvme_cancel_request);
296 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
297 enum nvme_ctrl_state new_state)
299 enum nvme_ctrl_state old_state;
301 bool changed = false;
303 spin_lock_irqsave(&ctrl->lock, flags);
305 old_state = ctrl->state;
307 case NVME_CTRL_ADMIN_ONLY:
309 case NVME_CTRL_CONNECTING:
319 case NVME_CTRL_RESETTING:
320 case NVME_CTRL_CONNECTING:
327 case NVME_CTRL_RESETTING:
331 case NVME_CTRL_ADMIN_ONLY:
338 case NVME_CTRL_CONNECTING:
341 case NVME_CTRL_RESETTING:
348 case NVME_CTRL_DELETING:
351 case NVME_CTRL_ADMIN_ONLY:
352 case NVME_CTRL_RESETTING:
353 case NVME_CTRL_CONNECTING:
362 case NVME_CTRL_DELETING:
374 ctrl->state = new_state;
376 spin_unlock_irqrestore(&ctrl->lock, flags);
377 if (changed && ctrl->state == NVME_CTRL_LIVE)
378 nvme_kick_requeue_lists(ctrl);
381 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
383 static void nvme_free_ns_head(struct kref *ref)
385 struct nvme_ns_head *head =
386 container_of(ref, struct nvme_ns_head, ref);
388 nvme_mpath_remove_disk(head);
389 ida_simple_remove(&head->subsys->ns_ida, head->instance);
390 list_del_init(&head->entry);
391 cleanup_srcu_struct(&head->srcu);
392 nvme_put_subsystem(head->subsys);
396 static void nvme_put_ns_head(struct nvme_ns_head *head)
398 kref_put(&head->ref, nvme_free_ns_head);
401 static void nvme_free_ns(struct kref *kref)
403 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
406 nvme_nvm_unregister(ns);
409 nvme_put_ns_head(ns->head);
410 nvme_put_ctrl(ns->ctrl);
414 static void nvme_put_ns(struct nvme_ns *ns)
416 kref_put(&ns->kref, nvme_free_ns);
419 static inline void nvme_clear_nvme_request(struct request *req)
421 if (!(req->rq_flags & RQF_DONTPREP)) {
422 nvme_req(req)->retries = 0;
423 nvme_req(req)->flags = 0;
424 req->rq_flags |= RQF_DONTPREP;
428 struct request *nvme_alloc_request(struct request_queue *q,
429 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
431 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
434 if (qid == NVME_QID_ANY) {
435 req = blk_mq_alloc_request(q, op, flags);
437 req = blk_mq_alloc_request_hctx(q, op, flags,
443 req->cmd_flags |= REQ_FAILFAST_DRIVER;
444 nvme_clear_nvme_request(req);
445 nvme_req(req)->cmd = cmd;
449 EXPORT_SYMBOL_GPL(nvme_alloc_request);
451 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
453 struct nvme_command c;
455 memset(&c, 0, sizeof(c));
457 c.directive.opcode = nvme_admin_directive_send;
458 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
459 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
460 c.directive.dtype = NVME_DIR_IDENTIFY;
461 c.directive.tdtype = NVME_DIR_STREAMS;
462 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
464 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
467 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
469 return nvme_toggle_streams(ctrl, false);
472 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
474 return nvme_toggle_streams(ctrl, true);
477 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
478 struct streams_directive_params *s, u32 nsid)
480 struct nvme_command c;
482 memset(&c, 0, sizeof(c));
483 memset(s, 0, sizeof(*s));
485 c.directive.opcode = nvme_admin_directive_recv;
486 c.directive.nsid = cpu_to_le32(nsid);
487 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
488 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
489 c.directive.dtype = NVME_DIR_STREAMS;
491 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
494 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
496 struct streams_directive_params s;
499 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
504 ret = nvme_enable_streams(ctrl);
508 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
512 ctrl->nssa = le16_to_cpu(s.nssa);
513 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
514 dev_info(ctrl->device, "too few streams (%u) available\n",
516 nvme_disable_streams(ctrl);
520 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
521 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
526 * Check if 'req' has a write hint associated with it. If it does, assign
527 * a valid namespace stream to the write.
529 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
530 struct request *req, u16 *control,
533 enum rw_hint streamid = req->write_hint;
535 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
539 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
542 *control |= NVME_RW_DTYPE_STREAMS;
543 *dsmgmt |= streamid << 16;
546 if (streamid < ARRAY_SIZE(req->q->write_hints))
547 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
550 static inline void nvme_setup_flush(struct nvme_ns *ns,
551 struct nvme_command *cmnd)
553 cmnd->common.opcode = nvme_cmd_flush;
554 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
557 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
558 struct nvme_command *cmnd)
560 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
561 struct nvme_dsm_range *range;
564 range = kmalloc_array(segments, sizeof(*range),
565 GFP_ATOMIC | __GFP_NOWARN);
568 * If we fail allocation our range, fallback to the controller
569 * discard page. If that's also busy, it's safe to return
570 * busy, as we know we can make progress once that's freed.
572 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
573 return BLK_STS_RESOURCE;
575 range = page_address(ns->ctrl->discard_page);
578 __rq_for_each_bio(bio, req) {
579 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
580 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
583 range[n].cattr = cpu_to_le32(0);
584 range[n].nlb = cpu_to_le32(nlb);
585 range[n].slba = cpu_to_le64(slba);
590 if (WARN_ON_ONCE(n != segments)) {
591 if (virt_to_page(range) == ns->ctrl->discard_page)
592 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
595 return BLK_STS_IOERR;
598 cmnd->dsm.opcode = nvme_cmd_dsm;
599 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
600 cmnd->dsm.nr = cpu_to_le32(segments - 1);
601 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
603 req->special_vec.bv_page = virt_to_page(range);
604 req->special_vec.bv_offset = offset_in_page(range);
605 req->special_vec.bv_len = sizeof(*range) * segments;
606 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
611 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
612 struct request *req, struct nvme_command *cmnd)
614 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
615 return nvme_setup_discard(ns, req, cmnd);
617 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
618 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
619 cmnd->write_zeroes.slba =
620 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
621 cmnd->write_zeroes.length =
622 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
623 cmnd->write_zeroes.control = 0;
627 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
628 struct request *req, struct nvme_command *cmnd)
630 struct nvme_ctrl *ctrl = ns->ctrl;
634 if (req->cmd_flags & REQ_FUA)
635 control |= NVME_RW_FUA;
636 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
637 control |= NVME_RW_LR;
639 if (req->cmd_flags & REQ_RAHEAD)
640 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
642 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
643 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
644 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
645 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
647 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
648 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
652 * If formated with metadata, the block layer always provides a
653 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
654 * we enable the PRACT bit for protection information or set the
655 * namespace capacity to zero to prevent any I/O.
