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]);
1262 effects |= nvme_known_admin_effects(opcode);
1265 * For simplicity, IO to all namespaces is quiesced even if the command
1266 * effects say only one namespace is affected.
1268 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1269 mutex_lock(&ctrl->scan_lock);
1270 nvme_start_freeze(ctrl);
1271 nvme_wait_freeze(ctrl);
1276 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1280 down_read(&ctrl->namespaces_rwsem);
1281 list_for_each_entry(ns, &ctrl->namespaces, list)
1282 if (ns->disk && nvme_revalidate_disk(ns->disk))
1283 nvme_set_queue_dying(ns);
1284 up_read(&ctrl->namespaces_rwsem);
1286 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1289 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1292 * Revalidate LBA changes prior to unfreezing. This is necessary to
1293 * prevent memory corruption if a logical block size was changed by
1296 if (effects & NVME_CMD_EFFECTS_LBCC)
1297 nvme_update_formats(ctrl);
1298 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1299 nvme_unfreeze(ctrl);
1300 mutex_unlock(&ctrl->scan_lock);
1302 if (effects & NVME_CMD_EFFECTS_CCC)
1303 nvme_init_identify(ctrl);
1304 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1305 nvme_queue_scan(ctrl);
1308 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1309 struct nvme_passthru_cmd __user *ucmd)
1311 struct nvme_passthru_cmd cmd;
1312 struct nvme_command c;
1313 unsigned timeout = 0;
1317 if (!capable(CAP_SYS_ADMIN))
1319 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1324 memset(&c, 0, sizeof(c));
1325 c.common.opcode = cmd.opcode;
1326 c.common.flags = cmd.flags;
1327 c.common.nsid = cpu_to_le32(cmd.nsid);
1328 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1329 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1330 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1331 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1332 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1333 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1334 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1335 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1338 timeout = msecs_to_jiffies(cmd.timeout_ms);
1340 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1341 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1342 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1343 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1344 0, &cmd.result, timeout);
1345 nvme_passthru_end(ctrl, effects);
1348 if (put_user(cmd.result, &ucmd->result))
1356 * Issue ioctl requests on the first available path. Note that unlike normal
1357 * block layer requests we will not retry failed request on another controller.
1359 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1360 struct nvme_ns_head **head, int *srcu_idx)
1362 #ifdef CONFIG_NVME_MULTIPATH
1363 if (disk->fops == &nvme_ns_head_ops) {
1366 *head = disk->private_data;
1367 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1368 ns = nvme_find_path(*head);
1370 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1376 return disk->private_data;
1379 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1382 srcu_read_unlock(&head->srcu, idx);
1385 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1386 unsigned int cmd, unsigned long arg)
1388 struct nvme_ns_head *head = NULL;
1389 void __user *argp = (void __user *)arg;
1393 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1395 return -EWOULDBLOCK;
1398 * Handle ioctls that apply to the controller instead of the namespace
1399 * seperately and drop the ns SRCU reference early. This avoids a
1400 * deadlock when deleting namespaces using the passthrough interface.
1402 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1403 struct nvme_ctrl *ctrl = ns->ctrl;
1405 nvme_get_ctrl(ns->ctrl);
1406 nvme_put_ns_from_disk(head, srcu_idx);
1408 if (cmd == NVME_IOCTL_ADMIN_CMD)
1409 ret = nvme_user_cmd(ctrl, NULL, argp);
1411 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1413 nvme_put_ctrl(ctrl);
1419 force_successful_syscall_return();
1420 ret = ns->head->ns_id;
1422 case NVME_IOCTL_IO_CMD:
1423 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1425 case NVME_IOCTL_SUBMIT_IO:
1426 ret = nvme_submit_io(ns, argp);
1430 ret = nvme_nvm_ioctl(ns, cmd, arg);
1435 nvme_put_ns_from_disk(head, srcu_idx);
1439 static int nvme_open(struct block_device *bdev, fmode_t mode)
1441 struct nvme_ns *ns = bdev->bd_disk->private_data;
1443 #ifdef CONFIG_NVME_MULTIPATH
1444 /* should never be called due to GENHD_FL_HIDDEN */
1445 if (WARN_ON_ONCE(ns->head->disk))
1448 if (!kref_get_unless_zero(&ns->kref))
1450 if (!try_module_get(ns->ctrl->ops->module))
1461 static void nvme_release(struct gendisk *disk, fmode_t mode)
1463 struct nvme_ns *ns = disk->private_data;
1465 module_put(ns->ctrl->ops->module);
1469 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1471 /* some standard values */
1472 geo->heads = 1 << 6;
1473 geo->sectors = 1 << 5;
1474 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1478 #ifdef CONFIG_BLK_DEV_INTEGRITY
1479 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1481 struct blk_integrity integrity;
1483 memset(&integrity, 0, sizeof(integrity));
1485 case NVME_NS_DPS_PI_TYPE3:
1486 integrity.profile = &t10_pi_type3_crc;
1487 integrity.tag_size = sizeof(u16) + sizeof(u32);
1488 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1490 case NVME_NS_DPS_PI_TYPE1:
1491 case NVME_NS_DPS_PI_TYPE2:
1492 integrity.profile = &t10_pi_type1_crc;
1493 integrity.tag_size = sizeof(u16);
1494 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1497 integrity.profile = NULL;
1500 integrity.tuple_size = ms;
1501 blk_integrity_register(disk, &integrity);
1502 blk_queue_max_integrity_segments(disk->queue, 1);
1505 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1508 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1510 static void nvme_set_chunk_size(struct nvme_ns *ns)
1512 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1513 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1516 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1518 struct nvme_ctrl *ctrl = ns->ctrl;
1519 struct request_queue *queue = disk->queue;
1520 u32 size = queue_logical_block_size(queue);
1522 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1523 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1527 if (ctrl->nr_streams && ns->sws && ns->sgs)
1528 size *= ns->sws * ns->sgs;
1530 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1531 NVME_DSM_MAX_RANGES);
1533 queue->limits.discard_alignment = 0;
1534 queue->limits.discard_granularity = size;
1536 /* If discard is already enabled, don't reset queue limits */
1537 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1540 blk_queue_max_discard_sectors(queue, UINT_MAX);
1541 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1543 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1544 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1547 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1550 unsigned short bs = 1 << ns->lba_shift;
1552 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1553 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1556 * Even though NVMe spec explicitly states that MDTS is not
1557 * applicable to the write-zeroes:- "The restriction does not apply to
1558 * commands that do not transfer data between the host and the
1559 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1560 * In order to be more cautious use controller's max_hw_sectors value
1561 * to configure the maximum sectors for the write-zeroes which is
1562 * configured based on the controller's MDTS field in the
1563 * nvme_init_identify() if available.
