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/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such are scan, aen handling, fw activation,
70 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
84 static DEFINE_IDA(nvme_subsystems_ida);
85 static LIST_HEAD(nvme_subsystems);
86 static DEFINE_MUTEX(nvme_subsystems_lock);
88 static DEFINE_IDA(nvme_instance_ida);
89 static dev_t nvme_chr_devt;
90 static struct class *nvme_class;
91 static struct class *nvme_subsys_class;
93 static int nvme_revalidate_disk(struct gendisk *disk);
94 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
95 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
98 static void nvme_set_queue_dying(struct nvme_ns *ns)
101 * Revalidating a dead namespace sets capacity to 0. This will end
102 * buffered writers dirtying pages that can't be synced.
104 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
106 revalidate_disk(ns->disk);
107 blk_set_queue_dying(ns->queue);
108 /* Forcibly unquiesce queues to avoid blocking dispatch */
109 blk_mq_unquiesce_queue(ns->queue);
112 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
115 * Only new queue scan work when admin and IO queues are both alive
117 if (ctrl->state == NVME_CTRL_LIVE)
118 queue_work(nvme_wq, &ctrl->scan_work);
121 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
123 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
125 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
129 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
131 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
135 ret = nvme_reset_ctrl(ctrl);
137 flush_work(&ctrl->reset_work);
138 if (ctrl->state != NVME_CTRL_LIVE &&
139 ctrl->state != NVME_CTRL_ADMIN_ONLY)
145 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
147 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
149 dev_info(ctrl->device,
150 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
152 flush_work(&ctrl->reset_work);
153 nvme_stop_ctrl(ctrl);
154 nvme_remove_namespaces(ctrl);
155 ctrl->ops->delete_ctrl(ctrl);
156 nvme_uninit_ctrl(ctrl);
160 static void nvme_delete_ctrl_work(struct work_struct *work)
162 struct nvme_ctrl *ctrl =
163 container_of(work, struct nvme_ctrl, delete_work);
165 nvme_do_delete_ctrl(ctrl);
168 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
170 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
172 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
176 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
178 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
183 * Keep a reference until nvme_do_delete_ctrl() complete,
184 * since ->delete_ctrl can free the controller.
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
190 nvme_do_delete_ctrl(ctrl);
195 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
197 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
200 static blk_status_t nvme_error_status(u16 status)
202 switch (status & 0x7ff) {
203 case NVME_SC_SUCCESS:
205 case NVME_SC_CAP_EXCEEDED:
206 return BLK_STS_NOSPC;
207 case NVME_SC_LBA_RANGE:
208 return BLK_STS_TARGET;
209 case NVME_SC_BAD_ATTRIBUTES:
210 case NVME_SC_ONCS_NOT_SUPPORTED:
211 case NVME_SC_INVALID_OPCODE:
212 case NVME_SC_INVALID_FIELD:
213 case NVME_SC_INVALID_NS:
214 return BLK_STS_NOTSUPP;
215 case NVME_SC_WRITE_FAULT:
216 case NVME_SC_READ_ERROR:
217 case NVME_SC_UNWRITTEN_BLOCK:
218 case NVME_SC_ACCESS_DENIED:
219 case NVME_SC_READ_ONLY:
220 case NVME_SC_COMPARE_FAILED:
221 return BLK_STS_MEDIUM;
222 case NVME_SC_GUARD_CHECK:
223 case NVME_SC_APPTAG_CHECK:
224 case NVME_SC_REFTAG_CHECK:
225 case NVME_SC_INVALID_PI:
226 return BLK_STS_PROTECTION;
227 case NVME_SC_RESERVATION_CONFLICT:
228 return BLK_STS_NEXUS;
229 case NVME_SC_HOST_PATH_ERROR:
230 return BLK_STS_TRANSPORT;
232 return BLK_STS_IOERR;
236 static inline bool nvme_req_needs_retry(struct request *req)
238 if (blk_noretry_request(req))
240 if (nvme_req(req)->status & NVME_SC_DNR)
242 if (nvme_req(req)->retries >= nvme_max_retries)
247 static void nvme_retry_req(struct request *req)
249 struct nvme_ns *ns = req->q->queuedata;
250 unsigned long delay = 0;
253 /* The mask and shift result must be <= 3 */
254 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
256 delay = ns->ctrl->crdt[crd - 1] * 100;
258 nvme_req(req)->retries++;
259 blk_mq_requeue_request(req, false);
260 blk_mq_delay_kick_requeue_list(req->q, delay);
263 void nvme_complete_rq(struct request *req)
265 blk_status_t status = nvme_error_status(nvme_req(req)->status);
267 trace_nvme_complete_rq(req);
269 if (nvme_req(req)->ctrl->kas)
270 nvme_req(req)->ctrl->comp_seen = true;
272 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
273 if ((req->cmd_flags & REQ_NVME_MPATH) &&
274 blk_path_error(status)) {
275 nvme_failover_req(req);
279 if (!blk_queue_dying(req->q)) {
285 nvme_trace_bio_complete(req, status);
286 blk_mq_end_request(req, status);
288 EXPORT_SYMBOL_GPL(nvme_complete_rq);
290 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
292 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
293 "Cancelling I/O %d", req->tag);
295 /* don't abort one completed request */
296 if (blk_mq_request_completed(req))
299 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
300 blk_mq_complete_request(req);
303 EXPORT_SYMBOL_GPL(nvme_cancel_request);
305 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
306 enum nvme_ctrl_state new_state)
308 enum nvme_ctrl_state old_state;
310 bool changed = false;
312 spin_lock_irqsave(&ctrl->lock, flags);
314 old_state = ctrl->state;
316 case NVME_CTRL_ADMIN_ONLY:
318 case NVME_CTRL_CONNECTING:
328 case NVME_CTRL_RESETTING:
329 case NVME_CTRL_CONNECTING:
336 case NVME_CTRL_RESETTING:
340 case NVME_CTRL_ADMIN_ONLY:
347 case NVME_CTRL_CONNECTING:
350 case NVME_CTRL_RESETTING:
357 case NVME_CTRL_DELETING:
360 case NVME_CTRL_ADMIN_ONLY:
361 case NVME_CTRL_RESETTING:
362 case NVME_CTRL_CONNECTING:
371 case NVME_CTRL_DELETING:
383 ctrl->state = new_state;
385 spin_unlock_irqrestore(&ctrl->lock, flags);
386 if (changed && ctrl->state == NVME_CTRL_LIVE)
387 nvme_kick_requeue_lists(ctrl);
390 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
392 static void nvme_free_ns_head(struct kref *ref)
394 struct nvme_ns_head *head =
395 container_of(ref, struct nvme_ns_head, ref);
397 nvme_mpath_remove_disk(head);
398 ida_simple_remove(&head->subsys->ns_ida, head->instance);
399 list_del_init(&head->entry);
400 cleanup_srcu_struct(&head->srcu);
401 nvme_put_subsystem(head->subsys);
405 static void nvme_put_ns_head(struct nvme_ns_head *head)
407 kref_put(&head->ref, nvme_free_ns_head);
410 static void nvme_free_ns(struct kref *kref)
412 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
415 nvme_nvm_unregister(ns);
418 nvme_put_ns_head(ns->head);
419 nvme_put_ctrl(ns->ctrl);
423 static void nvme_put_ns(struct nvme_ns *ns)
425 kref_put(&ns->kref, nvme_free_ns);
428 static inline void nvme_clear_nvme_request(struct request *req)
430 if (!(req->rq_flags & RQF_DONTPREP)) {
431 nvme_req(req)->retries = 0;
432 nvme_req(req)->flags = 0;
433 req->rq_flags |= RQF_DONTPREP;
437 struct request *nvme_alloc_request(struct request_queue *q,
438 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
440 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
443 if (qid == NVME_QID_ANY) {
444 req = blk_mq_alloc_request(q, op, flags);
446 req = blk_mq_alloc_request_hctx(q, op, flags,
452 req->cmd_flags |= REQ_FAILFAST_DRIVER;
453 nvme_clear_nvme_request(req);
454 nvme_req(req)->cmd = cmd;
458 EXPORT_SYMBOL_GPL(nvme_alloc_request);
460 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
462 struct nvme_command c;
464 memset(&c, 0, sizeof(c));
466 c.directive.opcode = nvme_admin_directive_send;
467 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
468 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
469 c.directive.dtype = NVME_DIR_IDENTIFY;
470 c.directive.tdtype = NVME_DIR_STREAMS;
471 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
473 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
476 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
478 return nvme_toggle_streams(ctrl, false);
481 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
483 return nvme_toggle_streams(ctrl, true);
486 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
487 struct streams_directive_params *s, u32 nsid)
489 struct nvme_command c;
491 memset(&c, 0, sizeof(c));
492 memset(s, 0, sizeof(*s));
494 c.directive.opcode = nvme_admin_directive_recv;
495 c.directive.nsid = cpu_to_le32(nsid);
496 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
497 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
498 c.directive.dtype = NVME_DIR_STREAMS;
500 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
503 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
505 struct streams_directive_params s;
508 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
513 ret = nvme_enable_streams(ctrl);
517 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
521 ctrl->nssa = le16_to_cpu(s.nssa);
522 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
523 dev_info(ctrl->device, "too few streams (%u) available\n",
525 nvme_disable_streams(ctrl);
529 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
530 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
535 * Check if 'req' has a write hint associated with it. If it does, assign
536 * a valid namespace stream to the write.