657 if (!blk_integrity_rq(req)) {
658 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
659 return BLK_STS_NOTSUPP;
660 control |= NVME_RW_PRINFO_PRACT;
661 } else if (req_op(req) == REQ_OP_WRITE) {
662 t10_pi_prepare(req, ns->pi_type);
665 switch (ns->pi_type) {
666 case NVME_NS_DPS_PI_TYPE3:
667 control |= NVME_RW_PRINFO_PRCHK_GUARD;
669 case NVME_NS_DPS_PI_TYPE1:
670 case NVME_NS_DPS_PI_TYPE2:
671 control |= NVME_RW_PRINFO_PRCHK_GUARD |
672 NVME_RW_PRINFO_PRCHK_REF;
673 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
678 cmnd->rw.control = cpu_to_le16(control);
679 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
683 void nvme_cleanup_cmd(struct request *req)
685 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
686 nvme_req(req)->status == 0) {
687 struct nvme_ns *ns = req->rq_disk->private_data;
689 t10_pi_complete(req, ns->pi_type,
690 blk_rq_bytes(req) >> ns->lba_shift);
692 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
693 struct nvme_ns *ns = req->rq_disk->private_data;
694 struct page *page = req->special_vec.bv_page;
696 if (page == ns->ctrl->discard_page)
697 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
699 kfree(page_address(page) + req->special_vec.bv_offset);
702 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
704 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
705 struct nvme_command *cmd)
707 blk_status_t ret = BLK_STS_OK;
709 nvme_clear_nvme_request(req);
711 memset(cmd, 0, sizeof(*cmd));
712 switch (req_op(req)) {
715 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
718 nvme_setup_flush(ns, cmd);
720 case REQ_OP_WRITE_ZEROES:
721 ret = nvme_setup_write_zeroes(ns, req, cmd);
724 ret = nvme_setup_discard(ns, req, cmd);
728 ret = nvme_setup_rw(ns, req, cmd);
732 return BLK_STS_IOERR;
735 cmd->common.command_id = req->tag;
736 trace_nvme_setup_cmd(req, cmd);
739 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
741 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
743 struct completion *waiting = rq->end_io_data;
745 rq->end_io_data = NULL;
749 static void nvme_execute_rq_polled(struct request_queue *q,
750 struct gendisk *bd_disk, struct request *rq, int at_head)
752 DECLARE_COMPLETION_ONSTACK(wait);
754 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
756 rq->cmd_flags |= REQ_HIPRI;
757 rq->end_io_data = &wait;
758 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
760 while (!completion_done(&wait)) {
761 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
767 * Returns 0 on success. If the result is negative, it's a Linux error code;
768 * if the result is positive, it's an NVM Express status code
770 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
771 union nvme_result *result, void *buffer, unsigned bufflen,
772 unsigned timeout, int qid, int at_head,
773 blk_mq_req_flags_t flags, bool poll)
778 req = nvme_alloc_request(q, cmd, flags, qid);
782 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
784 if (buffer && bufflen) {
785 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
791 nvme_execute_rq_polled(req->q, NULL, req, at_head);
793 blk_execute_rq(req->q, NULL, req, at_head);
795 *result = nvme_req(req)->result;
796 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
799 ret = nvme_req(req)->status;
801 blk_mq_free_request(req);
804 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
806 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
807 void *buffer, unsigned bufflen)
809 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
810 NVME_QID_ANY, 0, 0, false);
812 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
814 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
815 unsigned len, u32 seed, bool write)
817 struct bio_integrity_payload *bip;
821 buf = kmalloc(len, GFP_KERNEL);
826 if (write && copy_from_user(buf, ubuf, len))
829 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
835 bip->bip_iter.bi_size = len;
836 bip->bip_iter.bi_sector = seed;
837 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
838 offset_in_page(buf));
848 static int nvme_submit_user_cmd(struct request_queue *q,
849 struct nvme_command *cmd, void __user *ubuffer,
850 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
851 u32 meta_seed, u32 *result, unsigned timeout)
853 bool write = nvme_is_write(cmd);
854 struct nvme_ns *ns = q->queuedata;
855 struct gendisk *disk = ns ? ns->disk : NULL;
857 struct bio *bio = NULL;
861 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
865 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
866 nvme_req(req)->flags |= NVME_REQ_USERCMD;
868 if (ubuffer && bufflen) {
869 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
875 if (disk && meta_buffer && meta_len) {
876 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
882 req->cmd_flags |= REQ_INTEGRITY;
886 blk_execute_rq(req->q, disk, req, 0);
887 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
890 ret = nvme_req(req)->status;
892 *result = le32_to_cpu(nvme_req(req)->result.u32);
893 if (meta && !ret && !write) {
894 if (copy_to_user(meta_buffer, meta, meta_len))
900 blk_rq_unmap_user(bio);
902 blk_mq_free_request(req);
906 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
908 struct nvme_ctrl *ctrl = rq->end_io_data;
910 bool startka = false;
912 blk_mq_free_request(rq);
915 dev_err(ctrl->device,
916 "failed nvme_keep_alive_end_io error=%d\n",
921 ctrl->comp_seen = false;
922 spin_lock_irqsave(&ctrl->lock, flags);
923 if (ctrl->state == NVME_CTRL_LIVE ||
924 ctrl->state == NVME_CTRL_CONNECTING)
926 spin_unlock_irqrestore(&ctrl->lock, flags);
928 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
931 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
935 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
940 rq->timeout = ctrl->kato * HZ;
941 rq->end_io_data = ctrl;
943 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
948 static void nvme_keep_alive_work(struct work_struct *work)
950 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
951 struct nvme_ctrl, ka_work);
952 bool comp_seen = ctrl->comp_seen;
954 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
955 dev_dbg(ctrl->device,
956 "reschedule traffic based keep-alive timer\n");
957 ctrl->comp_seen = false;
958 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
962 if (nvme_keep_alive(ctrl)) {
963 /* allocation failure, reset the controller */
964 dev_err(ctrl->device, "keep-alive failed\n");
965 nvme_reset_ctrl(ctrl);
970 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
972 if (unlikely(ctrl->kato == 0))
975 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
978 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
980 if (unlikely(ctrl->kato == 0))
983 cancel_delayed_work_sync(&ctrl->ka_work);
985 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
987 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
989 struct nvme_command c = { };
992 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
993 c.identify.opcode = nvme_admin_identify;
994 c.identify.cns = NVME_ID_CNS_CTRL;
996 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1000 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1001 sizeof(struct nvme_id_ctrl));
1007 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1008 struct nvme_ns_ids *ids)
1010 struct nvme_command c = { };
1016 c.identify.opcode = nvme_admin_identify;
1017 c.identify.nsid = cpu_to_le32(nsid);
1018 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1020 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1024 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1025 NVME_IDENTIFY_DATA_SIZE);
1029 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1030 struct nvme_ns_id_desc *cur = data + pos;
1035 switch (cur->nidt) {
1036 case NVME_NIDT_EUI64:
1037 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1038 dev_warn(ctrl->device,
1039 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1043 len = NVME_NIDT_EUI64_LEN;
1044 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1046 case NVME_NIDT_NGUID:
1047 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1048 dev_warn(ctrl->device,
1049 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1053 len = NVME_NIDT_NGUID_LEN;
1054 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1056 case NVME_NIDT_UUID:
1057 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1058 dev_warn(ctrl->device,
1059 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1063 len = NVME_NIDT_UUID_LEN;
1064 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1067 /* Skip unknown types */
1072 len += sizeof(*cur);
1079 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1081 struct nvme_command c = { };
1083 c.identify.opcode = nvme_admin_identify;
1084 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1085 c.identify.nsid = cpu_to_le32(nsid);
1086 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1087 NVME_IDENTIFY_DATA_SIZE);
1090 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1093 struct nvme_id_ns *id;
1094 struct nvme_command c = { };
1097 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1098 c.identify.opcode = nvme_admin_identify;
1099 c.identify.nsid = cpu_to_le32(nsid);
1100 c.identify.cns = NVME_ID_CNS_NS;
1102 id = kmalloc(sizeof(*id), GFP_KERNEL);
1106 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1108 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1116 static int nvme_set_features(struct nvme_ctrl *dev, unsigned fid, unsigned dword11,
1117 void *buffer, size_t buflen, u32 *result)
1119 struct nvme_command c;
1120 union nvme_result res;
1123 memset(&c, 0, sizeof(c));
1124 c.features.opcode = nvme_admin_set_features;
1125 c.features.fid = cpu_to_le32(fid);
1126 c.features.dword11 = cpu_to_le32(dword11);
1128 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1129 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1130 if (ret >= 0 && result)
1131 *result = le32_to_cpu(res.u32);
1135 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1137 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1139 int status, nr_io_queues;
1141 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1147 * Degraded controllers might return an error when setting the queue
1148 * count. We still want to be able to bring them online and offer
1149 * access to the admin queue, as that might be only way to fix them up.
1152 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1155 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1156 *count = min(*count, nr_io_queues);
1161 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1163 #define NVME_AEN_SUPPORTED \
1164 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1166 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1168 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1171 if (!supported_aens)
1174 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1177 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1181 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1183 struct nvme_user_io io;
1184 struct nvme_command c;
1185 unsigned length, meta_len;
1186 void __user *metadata;
1188 if (copy_from_user(&io, uio, sizeof(io)))
1193 switch (io.opcode) {
1194 case nvme_cmd_write:
1196 case nvme_cmd_compare:
1202 length = (io.nblocks + 1) << ns->lba_shift;
1203 meta_len = (io.nblocks + 1) * ns->ms;
1204 metadata = (void __user *)(uintptr_t)io.metadata;
1209 } else if (meta_len) {
1210 if ((io.metadata & 3) || !io.metadata)
1214 memset(&c, 0, sizeof(c));
1215 c.rw.opcode = io.opcode;
1216 c.rw.flags = io.flags;
1217 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1218 c.rw.slba = cpu_to_le64(io.slba);
1219 c.rw.length = cpu_to_le16(io.nblocks);
1220 c.rw.control = cpu_to_le16(io.control);
1221 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1222 c.rw.reftag = cpu_to_le32(io.reftag);
1223 c.rw.apptag = cpu_to_le16(io.apptag);
1224 c.rw.appmask = cpu_to_le16(io.appmask);
1226 return nvme_submit_user_cmd(ns->queue, &c,
1227 (void __user *)(uintptr_t)io.addr, length,
1228 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1231 static u32 nvme_known_admin_effects(u8 opcode)
1234 case nvme_admin_format_nvm:
1235 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1236 NVME_CMD_EFFECTS_CSE_MASK;
1237 case nvme_admin_sanitize_nvm:
1238 return NVME_CMD_EFFECTS_CSE_MASK;
1245 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1252 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1253 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1254 dev_warn(ctrl->device,
1255 "IO command:%02x has unhandled effects:%08x\n",
1261 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1263 effects = nvme_known_admin_effects(opcode);
1266 * For simplicity, IO to all namespaces is quiesced even if the command
1267 * effects say only one namespace is affected.