1565 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1566 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1568 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1570 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1573 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1574 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1576 memset(ids, 0, sizeof(*ids));
1578 if (ctrl->vs >= NVME_VS(1, 1, 0))
1579 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1580 if (ctrl->vs >= NVME_VS(1, 2, 0))
1581 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1582 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1583 /* Don't treat error as fatal we potentially
1584 * already have a NGUID or EUI-64
1586 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1587 dev_warn(ctrl->device,
1588 "%s: Identify Descriptors failed\n", __func__);
1592 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1594 return !uuid_is_null(&ids->uuid) ||
1595 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1596 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1599 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1601 return uuid_equal(&a->uuid, &b->uuid) &&
1602 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1603 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1606 static void nvme_update_disk_info(struct gendisk *disk,
1607 struct nvme_ns *ns, struct nvme_id_ns *id)
1609 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1610 unsigned short bs = 1 << ns->lba_shift;
1612 if (ns->lba_shift > PAGE_SHIFT) {
1613 /* unsupported block size, set capacity to 0 later */
1616 blk_mq_freeze_queue(disk->queue);
1617 blk_integrity_unregister(disk);
1619 blk_queue_logical_block_size(disk->queue, bs);
1620 blk_queue_physical_block_size(disk->queue, bs);
1621 blk_queue_io_min(disk->queue, bs);
1623 if (ns->ms && !ns->ext &&
1624 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1625 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1626 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1627 ns->lba_shift > PAGE_SHIFT)
1630 set_capacity(disk, capacity);
1632 nvme_config_discard(disk, ns);
1633 nvme_config_write_zeroes(disk, ns);
1635 if (id->nsattr & (1 << 0))
1636 set_disk_ro(disk, true);
1638 set_disk_ro(disk, false);
1640 blk_mq_unfreeze_queue(disk->queue);
1643 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1645 struct nvme_ns *ns = disk->private_data;
1648 * If identify namespace failed, use default 512 byte block size so
1649 * block layer can use before failing read/write for 0 capacity.
1651 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1652 if (ns->lba_shift == 0)
1654 ns->noiob = le16_to_cpu(id->noiob);
1655 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1656 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1657 /* the PI implementation requires metadata equal t10 pi tuple size */
1658 if (ns->ms == sizeof(struct t10_pi_tuple))
1659 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1664 nvme_set_chunk_size(ns);
1665 nvme_update_disk_info(disk, ns, id);
1666 #ifdef CONFIG_NVME_MULTIPATH
1667 if (ns->head->disk) {
1668 nvme_update_disk_info(ns->head->disk, ns, id);
1669 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1674 static int nvme_revalidate_disk(struct gendisk *disk)
1676 struct nvme_ns *ns = disk->private_data;
1677 struct nvme_ctrl *ctrl = ns->ctrl;
1678 struct nvme_id_ns *id;
1679 struct nvme_ns_ids ids;
1682 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1683 set_capacity(disk, 0);
1687 id = nvme_identify_ns(ctrl, ns->head->ns_id);
1691 if (id->ncap == 0) {
1696 __nvme_revalidate_disk(disk, id);
1697 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1698 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1699 dev_err(ctrl->device,
1700 "identifiers changed for nsid %d\n", ns->head->ns_id);
1709 static char nvme_pr_type(enum pr_type type)
1712 case PR_WRITE_EXCLUSIVE:
1714 case PR_EXCLUSIVE_ACCESS:
1716 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1718 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1720 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1722 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1729 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1730 u64 key, u64 sa_key, u8 op)
1732 struct nvme_ns_head *head = NULL;
1734 struct nvme_command c;
1736 u8 data[16] = { 0, };
1738 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1740 return -EWOULDBLOCK;
1742 put_unaligned_le64(key, &data[0]);
1743 put_unaligned_le64(sa_key, &data[8]);
1745 memset(&c, 0, sizeof(c));
1746 c.common.opcode = op;
1747 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1748 c.common.cdw10 = cpu_to_le32(cdw10);
1750 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1751 nvme_put_ns_from_disk(head, srcu_idx);
1755 static int nvme_pr_register(struct block_device *bdev, u64 old,
1756 u64 new, unsigned flags)
1760 if (flags & ~PR_FL_IGNORE_KEY)
1763 cdw10 = old ? 2 : 0;
1764 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1765 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1766 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1769 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1770 enum pr_type type, unsigned flags)
1774 if (flags & ~PR_FL_IGNORE_KEY)
1777 cdw10 = nvme_pr_type(type) << 8;
1778 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1779 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1782 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1783 enum pr_type type, bool abort)
1785 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1786 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1789 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1791 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1792 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1795 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1797 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1798 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1801 static const struct pr_ops nvme_pr_ops = {
1802 .pr_register = nvme_pr_register,
1803 .pr_reserve = nvme_pr_reserve,
1804 .pr_release = nvme_pr_release,
1805 .pr_preempt = nvme_pr_preempt,
1806 .pr_clear = nvme_pr_clear,
1809 #ifdef CONFIG_BLK_SED_OPAL
1810 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1813 struct nvme_ctrl *ctrl = data;
1814 struct nvme_command cmd;
1816 memset(&cmd, 0, sizeof(cmd));
1818 cmd.common.opcode = nvme_admin_security_send;
1820 cmd.common.opcode = nvme_admin_security_recv;
1821 cmd.common.nsid = 0;
1822 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1823 cmd.common.cdw11 = cpu_to_le32(len);
1825 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1826 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1828 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1829 #endif /* CONFIG_BLK_SED_OPAL */
1831 static const struct block_device_operations nvme_fops = {
1832 .owner = THIS_MODULE,
1833 .ioctl = nvme_ioctl,
1834 .compat_ioctl = nvme_ioctl,
1836 .release = nvme_release,
1837 .getgeo = nvme_getgeo,
1838 .revalidate_disk= nvme_revalidate_disk,
1839 .pr_ops = &nvme_pr_ops,
1842 #ifdef CONFIG_NVME_MULTIPATH
1843 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1845 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1847 if (!kref_get_unless_zero(&head->ref))
1852 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1854 nvme_put_ns_head(disk->private_data);
1857 const struct block_device_operations nvme_ns_head_ops = {
1858 .owner = THIS_MODULE,
1859 .open = nvme_ns_head_open,
1860 .release = nvme_ns_head_release,
1861 .ioctl = nvme_ioctl,
1862 .compat_ioctl = nvme_ioctl,
1863 .getgeo = nvme_getgeo,
1864 .pr_ops = &nvme_pr_ops,
1866 #endif /* CONFIG_NVME_MULTIPATH */
1868 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1870 unsigned long timeout =
1871 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1872 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1875 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1878 if ((csts & NVME_CSTS_RDY) == bit)
1882 if (fatal_signal_pending(current))
1884 if (time_after(jiffies, timeout)) {
1885 dev_err(ctrl->device,
1886 "Device not ready; aborting %s\n", enabled ?
1887 "initialisation" : "reset");
1896 * If the device has been passed off to us in an enabled state, just clear
1897 * the enabled bit. The spec says we should set the 'shutdown notification
1898 * bits', but doing so may cause the device to complete commands to the
1899 * admin queue ... and we don't know what memory that might be pointing at!
1901 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1905 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1906 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1908 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1912 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1913 msleep(NVME_QUIRK_DELAY_AMOUNT);
1915 return nvme_wait_ready(ctrl, cap, false);
1917 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1919 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1922 * Default to a 4K page size, with the intention to update this
1923 * path in the future to accomodate architectures with differing
1924 * kernel and IO page sizes.