538 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
539 struct request *req, u16 *control,
542 enum rw_hint streamid = req->write_hint;
544 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
548 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
551 *control |= NVME_RW_DTYPE_STREAMS;
552 *dsmgmt |= streamid << 16;
555 if (streamid < ARRAY_SIZE(req->q->write_hints))
556 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
559 static inline void nvme_setup_flush(struct nvme_ns *ns,
560 struct nvme_command *cmnd)
562 cmnd->common.opcode = nvme_cmd_flush;
563 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
566 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
567 struct nvme_command *cmnd)
569 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
570 struct nvme_dsm_range *range;
573 range = kmalloc_array(segments, sizeof(*range),
574 GFP_ATOMIC | __GFP_NOWARN);
577 * If we fail allocation our range, fallback to the controller
578 * discard page. If that's also busy, it's safe to return
579 * busy, as we know we can make progress once that's freed.
581 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
582 return BLK_STS_RESOURCE;
584 range = page_address(ns->ctrl->discard_page);
587 __rq_for_each_bio(bio, req) {
588 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
589 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
592 range[n].cattr = cpu_to_le32(0);
593 range[n].nlb = cpu_to_le32(nlb);
594 range[n].slba = cpu_to_le64(slba);
599 if (WARN_ON_ONCE(n != segments)) {
600 if (virt_to_page(range) == ns->ctrl->discard_page)
601 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
604 return BLK_STS_IOERR;
607 cmnd->dsm.opcode = nvme_cmd_dsm;
608 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
609 cmnd->dsm.nr = cpu_to_le32(segments - 1);
610 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
612 req->special_vec.bv_page = virt_to_page(range);
613 req->special_vec.bv_offset = offset_in_page(range);
614 req->special_vec.bv_len = sizeof(*range) * segments;
615 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
620 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
621 struct request *req, struct nvme_command *cmnd)
623 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
624 return nvme_setup_discard(ns, req, cmnd);
626 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
627 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
628 cmnd->write_zeroes.slba =
629 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
630 cmnd->write_zeroes.length =
631 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
632 cmnd->write_zeroes.control = 0;
636 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
637 struct request *req, struct nvme_command *cmnd)
639 struct nvme_ctrl *ctrl = ns->ctrl;
643 if (req->cmd_flags & REQ_FUA)
644 control |= NVME_RW_FUA;
645 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
646 control |= NVME_RW_LR;
648 if (req->cmd_flags & REQ_RAHEAD)
649 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
651 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
652 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
653 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
654 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
656 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
657 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
661 * If formated with metadata, the block layer always provides a
662 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
663 * we enable the PRACT bit for protection information or set the
664 * namespace capacity to zero to prevent any I/O.
666 if (!blk_integrity_rq(req)) {
667 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
668 return BLK_STS_NOTSUPP;
669 control |= NVME_RW_PRINFO_PRACT;
670 } else if (req_op(req) == REQ_OP_WRITE) {
671 t10_pi_prepare(req, ns->pi_type);
674 switch (ns->pi_type) {
675 case NVME_NS_DPS_PI_TYPE3:
676 control |= NVME_RW_PRINFO_PRCHK_GUARD;
678 case NVME_NS_DPS_PI_TYPE1:
679 case NVME_NS_DPS_PI_TYPE2:
680 control |= NVME_RW_PRINFO_PRCHK_GUARD |
681 NVME_RW_PRINFO_PRCHK_REF;
682 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
687 cmnd->rw.control = cpu_to_le16(control);
688 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
692 void nvme_cleanup_cmd(struct request *req)
694 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
695 nvme_req(req)->status == 0) {
696 struct nvme_ns *ns = req->rq_disk->private_data;
698 t10_pi_complete(req, ns->pi_type,
699 blk_rq_bytes(req) >> ns->lba_shift);
701 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
702 struct nvme_ns *ns = req->rq_disk->private_data;
703 struct page *page = req->special_vec.bv_page;
705 if (page == ns->ctrl->discard_page)
706 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
708 kfree(page_address(page) + req->special_vec.bv_offset);
711 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
713 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
714 struct nvme_command *cmd)
716 blk_status_t ret = BLK_STS_OK;
718 nvme_clear_nvme_request(req);
720 memset(cmd, 0, sizeof(*cmd));
721 switch (req_op(req)) {
724 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
727 nvme_setup_flush(ns, cmd);
729 case REQ_OP_WRITE_ZEROES:
730 ret = nvme_setup_write_zeroes(ns, req, cmd);
733 ret = nvme_setup_discard(ns, req, cmd);
737 ret = nvme_setup_rw(ns, req, cmd);
741 return BLK_STS_IOERR;
744 cmd->common.command_id = req->tag;
745 trace_nvme_setup_cmd(req, cmd);
748 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
750 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
752 struct completion *waiting = rq->end_io_data;
754 rq->end_io_data = NULL;
758 static void nvme_execute_rq_polled(struct request_queue *q,
759 struct gendisk *bd_disk, struct request *rq, int at_head)
761 DECLARE_COMPLETION_ONSTACK(wait);
763 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
765 rq->cmd_flags |= REQ_HIPRI;
766 rq->end_io_data = &wait;
767 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
769 while (!completion_done(&wait)) {
770 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
776 * Returns 0 on success. If the result is negative, it's a Linux error code;
777 * if the result is positive, it's an NVM Express status code
779 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
780 union nvme_result *result, void *buffer, unsigned bufflen,
781 unsigned timeout, int qid, int at_head,
782 blk_mq_req_flags_t flags, bool poll)
787 req = nvme_alloc_request(q, cmd, flags, qid);
791 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
793 if (buffer && bufflen) {
794 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
800 nvme_execute_rq_polled(req->q, NULL, req, at_head);
802 blk_execute_rq(req->q, NULL, req, at_head);
804 *result = nvme_req(req)->result;
805 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
808 ret = nvme_req(req)->status;
810 blk_mq_free_request(req);
813 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
815 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
816 void *buffer, unsigned bufflen)
818 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
819 NVME_QID_ANY, 0, 0, false);
821 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
823 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
824 unsigned len, u32 seed, bool write)
826 struct bio_integrity_payload *bip;
830 buf = kmalloc(len, GFP_KERNEL);
835 if (write && copy_from_user(buf, ubuf, len))
838 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
844 bip->bip_iter.bi_size = len;
845 bip->bip_iter.bi_sector = seed;
846 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
847 offset_in_page(buf));
857 static int nvme_submit_user_cmd(struct request_queue *q,
858 struct nvme_command *cmd, void __user *ubuffer,
859 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
860 u32 meta_seed, u32 *result, unsigned timeout)
862 bool write = nvme_is_write(cmd);
863 struct nvme_ns *ns = q->queuedata;
864 struct gendisk *disk = ns ? ns->disk : NULL;
866 struct bio *bio = NULL;
870 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
874 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
875 nvme_req(req)->flags |= NVME_REQ_USERCMD;
877 if (ubuffer && bufflen) {
878 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
884 if (disk && meta_buffer && meta_len) {
885 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
891 req->cmd_flags |= REQ_INTEGRITY;