1269 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1270 mutex_lock(&ctrl->scan_lock);
1271 nvme_start_freeze(ctrl);
1272 nvme_wait_freeze(ctrl);
1277 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1281 down_read(&ctrl->namespaces_rwsem);
1282 list_for_each_entry(ns, &ctrl->namespaces, list)
1283 if (ns->disk && nvme_revalidate_disk(ns->disk))
1284 nvme_set_queue_dying(ns);
1285 up_read(&ctrl->namespaces_rwsem);
1287 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1290 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1293 * Revalidate LBA changes prior to unfreezing. This is necessary to
1294 * prevent memory corruption if a logical block size was changed by
1297 if (effects & NVME_CMD_EFFECTS_LBCC)
1298 nvme_update_formats(ctrl);
1299 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1300 nvme_unfreeze(ctrl);
1301 mutex_unlock(&ctrl->scan_lock);
1303 if (effects & NVME_CMD_EFFECTS_CCC)
1304 nvme_init_identify(ctrl);
1305 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1306 nvme_queue_scan(ctrl);
1309 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1310 struct nvme_passthru_cmd __user *ucmd)
1312 struct nvme_passthru_cmd cmd;
1313 struct nvme_command c;
1314 unsigned timeout = 0;
1318 if (!capable(CAP_SYS_ADMIN))
1320 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1325 memset(&c, 0, sizeof(c));
1326 c.common.opcode = cmd.opcode;
1327 c.common.flags = cmd.flags;
1328 c.common.nsid = cpu_to_le32(cmd.nsid);
1329 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1330 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1331 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1332 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1333 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1334 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1335 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1336 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1339 timeout = msecs_to_jiffies(cmd.timeout_ms);
1341 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1342 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1343 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1344 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1345 0, &cmd.result, timeout);
1346 nvme_passthru_end(ctrl, effects);
1349 if (put_user(cmd.result, &ucmd->result))
1357 * Issue ioctl requests on the first available path. Note that unlike normal
1358 * block layer requests we will not retry failed request on another controller.
1360 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1361 struct nvme_ns_head **head, int *srcu_idx)
1363 #ifdef CONFIG_NVME_MULTIPATH
1364 if (disk->fops == &nvme_ns_head_ops) {
1365 *head = disk->private_data;
1366 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1367 return nvme_find_path(*head);
1372 return disk->private_data;
1375 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1378 srcu_read_unlock(&head->srcu, idx);
1381 static int nvme_ns_ioctl(struct nvme_ns *ns, unsigned cmd, unsigned long arg)
1385 force_successful_syscall_return();
1386 return ns->head->ns_id;
1387 case NVME_IOCTL_ADMIN_CMD:
1388 return nvme_user_cmd(ns->ctrl, NULL, (void __user *)arg);
1389 case NVME_IOCTL_IO_CMD:
1390 return nvme_user_cmd(ns->ctrl, ns, (void __user *)arg);
1391 case NVME_IOCTL_SUBMIT_IO:
1392 return nvme_submit_io(ns, (void __user *)arg);
1396 return nvme_nvm_ioctl(ns, cmd, arg);
1398 if (is_sed_ioctl(cmd))
1399 return sed_ioctl(ns->ctrl->opal_dev, cmd,
1400 (void __user *) arg);
1405 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1406 unsigned int cmd, unsigned long arg)
1408 struct nvme_ns_head *head = NULL;
1412 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1416 ret = nvme_ns_ioctl(ns, cmd, arg);
1417 nvme_put_ns_from_disk(head, srcu_idx);
1421 static int nvme_open(struct block_device *bdev, fmode_t mode)
1423 struct nvme_ns *ns = bdev->bd_disk->private_data;
1425 #ifdef CONFIG_NVME_MULTIPATH
1426 /* should never be called due to GENHD_FL_HIDDEN */
1427 if (WARN_ON_ONCE(ns->head->disk))
1430 if (!kref_get_unless_zero(&ns->kref))
1432 if (!try_module_get(ns->ctrl->ops->module))
1443 static void nvme_release(struct gendisk *disk, fmode_t mode)
1445 struct nvme_ns *ns = disk->private_data;
1447 module_put(ns->ctrl->ops->module);
1451 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1453 /* some standard values */
1454 geo->heads = 1 << 6;
1455 geo->sectors = 1 << 5;
1456 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1460 #ifdef CONFIG_BLK_DEV_INTEGRITY
1461 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1463 struct blk_integrity integrity;
1465 memset(&integrity, 0, sizeof(integrity));
1467 case NVME_NS_DPS_PI_TYPE3:
1468 integrity.profile = &t10_pi_type3_crc;
1469 integrity.tag_size = sizeof(u16) + sizeof(u32);
1470 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1472 case NVME_NS_DPS_PI_TYPE1:
1473 case NVME_NS_DPS_PI_TYPE2:
1474 integrity.profile = &t10_pi_type1_crc;
1475 integrity.tag_size = sizeof(u16);
1476 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1479 integrity.profile = NULL;
1482 integrity.tuple_size = ms;
1483 blk_integrity_register(disk, &integrity);
1484 blk_queue_max_integrity_segments(disk->queue, 1);
1487 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1490 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1492 static void nvme_set_chunk_size(struct nvme_ns *ns)
1494 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1495 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1498 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1500 struct nvme_ctrl *ctrl = ns->ctrl;
1501 struct request_queue *queue = disk->queue;
1502 u32 size = queue_logical_block_size(queue);
1504 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1505 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1509 if (ctrl->nr_streams && ns->sws && ns->sgs)
1510 size *= ns->sws * ns->sgs;
1512 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1513 NVME_DSM_MAX_RANGES);
1515 queue->limits.discard_alignment = 0;
1516 queue->limits.discard_granularity = size;
1518 /* If discard is already enabled, don't reset queue limits */
1519 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1522 blk_queue_max_discard_sectors(queue, UINT_MAX);
1523 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1525 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1526 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1529 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1532 unsigned short bs = 1 << ns->lba_shift;
1534 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1535 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1538 * Even though NVMe spec explicitly states that MDTS is not
1539 * applicable to the write-zeroes:- "The restriction does not apply to
1540 * commands that do not transfer data between the host and the
1541 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1542 * In order to be more cautious use controller's max_hw_sectors value
1543 * to configure the maximum sectors for the write-zeroes which is
1544 * configured based on the controller's MDTS field in the
1545 * nvme_init_identify() if available.
1547 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1548 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1550 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1552 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1555 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1556 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1558 memset(ids, 0, sizeof(*ids));
1560 if (ctrl->vs >= NVME_VS(1, 1, 0))
1561 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1562 if (ctrl->vs >= NVME_VS(1, 2, 0))
1563 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1564 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1565 /* Don't treat error as fatal we potentially
1566 * already have a NGUID or EUI-64
1568 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1569 dev_warn(ctrl->device,
1570 "%s: Identify Descriptors failed\n", __func__);
1574 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1576 return !uuid_is_null(&ids->uuid) ||
1577 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1578 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1581 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1583 return uuid_equal(&a->uuid, &b->uuid) &&
1584 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1585 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1588 static void nvme_update_disk_info(struct gendisk *disk,
1589 struct nvme_ns *ns, struct nvme_id_ns *id)
1591 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1592 unsigned short bs = 1 << ns->lba_shift;
1594 if (ns->lba_shift > PAGE_SHIFT) {
1595 /* unsupported block size, set capacity to 0 later */
1598 blk_mq_freeze_queue(disk->queue);
1599 blk_integrity_unregister(disk);
1601 blk_queue_logical_block_size(disk->queue, bs);
1602 blk_queue_physical_block_size(disk->queue, bs);
1603 blk_queue_io_min(disk->queue, bs);
1605 if (ns->ms && !ns->ext &&
1606 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1607 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1608 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1609 ns->lba_shift > PAGE_SHIFT)
1612 set_capacity(disk, capacity);
1614 nvme_config_discard(disk, ns);
1615 nvme_config_write_zeroes(disk, ns);
1617 if (id->nsattr & (1 << 0))
1618 set_disk_ro(disk, true);
1620 set_disk_ro(disk, false);
1622 blk_mq_unfreeze_queue(disk->queue);
1625 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1627 struct nvme_ns *ns = disk->private_data;
1630 * If identify namespace failed, use default 512 byte block size so
1631 * block layer can use before failing read/write for 0 capacity.