1926 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1929 if (page_shift < dev_page_min) {
1930 dev_err(ctrl->device,
1931 "Minimum device page size %u too large for host (%u)\n",
1932 1 << dev_page_min, 1 << page_shift);
1936 ctrl->page_size = 1 << page_shift;
1938 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1939 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1940 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1941 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1942 ctrl->ctrl_config |= NVME_CC_ENABLE;
1944 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1947 return nvme_wait_ready(ctrl, cap, true);
1949 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
1951 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
1953 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
1957 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1958 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
1960 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1964 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1965 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
1969 if (fatal_signal_pending(current))
1971 if (time_after(jiffies, timeout)) {
1972 dev_err(ctrl->device,
1973 "Device shutdown incomplete; abort shutdown\n");
1980 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
1982 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1983 struct request_queue *q)
1987 if (ctrl->max_hw_sectors) {
1989 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
1991 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1992 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1993 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1995 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1996 is_power_of_2(ctrl->max_hw_sectors))
1997 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
1998 blk_queue_virt_boundary(q, ctrl->page_size - 1);
1999 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2001 blk_queue_write_cache(q, vwc, vwc);
2004 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2009 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2012 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2013 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2016 dev_warn_once(ctrl->device,
2017 "could not set timestamp (%d)\n", ret);
2021 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2023 struct nvme_feat_host_behavior *host;
2026 /* Don't bother enabling the feature if retry delay is not reported */
2030 host = kzalloc(sizeof(*host), GFP_KERNEL);
2034 host->acre = NVME_ENABLE_ACRE;
2035 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2036 host, sizeof(*host), NULL);
2041 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2044 * APST (Autonomous Power State Transition) lets us program a
2045 * table of power state transitions that the controller will
2046 * perform automatically. We configure it with a simple
2047 * heuristic: we are willing to spend at most 2% of the time
2048 * transitioning between power states. Therefore, when running
2049 * in any given state, we will enter the next lower-power
2050 * non-operational state after waiting 50 * (enlat + exlat)
2051 * microseconds, as long as that state's exit latency is under
2052 * the requested maximum latency.
2054 * We will not autonomously enter any non-operational state for
2055 * which the total latency exceeds ps_max_latency_us. Users
2056 * can set ps_max_latency_us to zero to turn off APST.
2060 struct nvme_feat_auto_pst *table;
2066 * If APST isn't supported or if we haven't been initialized yet,
2067 * then don't do anything.
2072 if (ctrl->npss > 31) {
2073 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2077 table = kzalloc(sizeof(*table), GFP_KERNEL);
2081 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2082 /* Turn off APST. */
2084 dev_dbg(ctrl->device, "APST disabled\n");
2086 __le64 target = cpu_to_le64(0);
2090 * Walk through all states from lowest- to highest-power.
2091 * According to the spec, lower-numbered states use more
2092 * power. NPSS, despite the name, is the index of the
2093 * lowest-power state, not the number of states.
2095 for (state = (int)ctrl->npss; state >= 0; state--) {
2096 u64 total_latency_us, exit_latency_us, transition_ms;
2099 table->entries[state] = target;
2102 * Don't allow transitions to the deepest state
2103 * if it's quirked off.
2105 if (state == ctrl->npss &&
2106 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2110 * Is this state a useful non-operational state for
2111 * higher-power states to autonomously transition to?
2113 if (!(ctrl->psd[state].flags &
2114 NVME_PS_FLAGS_NON_OP_STATE))
2118 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2119 if (exit_latency_us > ctrl->ps_max_latency_us)
2124 le32_to_cpu(ctrl->psd[state].entry_lat);
2127 * This state is good. Use it as the APST idle
2128 * target for higher power states.
2130 transition_ms = total_latency_us + 19;
2131 do_div(transition_ms, 20);
2132 if (transition_ms > (1 << 24) - 1)
2133 transition_ms = (1 << 24) - 1;
2135 target = cpu_to_le64((state << 3) |
2136 (transition_ms << 8));
2141 if (total_latency_us > max_lat_us)
2142 max_lat_us = total_latency_us;
2148 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2150 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2151 max_ps, max_lat_us, (int)sizeof(*table), table);
2155 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2156 table, sizeof(*table), NULL);
2158 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2164 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2166 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2170 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2171 case PM_QOS_LATENCY_ANY:
2179 if (ctrl->ps_max_latency_us != latency) {
2180 ctrl->ps_max_latency_us = latency;
2181 nvme_configure_apst(ctrl);
2185 struct nvme_core_quirk_entry {
2187 * NVMe model and firmware strings are padded with spaces. For
2188 * simplicity, strings in the quirk table are padded with NULLs
2194 unsigned long quirks;
2197 static const struct nvme_core_quirk_entry core_quirks[] = {
2200 * This Toshiba device seems to die using any APST states. See:
2201 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2204 .mn = "THNSF5256GPUK TOSHIBA",
2205 .quirks = NVME_QUIRK_NO_APST,
2209 /* match is null-terminated but idstr is space-padded. */
2210 static bool string_matches(const char *idstr, const char *match, size_t len)
2217 matchlen = strlen(match);
2218 WARN_ON_ONCE(matchlen > len);
2220 if (memcmp(idstr, match, matchlen))
2223 for (; matchlen < len; matchlen++)
2224 if (idstr[matchlen] != ' ')
2230 static bool quirk_matches(const struct nvme_id_ctrl *id,
2231 const struct nvme_core_quirk_entry *q)
2233 return q->vid == le16_to_cpu(id->vid) &&
2234 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2235 string_matches(id->fr, q->fr, sizeof(id->fr));
2238 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2239 struct nvme_id_ctrl *id)
2244 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2245 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2246 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2247 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2251 if (ctrl->vs >= NVME_VS(1, 2, 1))
2252 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2255 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2256 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2257 "nqn.2014.08.org.nvmexpress:%04x%04x",