895 blk_execute_rq(req->q, disk, req, 0);
896 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
899 ret = nvme_req(req)->status;
901 *result = le32_to_cpu(nvme_req(req)->result.u32);
902 if (meta && !ret && !write) {
903 if (copy_to_user(meta_buffer, meta, meta_len))
909 blk_rq_unmap_user(bio);
911 blk_mq_free_request(req);
915 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
917 struct nvme_ctrl *ctrl = rq->end_io_data;
919 bool startka = false;
921 blk_mq_free_request(rq);
924 dev_err(ctrl->device,
925 "failed nvme_keep_alive_end_io error=%d\n",
930 ctrl->comp_seen = false;
931 spin_lock_irqsave(&ctrl->lock, flags);
932 if (ctrl->state == NVME_CTRL_LIVE ||
933 ctrl->state == NVME_CTRL_CONNECTING)
935 spin_unlock_irqrestore(&ctrl->lock, flags);
937 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
940 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
944 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
949 rq->timeout = ctrl->kato * HZ;
950 rq->end_io_data = ctrl;
952 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
957 static void nvme_keep_alive_work(struct work_struct *work)
959 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
960 struct nvme_ctrl, ka_work);
961 bool comp_seen = ctrl->comp_seen;
963 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
964 dev_dbg(ctrl->device,
965 "reschedule traffic based keep-alive timer\n");
966 ctrl->comp_seen = false;
967 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
971 if (nvme_keep_alive(ctrl)) {
972 /* allocation failure, reset the controller */
973 dev_err(ctrl->device, "keep-alive failed\n");
974 nvme_reset_ctrl(ctrl);
979 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
981 if (unlikely(ctrl->kato == 0))
984 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
987 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
989 if (unlikely(ctrl->kato == 0))
992 cancel_delayed_work_sync(&ctrl->ka_work);
994 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
996 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
998 struct nvme_command c = { };
1001 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1002 c.identify.opcode = nvme_admin_identify;
1003 c.identify.cns = NVME_ID_CNS_CTRL;
1005 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1009 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1010 sizeof(struct nvme_id_ctrl));
1016 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1017 struct nvme_ns_ids *ids)
1019 struct nvme_command c = { };
1025 c.identify.opcode = nvme_admin_identify;
1026 c.identify.nsid = cpu_to_le32(nsid);
1027 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1029 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1033 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1034 NVME_IDENTIFY_DATA_SIZE);
1038 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1039 struct nvme_ns_id_desc *cur = data + pos;
1044 switch (cur->nidt) {
1045 case NVME_NIDT_EUI64:
1046 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1047 dev_warn(ctrl->device,
1048 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1052 len = NVME_NIDT_EUI64_LEN;
1053 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1055 case NVME_NIDT_NGUID:
1056 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1057 dev_warn(ctrl->device,
1058 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1062 len = NVME_NIDT_NGUID_LEN;
1063 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1065 case NVME_NIDT_UUID:
1066 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1067 dev_warn(ctrl->device,
1068 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1072 len = NVME_NIDT_UUID_LEN;
1073 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1076 /* Skip unknown types */
1081 len += sizeof(*cur);
1088 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1090 struct nvme_command c = { };
1092 c.identify.opcode = nvme_admin_identify;
1093 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1094 c.identify.nsid = cpu_to_le32(nsid);
1095 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1096 NVME_IDENTIFY_DATA_SIZE);
1099 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1100 unsigned nsid, struct nvme_id_ns **id)
1102 struct nvme_command c = { };
1105 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1106 c.identify.opcode = nvme_admin_identify;
1107 c.identify.nsid = cpu_to_le32(nsid);
1108 c.identify.cns = NVME_ID_CNS_NS;
1110 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1114 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1116 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1123 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1124 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1126 struct nvme_command c;
1127 union nvme_result res;
1130 memset(&c, 0, sizeof(c));
1131 c.features.opcode = op;
1132 c.features.fid = cpu_to_le32(fid);
1133 c.features.dword11 = cpu_to_le32(dword11);
1135 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1136 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1137 if (ret >= 0 && result)
1138 *result = le32_to_cpu(res.u32);
1142 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1143 unsigned int dword11, void *buffer, size_t buflen,
1146 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1149 EXPORT_SYMBOL_GPL(nvme_set_features);
1151 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1152 unsigned int dword11, void *buffer, size_t buflen,
1155 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1158 EXPORT_SYMBOL_GPL(nvme_get_features);
1160 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1162 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1164 int status, nr_io_queues;
1166 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1172 * Degraded controllers might return an error when setting the queue
1173 * count. We still want to be able to bring them online and offer
1174 * access to the admin queue, as that might be only way to fix them up.
1177 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1180 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1181 *count = min(*count, nr_io_queues);
1186 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1188 #define NVME_AEN_SUPPORTED \
1189 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1191 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1193 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1196 if (!supported_aens)
1199 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1202 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1206 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1208 struct nvme_user_io io;
1209 struct nvme_command c;
1210 unsigned length, meta_len;
1211 void __user *metadata;
1213 if (copy_from_user(&io, uio, sizeof(io)))
1218 switch (io.opcode) {
1219 case nvme_cmd_write:
1221 case nvme_cmd_compare:
1227 length = (io.nblocks + 1) << ns->lba_shift;
1228 meta_len = (io.nblocks + 1) * ns->ms;
1229 metadata = (void __user *)(uintptr_t)io.metadata;
1234 } else if (meta_len) {
1235 if ((io.metadata & 3) || !io.metadata)
1239 memset(&c, 0, sizeof(c));
1240 c.rw.opcode = io.opcode;
1241 c.rw.flags = io.flags;
1242 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1243 c.rw.slba = cpu_to_le64(io.slba);
1244 c.rw.length = cpu_to_le16(io.nblocks);
1245 c.rw.control = cpu_to_le16(io.control);
1246 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1247 c.rw.reftag = cpu_to_le32(io.reftag);
1248 c.rw.apptag = cpu_to_le16(io.apptag);
1249 c.rw.appmask = cpu_to_le16(io.appmask);
1251 return nvme_submit_user_cmd(ns->queue, &c,
1252 (void __user *)(uintptr_t)io.addr, length,
1253 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1256 static u32 nvme_known_admin_effects(u8 opcode)
1259 case nvme_admin_format_nvm:
1260 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1261 NVME_CMD_EFFECTS_CSE_MASK;
1262 case nvme_admin_sanitize_nvm:
1263 return NVME_CMD_EFFECTS_CSE_MASK;
1270 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1277 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1278 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1279 dev_warn(ctrl->device,
1280 "IO command:%02x has unhandled effects:%08x\n",
1286 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1287 effects |= nvme_known_admin_effects(opcode);
1290 * For simplicity, IO to all namespaces is quiesced even if the command
1291 * effects say only one namespace is affected.