1633 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1634 if (ns->lba_shift == 0)
1636 ns->noiob = le16_to_cpu(id->noiob);
1637 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1638 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1639 /* the PI implementation requires metadata equal t10 pi tuple size */
1640 if (ns->ms == sizeof(struct t10_pi_tuple))
1641 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1646 nvme_set_chunk_size(ns);
1647 nvme_update_disk_info(disk, ns, id);
1648 #ifdef CONFIG_NVME_MULTIPATH
1649 if (ns->head->disk) {
1650 nvme_update_disk_info(ns->head->disk, ns, id);
1651 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1656 static int nvme_revalidate_disk(struct gendisk *disk)
1658 struct nvme_ns *ns = disk->private_data;
1659 struct nvme_ctrl *ctrl = ns->ctrl;
1660 struct nvme_id_ns *id;
1661 struct nvme_ns_ids ids;
1664 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1665 set_capacity(disk, 0);
1669 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1673 if (id->ncap == 0) {
1678 __nvme_revalidate_disk(disk, id);
1679 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1680 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1681 dev_err(ctrl->device,
1682 "identifiers changed for nsid %d\n", ns->head->ns_id);
1691 static char nvme_pr_type(enum pr_type type)
1694 case PR_WRITE_EXCLUSIVE:
1696 case PR_EXCLUSIVE_ACCESS:
1698 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1700 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1702 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1704 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1711 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1712 u64 key, u64 sa_key, u8 op)
1714 struct nvme_ns_head *head = NULL;
1716 struct nvme_command c;
1718 u8 data[16] = { 0, };
1720 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1722 return -EWOULDBLOCK;
1724 put_unaligned_le64(key, &data[0]);
1725 put_unaligned_le64(sa_key, &data[8]);
1727 memset(&c, 0, sizeof(c));
1728 c.common.opcode = op;
1729 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1730 c.common.cdw10 = cpu_to_le32(cdw10);
1732 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1733 nvme_put_ns_from_disk(head, srcu_idx);
1737 static int nvme_pr_register(struct block_device *bdev, u64 old,
1738 u64 new, unsigned flags)
1742 if (flags & ~PR_FL_IGNORE_KEY)
1745 cdw10 = old ? 2 : 0;
1746 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1747 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1748 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1751 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1752 enum pr_type type, unsigned flags)
1756 if (flags & ~PR_FL_IGNORE_KEY)
1759 cdw10 = nvme_pr_type(type) << 8;
1760 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1761 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1764 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1765 enum pr_type type, bool abort)
1767 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1768 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1771 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1773 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1774 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1777 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1779 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1780 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1783 static const struct pr_ops nvme_pr_ops = {
1784 .pr_register = nvme_pr_register,
1785 .pr_reserve = nvme_pr_reserve,
1786 .pr_release = nvme_pr_release,
1787 .pr_preempt = nvme_pr_preempt,
1788 .pr_clear = nvme_pr_clear,
1791 #ifdef CONFIG_BLK_SED_OPAL
1792 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1795 struct nvme_ctrl *ctrl = data;
1796 struct nvme_command cmd;
1798 memset(&cmd, 0, sizeof(cmd));
1800 cmd.common.opcode = nvme_admin_security_send;
1802 cmd.common.opcode = nvme_admin_security_recv;
1803 cmd.common.nsid = 0;
1804 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1805 cmd.common.cdw11 = cpu_to_le32(len);
1807 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1808 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1810 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1811 #endif /* CONFIG_BLK_SED_OPAL */
1813 static const struct block_device_operations nvme_fops = {
1814 .owner = THIS_MODULE,
1815 .ioctl = nvme_ioctl,
1816 .compat_ioctl = nvme_ioctl,
1818 .release = nvme_release,
1819 .getgeo = nvme_getgeo,
1820 .revalidate_disk= nvme_revalidate_disk,
1821 .pr_ops = &nvme_pr_ops,
1824 #ifdef CONFIG_NVME_MULTIPATH
1825 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1827 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1829 if (!kref_get_unless_zero(&head->ref))
1834 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1836 nvme_put_ns_head(disk->private_data);
1839 const struct block_device_operations nvme_ns_head_ops = {
1840 .owner = THIS_MODULE,
1841 .open = nvme_ns_head_open,
1842 .release = nvme_ns_head_release,
1843 .ioctl = nvme_ioctl,
1844 .compat_ioctl = nvme_ioctl,
1845 .getgeo = nvme_getgeo,
1846 .pr_ops = &nvme_pr_ops,
1848 #endif /* CONFIG_NVME_MULTIPATH */
1850 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1852 unsigned long timeout =
1853 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1854 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1857 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1860 if ((csts & NVME_CSTS_RDY) == bit)
1864 if (fatal_signal_pending(current))
1866 if (time_after(jiffies, timeout)) {
1867 dev_err(ctrl->device,
1868 "Device not ready; aborting %s\n", enabled ?
1869 "initialisation" : "reset");
1878 * If the device has been passed off to us in an enabled state, just clear
1879 * the enabled bit. The spec says we should set the 'shutdown notification
1880 * bits', but doing so may cause the device to complete commands to the
1881 * admin queue ... and we don't know what memory that might be pointing at!
1883 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1887 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1888 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1890 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1894 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1895 msleep(NVME_QUIRK_DELAY_AMOUNT);
1897 return nvme_wait_ready(ctrl, cap, false);
1899 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1901 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1904 * Default to a 4K page size, with the intention to update this
1905 * path in the future to accomodate architectures with differing
1906 * kernel and IO page sizes.
1908 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1911 if (page_shift < dev_page_min) {
1912 dev_err(ctrl->device,
1913 "Minimum device page size %u too large for host (%u)\n",
1914 1 << dev_page_min, 1 << page_shift);
1918 ctrl->page_size = 1 << page_shift;
1920 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1921 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1922 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1923 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1924 ctrl->ctrl_config |= NVME_CC_ENABLE;
1926 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1929 return nvme_wait_ready(ctrl, cap, true);
1931 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1933 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1935 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1939 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1940 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1942 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1946 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1947 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1951 if (fatal_signal_pending(current))
1953 if (time_after(jiffies, timeout)) {
1954 dev_err(ctrl->device,
1955 "Device shutdown incomplete; abort shutdown\n");
1962 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1964 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1965 struct request_queue *q)
1969 if (ctrl->max_hw_sectors) {
1971 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1973 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1974 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1975 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1977 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1978 is_power_of_2(ctrl->max_hw_sectors))
1979 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1980 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1981 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
1983 blk_queue_write_cache(q, vwc, vwc);
1986 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
1991 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
1994 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
1995 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
1998 dev_warn_once(ctrl->device,
1999 "could not set timestamp (%d)\n", ret);
2003 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2005 struct nvme_feat_host_behavior *host;
2008 /* Don't bother enabling the feature if retry delay is not reported */
2012 host = kzalloc(sizeof(*host), GFP_KERNEL);
2016 host->acre = NVME_ENABLE_ACRE;
2017 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2018 host, sizeof(*host), NULL);
2023 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2026 * APST (Autonomous Power State Transition) lets us program a
2027 * table of power state transitions that the controller will
2028 * perform automatically. We configure it with a simple
2029 * heuristic: we are willing to spend at most 2% of the time
2030 * transitioning between power states. Therefore, when running
2031 * in any given state, we will enter the next lower-power
2032 * non-operational state after waiting 50 * (enlat + exlat)
2033 * microseconds, as long as that state's exit latency is under
2034 * the requested maximum latency.
2036 * We will not autonomously enter any non-operational state for
2037 * which the total latency exceeds ps_max_latency_us. Users
2038 * can set ps_max_latency_us to zero to turn off APST.
2042 struct nvme_feat_auto_pst *table;
2048 * If APST isn't supported or if we haven't been initialized yet,
2049 * then don't do anything.
2054 if (ctrl->npss > 31) {
2055 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2059 table = kzalloc(sizeof(*table), GFP_KERNEL);
2063 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2064 /* Turn off APST. */
2066 dev_dbg(ctrl->device, "APST disabled\n");
2068 __le64 target = cpu_to_le64(0);
2072 * Walk through all states from lowest- to highest-power.
2073 * According to the spec, lower-numbered states use more
2074 * power. NPSS, despite the name, is the index of the
2075 * lowest-power state, not the number of states.
2077 for (state = (int)ctrl->npss; state >= 0; state--) {
2078 u64 total_latency_us, exit_latency_us, transition_ms;
2081 table->entries[state] = target;
2084 * Don't allow transitions to the deepest state
2085 * if it's quirked off.
2087 if (state == ctrl->npss &&
2088 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2092 * Is this state a useful non-operational state for
2093 * higher-power states to autonomously transition to?
2095 if (!(ctrl->psd[state].flags &
2096 NVME_PS_FLAGS_NON_OP_STATE))
2100 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2101 if (exit_latency_us > ctrl->ps_max_latency_us)
2106 le32_to_cpu(ctrl->psd[state].entry_lat);
2109 * This state is good. Use it as the APST idle
2110 * target for higher power states.