2258 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2259 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2260 off += sizeof(id->sn);
2261 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2262 off += sizeof(id->mn);
2263 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2266 static void __nvme_release_subsystem(struct nvme_subsystem *subsys)
2268 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2272 static void nvme_release_subsystem(struct device *dev)
2274 __nvme_release_subsystem(container_of(dev, struct nvme_subsystem, dev));
2277 static void nvme_destroy_subsystem(struct kref *ref)
2279 struct nvme_subsystem *subsys =
2280 container_of(ref, struct nvme_subsystem, ref);
2282 mutex_lock(&nvme_subsystems_lock);
2283 list_del(&subsys->entry);
2284 mutex_unlock(&nvme_subsystems_lock);
2286 ida_destroy(&subsys->ns_ida);
2287 device_del(&subsys->dev);
2288 put_device(&subsys->dev);
2291 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2293 kref_put(&subsys->ref, nvme_destroy_subsystem);
2296 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2298 struct nvme_subsystem *subsys;
2300 lockdep_assert_held(&nvme_subsystems_lock);
2302 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2303 if (strcmp(subsys->subnqn, subsysnqn))
2305 if (!kref_get_unless_zero(&subsys->ref))
2313 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2314 struct device_attribute subsys_attr_##_name = \
2315 __ATTR(_name, _mode, _show, NULL)
2317 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2318 struct device_attribute *attr,
2321 struct nvme_subsystem *subsys =
2322 container_of(dev, struct nvme_subsystem, dev);
2324 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2326 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2328 #define nvme_subsys_show_str_function(field) \
2329 static ssize_t subsys_##field##_show(struct device *dev, \
2330 struct device_attribute *attr, char *buf) \
2332 struct nvme_subsystem *subsys = \
2333 container_of(dev, struct nvme_subsystem, dev); \
2334 return sprintf(buf, "%.*s\n", \
2335 (int)sizeof(subsys->field), subsys->field); \
2337 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2339 nvme_subsys_show_str_function(model);
2340 nvme_subsys_show_str_function(serial);
2341 nvme_subsys_show_str_function(firmware_rev);
2343 static struct attribute *nvme_subsys_attrs[] = {
2344 &subsys_attr_model.attr,
2345 &subsys_attr_serial.attr,
2346 &subsys_attr_firmware_rev.attr,
2347 &subsys_attr_subsysnqn.attr,
2348 #ifdef CONFIG_NVME_MULTIPATH
2349 &subsys_attr_iopolicy.attr,
2354 static struct attribute_group nvme_subsys_attrs_group = {
2355 .attrs = nvme_subsys_attrs,
2358 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2359 &nvme_subsys_attrs_group,
2363 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2364 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2366 struct nvme_ctrl *tmp;
2368 lockdep_assert_held(&nvme_subsystems_lock);
2370 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2371 if (ctrl->state == NVME_CTRL_DELETING ||
2372 ctrl->state == NVME_CTRL_DEAD)
2375 if (tmp->cntlid == ctrl->cntlid) {
2376 dev_err(ctrl->device,
2377 "Duplicate cntlid %u with %s, rejecting\n",
2378 ctrl->cntlid, dev_name(tmp->device));
2382 if ((id->cmic & (1 << 1)) ||
2383 (ctrl->opts && ctrl->opts->discovery_nqn))
2386 dev_err(ctrl->device,
2387 "Subsystem does not support multiple controllers\n");
2394 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2396 struct nvme_subsystem *subsys, *found;
2399 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2402 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2407 subsys->instance = ret;
2408 mutex_init(&subsys->lock);
2409 kref_init(&subsys->ref);
2410 INIT_LIST_HEAD(&subsys->ctrls);
2411 INIT_LIST_HEAD(&subsys->nsheads);
2412 nvme_init_subnqn(subsys, ctrl, id);
2413 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2414 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2415 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2416 subsys->vendor_id = le16_to_cpu(id->vid);
2417 subsys->cmic = id->cmic;
2418 #ifdef CONFIG_NVME_MULTIPATH
2419 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2422 subsys->dev.class = nvme_subsys_class;
2423 subsys->dev.release = nvme_release_subsystem;
2424 subsys->dev.groups = nvme_subsys_attrs_groups;
2425 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2426 device_initialize(&subsys->dev);
2428 mutex_lock(&nvme_subsystems_lock);
2429 found = __nvme_find_get_subsystem(subsys->subnqn);
2431 __nvme_release_subsystem(subsys);
2434 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2436 goto out_put_subsystem;
2439 ret = device_add(&subsys->dev);
2441 dev_err(ctrl->device,
2442 "failed to register subsystem device.\n");
2445 ida_init(&subsys->ns_ida);
2446 list_add_tail(&subsys->entry, &nvme_subsystems);
2449 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2450 dev_name(ctrl->device))) {
2451 dev_err(ctrl->device,
2452 "failed to create sysfs link from subsystem.\n");
2453 goto out_put_subsystem;
2456 ctrl->subsys = subsys;
2457 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2458 mutex_unlock(&nvme_subsystems_lock);
2462 nvme_put_subsystem(subsys);
2464 mutex_unlock(&nvme_subsystems_lock);
2465 put_device(&subsys->dev);
2469 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2470 void *log, size_t size, u64 offset)
2472 struct nvme_command c = { };
2473 unsigned long dwlen = size / 4 - 1;
2475 c.get_log_page.opcode = nvme_admin_get_log_page;
2476 c.get_log_page.nsid = cpu_to_le32(nsid);
2477 c.get_log_page.lid = log_page;
2478 c.get_log_page.lsp = lsp;
2479 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2480 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2481 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2482 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2484 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2487 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2492 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2497 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2498 ctrl->effects, sizeof(*ctrl->effects), 0);
2500 kfree(ctrl->effects);
2501 ctrl->effects = NULL;
2507 * Initialize the cached copies of the Identify data and various controller
2508 * register in our nvme_ctrl structure. This should be called as soon as
2509 * the admin queue is fully up and running.
2511 int nvme_init_identify(struct nvme_ctrl *ctrl)
2513 struct nvme_id_ctrl *id;
2515 int ret, page_shift;
2517 bool prev_apst_enabled;
2519 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2521 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2525 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2527 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2530 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2532 if (ctrl->vs >= NVME_VS(1, 1, 0))
2533 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2535 ret = nvme_identify_ctrl(ctrl, &id);
2537 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2541 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2542 ret = nvme_get_effects_log(ctrl);
2547 if (!ctrl->identified) {
2550 ret = nvme_init_subsystem(ctrl, id);
2555 * Check for quirks. Quirk can depend on firmware version,
2556 * so, in principle, the set of quirks present can change
2557 * across a reset. As a possible future enhancement, we
2558 * could re-scan for quirks every time we reinitialize
2559 * the device, but we'd have to make sure that the driver
2560 * behaves intelligently if the quirks change.