1293 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1294 mutex_lock(&ctrl->scan_lock);
1295 nvme_start_freeze(ctrl);
1296 nvme_wait_freeze(ctrl);
1301 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1305 down_read(&ctrl->namespaces_rwsem);
1306 list_for_each_entry(ns, &ctrl->namespaces, list)
1307 if (ns->disk && nvme_revalidate_disk(ns->disk))
1308 nvme_set_queue_dying(ns);
1309 up_read(&ctrl->namespaces_rwsem);
1311 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1314 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1317 * Revalidate LBA changes prior to unfreezing. This is necessary to
1318 * prevent memory corruption if a logical block size was changed by
1321 if (effects & NVME_CMD_EFFECTS_LBCC)
1322 nvme_update_formats(ctrl);
1323 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1324 nvme_unfreeze(ctrl);
1325 mutex_unlock(&ctrl->scan_lock);
1327 if (effects & NVME_CMD_EFFECTS_CCC)
1328 nvme_init_identify(ctrl);
1329 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1330 nvme_queue_scan(ctrl);
1333 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1334 struct nvme_passthru_cmd __user *ucmd)
1336 struct nvme_passthru_cmd cmd;
1337 struct nvme_command c;
1338 unsigned timeout = 0;
1342 if (!capable(CAP_SYS_ADMIN))
1344 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1349 memset(&c, 0, sizeof(c));
1350 c.common.opcode = cmd.opcode;
1351 c.common.flags = cmd.flags;
1352 c.common.nsid = cpu_to_le32(cmd.nsid);
1353 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1354 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1355 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1356 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1357 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1358 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1359 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1360 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1363 timeout = msecs_to_jiffies(cmd.timeout_ms);
1365 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1366 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1367 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1368 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1369 0, &cmd.result, timeout);
1370 nvme_passthru_end(ctrl, effects);
1373 if (put_user(cmd.result, &ucmd->result))
1381 * Issue ioctl requests on the first available path. Note that unlike normal
1382 * block layer requests we will not retry failed request on another controller.
1384 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1385 struct nvme_ns_head **head, int *srcu_idx)
1387 #ifdef CONFIG_NVME_MULTIPATH
1388 if (disk->fops == &nvme_ns_head_ops) {
1391 *head = disk->private_data;
1392 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1393 ns = nvme_find_path(*head);
1395 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1401 return disk->private_data;
1404 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1407 srcu_read_unlock(&head->srcu, idx);
1410 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1411 unsigned int cmd, unsigned long arg)
1413 struct nvme_ns_head *head = NULL;
1414 void __user *argp = (void __user *)arg;
1418 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1420 return -EWOULDBLOCK;
1423 * Handle ioctls that apply to the controller instead of the namespace
1424 * seperately and drop the ns SRCU reference early. This avoids a
1425 * deadlock when deleting namespaces using the passthrough interface.
1427 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1428 struct nvme_ctrl *ctrl = ns->ctrl;
1430 nvme_get_ctrl(ns->ctrl);
1431 nvme_put_ns_from_disk(head, srcu_idx);
1433 if (cmd == NVME_IOCTL_ADMIN_CMD)
1434 ret = nvme_user_cmd(ctrl, NULL, argp);
1436 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1438 nvme_put_ctrl(ctrl);
1444 force_successful_syscall_return();
1445 ret = ns->head->ns_id;
1447 case NVME_IOCTL_IO_CMD:
1448 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1450 case NVME_IOCTL_SUBMIT_IO:
1451 ret = nvme_submit_io(ns, argp);
1455 ret = nvme_nvm_ioctl(ns, cmd, arg);
1460 nvme_put_ns_from_disk(head, srcu_idx);
1464 static int nvme_open(struct block_device *bdev, fmode_t mode)
1466 struct nvme_ns *ns = bdev->bd_disk->private_data;
1468 #ifdef CONFIG_NVME_MULTIPATH
1469 /* should never be called due to GENHD_FL_HIDDEN */
1470 if (WARN_ON_ONCE(ns->head->disk))
1473 if (!kref_get_unless_zero(&ns->kref))
1475 if (!try_module_get(ns->ctrl->ops->module))
1486 static void nvme_release(struct gendisk *disk, fmode_t mode)
1488 struct nvme_ns *ns = disk->private_data;
1490 module_put(ns->ctrl->ops->module);
1494 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1496 /* some standard values */
1497 geo->heads = 1 << 6;
1498 geo->sectors = 1 << 5;
1499 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1503 #ifdef CONFIG_BLK_DEV_INTEGRITY
1504 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1506 struct blk_integrity integrity;
1508 memset(&integrity, 0, sizeof(integrity));
1510 case NVME_NS_DPS_PI_TYPE3:
1511 integrity.profile = &t10_pi_type3_crc;
1512 integrity.tag_size = sizeof(u16) + sizeof(u32);
1513 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1515 case NVME_NS_DPS_PI_TYPE1:
1516 case NVME_NS_DPS_PI_TYPE2:
1517 integrity.profile = &t10_pi_type1_crc;
1518 integrity.tag_size = sizeof(u16);
1519 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1522 integrity.profile = NULL;
1525 integrity.tuple_size = ms;
1526 blk_integrity_register(disk, &integrity);
1527 blk_queue_max_integrity_segments(disk->queue, 1);
1530 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1533 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1535 static void nvme_set_chunk_size(struct nvme_ns *ns)
1537 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1538 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1541 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1543 struct nvme_ctrl *ctrl = ns->ctrl;
1544 struct request_queue *queue = disk->queue;
1545 u32 size = queue_logical_block_size(queue);
1547 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1548 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1552 if (ctrl->nr_streams && ns->sws && ns->sgs)
1553 size *= ns->sws * ns->sgs;
1555 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1556 NVME_DSM_MAX_RANGES);
1558 queue->limits.discard_alignment = 0;
1559 queue->limits.discard_granularity = size;
1561 /* If discard is already enabled, don't reset queue limits */
1562 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1565 blk_queue_max_discard_sectors(queue, UINT_MAX);
1566 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1568 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1569 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1572 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1575 unsigned short bs = 1 << ns->lba_shift;
1577 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1578 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1581 * Even though NVMe spec explicitly states that MDTS is not
1582 * applicable to the write-zeroes:- "The restriction does not apply to
1583 * commands that do not transfer data between the host and the
1584 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1585 * In order to be more cautious use controller's max_hw_sectors value
1586 * to configure the maximum sectors for the write-zeroes which is
1587 * configured based on the controller's MDTS field in the
1588 * nvme_init_identify() if available.