2112 transition_ms = total_latency_us + 19;
2113 do_div(transition_ms, 20);
2114 if (transition_ms > (1 << 24) - 1)
2115 transition_ms = (1 << 24) - 1;
2117 target = cpu_to_le64((state << 3) |
2118 (transition_ms << 8));
2123 if (total_latency_us > max_lat_us)
2124 max_lat_us = total_latency_us;
2130 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2132 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2133 max_ps, max_lat_us, (int)sizeof(*table), table);
2137 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2138 table, sizeof(*table), NULL);
2140 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2146 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2148 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2152 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2153 case PM_QOS_LATENCY_ANY:
2161 if (ctrl->ps_max_latency_us != latency) {
2162 ctrl->ps_max_latency_us = latency;
2163 nvme_configure_apst(ctrl);
2167 struct nvme_core_quirk_entry {
2169 * NVMe model and firmware strings are padded with spaces. For
2170 * simplicity, strings in the quirk table are padded with NULLs
2176 unsigned long quirks;
2179 static const struct nvme_core_quirk_entry core_quirks[] = {
2182 * This Toshiba device seems to die using any APST states. See:
2183 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2186 .mn = "THNSF5256GPUK TOSHIBA",
2187 .quirks = NVME_QUIRK_NO_APST,
2191 /* match is null-terminated but idstr is space-padded. */
2192 static bool string_matches(const char *idstr, const char *match, size_t len)
2199 matchlen = strlen(match);
2200 WARN_ON_ONCE(matchlen > len);
2202 if (memcmp(idstr, match, matchlen))
2205 for (; matchlen < len; matchlen++)
2206 if (idstr[matchlen] != ' ')
2212 static bool quirk_matches(const struct nvme_id_ctrl *id,
2213 const struct nvme_core_quirk_entry *q)
2215 return q->vid == le16_to_cpu(id->vid) &&
2216 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2217 string_matches(id->fr, q->fr, sizeof(id->fr));
2220 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2221 struct nvme_id_ctrl *id)
2226 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2227 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2228 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2229 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2233 if (ctrl->vs >= NVME_VS(1, 2, 1))
2234 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2237 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2238 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2239 "nqn.2014.08.org.nvmexpress:%04x%04x",
2240 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2241 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2242 off += sizeof(id->sn);
2243 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2244 off += sizeof(id->mn);
2245 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2248 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2250 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2254 static void nvme_release_subsystem(struct device *dev)
2256 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2259 static void nvme_destroy_subsystem(struct kref *ref)
2261 struct nvme_subsystem *subsys =
2262 container_of(ref, struct nvme_subsystem, ref);
2264 mutex_lock(&nvme_subsystems_lock);
2265 list_del(&subsys->entry);
2266 mutex_unlock(&nvme_subsystems_lock);
2268 ida_destroy(&subsys->ns_ida);
2269 device_del(&subsys->dev);
2270 put_device(&subsys->dev);
2273 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2275 kref_put(&subsys->ref, nvme_destroy_subsystem);
2278 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2280 struct nvme_subsystem *subsys;
2282 lockdep_assert_held(&nvme_subsystems_lock);
2284 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2285 if (strcmp(subsys->subnqn, subsysnqn))
2287 if (!kref_get_unless_zero(&subsys->ref))
2295 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2296 struct device_attribute subsys_attr_##_name = \
2297 __ATTR(_name, _mode, _show, NULL)
2299 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2300 struct device_attribute *attr,
2303 struct nvme_subsystem *subsys =
2304 container_of(dev, struct nvme_subsystem, dev);
2306 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2308 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2310 #define nvme_subsys_show_str_function(field) \
2311 static ssize_t subsys_##field##_show(struct device *dev, \
2312 struct device_attribute *attr, char *buf) \
2314 struct nvme_subsystem *subsys = \
2315 container_of(dev, struct nvme_subsystem, dev); \
2316 return sprintf(buf, "%.*s\n", \
2317 (int)sizeof(subsys->field), subsys->field); \
2319 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2321 nvme_subsys_show_str_function(model);
2322 nvme_subsys_show_str_function(serial);
2323 nvme_subsys_show_str_function(firmware_rev);
2325 static struct attribute *nvme_subsys_attrs[] = {
2326 &subsys_attr_model.attr,
2327 &subsys_attr_serial.attr,
2328 &subsys_attr_firmware_rev.attr,
2329 &subsys_attr_subsysnqn.attr,
2330 #ifdef CONFIG_NVME_MULTIPATH
2331 &subsys_attr_iopolicy.attr,
2336 static struct attribute_group nvme_subsys_attrs_group = {
2337 .attrs = nvme_subsys_attrs,
2340 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2341 &nvme_subsys_attrs_group,
2345 static int nvme_active_ctrls(struct nvme_subsystem *subsys)
2348 struct nvme_ctrl *ctrl;
2350 mutex_lock(&subsys->lock);
2351 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
2352 if (ctrl->state != NVME_CTRL_DELETING &&
2353 ctrl->state != NVME_CTRL_DEAD)
2356 mutex_unlock(&subsys->lock);
2361 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2363 struct nvme_subsystem *subsys, *found;
2366 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2369 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2374 subsys->instance = ret;
2375 mutex_init(&subsys->lock);
2376 kref_init(&subsys->ref);
2377 INIT_LIST_HEAD(&subsys->ctrls);
2378 INIT_LIST_HEAD(&subsys->nsheads);
2379 nvme_init_subnqn(subsys, ctrl, id);
2380 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2381 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2382 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2383 subsys->vendor_id = le16_to_cpu(id->vid);
2384 subsys->cmic = id->cmic;
2385 #ifdef CONFIG_NVME_MULTIPATH
2386 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2389 subsys->dev.class = nvme_subsys_class;
2390 subsys->dev.release = nvme_release_subsystem;
2391 subsys->dev.groups = nvme_subsys_attrs_groups;
2392 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2393 device_initialize(&subsys->dev);
2395 mutex_lock(&nvme_subsystems_lock);
2396 found = __nvme_find_get_subsystem(subsys->subnqn);
2399 * Verify that the subsystem actually supports multiple
2400 * controllers, else bail out.
2402 if (!(ctrl->opts && ctrl->opts->discovery_nqn) &&
2403 nvme_active_ctrls(found) && !(id->cmic & (1 << 1))) {
2404 dev_err(ctrl->device,
2405 "ignoring ctrl due to duplicate subnqn (%s).\n",
2407 nvme_put_subsystem(found);
2412 __nvme_release_subsystem(subsys);
2415 ret = device_add(&subsys->dev);
2417 dev_err(ctrl->device,
2418 "failed to register subsystem device.\n");
2421 ida_init(&subsys->ns_ida);
2422 list_add_tail(&subsys->entry, &nvme_subsystems);
2425 ctrl->subsys = subsys;
2426 mutex_unlock(&nvme_subsystems_lock);
2428 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2429 dev_name(ctrl->device))) {
2430 dev_err(ctrl->device,
2431 "failed to create sysfs link from subsystem.\n");
2432 /* the transport driver will eventually put the subsystem */
2436 mutex_lock(&subsys->lock);
2437 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2438 mutex_unlock(&subsys->lock);
2443 mutex_unlock(&nvme_subsystems_lock);
2444 put_device(&subsys->dev);
2448 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2449 void *log, size_t size, u64 offset)
2451 struct nvme_command c = { };
2452 unsigned long dwlen = size / 4 - 1;
2454 c.get_log_page.opcode = nvme_admin_get_log_page;
2455 c.get_log_page.nsid = cpu_to_le32(nsid);
2456 c.get_log_page.lid = log_page;
2457 c.get_log_page.lsp = lsp;
2458 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2459 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2460 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2461 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2463 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2466 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2471 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2476 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2477 ctrl->effects, sizeof(*ctrl->effects), 0);
2479 kfree(ctrl->effects);
2480 ctrl->effects = NULL;
2486 * Initialize the cached copies of the Identify data and various controller
2487 * register in our nvme_ctrl structure. This should be called as soon as
2488 * the admin queue is fully up and running.
2490 int nvme_init_identify(struct nvme_ctrl *ctrl)
2492 struct nvme_id_ctrl *id;
2494 int ret, page_shift;
2496 bool prev_apst_enabled;
2498 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2500 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2504 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2506 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2509 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2511 if (ctrl->vs >= NVME_VS(1, 1, 0))
2512 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2514 ret = nvme_identify_ctrl(ctrl, &id);
2516 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2520 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2521 ret = nvme_get_effects_log(ctrl);
2526 if (!ctrl->identified) {
2529 ret = nvme_init_subsystem(ctrl, id);
2534 * Check for quirks. Quirk can depend on firmware version,
2535 * so, in principle, the set of quirks present can change
2536 * across a reset. As a possible future enhancement, we
2537 * could re-scan for quirks every time we reinitialize
2538 * the device, but we'd have to make sure that the driver
2539 * behaves intelligently if the quirks change.