2562 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2563 if (quirk_matches(id, &core_quirks[i]))
2564 ctrl->quirks |= core_quirks[i].quirks;
2568 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2569 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2570 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2573 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2574 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2575 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2577 ctrl->oacs = le16_to_cpu(id->oacs);
2578 ctrl->oncs = le16_to_cpu(id->oncs);
2579 ctrl->mtfa = le16_to_cpu(id->mtfa);
2580 ctrl->oaes = le32_to_cpu(id->oaes);
2581 atomic_set(&ctrl->abort_limit, id->acl + 1);
2582 ctrl->vwc = id->vwc;
2584 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2586 max_hw_sectors = UINT_MAX;
2587 ctrl->max_hw_sectors =
2588 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2590 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2591 ctrl->sgls = le32_to_cpu(id->sgls);
2592 ctrl->kas = le16_to_cpu(id->kas);
2593 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2594 ctrl->ctratt = le32_to_cpu(id->ctratt);
2598 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2600 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2601 shutdown_timeout, 60);
2603 if (ctrl->shutdown_timeout != shutdown_timeout)
2604 dev_info(ctrl->device,
2605 "Shutdown timeout set to %u seconds\n",
2606 ctrl->shutdown_timeout);
2608 ctrl->shutdown_timeout = shutdown_timeout;
2610 ctrl->npss = id->npss;
2611 ctrl->apsta = id->apsta;
2612 prev_apst_enabled = ctrl->apst_enabled;
2613 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2614 if (force_apst && id->apsta) {
2615 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2616 ctrl->apst_enabled = true;
2618 ctrl->apst_enabled = false;
2621 ctrl->apst_enabled = id->apsta;
2623 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2625 if (ctrl->ops->flags & NVME_F_FABRICS) {
2626 ctrl->icdoff = le16_to_cpu(id->icdoff);
2627 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2628 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2629 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2632 * In fabrics we need to verify the cntlid matches the
2635 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2640 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2641 dev_err(ctrl->device,
2642 "keep-alive support is mandatory for fabrics\n");
2647 ctrl->cntlid = le16_to_cpu(id->cntlid);
2648 ctrl->hmpre = le32_to_cpu(id->hmpre);
2649 ctrl->hmmin = le32_to_cpu(id->hmmin);
2650 ctrl->hmminds = le32_to_cpu(id->hmminds);
2651 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2654 ret = nvme_mpath_init(ctrl, id);
2660 if (ctrl->apst_enabled && !prev_apst_enabled)
2661 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2662 else if (!ctrl->apst_enabled && prev_apst_enabled)
2663 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2665 ret = nvme_configure_apst(ctrl);
2669 ret = nvme_configure_timestamp(ctrl);
2673 ret = nvme_configure_directives(ctrl);
2677 ret = nvme_configure_acre(ctrl);
2681 ctrl->identified = true;
2689 EXPORT_SYMBOL_GPL(nvme_init_identify);
2691 static int nvme_dev_open(struct inode *inode, struct file *file)
2693 struct nvme_ctrl *ctrl =
2694 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2696 switch (ctrl->state) {
2697 case NVME_CTRL_LIVE:
2698 case NVME_CTRL_ADMIN_ONLY:
2701 return -EWOULDBLOCK;
2704 file->private_data = ctrl;
2708 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2713 down_read(&ctrl->namespaces_rwsem);
2714 if (list_empty(&ctrl->namespaces)) {
2719 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2720 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2721 dev_warn(ctrl->device,
2722 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2727 dev_warn(ctrl->device,
2728 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2729 kref_get(&ns->kref);
2730 up_read(&ctrl->namespaces_rwsem);
2732 ret = nvme_user_cmd(ctrl, ns, argp);
2737 up_read(&ctrl->namespaces_rwsem);
2741 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2744 struct nvme_ctrl *ctrl = file->private_data;
2745 void __user *argp = (void __user *)arg;
2748 case NVME_IOCTL_ADMIN_CMD:
2749 return nvme_user_cmd(ctrl, NULL, argp);
2750 case NVME_IOCTL_IO_CMD:
2751 return nvme_dev_user_cmd(ctrl, argp);
2752 case NVME_IOCTL_RESET:
2753 dev_warn(ctrl->device, "resetting controller\n");
2754 return nvme_reset_ctrl_sync(ctrl);
2755 case NVME_IOCTL_SUBSYS_RESET:
2756 return nvme_reset_subsystem(ctrl);
2757 case NVME_IOCTL_RESCAN:
2758 nvme_queue_scan(ctrl);
2765 static const struct file_operations nvme_dev_fops = {
2766 .owner = THIS_MODULE,
2767 .open = nvme_dev_open,
2768 .unlocked_ioctl = nvme_dev_ioctl,
2769 .compat_ioctl = nvme_dev_ioctl,
2772 static ssize_t nvme_sysfs_reset(struct device *dev,
2773 struct device_attribute *attr, const char *buf,
2776 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2779 ret = nvme_reset_ctrl_sync(ctrl);
2784 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2786 static ssize_t nvme_sysfs_rescan(struct device *dev,
2787 struct device_attribute *attr, const char *buf,
2790 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2792 nvme_queue_scan(ctrl);
2795 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2797 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2799 struct gendisk *disk = dev_to_disk(dev);
2801 if (disk->fops == &nvme_fops)
2802 return nvme_get_ns_from_dev(dev)->head;
2804 return disk->private_data;
2807 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2810 struct nvme_ns_head *head = dev_to_ns_head(dev);
2811 struct nvme_ns_ids *ids = &head->ids;
2812 struct nvme_subsystem *subsys = head->subsys;
2813 int serial_len = sizeof(subsys->serial);
2814 int model_len = sizeof(subsys->model);
2816 if (!uuid_is_null(&ids->uuid))
2817 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2819 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2820 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2822 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2823 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2825 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2826 subsys->serial[serial_len - 1] == '\0'))
2828 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2829 subsys->model[model_len - 1] == '\0'))
2832 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2833 serial_len, subsys->serial, model_len, subsys->model,
2836 static DEVICE_ATTR_RO(wwid);
2838 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2841 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2843 static DEVICE_ATTR_RO(nguid);
2845 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2848 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2850 /* For backward compatibility expose the NGUID to userspace if
2851 * we have no UUID set
2853 if (uuid_is_null(&ids->uuid)) {
2854 printk_ratelimited(KERN_WARNING
2855 "No UUID available providing old NGUID\n");
2856 return sprintf(buf, "%pU\n", ids->nguid);
2858 return sprintf(buf, "%pU\n", &ids->uuid);
2860 static DEVICE_ATTR_RO(uuid);
2862 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2865 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2867 static DEVICE_ATTR_RO(eui);
2869 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2872 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2874 static DEVICE_ATTR_RO(nsid);
2876 static struct attribute *nvme_ns_id_attrs[] = {
2877 &dev_attr_wwid.attr,
2878 &dev_attr_uuid.attr,
2879 &dev_attr_nguid.attr,
2881 &dev_attr_nsid.attr,
2882 #ifdef CONFIG_NVME_MULTIPATH
2883 &dev_attr_ana_grpid.attr,
2884 &dev_attr_ana_state.attr,
2889 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2890 struct attribute *a, int n)
2892 struct device *dev = container_of(kobj, struct device, kobj);
2893 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2895 if (a == &dev_attr_uuid.attr) {
2896 if (uuid_is_null(&ids->uuid) &&
2897 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2900 if (a == &dev_attr_nguid.attr) {
2901 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2904 if (a == &dev_attr_eui.attr) {
2905 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2908 #ifdef CONFIG_NVME_MULTIPATH
2909 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2910 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2912 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2919 static const struct attribute_group nvme_ns_id_attr_group = {