1590 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1591 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1593 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1595 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1598 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1599 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1601 memset(ids, 0, sizeof(*ids));
1603 if (ctrl->vs >= NVME_VS(1, 1, 0))
1604 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1605 if (ctrl->vs >= NVME_VS(1, 2, 0))
1606 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1607 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1608 /* Don't treat error as fatal we potentially
1609 * already have a NGUID or EUI-64
1611 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1612 dev_warn(ctrl->device,
1613 "%s: Identify Descriptors failed\n", __func__);
1617 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1619 return !uuid_is_null(&ids->uuid) ||
1620 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1621 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1624 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1626 return uuid_equal(&a->uuid, &b->uuid) &&
1627 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1628 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1631 static void nvme_update_disk_info(struct gendisk *disk,
1632 struct nvme_ns *ns, struct nvme_id_ns *id)
1634 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1635 unsigned short bs = 1 << ns->lba_shift;
1636 u32 atomic_bs, phys_bs, io_opt;
1638 if (ns->lba_shift > PAGE_SHIFT) {
1639 /* unsupported block size, set capacity to 0 later */
1642 blk_mq_freeze_queue(disk->queue);
1643 blk_integrity_unregister(disk);
1645 if (id->nabo == 0) {
1647 * Bit 1 indicates whether NAWUPF is defined for this namespace
1648 * and whether it should be used instead of AWUPF. If NAWUPF ==
1649 * 0 then AWUPF must be used instead.
1651 if (id->nsfeat & (1 << 1) && id->nawupf)
1652 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1654 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1660 if (id->nsfeat & (1 << 4)) {
1661 /* NPWG = Namespace Preferred Write Granularity */
1662 phys_bs *= 1 + le16_to_cpu(id->npwg);
1663 /* NOWS = Namespace Optimal Write Size */
1664 io_opt *= 1 + le16_to_cpu(id->nows);
1667 blk_queue_logical_block_size(disk->queue, bs);
1669 * Linux filesystems assume writing a single physical block is
1670 * an atomic operation. Hence limit the physical block size to the
1671 * value of the Atomic Write Unit Power Fail parameter.
1673 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1674 blk_queue_io_min(disk->queue, phys_bs);
1675 blk_queue_io_opt(disk->queue, io_opt);
1677 if (ns->ms && !ns->ext &&
1678 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1679 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1680 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1681 ns->lba_shift > PAGE_SHIFT)
1684 set_capacity(disk, capacity);
1686 nvme_config_discard(disk, ns);
1687 nvme_config_write_zeroes(disk, ns);
1689 if (id->nsattr & (1 << 0))
1690 set_disk_ro(disk, true);
1692 set_disk_ro(disk, false);
1694 blk_mq_unfreeze_queue(disk->queue);
1697 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1699 struct nvme_ns *ns = disk->private_data;
1702 * If identify namespace failed, use default 512 byte block size so
1703 * block layer can use before failing read/write for 0 capacity.
1705 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1706 if (ns->lba_shift == 0)
1708 ns->noiob = le16_to_cpu(id->noiob);
1709 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1710 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1711 /* the PI implementation requires metadata equal t10 pi tuple size */
1712 if (ns->ms == sizeof(struct t10_pi_tuple))
1713 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1718 nvme_set_chunk_size(ns);
1719 nvme_update_disk_info(disk, ns, id);
1720 #ifdef CONFIG_NVME_MULTIPATH
1721 if (ns->head->disk) {
1722 nvme_update_disk_info(ns->head->disk, ns, id);
1723 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1728 static int nvme_revalidate_disk(struct gendisk *disk)
1730 struct nvme_ns *ns = disk->private_data;
1731 struct nvme_ctrl *ctrl = ns->ctrl;
1732 struct nvme_id_ns *id;
1733 struct nvme_ns_ids ids;
1736 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1737 set_capacity(disk, 0);
1741 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1745 if (id->ncap == 0) {
1750 __nvme_revalidate_disk(disk, id);
1751 nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1752 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1753 dev_err(ctrl->device,
1754 "identifiers changed for nsid %d\n", ns->head->ns_id);
1762 ret = blk_status_to_errno(nvme_error_status(ret));
1766 static char nvme_pr_type(enum pr_type type)
1769 case PR_WRITE_EXCLUSIVE:
1771 case PR_EXCLUSIVE_ACCESS:
1773 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1775 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1777 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1779 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1786 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1787 u64 key, u64 sa_key, u8 op)
1789 struct nvme_ns_head *head = NULL;
1791 struct nvme_command c;
1793 u8 data[16] = { 0, };
1795 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1797 return -EWOULDBLOCK;
1799 put_unaligned_le64(key, &data[0]);
1800 put_unaligned_le64(sa_key, &data[8]);
1802 memset(&c, 0, sizeof(c));
1803 c.common.opcode = op;
1804 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1805 c.common.cdw10 = cpu_to_le32(cdw10);
1807 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1808 nvme_put_ns_from_disk(head, srcu_idx);
1812 static int nvme_pr_register(struct block_device *bdev, u64 old,
1813 u64 new, unsigned flags)
1817 if (flags & ~PR_FL_IGNORE_KEY)
1820 cdw10 = old ? 2 : 0;
1821 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1822 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1823 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1826 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1827 enum pr_type type, unsigned flags)
1831 if (flags & ~PR_FL_IGNORE_KEY)
1834 cdw10 = nvme_pr_type(type) << 8;
1835 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1836 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1839 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1840 enum pr_type type, bool abort)
1842 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1843 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1846 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1848 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1849 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1852 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1854 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1855 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1858 static const struct pr_ops nvme_pr_ops = {
1859 .pr_register = nvme_pr_register,
1860 .pr_reserve = nvme_pr_reserve,
1861 .pr_release = nvme_pr_release,
1862 .pr_preempt = nvme_pr_preempt,
1863 .pr_clear = nvme_pr_clear,
1866 #ifdef CONFIG_BLK_SED_OPAL
1867 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1870 struct nvme_ctrl *ctrl = data;
1871 struct nvme_command cmd;
1873 memset(&cmd, 0, sizeof(cmd));
1875 cmd.common.opcode = nvme_admin_security_send;
1877 cmd.common.opcode = nvme_admin_security_recv;
1878 cmd.common.nsid = 0;
1879 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1880 cmd.common.cdw11 = cpu_to_le32(len);
1882 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1883 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1885 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1886 #endif /* CONFIG_BLK_SED_OPAL */
1888 static const struct block_device_operations nvme_fops = {
1889 .owner = THIS_MODULE,
1890 .ioctl = nvme_ioctl,
1891 .compat_ioctl = nvme_ioctl,
1893 .release = nvme_release,
1894 .getgeo = nvme_getgeo,
1895 .revalidate_disk= nvme_revalidate_disk,
1896 .pr_ops = &nvme_pr_ops,
1899 #ifdef CONFIG_NVME_MULTIPATH
1900 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1902 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1904 if (!kref_get_unless_zero(&head->ref))
1909 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1911 nvme_put_ns_head(disk->private_data);
1914 const struct block_device_operations nvme_ns_head_ops = {
1915 .owner = THIS_MODULE,
1916 .open = nvme_ns_head_open,
1917 .release = nvme_ns_head_release,
1918 .ioctl = nvme_ioctl,
1919 .compat_ioctl = nvme_ioctl,
1920 .getgeo = nvme_getgeo,
1921 .pr_ops = &nvme_pr_ops,
1923 #endif /* CONFIG_NVME_MULTIPATH */
1925 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1927 unsigned long timeout =
1928 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1929 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1932 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1935 if ((csts & NVME_CSTS_RDY) == bit)
1939 if (fatal_signal_pending(current))
1941 if (time_after(jiffies, timeout)) {
1942 dev_err(ctrl->device,
1943 "Device not ready; aborting %s\n", enabled ?
1944 "initialisation" : "reset");
1953 * If the device has been passed off to us in an enabled state, just clear
1954 * the enabled bit. The spec says we should set the 'shutdown notification
1955 * bits', but doing so may cause the device to complete commands to the
1956 * admin queue ... and we don't know what memory that might be pointing at!