2541 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2542 if (quirk_matches(id, &core_quirks[i]))
2543 ctrl->quirks |= core_quirks[i].quirks;
2547 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2548 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2549 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2552 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2553 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2554 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2556 ctrl->oacs = le16_to_cpu(id->oacs);
2557 ctrl->oncs = le16_to_cpu(id->oncs);
2558 ctrl->oaes = le32_to_cpu(id->oaes);
2559 atomic_set(&ctrl->abort_limit, id->acl + 1);
2560 ctrl->vwc = id->vwc;
2562 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2564 max_hw_sectors = UINT_MAX;
2565 ctrl->max_hw_sectors =
2566 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2568 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2569 ctrl->sgls = le32_to_cpu(id->sgls);
2570 ctrl->kas = le16_to_cpu(id->kas);
2571 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2572 ctrl->ctratt = le32_to_cpu(id->ctratt);
2576 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2578 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2579 shutdown_timeout, 60);
2581 if (ctrl->shutdown_timeout != shutdown_timeout)
2582 dev_info(ctrl->device,
2583 "Shutdown timeout set to %u seconds\n",
2584 ctrl->shutdown_timeout);
2586 ctrl->shutdown_timeout = shutdown_timeout;
2588 ctrl->npss = id->npss;
2589 ctrl->apsta = id->apsta;
2590 prev_apst_enabled = ctrl->apst_enabled;
2591 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2592 if (force_apst && id->apsta) {
2593 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2594 ctrl->apst_enabled = true;
2596 ctrl->apst_enabled = false;
2599 ctrl->apst_enabled = id->apsta;
2601 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2603 if (ctrl->ops->flags & NVME_F_FABRICS) {
2604 ctrl->icdoff = le16_to_cpu(id->icdoff);
2605 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2606 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2607 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2610 * In fabrics we need to verify the cntlid matches the
2613 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2618 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2619 dev_err(ctrl->device,
2620 "keep-alive support is mandatory for fabrics\n");
2625 ctrl->cntlid = le16_to_cpu(id->cntlid);
2626 ctrl->hmpre = le32_to_cpu(id->hmpre);
2627 ctrl->hmmin = le32_to_cpu(id->hmmin);
2628 ctrl->hmminds = le32_to_cpu(id->hmminds);
2629 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2632 ret = nvme_mpath_init(ctrl, id);
2638 if (ctrl->apst_enabled && !prev_apst_enabled)
2639 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2640 else if (!ctrl->apst_enabled && prev_apst_enabled)
2641 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2643 ret = nvme_configure_apst(ctrl);
2647 ret = nvme_configure_timestamp(ctrl);
2651 ret = nvme_configure_directives(ctrl);
2655 ret = nvme_configure_acre(ctrl);
2659 ctrl->identified = true;
2667 EXPORT_SYMBOL_GPL(nvme_init_identify);
2669 static int nvme_dev_open(struct inode *inode, struct file *file)
2671 struct nvme_ctrl *ctrl =
2672 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2674 switch (ctrl->state) {
2675 case NVME_CTRL_LIVE:
2676 case NVME_CTRL_ADMIN_ONLY:
2679 return -EWOULDBLOCK;
2682 file->private_data = ctrl;
2686 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2691 down_read(&ctrl->namespaces_rwsem);
2692 if (list_empty(&ctrl->namespaces)) {
2697 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2698 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2699 dev_warn(ctrl->device,
2700 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2705 dev_warn(ctrl->device,
2706 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2707 kref_get(&ns->kref);
2708 up_read(&ctrl->namespaces_rwsem);
2710 ret = nvme_user_cmd(ctrl, ns, argp);
2715 up_read(&ctrl->namespaces_rwsem);
2719 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2722 struct nvme_ctrl *ctrl = file->private_data;
2723 void __user *argp = (void __user *)arg;
2726 case NVME_IOCTL_ADMIN_CMD:
2727 return nvme_user_cmd(ctrl, NULL, argp);
2728 case NVME_IOCTL_IO_CMD:
2729 return nvme_dev_user_cmd(ctrl, argp);
2730 case NVME_IOCTL_RESET:
2731 dev_warn(ctrl->device, "resetting controller\n");
2732 return nvme_reset_ctrl_sync(ctrl);
2733 case NVME_IOCTL_SUBSYS_RESET:
2734 return nvme_reset_subsystem(ctrl);
2735 case NVME_IOCTL_RESCAN:
2736 nvme_queue_scan(ctrl);
2743 static const struct file_operations nvme_dev_fops = {
2744 .owner = THIS_MODULE,
2745 .open = nvme_dev_open,
2746 .unlocked_ioctl = nvme_dev_ioctl,
2747 .compat_ioctl = nvme_dev_ioctl,
2750 static ssize_t nvme_sysfs_reset(struct device *dev,
2751 struct device_attribute *attr, const char *buf,
2754 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2757 ret = nvme_reset_ctrl_sync(ctrl);
2762 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2764 static ssize_t nvme_sysfs_rescan(struct device *dev,
2765 struct device_attribute *attr, const char *buf,
2768 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2770 nvme_queue_scan(ctrl);
2773 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2775 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2777 struct gendisk *disk = dev_to_disk(dev);
2779 if (disk->fops == &nvme_fops)
2780 return nvme_get_ns_from_dev(dev)->head;
2782 return disk->private_data;
2785 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2788 struct nvme_ns_head *head = dev_to_ns_head(dev);
2789 struct nvme_ns_ids *ids = &head->ids;
2790 struct nvme_subsystem *subsys = head->subsys;
2791 int serial_len = sizeof(subsys->serial);
2792 int model_len = sizeof(subsys->model);
2794 if (!uuid_is_null(&ids->uuid))
2795 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2797 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2798 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2800 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2801 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2803 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2804 subsys->serial[serial_len - 1] == '\0'))
2806 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2807 subsys->model[model_len - 1] == '\0'))
2810 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2811 serial_len, subsys->serial, model_len, subsys->model,
2814 static DEVICE_ATTR_RO(wwid);
2816 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2819 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2821 static DEVICE_ATTR_RO(nguid);
2823 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2826 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2828 /* For backward compatibility expose the NGUID to userspace if
2829 * we have no UUID set
2831 if (uuid_is_null(&ids->uuid)) {
2832 printk_ratelimited(KERN_WARNING
2833 "No UUID available providing old NGUID\n");
2834 return sprintf(buf, "%pU\n", ids->nguid);
2836 return sprintf(buf, "%pU\n", &ids->uuid);
2838 static DEVICE_ATTR_RO(uuid);
2840 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2843 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2845 static DEVICE_ATTR_RO(eui);
2847 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2850 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2852 static DEVICE_ATTR_RO(nsid);
2854 static struct attribute *nvme_ns_id_attrs[] = {
2855 &dev_attr_wwid.attr,
2856 &dev_attr_uuid.attr,
2857 &dev_attr_nguid.attr,
2859 &dev_attr_nsid.attr,
2860 #ifdef CONFIG_NVME_MULTIPATH
2861 &dev_attr_ana_grpid.attr,
2862 &dev_attr_ana_state.attr,
2867 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2868 struct attribute *a, int n)
2870 struct device *dev = container_of(kobj, struct device, kobj);
2871 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2873 if (a == &dev_attr_uuid.attr) {
2874 if (uuid_is_null(&ids->uuid) &&
2875 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2878 if (a == &dev_attr_nguid.attr) {
2879 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2882 if (a == &dev_attr_eui.attr) {
2883 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2886 #ifdef CONFIG_NVME_MULTIPATH
2887 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2888 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2890 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2897 static const struct attribute_group nvme_ns_id_attr_group = {
2898 .attrs = nvme_ns_id_attrs,
2899 .is_visible = nvme_ns_id_attrs_are_visible,
2902 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2903 &nvme_ns_id_attr_group,
2905 &nvme_nvm_attr_group,
2910 #define nvme_show_str_function(field) \
2911 static ssize_t field##_show(struct device *dev, \
2912 struct device_attribute *attr, char *buf) \
2914 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2915 return sprintf(buf, "%.*s\n", \
2916 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2918 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2920 nvme_show_str_function(model);
2921 nvme_show_str_function(serial);
2922 nvme_show_str_function(firmware_rev);
2924 #define nvme_show_int_function(field) \
2925 static ssize_t field##_show(struct device *dev, \
2926 struct device_attribute *attr, char *buf) \
2928 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2929 return sprintf(buf, "%d\n", ctrl->field); \
2931 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2933 nvme_show_int_function(cntlid);
2934 nvme_show_int_function(numa_node);
2936 static ssize_t nvme_sysfs_delete(struct device *dev,
2937 struct device_attribute *attr, const char *buf,
2940 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2942 if (device_remove_file_self(dev, attr))
2943 nvme_delete_ctrl_sync(ctrl);
2946 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2948 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2949 struct device_attribute *attr,
2952 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2954 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2956 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2958 static ssize_t nvme_sysfs_show_state(struct device *dev,
2959 struct device_attribute *attr,
2962 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2963 static const char *const state_name[] = {
2964 [NVME_CTRL_NEW] = "new",
2965 [NVME_CTRL_LIVE] = "live",
2966 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2967 [NVME_CTRL_RESETTING] = "resetting",
2968 [NVME_CTRL_CONNECTING] = "connecting",
2969 [NVME_CTRL_DELETING] = "deleting",
2970 [NVME_CTRL_DEAD] = "dead",
2973 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2974 state_name[ctrl->state])
2975 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2977 return sprintf(buf, "unknown state\n");