2920 .attrs = nvme_ns_id_attrs,
2921 .is_visible = nvme_ns_id_attrs_are_visible,
2924 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2925 &nvme_ns_id_attr_group,
2927 &nvme_nvm_attr_group,
2932 #define nvme_show_str_function(field) \
2933 static ssize_t field##_show(struct device *dev, \
2934 struct device_attribute *attr, char *buf) \
2936 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2937 return sprintf(buf, "%.*s\n", \
2938 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2940 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2942 nvme_show_str_function(model);
2943 nvme_show_str_function(serial);
2944 nvme_show_str_function(firmware_rev);
2946 #define nvme_show_int_function(field) \
2947 static ssize_t field##_show(struct device *dev, \
2948 struct device_attribute *attr, char *buf) \
2950 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2951 return sprintf(buf, "%d\n", ctrl->field); \
2953 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2955 nvme_show_int_function(cntlid);
2956 nvme_show_int_function(numa_node);
2958 static ssize_t nvme_sysfs_delete(struct device *dev,
2959 struct device_attribute *attr, const char *buf,
2962 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2964 if (device_remove_file_self(dev, attr))
2965 nvme_delete_ctrl_sync(ctrl);
2968 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
2970 static ssize_t nvme_sysfs_show_transport(struct device *dev,
2971 struct device_attribute *attr,
2974 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2976 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
2978 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
2980 static ssize_t nvme_sysfs_show_state(struct device *dev,
2981 struct device_attribute *attr,
2984 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2985 static const char *const state_name[] = {
2986 [NVME_CTRL_NEW] = "new",
2987 [NVME_CTRL_LIVE] = "live",
2988 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
2989 [NVME_CTRL_RESETTING] = "resetting",
2990 [NVME_CTRL_CONNECTING] = "connecting",
2991 [NVME_CTRL_DELETING] = "deleting",
2992 [NVME_CTRL_DEAD] = "dead",
2995 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
2996 state_name[ctrl->state])
2997 return sprintf(buf, "%s\n", state_name[ctrl->state]);
2999 return sprintf(buf, "unknown state\n");
3002 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3004 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3005 struct device_attribute *attr,
3008 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3010 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3012 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3014 static ssize_t nvme_sysfs_show_address(struct device *dev,
3015 struct device_attribute *attr,
3018 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3020 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3022 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3024 static struct attribute *nvme_dev_attrs[] = {
3025 &dev_attr_reset_controller.attr,
3026 &dev_attr_rescan_controller.attr,
3027 &dev_attr_model.attr,
3028 &dev_attr_serial.attr,
3029 &dev_attr_firmware_rev.attr,
3030 &dev_attr_cntlid.attr,
3031 &dev_attr_delete_controller.attr,
3032 &dev_attr_transport.attr,
3033 &dev_attr_subsysnqn.attr,
3034 &dev_attr_address.attr,
3035 &dev_attr_state.attr,
3036 &dev_attr_numa_node.attr,
3040 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3041 struct attribute *a, int n)
3043 struct device *dev = container_of(kobj, struct device, kobj);
3044 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3046 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3048 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3054 static struct attribute_group nvme_dev_attrs_group = {
3055 .attrs = nvme_dev_attrs,
3056 .is_visible = nvme_dev_attrs_are_visible,
3059 static const struct attribute_group *nvme_dev_attr_groups[] = {
3060 &nvme_dev_attrs_group,
3064 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3067 struct nvme_ns_head *h;
3069 lockdep_assert_held(&subsys->lock);
3071 list_for_each_entry(h, &subsys->nsheads, entry) {
3072 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3079 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3080 struct nvme_ns_head *new)
3082 struct nvme_ns_head *h;
3084 lockdep_assert_held(&subsys->lock);
3086 list_for_each_entry(h, &subsys->nsheads, entry) {
3087 if (nvme_ns_ids_valid(&new->ids) &&
3088 !list_empty(&h->list) &&
3089 nvme_ns_ids_equal(&new->ids, &h->ids))
3096 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3097 unsigned nsid, struct nvme_id_ns *id)
3099 struct nvme_ns_head *head;
3100 size_t size = sizeof(*head);
3103 #ifdef CONFIG_NVME_MULTIPATH
3104 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3107 head = kzalloc(size, GFP_KERNEL);
3110 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3113 head->instance = ret;
3114 INIT_LIST_HEAD(&head->list);
3115 ret = init_srcu_struct(&head->srcu);
3117 goto out_ida_remove;
3118 head->subsys = ctrl->subsys;
3120 kref_init(&head->ref);
3122 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3124 ret = __nvme_check_ids(ctrl->subsys, head);
3126 dev_err(ctrl->device,
3127 "duplicate IDs for nsid %d\n", nsid);
3128 goto out_cleanup_srcu;
3131 ret = nvme_mpath_alloc_disk(ctrl, head);
3133 goto out_cleanup_srcu;
3135 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3137 kref_get(&ctrl->subsys->ref);
3141 cleanup_srcu_struct(&head->srcu);
3143 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3147 return ERR_PTR(ret);
3150 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3151 struct nvme_id_ns *id)
3153 struct nvme_ctrl *ctrl = ns->ctrl;
3154 bool is_shared = id->nmic & (1 << 0);
3155 struct nvme_ns_head *head = NULL;
3158 mutex_lock(&ctrl->subsys->lock);
3160 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3162 head = nvme_alloc_ns_head(ctrl, nsid, id);
3164 ret = PTR_ERR(head);
3168 struct nvme_ns_ids ids;
3170 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3171 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3172 dev_err(ctrl->device,
3173 "IDs don't match for shared namespace %d\n",
3180 list_add_tail(&ns->siblings, &head->list);
3184 mutex_unlock(&ctrl->subsys->lock);
3188 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3190 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3191 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3193 return nsa->head->ns_id - nsb->head->ns_id;
3196 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3198 struct nvme_ns *ns, *ret = NULL;
3200 down_read(&ctrl->namespaces_rwsem);
3201 list_for_each_entry(ns, &ctrl->namespaces, list) {
3202 if (ns->head->ns_id == nsid) {
3203 if (!kref_get_unless_zero(&ns->kref))
3208 if (ns->head->ns_id > nsid)
3211 up_read(&ctrl->namespaces_rwsem);
3215 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3217 struct streams_directive_params s;
3220 if (!ctrl->nr_streams)
3223 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3227 ns->sws = le32_to_cpu(s.sws);
3228 ns->sgs = le16_to_cpu(s.sgs);
3231 unsigned int bs = 1 << ns->lba_shift;
3233 blk_queue_io_min(ns->queue, bs * ns->sws);
3235 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3241 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3244 struct gendisk *disk;
3245 struct nvme_id_ns *id;
3246 char disk_name[DISK_NAME_LEN];
3247 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3249 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3253 ns->queue = blk_mq_init_queue(ctrl->tagset);
3254 if (IS_ERR(ns->queue)) {
3255 ret = PTR_ERR(ns->queue);
3259 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3260 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3261 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3263 ns->queue->queuedata = ns;
3266 kref_init(&ns->kref);
3267 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3269 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3270 nvme_set_queue_limits(ctrl, ns->queue);
3272 id = nvme_identify_ns(ctrl, nsid);
3275 goto out_free_queue;
3278 if (id->ncap == 0) {
3283 ret = nvme_init_ns_head(ns, nsid, id);
3286 nvme_setup_streams_ns(ctrl, ns);
3287 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3289 disk = alloc_disk_node(0, node);
3295 disk->fops = &nvme_fops;
3296 disk->private_data = ns;
3297 disk->queue = ns->queue;
3298 disk->flags = flags;