1958 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
1962 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1963 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1965 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1969 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1970 msleep(NVME_QUIRK_DELAY_AMOUNT);
1972 return nvme_wait_ready(ctrl, ctrl->cap, false);
1974 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1976 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
1979 * Default to a 4K page size, with the intention to update this
1980 * path in the future to accomodate architectures with differing
1981 * kernel and IO page sizes.
1983 unsigned dev_page_min, page_shift = 12;
1986 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
1988 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
1991 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
1993 if (page_shift < dev_page_min) {
1994 dev_err(ctrl->device,
1995 "Minimum device page size %u too large for host (%u)\n",
1996 1 << dev_page_min, 1 << page_shift);
2000 ctrl->page_size = 1 << page_shift;
2002 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2003 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2004 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2005 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2006 ctrl->ctrl_config |= NVME_CC_ENABLE;
2008 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2011 return nvme_wait_ready(ctrl, ctrl->cap, true);
2013 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2015 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2017 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2021 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2022 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2024 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2028 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2029 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2033 if (fatal_signal_pending(current))
2035 if (time_after(jiffies, timeout)) {
2036 dev_err(ctrl->device,
2037 "Device shutdown incomplete; abort shutdown\n");
2044 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2046 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2047 struct request_queue *q)
2051 if (ctrl->max_hw_sectors) {
2053 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2055 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2056 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2057 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2059 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2060 is_power_of_2(ctrl->max_hw_sectors))
2061 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2062 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2063 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2065 blk_queue_write_cache(q, vwc, vwc);
2068 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2073 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2076 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2077 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2080 dev_warn_once(ctrl->device,
2081 "could not set timestamp (%d)\n", ret);
2085 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2087 struct nvme_feat_host_behavior *host;
2090 /* Don't bother enabling the feature if retry delay is not reported */
2094 host = kzalloc(sizeof(*host), GFP_KERNEL);
2098 host->acre = NVME_ENABLE_ACRE;
2099 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2100 host, sizeof(*host), NULL);
2105 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2108 * APST (Autonomous Power State Transition) lets us program a
2109 * table of power state transitions that the controller will
2110 * perform automatically. We configure it with a simple
2111 * heuristic: we are willing to spend at most 2% of the time
2112 * transitioning between power states. Therefore, when running
2113 * in any given state, we will enter the next lower-power
2114 * non-operational state after waiting 50 * (enlat + exlat)
2115 * microseconds, as long as that state's exit latency is under
2116 * the requested maximum latency.
2118 * We will not autonomously enter any non-operational state for
2119 * which the total latency exceeds ps_max_latency_us. Users
2120 * can set ps_max_latency_us to zero to turn off APST.
2124 struct nvme_feat_auto_pst *table;
2130 * If APST isn't supported or if we haven't been initialized yet,
2131 * then don't do anything.
2136 if (ctrl->npss > 31) {
2137 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2141 table = kzalloc(sizeof(*table), GFP_KERNEL);
2145 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2146 /* Turn off APST. */
2148 dev_dbg(ctrl->device, "APST disabled\n");
2150 __le64 target = cpu_to_le64(0);
2154 * Walk through all states from lowest- to highest-power.
2155 * According to the spec, lower-numbered states use more
2156 * power. NPSS, despite the name, is the index of the
2157 * lowest-power state, not the number of states.
2159 for (state = (int)ctrl->npss; state >= 0; state--) {
2160 u64 total_latency_us, exit_latency_us, transition_ms;
2163 table->entries[state] = target;
2166 * Don't allow transitions to the deepest state
2167 * if it's quirked off.
2169 if (state == ctrl->npss &&
2170 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2174 * Is this state a useful non-operational state for
2175 * higher-power states to autonomously transition to?
2177 if (!(ctrl->psd[state].flags &
2178 NVME_PS_FLAGS_NON_OP_STATE))
2182 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2183 if (exit_latency_us > ctrl->ps_max_latency_us)
2188 le32_to_cpu(ctrl->psd[state].entry_lat);
2191 * This state is good. Use it as the APST idle
2192 * target for higher power states.
2194 transition_ms = total_latency_us + 19;
2195 do_div(transition_ms, 20);
2196 if (transition_ms > (1 << 24) - 1)
2197 transition_ms = (1 << 24) - 1;
2199 target = cpu_to_le64((state << 3) |
2200 (transition_ms << 8));
2205 if (total_latency_us > max_lat_us)
2206 max_lat_us = total_latency_us;
2212 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2214 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2215 max_ps, max_lat_us, (int)sizeof(*table), table);
2219 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2220 table, sizeof(*table), NULL);
2222 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2228 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2230 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2234 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2235 case PM_QOS_LATENCY_ANY:
2243 if (ctrl->ps_max_latency_us != latency) {
2244 ctrl->ps_max_latency_us = latency;
2245 nvme_configure_apst(ctrl);
2249 struct nvme_core_quirk_entry {
2251 * NVMe model and firmware strings are padded with spaces. For
2252 * simplicity, strings in the quirk table are padded with NULLs
2258 unsigned long quirks;
2261 static const struct nvme_core_quirk_entry core_quirks[] = {
2264 * This Toshiba device seems to die using any APST states. See:
2265 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2268 .mn = "THNSF5256GPUK TOSHIBA",
2269 .quirks = NVME_QUIRK_NO_APST,
2273 /* match is null-terminated but idstr is space-padded. */
2274 static bool string_matches(const char *idstr, const char *match, size_t len)
2281 matchlen = strlen(match);
2282 WARN_ON_ONCE(matchlen > len);
2284 if (memcmp(idstr, match, matchlen))
2287 for (; matchlen < len; matchlen++)
2288 if (idstr[matchlen] != ' ')
2294 static bool quirk_matches(const struct nvme_id_ctrl *id,
2295 const struct nvme_core_quirk_entry *q)
2297 return q->vid == le16_to_cpu(id->vid) &&
2298 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2299 string_matches(id->fr, q->fr, sizeof(id->fr));
2302 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2303 struct nvme_id_ctrl *id)