2980 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
2982 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
2983 struct device_attribute *attr,
2986 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2988 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
2990 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
2992 static ssize_t nvme_sysfs_show_address(struct device *dev,
2993 struct device_attribute *attr,
2996 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2998 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3000 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3002 static struct attribute *nvme_dev_attrs[] = {
3003 &dev_attr_reset_controller.attr,
3004 &dev_attr_rescan_controller.attr,
3005 &dev_attr_model.attr,
3006 &dev_attr_serial.attr,
3007 &dev_attr_firmware_rev.attr,
3008 &dev_attr_cntlid.attr,
3009 &dev_attr_delete_controller.attr,
3010 &dev_attr_transport.attr,
3011 &dev_attr_subsysnqn.attr,
3012 &dev_attr_address.attr,
3013 &dev_attr_state.attr,
3014 &dev_attr_numa_node.attr,
3018 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3019 struct attribute *a, int n)
3021 struct device *dev = container_of(kobj, struct device, kobj);
3022 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3024 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3026 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3032 static struct attribute_group nvme_dev_attrs_group = {
3033 .attrs = nvme_dev_attrs,
3034 .is_visible = nvme_dev_attrs_are_visible,
3037 static const struct attribute_group *nvme_dev_attr_groups[] = {
3038 &nvme_dev_attrs_group,
3042 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3045 struct nvme_ns_head *h;
3047 lockdep_assert_held(&subsys->lock);
3049 list_for_each_entry(h, &subsys->nsheads, entry) {
3050 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3057 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3058 struct nvme_ns_head *new)
3060 struct nvme_ns_head *h;
3062 lockdep_assert_held(&subsys->lock);
3064 list_for_each_entry(h, &subsys->nsheads, entry) {
3065 if (nvme_ns_ids_valid(&new->ids) &&
3066 !list_empty(&h->list) &&
3067 nvme_ns_ids_equal(&new->ids, &h->ids))
3074 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3075 unsigned nsid, struct nvme_id_ns *id)
3077 struct nvme_ns_head *head;
3078 size_t size = sizeof(*head);
3081 #ifdef CONFIG_NVME_MULTIPATH
3082 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3085 head = kzalloc(size, GFP_KERNEL);
3088 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3091 head->instance = ret;
3092 INIT_LIST_HEAD(&head->list);
3093 ret = init_srcu_struct(&head->srcu);
3095 goto out_ida_remove;
3096 head->subsys = ctrl->subsys;
3098 kref_init(&head->ref);
3100 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3102 ret = __nvme_check_ids(ctrl->subsys, head);
3104 dev_err(ctrl->device,
3105 "duplicate IDs for nsid %d\n", nsid);
3106 goto out_cleanup_srcu;
3109 ret = nvme_mpath_alloc_disk(ctrl, head);
3111 goto out_cleanup_srcu;
3113 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3115 kref_get(&ctrl->subsys->ref);
3119 cleanup_srcu_struct(&head->srcu);
3121 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3125 return ERR_PTR(ret);
3128 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3129 struct nvme_id_ns *id)
3131 struct nvme_ctrl *ctrl = ns->ctrl;
3132 bool is_shared = id->nmic & (1 << 0);
3133 struct nvme_ns_head *head = NULL;
3136 mutex_lock(&ctrl->subsys->lock);
3138 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3140 head = nvme_alloc_ns_head(ctrl, nsid, id);
3142 ret = PTR_ERR(head);
3146 struct nvme_ns_ids ids;
3148 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3149 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3150 dev_err(ctrl->device,
3151 "IDs don't match for shared namespace %d\n",
3158 list_add_tail(&ns->siblings, &head->list);
3162 mutex_unlock(&ctrl->subsys->lock);
3166 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3168 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3169 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3171 return nsa->head->ns_id - nsb->head->ns_id;
3174 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3176 struct nvme_ns *ns, *ret = NULL;
3178 down_read(&ctrl->namespaces_rwsem);
3179 list_for_each_entry(ns, &ctrl->namespaces, list) {
3180 if (ns->head->ns_id == nsid) {
3181 if (!kref_get_unless_zero(&ns->kref))
3186 if (ns->head->ns_id > nsid)
3189 up_read(&ctrl->namespaces_rwsem);
3193 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3195 struct streams_directive_params s;
3198 if (!ctrl->nr_streams)
3201 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3205 ns->sws = le32_to_cpu(s.sws);
3206 ns->sgs = le16_to_cpu(s.sgs);
3209 unsigned int bs = 1 << ns->lba_shift;
3211 blk_queue_io_min(ns->queue, bs * ns->sws);
3213 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3219 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3222 struct gendisk *disk;
3223 struct nvme_id_ns *id;
3224 char disk_name[DISK_NAME_LEN];
3225 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3227 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3231 ns->queue = blk_mq_init_queue(ctrl->tagset);
3232 if (IS_ERR(ns->queue)) {
3233 ret = PTR_ERR(ns->queue);
3237 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3238 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3239 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3241 ns->queue->queuedata = ns;
3244 kref_init(&ns->kref);
3245 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3247 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3248 nvme_set_queue_limits(ctrl, ns->queue);
3250 id = nvme_identify_ns(ctrl, nsid);
3253 goto out_free_queue;
3256 if (id->ncap == 0) {
3261 ret = nvme_init_ns_head(ns, nsid, id);
3264 nvme_setup_streams_ns(ctrl, ns);
3265 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3267 disk = alloc_disk_node(0, node);
3273 disk->fops = &nvme_fops;
3274 disk->private_data = ns;
3275 disk->queue = ns->queue;
3276 disk->flags = flags;
3277 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3280 __nvme_revalidate_disk(disk, id);
3282 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3283 ret = nvme_nvm_register(ns, disk_name, node);
3285 dev_warn(ctrl->device, "LightNVM init failure\n");
3290 down_write(&ctrl->namespaces_rwsem);
3291 list_add_tail(&ns->list, &ctrl->namespaces);
3292 up_write(&ctrl->namespaces_rwsem);
3294 nvme_get_ctrl(ctrl);
3296 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3298 nvme_mpath_add_disk(ns, id);
3299 nvme_fault_inject_init(ns);
3306 mutex_lock(&ctrl->subsys->lock);
3307 list_del_rcu(&ns->siblings);
3308 mutex_unlock(&ctrl->subsys->lock);
3309 nvme_put_ns_head(ns->head);
3313 blk_cleanup_queue(ns->queue);
3319 static void nvme_ns_remove(struct nvme_ns *ns)
3321 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3324 nvme_fault_inject_fini(ns);
3325 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3326 del_gendisk(ns->disk);
3327 blk_cleanup_queue(ns->queue);
3328 if (blk_get_integrity(ns->disk))
3329 blk_integrity_unregister(ns->disk);
3332 mutex_lock(&ns->ctrl->subsys->lock);
3333 list_del_rcu(&ns->siblings);
3334 nvme_mpath_clear_current_path(ns);
3335 mutex_unlock(&ns->ctrl->subsys->lock);
3337 down_write(&ns->ctrl->namespaces_rwsem);
3338 list_del_init(&ns->list);
3339 up_write(&ns->ctrl->namespaces_rwsem);
3341 synchronize_srcu(&ns->head->srcu);
3342 nvme_mpath_check_last_path(ns);
3346 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3350 ns = nvme_find_get_ns(ctrl, nsid);
3352 if (ns->disk && revalidate_disk(ns->disk))
3356 nvme_alloc_ns(ctrl, nsid);
3359 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3362 struct nvme_ns *ns, *next;
3365 down_write(&ctrl->namespaces_rwsem);
3366 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3367 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3368 list_move_tail(&ns->list, &rm_list);
3370 up_write(&ctrl->namespaces_rwsem);
3372 list_for_each_entry_safe(ns, next, &rm_list, list)
3377 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3381 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3384 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3388 for (i = 0; i < num_lists; i++) {
3389 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3393 for (j = 0; j < min(nn, 1024U); j++) {
3394 nsid = le32_to_cpu(ns_list[j]);
3398 nvme_validate_ns(ctrl, nsid);
3400 while (++prev < nsid) {
3401 ns = nvme_find_get_ns(ctrl, prev);
3411 nvme_remove_invalid_namespaces(ctrl, prev);
3417 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3421 for (i = 1; i <= nn; i++)
3422 nvme_validate_ns(ctrl, i);
3424 nvme_remove_invalid_namespaces(ctrl, nn);
3427 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3429 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3433 log = kzalloc(log_size, GFP_KERNEL);
3438 * We need to read the log to clear the AEN, but we don't want to rely
3439 * on it for the changed namespace information as userspace could have
3440 * raced with us in reading the log page, which could cause us to miss
3443 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3446 dev_warn(ctrl->device,
3447 "reading changed ns log failed: %d\n", error);
3452 static void nvme_scan_work(struct work_struct *work)
3454 struct nvme_ctrl *ctrl =
3455 container_of(work, struct nvme_ctrl, scan_work);
3456 struct nvme_id_ctrl *id;
3459 if (ctrl->state != NVME_CTRL_LIVE)
3462 WARN_ON_ONCE(!ctrl->tagset);
3464 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3465 dev_info(ctrl->device, "rescanning namespaces.\n");
3466 nvme_clear_changed_ns_log(ctrl);
3469 if (nvme_identify_ctrl(ctrl, &id))
3472 mutex_lock(&ctrl->scan_lock);
3473 nn = le32_to_cpu(id->nn);
3474 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3475 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3476 if (!nvme_scan_ns_list(ctrl, nn))
3479 nvme_scan_ns_sequential(ctrl, nn);
3481 mutex_unlock(&ctrl->scan_lock);
3483 down_write(&ctrl->namespaces_rwsem);
3484 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3485 up_write(&ctrl->namespaces_rwsem);
3489 * This function iterates the namespace list unlocked to allow recovery from
3490 * controller failure. It is up to the caller to ensure the namespace list is
3491 * not modified by scan work while this function is executing.
3493 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3495 struct nvme_ns *ns, *next;
3498 /* prevent racing with ns scanning */
3499 flush_work(&ctrl->scan_work);
3502 * The dead states indicates the controller was not gracefully
3503 * disconnected. In that case, we won't be able to flush any data while
3504 * removing the namespaces' disks; fail all the queues now to avoid
3505 * potentially having to clean up the failed sync later.