3299 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3302 __nvme_revalidate_disk(disk, id);
3304 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3305 ret = nvme_nvm_register(ns, disk_name, node);
3307 dev_warn(ctrl->device, "LightNVM init failure\n");
3312 down_write(&ctrl->namespaces_rwsem);
3313 list_add_tail(&ns->list, &ctrl->namespaces);
3314 up_write(&ctrl->namespaces_rwsem);
3316 nvme_get_ctrl(ctrl);
3318 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3320 nvme_mpath_add_disk(ns, id);
3321 nvme_fault_inject_init(ns);
3328 mutex_lock(&ctrl->subsys->lock);
3329 list_del_rcu(&ns->siblings);
3330 mutex_unlock(&ctrl->subsys->lock);
3331 nvme_put_ns_head(ns->head);
3335 blk_cleanup_queue(ns->queue);
3341 static void nvme_ns_remove(struct nvme_ns *ns)
3343 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3346 nvme_fault_inject_fini(ns);
3347 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3348 del_gendisk(ns->disk);
3349 blk_cleanup_queue(ns->queue);
3350 if (blk_get_integrity(ns->disk))
3351 blk_integrity_unregister(ns->disk);
3354 mutex_lock(&ns->ctrl->subsys->lock);
3355 list_del_rcu(&ns->siblings);
3356 nvme_mpath_clear_current_path(ns);
3357 mutex_unlock(&ns->ctrl->subsys->lock);
3359 down_write(&ns->ctrl->namespaces_rwsem);
3360 list_del_init(&ns->list);
3361 up_write(&ns->ctrl->namespaces_rwsem);
3363 synchronize_srcu(&ns->head->srcu);
3364 nvme_mpath_check_last_path(ns);
3368 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3372 ns = nvme_find_get_ns(ctrl, nsid);
3374 if (ns->disk && revalidate_disk(ns->disk))
3378 nvme_alloc_ns(ctrl, nsid);
3381 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3384 struct nvme_ns *ns, *next;
3387 down_write(&ctrl->namespaces_rwsem);
3388 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3389 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3390 list_move_tail(&ns->list, &rm_list);
3392 up_write(&ctrl->namespaces_rwsem);
3394 list_for_each_entry_safe(ns, next, &rm_list, list)
3399 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3403 unsigned i, j, nsid, prev = 0, num_lists = DIV_ROUND_UP(nn, 1024);
3406 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3410 for (i = 0; i < num_lists; i++) {
3411 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3415 for (j = 0; j < min(nn, 1024U); j++) {
3416 nsid = le32_to_cpu(ns_list[j]);
3420 nvme_validate_ns(ctrl, nsid);
3422 while (++prev < nsid) {
3423 ns = nvme_find_get_ns(ctrl, prev);
3433 nvme_remove_invalid_namespaces(ctrl, prev);
3439 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3443 for (i = 1; i <= nn; i++)
3444 nvme_validate_ns(ctrl, i);
3446 nvme_remove_invalid_namespaces(ctrl, nn);
3449 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3451 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3455 log = kzalloc(log_size, GFP_KERNEL);
3460 * We need to read the log to clear the AEN, but we don't want to rely
3461 * on it for the changed namespace information as userspace could have
3462 * raced with us in reading the log page, which could cause us to miss
3465 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3468 dev_warn(ctrl->device,
3469 "reading changed ns log failed: %d\n", error);
3474 static void nvme_scan_work(struct work_struct *work)
3476 struct nvme_ctrl *ctrl =
3477 container_of(work, struct nvme_ctrl, scan_work);
3478 struct nvme_id_ctrl *id;
3481 if (ctrl->state != NVME_CTRL_LIVE)
3484 WARN_ON_ONCE(!ctrl->tagset);
3486 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3487 dev_info(ctrl->device, "rescanning namespaces.\n");
3488 nvme_clear_changed_ns_log(ctrl);
3491 if (nvme_identify_ctrl(ctrl, &id))
3494 mutex_lock(&ctrl->scan_lock);
3495 nn = le32_to_cpu(id->nn);
3496 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3497 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3498 if (!nvme_scan_ns_list(ctrl, nn))
3501 nvme_scan_ns_sequential(ctrl, nn);
3503 mutex_unlock(&ctrl->scan_lock);
3505 down_write(&ctrl->namespaces_rwsem);
3506 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3507 up_write(&ctrl->namespaces_rwsem);
3511 * This function iterates the namespace list unlocked to allow recovery from
3512 * controller failure. It is up to the caller to ensure the namespace list is
3513 * not modified by scan work while this function is executing.
3515 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3517 struct nvme_ns *ns, *next;
3520 /* prevent racing with ns scanning */
3521 flush_work(&ctrl->scan_work);
3524 * The dead states indicates the controller was not gracefully
3525 * disconnected. In that case, we won't be able to flush any data while
3526 * removing the namespaces' disks; fail all the queues now to avoid
3527 * potentially having to clean up the failed sync later.
3529 if (ctrl->state == NVME_CTRL_DEAD)
3530 nvme_kill_queues(ctrl);
3532 down_write(&ctrl->namespaces_rwsem);
3533 list_splice_init(&ctrl->namespaces, &ns_list);
3534 up_write(&ctrl->namespaces_rwsem);
3536 list_for_each_entry_safe(ns, next, &ns_list, list)
3539 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3541 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3543 char *envp[2] = { NULL, NULL };
3544 u32 aen_result = ctrl->aen_result;
3546 ctrl->aen_result = 0;
3550 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3553 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3557 static void nvme_async_event_work(struct work_struct *work)
3559 struct nvme_ctrl *ctrl =
3560 container_of(work, struct nvme_ctrl, async_event_work);
3562 nvme_aen_uevent(ctrl);
3563 ctrl->ops->submit_async_event(ctrl);
3566 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3571 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3577 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3580 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3582 struct nvme_fw_slot_info_log *log;
3584 log = kmalloc(sizeof(*log), GFP_KERNEL);
3588 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3590 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3594 static void nvme_fw_act_work(struct work_struct *work)
3596 struct nvme_ctrl *ctrl = container_of(work,
3597 struct nvme_ctrl, fw_act_work);
3598 unsigned long fw_act_timeout;
3601 fw_act_timeout = jiffies +
3602 msecs_to_jiffies(ctrl->mtfa * 100);
3604 fw_act_timeout = jiffies +
3605 msecs_to_jiffies(admin_timeout * 1000);
3607 nvme_stop_queues(ctrl);
3608 while (nvme_ctrl_pp_status(ctrl)) {
3609 if (time_after(jiffies, fw_act_timeout)) {
3610 dev_warn(ctrl->device,
3611 "Fw activation timeout, reset controller\n");
3612 nvme_reset_ctrl(ctrl);
3618 if (ctrl->state != NVME_CTRL_LIVE)
3621 nvme_start_queues(ctrl);
3622 /* read FW slot information to clear the AER */
3623 nvme_get_fw_slot_info(ctrl);
3626 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3628 u32 aer_notice_type = (result & 0xff00) >> 8;
3630 trace_nvme_async_event(ctrl, aer_notice_type);
3632 switch (aer_notice_type) {
3633 case NVME_AER_NOTICE_NS_CHANGED:
3634 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3635 nvme_queue_scan(ctrl);
3637 case NVME_AER_NOTICE_FW_ACT_STARTING:
3638 queue_work(nvme_wq, &ctrl->fw_act_work);
3640 #ifdef CONFIG_NVME_MULTIPATH
3641 case NVME_AER_NOTICE_ANA:
3642 if (!ctrl->ana_log_buf)
3644 queue_work(nvme_wq, &ctrl->ana_work);
3648 dev_warn(ctrl->device, "async event result %08x\n", result);
3652 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3653 volatile union nvme_result *res)
3655 u32 result = le32_to_cpu(res->u32);
3656 u32 aer_type = result & 0x07;
3658 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3662 case NVME_AER_NOTICE:
3663 nvme_handle_aen_notice(ctrl, result);
3665 case NVME_AER_ERROR:
3666 case NVME_AER_SMART:
3669 trace_nvme_async_event(ctrl, aer_type);
3670 ctrl->aen_result = result;
3675 queue_work(nvme_wq, &ctrl->async_event_work);
3677 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3679 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3681 nvme_mpath_stop(ctrl);
3682 nvme_stop_keep_alive(ctrl);
3683 flush_work(&ctrl->async_event_work);
3684 cancel_work_sync(&ctrl->fw_act_work);
3686 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3688 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3691 nvme_start_keep_alive(ctrl);
3693 if (ctrl->queue_count > 1) {
3694 nvme_queue_scan(ctrl);
3695 nvme_enable_aen(ctrl);
3696 queue_work(nvme_wq, &ctrl->async_event_work);
3697 nvme_start_queues(ctrl);