2308 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2309 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2310 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2311 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2315 if (ctrl->vs >= NVME_VS(1, 2, 1))
2316 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2319 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2320 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2321 "nqn.2014.08.org.nvmexpress:%04x%04x",
2322 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2323 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2324 off += sizeof(id->sn);
2325 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2326 off += sizeof(id->mn);
2327 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2330 static void nvme_release_subsystem(struct device *dev)
2332 struct nvme_subsystem *subsys =
2333 container_of(dev, struct nvme_subsystem, dev);
2335 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2339 static void nvme_destroy_subsystem(struct kref *ref)
2341 struct nvme_subsystem *subsys =
2342 container_of(ref, struct nvme_subsystem, ref);
2344 mutex_lock(&nvme_subsystems_lock);
2345 list_del(&subsys->entry);
2346 mutex_unlock(&nvme_subsystems_lock);
2348 ida_destroy(&subsys->ns_ida);
2349 device_del(&subsys->dev);
2350 put_device(&subsys->dev);
2353 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2355 kref_put(&subsys->ref, nvme_destroy_subsystem);
2358 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2360 struct nvme_subsystem *subsys;
2362 lockdep_assert_held(&nvme_subsystems_lock);
2364 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2365 if (strcmp(subsys->subnqn, subsysnqn))
2367 if (!kref_get_unless_zero(&subsys->ref))
2375 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2376 struct device_attribute subsys_attr_##_name = \
2377 __ATTR(_name, _mode, _show, NULL)
2379 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2380 struct device_attribute *attr,
2383 struct nvme_subsystem *subsys =
2384 container_of(dev, struct nvme_subsystem, dev);
2386 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2388 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2390 #define nvme_subsys_show_str_function(field) \
2391 static ssize_t subsys_##field##_show(struct device *dev, \
2392 struct device_attribute *attr, char *buf) \
2394 struct nvme_subsystem *subsys = \
2395 container_of(dev, struct nvme_subsystem, dev); \
2396 return sprintf(buf, "%.*s\n", \
2397 (int)sizeof(subsys->field), subsys->field); \
2399 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2401 nvme_subsys_show_str_function(model);
2402 nvme_subsys_show_str_function(serial);
2403 nvme_subsys_show_str_function(firmware_rev);
2405 static struct attribute *nvme_subsys_attrs[] = {
2406 &subsys_attr_model.attr,
2407 &subsys_attr_serial.attr,
2408 &subsys_attr_firmware_rev.attr,
2409 &subsys_attr_subsysnqn.attr,
2410 #ifdef CONFIG_NVME_MULTIPATH
2411 &subsys_attr_iopolicy.attr,
2416 static struct attribute_group nvme_subsys_attrs_group = {
2417 .attrs = nvme_subsys_attrs,
2420 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2421 &nvme_subsys_attrs_group,
2425 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2426 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2428 struct nvme_ctrl *tmp;
2430 lockdep_assert_held(&nvme_subsystems_lock);
2432 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2433 if (tmp->state == NVME_CTRL_DELETING ||
2434 tmp->state == NVME_CTRL_DEAD)
2437 if (tmp->cntlid == ctrl->cntlid) {
2438 dev_err(ctrl->device,
2439 "Duplicate cntlid %u with %s, rejecting\n",
2440 ctrl->cntlid, dev_name(tmp->device));
2444 if ((id->cmic & (1 << 1)) ||
2445 (ctrl->opts && ctrl->opts->discovery_nqn))
2448 dev_err(ctrl->device,
2449 "Subsystem does not support multiple controllers\n");
2456 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2458 struct nvme_subsystem *subsys, *found;
2461 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2464 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2469 subsys->instance = ret;
2470 mutex_init(&subsys->lock);
2471 kref_init(&subsys->ref);
2472 INIT_LIST_HEAD(&subsys->ctrls);
2473 INIT_LIST_HEAD(&subsys->nsheads);
2474 nvme_init_subnqn(subsys, ctrl, id);
2475 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2476 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2477 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2478 subsys->vendor_id = le16_to_cpu(id->vid);
2479 subsys->cmic = id->cmic;
2480 subsys->awupf = le16_to_cpu(id->awupf);
2481 #ifdef CONFIG_NVME_MULTIPATH
2482 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2485 subsys->dev.class = nvme_subsys_class;
2486 subsys->dev.release = nvme_release_subsystem;
2487 subsys->dev.groups = nvme_subsys_attrs_groups;
2488 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2489 device_initialize(&subsys->dev);
2491 mutex_lock(&nvme_subsystems_lock);
2492 found = __nvme_find_get_subsystem(subsys->subnqn);
2494 put_device(&subsys->dev);
2497 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2499 goto out_put_subsystem;
2502 ret = device_add(&subsys->dev);
2504 dev_err(ctrl->device,
2505 "failed to register subsystem device.\n");
2508 ida_init(&subsys->ns_ida);
2509 list_add_tail(&subsys->entry, &nvme_subsystems);
2512 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2513 dev_name(ctrl->device))) {
2514 dev_err(ctrl->device,
2515 "failed to create sysfs link from subsystem.\n");
2516 goto out_put_subsystem;
2519 ctrl->subsys = subsys;
2520 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2521 mutex_unlock(&nvme_subsystems_lock);
2525 nvme_put_subsystem(subsys);
2527 mutex_unlock(&nvme_subsystems_lock);
2528 put_device(&subsys->dev);
2532 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2533 void *log, size_t size, u64 offset)
2535 struct nvme_command c = { };
2536 unsigned long dwlen = size / 4 - 1;
2538 c.get_log_page.opcode = nvme_admin_get_log_page;
2539 c.get_log_page.nsid = cpu_to_le32(nsid);
2540 c.get_log_page.lid = log_page;
2541 c.get_log_page.lsp = lsp;
2542 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2543 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2544 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2545 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2547 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2550 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2555 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2560 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2561 ctrl->effects, sizeof(*ctrl->effects), 0);
2563 kfree(ctrl->effects);
2564 ctrl->effects = NULL;
2570 * Initialize the cached copies of the Identify data and various controller
2571 * register in our nvme_ctrl structure. This should be called as soon as
2572 * the admin queue is fully up and running.
2574 int nvme_init_identify(struct nvme_ctrl *ctrl)
2576 struct nvme_id_ctrl *id;
2577 int ret, page_shift;
2579 bool prev_apst_enabled;
2581 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2583 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2586 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2587 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2589 if (ctrl->vs >= NVME_VS(1, 1, 0))
2590 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2592 ret = nvme_identify_ctrl(ctrl, &id);
2594 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2598 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2599 ret = nvme_get_effects_log(ctrl);
2604 if (!ctrl->identified) {
2607 ret = nvme_init_subsystem(ctrl, id);
2612 * Check for quirks. Quirk can depend on firmware version,
2613 * so, in principle, the set of quirks present can change
2614 * across a reset. As a possible future enhancement, we
2615 * could re-scan for quirks every time we reinitialize
2616 * the device, but we'd have to make sure that the driver
2617 * behaves intelligently if the quirks change.