3507 if (ctrl->state == NVME_CTRL_DEAD)
3508 nvme_kill_queues(ctrl);
3510 down_write(&ctrl->namespaces_rwsem);
3511 list_splice_init(&ctrl->namespaces, &ns_list);
3512 up_write(&ctrl->namespaces_rwsem);
3514 list_for_each_entry_safe(ns, next, &ns_list, list)
3517 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3519 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3521 char *envp[2] = { NULL, NULL };
3522 u32 aen_result = ctrl->aen_result;
3524 ctrl->aen_result = 0;
3528 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3531 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3535 static void nvme_async_event_work(struct work_struct *work)
3537 struct nvme_ctrl *ctrl =
3538 container_of(work, struct nvme_ctrl, async_event_work);
3540 nvme_aen_uevent(ctrl);
3541 ctrl->ops->submit_async_event(ctrl);
3544 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3549 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3555 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3558 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3560 struct nvme_fw_slot_info_log *log;
3562 log = kmalloc(sizeof(*log), GFP_KERNEL);
3566 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3568 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3572 static void nvme_fw_act_work(struct work_struct *work)
3574 struct nvme_ctrl *ctrl = container_of(work,
3575 struct nvme_ctrl, fw_act_work);
3576 unsigned long fw_act_timeout;
3579 fw_act_timeout = jiffies +
3580 msecs_to_jiffies(ctrl->mtfa * 100);
3582 fw_act_timeout = jiffies +
3583 msecs_to_jiffies(admin_timeout * 1000);
3585 nvme_stop_queues(ctrl);
3586 while (nvme_ctrl_pp_status(ctrl)) {
3587 if (time_after(jiffies, fw_act_timeout)) {
3588 dev_warn(ctrl->device,
3589 "Fw activation timeout, reset controller\n");
3590 nvme_reset_ctrl(ctrl);
3596 if (ctrl->state != NVME_CTRL_LIVE)
3599 nvme_start_queues(ctrl);
3600 /* read FW slot information to clear the AER */
3601 nvme_get_fw_slot_info(ctrl);
3604 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3606 u32 aer_notice_type = (result & 0xff00) >> 8;
3608 switch (aer_notice_type) {
3609 case NVME_AER_NOTICE_NS_CHANGED:
3610 trace_nvme_async_event(ctrl, aer_notice_type);
3611 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3612 nvme_queue_scan(ctrl);
3614 case NVME_AER_NOTICE_FW_ACT_STARTING:
3615 trace_nvme_async_event(ctrl, aer_notice_type);
3616 queue_work(nvme_wq, &ctrl->fw_act_work);
3618 #ifdef CONFIG_NVME_MULTIPATH
3619 case NVME_AER_NOTICE_ANA:
3620 trace_nvme_async_event(ctrl, aer_notice_type);
3621 if (!ctrl->ana_log_buf)
3623 queue_work(nvme_wq, &ctrl->ana_work);
3627 dev_warn(ctrl->device, "async event result %08x\n", result);
3631 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3632 volatile union nvme_result *res)
3634 u32 result = le32_to_cpu(res->u32);
3635 u32 aer_type = result & 0x07;
3637 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3641 case NVME_AER_NOTICE:
3642 nvme_handle_aen_notice(ctrl, result);
3644 case NVME_AER_ERROR:
3645 case NVME_AER_SMART:
3648 trace_nvme_async_event(ctrl, aer_type);
3649 ctrl->aen_result = result;
3654 queue_work(nvme_wq, &ctrl->async_event_work);
3656 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3658 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3660 nvme_mpath_stop(ctrl);
3661 nvme_stop_keep_alive(ctrl);
3662 flush_work(&ctrl->async_event_work);
3663 cancel_work_sync(&ctrl->fw_act_work);
3665 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3667 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3670 nvme_start_keep_alive(ctrl);
3672 if (ctrl->queue_count > 1) {
3673 nvme_queue_scan(ctrl);
3674 nvme_enable_aen(ctrl);
3675 queue_work(nvme_wq, &ctrl->async_event_work);
3676 nvme_start_queues(ctrl);
3679 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3681 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3683 cdev_device_del(&ctrl->cdev, ctrl->device);
3685 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3687 static void nvme_free_ctrl(struct device *dev)
3689 struct nvme_ctrl *ctrl =
3690 container_of(dev, struct nvme_ctrl, ctrl_device);
3691 struct nvme_subsystem *subsys = ctrl->subsys;
3693 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3694 kfree(ctrl->effects);
3695 nvme_mpath_uninit(ctrl);
3696 __free_page(ctrl->discard_page);
3699 mutex_lock(&subsys->lock);
3700 list_del(&ctrl->subsys_entry);
3701 mutex_unlock(&subsys->lock);
3702 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3705 ctrl->ops->free_ctrl(ctrl);
3708 nvme_put_subsystem(subsys);
3712 * Initialize a NVMe controller structures. This needs to be called during
3713 * earliest initialization so that we have the initialized structured around
3716 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3717 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3721 ctrl->state = NVME_CTRL_NEW;
3722 spin_lock_init(&ctrl->lock);
3723 mutex_init(&ctrl->scan_lock);
3724 INIT_LIST_HEAD(&ctrl->namespaces);
3725 init_rwsem(&ctrl->namespaces_rwsem);
3728 ctrl->quirks = quirks;
3729 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3730 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3731 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3732 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3734 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3735 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3736 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3738 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3740 ctrl->discard_page = alloc_page(GFP_KERNEL);
3741 if (!ctrl->discard_page) {
3746 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3749 ctrl->instance = ret;
3751 device_initialize(&ctrl->ctrl_device);
3752 ctrl->device = &ctrl->ctrl_device;
3753 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3754 ctrl->device->class = nvme_class;
3755 ctrl->device->parent = ctrl->dev;
3756 ctrl->device->groups = nvme_dev_attr_groups;
3757 ctrl->device->release = nvme_free_ctrl;
3758 dev_set_drvdata(ctrl->device, ctrl);
3759 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3761 goto out_release_instance;
3763 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3764 ctrl->cdev.owner = ops->module;
3765 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3770 * Initialize latency tolerance controls. The sysfs files won't
3771 * be visible to userspace unless the device actually supports APST.
3773 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3774 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3775 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3779 kfree_const(ctrl->device->kobj.name);
3780 out_release_instance:
3781 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3783 if (ctrl->discard_page)
3784 __free_page(ctrl->discard_page);
3787 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3790 * nvme_kill_queues(): Ends all namespace queues
3791 * @ctrl: the dead controller that needs to end
3793 * Call this function when the driver determines it is unable to get the
3794 * controller in a state capable of servicing IO.
3796 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3800 down_read(&ctrl->namespaces_rwsem);
3802 /* Forcibly unquiesce queues to avoid blocking dispatch */
3803 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3804 blk_mq_unquiesce_queue(ctrl->admin_q);
3806 list_for_each_entry(ns, &ctrl->namespaces, list)
3807 nvme_set_queue_dying(ns);
3809 up_read(&ctrl->namespaces_rwsem);
3811 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3813 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3817 down_read(&ctrl->namespaces_rwsem);
3818 list_for_each_entry(ns, &ctrl->namespaces, list)
3819 blk_mq_unfreeze_queue(ns->queue);
3820 up_read(&ctrl->namespaces_rwsem);
3822 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3824 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3828 down_read(&ctrl->namespaces_rwsem);
3829 list_for_each_entry(ns, &ctrl->namespaces, list) {
3830 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3834 up_read(&ctrl->namespaces_rwsem);
3836 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3838 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3842 down_read(&ctrl->namespaces_rwsem);
3843 list_for_each_entry(ns, &ctrl->namespaces, list)
3844 blk_mq_freeze_queue_wait(ns->queue);
3845 up_read(&ctrl->namespaces_rwsem);
3847 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3849 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3853 down_read(&ctrl->namespaces_rwsem);
3854 list_for_each_entry(ns, &ctrl->namespaces, list)
3855 blk_freeze_queue_start(ns->queue);
3856 up_read(&ctrl->namespaces_rwsem);
3858 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3860 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3864 down_read(&ctrl->namespaces_rwsem);
3865 list_for_each_entry(ns, &ctrl->namespaces, list)
3866 blk_mq_quiesce_queue(ns->queue);
3867 up_read(&ctrl->namespaces_rwsem);
3869 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3871 void nvme_start_queues(struct nvme_ctrl *ctrl)
3875 down_read(&ctrl->namespaces_rwsem);
3876 list_for_each_entry(ns, &ctrl->namespaces, list)
3877 blk_mq_unquiesce_queue(ns->queue);
3878 up_read(&ctrl->namespaces_rwsem);
3880 EXPORT_SYMBOL_GPL(nvme_start_queues);
3883 * Check we didn't inadvertently grow the command structure sizes:
3885 static inline void _nvme_check_size(void)
3887 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3888 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3889 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3890 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3891 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3892 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3893 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3894 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3895 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3896 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3897 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3898 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3899 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3900 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3901 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3902 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3903 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3907 static int __init nvme_core_init(void)
3909 int result = -ENOMEM;
3913 nvme_wq = alloc_workqueue("nvme-wq",
3914 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3918 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3919 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3923 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3924 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3925 if (!nvme_delete_wq)
3926 goto destroy_reset_wq;
3928 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3930 goto destroy_delete_wq;
3932 nvme_class = class_create(THIS_MODULE, "nvme");
3933 if (IS_ERR(nvme_class)) {
3934 result = PTR_ERR(nvme_class);
3935 goto unregister_chrdev;
3938 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3939 if (IS_ERR(nvme_subsys_class)) {
3940 result = PTR_ERR(nvme_subsys_class);
3946 class_destroy(nvme_class);
3948 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3950 destroy_workqueue(nvme_delete_wq);
3952 destroy_workqueue(nvme_reset_wq);
3954 destroy_workqueue(nvme_wq);
3959 static void __exit nvme_core_exit(void)
3961 ida_destroy(&nvme_subsystems_ida);
3962 class_destroy(nvme_subsys_class);
3963 class_destroy(nvme_class);
3964 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3965 destroy_workqueue(nvme_delete_wq);
3966 destroy_workqueue(nvme_reset_wq);
3967 destroy_workqueue(nvme_wq);
3970 MODULE_LICENSE("GPL");
3971 MODULE_VERSION("1.0");
3972 module_init(nvme_core_init);
3973 module_exit(nvme_core_exit);