3700 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3702 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3704 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3705 cdev_device_del(&ctrl->cdev, ctrl->device);
3707 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3709 static void nvme_free_ctrl(struct device *dev)
3711 struct nvme_ctrl *ctrl =
3712 container_of(dev, struct nvme_ctrl, ctrl_device);
3713 struct nvme_subsystem *subsys = ctrl->subsys;
3715 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3716 kfree(ctrl->effects);
3717 nvme_mpath_uninit(ctrl);
3718 __free_page(ctrl->discard_page);
3721 mutex_lock(&nvme_subsystems_lock);
3722 list_del(&ctrl->subsys_entry);
3723 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3724 mutex_unlock(&nvme_subsystems_lock);
3727 ctrl->ops->free_ctrl(ctrl);
3730 nvme_put_subsystem(subsys);
3734 * Initialize a NVMe controller structures. This needs to be called during
3735 * earliest initialization so that we have the initialized structured around
3738 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3739 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3743 ctrl->state = NVME_CTRL_NEW;
3744 spin_lock_init(&ctrl->lock);
3745 mutex_init(&ctrl->scan_lock);
3746 INIT_LIST_HEAD(&ctrl->namespaces);
3747 init_rwsem(&ctrl->namespaces_rwsem);
3750 ctrl->quirks = quirks;
3751 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3752 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3753 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3754 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3756 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3757 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3758 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3760 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3762 ctrl->discard_page = alloc_page(GFP_KERNEL);
3763 if (!ctrl->discard_page) {
3768 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3771 ctrl->instance = ret;
3773 device_initialize(&ctrl->ctrl_device);
3774 ctrl->device = &ctrl->ctrl_device;
3775 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3776 ctrl->device->class = nvme_class;
3777 ctrl->device->parent = ctrl->dev;
3778 ctrl->device->groups = nvme_dev_attr_groups;
3779 ctrl->device->release = nvme_free_ctrl;
3780 dev_set_drvdata(ctrl->device, ctrl);
3781 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3783 goto out_release_instance;
3785 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3786 ctrl->cdev.owner = ops->module;
3787 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3792 * Initialize latency tolerance controls. The sysfs files won't
3793 * be visible to userspace unless the device actually supports APST.
3795 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3796 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3797 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3801 kfree_const(ctrl->device->kobj.name);
3802 out_release_instance:
3803 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3805 if (ctrl->discard_page)
3806 __free_page(ctrl->discard_page);
3809 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3812 * nvme_kill_queues(): Ends all namespace queues
3813 * @ctrl: the dead controller that needs to end
3815 * Call this function when the driver determines it is unable to get the
3816 * controller in a state capable of servicing IO.
3818 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3822 down_read(&ctrl->namespaces_rwsem);
3824 /* Forcibly unquiesce queues to avoid blocking dispatch */
3825 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3826 blk_mq_unquiesce_queue(ctrl->admin_q);
3828 list_for_each_entry(ns, &ctrl->namespaces, list)
3829 nvme_set_queue_dying(ns);
3831 up_read(&ctrl->namespaces_rwsem);
3833 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3835 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3839 down_read(&ctrl->namespaces_rwsem);
3840 list_for_each_entry(ns, &ctrl->namespaces, list)
3841 blk_mq_unfreeze_queue(ns->queue);
3842 up_read(&ctrl->namespaces_rwsem);
3844 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3846 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3850 down_read(&ctrl->namespaces_rwsem);
3851 list_for_each_entry(ns, &ctrl->namespaces, list) {
3852 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3856 up_read(&ctrl->namespaces_rwsem);
3858 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3860 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3864 down_read(&ctrl->namespaces_rwsem);
3865 list_for_each_entry(ns, &ctrl->namespaces, list)
3866 blk_mq_freeze_queue_wait(ns->queue);
3867 up_read(&ctrl->namespaces_rwsem);
3869 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3871 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3875 down_read(&ctrl->namespaces_rwsem);
3876 list_for_each_entry(ns, &ctrl->namespaces, list)
3877 blk_freeze_queue_start(ns->queue);
3878 up_read(&ctrl->namespaces_rwsem);
3880 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3882 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3886 down_read(&ctrl->namespaces_rwsem);
3887 list_for_each_entry(ns, &ctrl->namespaces, list)
3888 blk_mq_quiesce_queue(ns->queue);
3889 up_read(&ctrl->namespaces_rwsem);
3891 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3893 void nvme_start_queues(struct nvme_ctrl *ctrl)
3897 down_read(&ctrl->namespaces_rwsem);
3898 list_for_each_entry(ns, &ctrl->namespaces, list)
3899 blk_mq_unquiesce_queue(ns->queue);
3900 up_read(&ctrl->namespaces_rwsem);
3902 EXPORT_SYMBOL_GPL(nvme_start_queues);
3905 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3909 down_read(&ctrl->namespaces_rwsem);
3910 list_for_each_entry(ns, &ctrl->namespaces, list)
3911 blk_sync_queue(ns->queue);
3912 up_read(&ctrl->namespaces_rwsem);
3914 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3917 * Check we didn't inadvertently grow the command structure sizes:
3919 static inline void _nvme_check_size(void)
3921 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3922 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3923 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3924 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3925 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3926 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3927 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3928 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3929 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3930 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3931 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3932 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3933 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3934 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3935 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3936 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
3937 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
3941 static int __init nvme_core_init(void)
3943 int result = -ENOMEM;
3947 nvme_wq = alloc_workqueue("nvme-wq",
3948 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3952 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
3953 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3957 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
3958 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
3959 if (!nvme_delete_wq)
3960 goto destroy_reset_wq;
3962 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
3964 goto destroy_delete_wq;
3966 nvme_class = class_create(THIS_MODULE, "nvme");
3967 if (IS_ERR(nvme_class)) {
3968 result = PTR_ERR(nvme_class);
3969 goto unregister_chrdev;
3972 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
3973 if (IS_ERR(nvme_subsys_class)) {
3974 result = PTR_ERR(nvme_subsys_class);
3980 class_destroy(nvme_class);
3982 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3984 destroy_workqueue(nvme_delete_wq);
3986 destroy_workqueue(nvme_reset_wq);
3988 destroy_workqueue(nvme_wq);
3993 static void __exit nvme_core_exit(void)
3995 ida_destroy(&nvme_subsystems_ida);
3996 class_destroy(nvme_subsys_class);
3997 class_destroy(nvme_class);
3998 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
3999 destroy_workqueue(nvme_delete_wq);
4000 destroy_workqueue(nvme_reset_wq);
4001 destroy_workqueue(nvme_wq);
4004 MODULE_LICENSE("GPL");
4005 MODULE_VERSION("1.0");
4006 module_init(nvme_core_init);
4007 module_exit(nvme_core_exit);