2619 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2620 if (quirk_matches(id, &core_quirks[i]))
2621 ctrl->quirks |= core_quirks[i].quirks;
2625 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2626 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2627 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2630 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2631 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2632 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2634 ctrl->oacs = le16_to_cpu(id->oacs);
2635 ctrl->oncs = le16_to_cpu(id->oncs);
2636 ctrl->mtfa = le16_to_cpu(id->mtfa);
2637 ctrl->oaes = le32_to_cpu(id->oaes);
2638 atomic_set(&ctrl->abort_limit, id->acl + 1);
2639 ctrl->vwc = id->vwc;
2641 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2643 max_hw_sectors = UINT_MAX;
2644 ctrl->max_hw_sectors =
2645 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2647 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2648 ctrl->sgls = le32_to_cpu(id->sgls);
2649 ctrl->kas = le16_to_cpu(id->kas);
2650 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2651 ctrl->ctratt = le32_to_cpu(id->ctratt);
2655 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2657 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2658 shutdown_timeout, 60);
2660 if (ctrl->shutdown_timeout != shutdown_timeout)
2661 dev_info(ctrl->device,
2662 "Shutdown timeout set to %u seconds\n",
2663 ctrl->shutdown_timeout);
2665 ctrl->shutdown_timeout = shutdown_timeout;
2667 ctrl->npss = id->npss;
2668 ctrl->apsta = id->apsta;
2669 prev_apst_enabled = ctrl->apst_enabled;
2670 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2671 if (force_apst && id->apsta) {
2672 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2673 ctrl->apst_enabled = true;
2675 ctrl->apst_enabled = false;
2678 ctrl->apst_enabled = id->apsta;
2680 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2682 if (ctrl->ops->flags & NVME_F_FABRICS) {
2683 ctrl->icdoff = le16_to_cpu(id->icdoff);
2684 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2685 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2686 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2689 * In fabrics we need to verify the cntlid matches the
2692 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2697 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2698 dev_err(ctrl->device,
2699 "keep-alive support is mandatory for fabrics\n");
2704 ctrl->cntlid = le16_to_cpu(id->cntlid);
2705 ctrl->hmpre = le32_to_cpu(id->hmpre);
2706 ctrl->hmmin = le32_to_cpu(id->hmmin);
2707 ctrl->hmminds = le32_to_cpu(id->hmminds);
2708 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2711 ret = nvme_mpath_init(ctrl, id);
2717 if (ctrl->apst_enabled && !prev_apst_enabled)
2718 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2719 else if (!ctrl->apst_enabled && prev_apst_enabled)
2720 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2722 ret = nvme_configure_apst(ctrl);
2726 ret = nvme_configure_timestamp(ctrl);
2730 ret = nvme_configure_directives(ctrl);
2734 ret = nvme_configure_acre(ctrl);
2738 ctrl->identified = true;
2746 EXPORT_SYMBOL_GPL(nvme_init_identify);
2748 static int nvme_dev_open(struct inode *inode, struct file *file)
2750 struct nvme_ctrl *ctrl =
2751 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2753 switch (ctrl->state) {
2754 case NVME_CTRL_LIVE:
2755 case NVME_CTRL_ADMIN_ONLY:
2758 return -EWOULDBLOCK;
2761 file->private_data = ctrl;
2765 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2770 down_read(&ctrl->namespaces_rwsem);
2771 if (list_empty(&ctrl->namespaces)) {
2776 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2777 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2778 dev_warn(ctrl->device,
2779 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2784 dev_warn(ctrl->device,
2785 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2786 kref_get(&ns->kref);
2787 up_read(&ctrl->namespaces_rwsem);
2789 ret = nvme_user_cmd(ctrl, ns, argp);
2794 up_read(&ctrl->namespaces_rwsem);
2798 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2801 struct nvme_ctrl *ctrl = file->private_data;
2802 void __user *argp = (void __user *)arg;
2805 case NVME_IOCTL_ADMIN_CMD:
2806 return nvme_user_cmd(ctrl, NULL, argp);
2807 case NVME_IOCTL_IO_CMD:
2808 return nvme_dev_user_cmd(ctrl, argp);
2809 case NVME_IOCTL_RESET:
2810 dev_warn(ctrl->device, "resetting controller\n");
2811 return nvme_reset_ctrl_sync(ctrl);
2812 case NVME_IOCTL_SUBSYS_RESET:
2813 return nvme_reset_subsystem(ctrl);
2814 case NVME_IOCTL_RESCAN:
2815 nvme_queue_scan(ctrl);
2822 static const struct file_operations nvme_dev_fops = {
2823 .owner = THIS_MODULE,
2824 .open = nvme_dev_open,
2825 .unlocked_ioctl = nvme_dev_ioctl,
2826 .compat_ioctl = nvme_dev_ioctl,
2829 static ssize_t nvme_sysfs_reset(struct device *dev,
2830 struct device_attribute *attr, const char *buf,
2833 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2836 ret = nvme_reset_ctrl_sync(ctrl);
2841 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2843 static ssize_t nvme_sysfs_rescan(struct device *dev,
2844 struct device_attribute *attr, const char *buf,
2847 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2849 nvme_queue_scan(ctrl);
2852 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2854 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2856 struct gendisk *disk = dev_to_disk(dev);
2858 if (disk->fops == &nvme_fops)
2859 return nvme_get_ns_from_dev(dev)->head;
2861 return disk->private_data;
2864 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2867 struct nvme_ns_head *head = dev_to_ns_head(dev);
2868 struct nvme_ns_ids *ids = &head->ids;
2869 struct nvme_subsystem *subsys = head->subsys;
2870 int serial_len = sizeof(subsys->serial);
2871 int model_len = sizeof(subsys->model);
2873 if (!uuid_is_null(&ids->uuid))
2874 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2876 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2877 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2879 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2880 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2882 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2883 subsys->serial[serial_len - 1] == '\0'))
2885 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2886 subsys->model[model_len - 1] == '\0'))
2889 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2890 serial_len, subsys->serial, model_len, subsys->model,
2893 static DEVICE_ATTR_RO(wwid);
2895 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2898 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2900 static DEVICE_ATTR_RO(nguid);
2902 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2905 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2907 /* For backward compatibility expose the NGUID to userspace if
2908 * we have no UUID set
2910 if (uuid_is_null(&ids->uuid)) {
2911 printk_ratelimited(KERN_WARNING
2912 "No UUID available providing old NGUID\n");
2913 return sprintf(buf, "%pU\n", ids->nguid);
2915 return sprintf(buf, "%pU\n", &ids->uuid);
2917 static DEVICE_ATTR_RO(uuid);
2919 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2922 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2924 static DEVICE_ATTR_RO(eui);
2926 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2929 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2931 static DEVICE_ATTR_RO(nsid);
2933 static struct attribute *nvme_ns_id_attrs[] = {
2934 &dev_attr_wwid.attr,
2935 &dev_attr_uuid.attr,
2936 &dev_attr_nguid.attr,
2938 &dev_attr_nsid.attr,
2939 #ifdef CONFIG_NVME_MULTIPATH
2940 &dev_attr_ana_grpid.attr,
2941 &dev_attr_ana_state.attr,
2946 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2947 struct attribute *a, int n)
2949 struct device *dev = container_of(kobj, struct device, kobj);
2950 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2952 if (a == &dev_attr_uuid.attr) {
2953 if (uuid_is_null(&ids->uuid) &&
2954 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2957 if (a == &dev_attr_nguid.attr) {
2958 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2961 if (a == &dev_attr_eui.attr) {
2962 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2965 #ifdef CONFIG_NVME_MULTIPATH
2966 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2967 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2969 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2976 static const struct attribute_group nvme_ns_id_attr_group = {
2977 .attrs = nvme_ns_id_attrs,
2978 .is_visible = nvme_ns_id_attrs_are_visible,
2981 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2982 &nvme_ns_id_attr_group,
2984 &nvme_nvm_attr_group,
2989 #define nvme_show_str_function(field) \
2990 static ssize_t field##_show(struct device *dev, \
2991 struct device_attribute *attr, char *buf) \
2993 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2994 return sprintf(buf, "%.*s\n", \
2995 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2997 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2999 nvme_show_str_function(model);
3000 nvme_show_str_function(serial);
3001 nvme_show_str_function(firmware_rev);
3003 #define nvme_show_int_function(field) \
3004 static ssize_t field##_show(struct device *dev, \
3005 struct device_attribute *attr, char *buf) \
3007 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3008 return sprintf(buf, "%d\n", ctrl->field); \
3010 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3012 nvme_show_int_function(cntlid);
3013 nvme_show_int_function(numa_node);
3015 static ssize_t nvme_sysfs_delete(struct device *dev,
3016 struct device_attribute *attr, const char *buf,
3019 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3021 if (device_remove_file_self(dev, attr))
3022 nvme_delete_ctrl_sync(ctrl);
3025 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3027 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3028 struct device_attribute *attr,
3031 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3033 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3035 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3037 static ssize_t nvme_sysfs_show_state(struct device *dev,
3038 struct device_attribute *attr,
3041 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3042 static const char *const state_name[] = {
3043 [NVME_CTRL_NEW] = "new",
3044 [NVME_CTRL_LIVE] = "live",
3045 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3046 [NVME_CTRL_RESETTING] = "resetting",
3047 [NVME_CTRL_CONNECTING] = "connecting",
3048 [NVME_CTRL_DELETING] = "deleting",
3049 [NVME_CTRL_DEAD] = "dead",
3052 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3053 state_name[ctrl->state])
3054 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3056 return sprintf(buf, "unknown state\n");
3059 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3061 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3062 struct device_attribute *attr,
3065 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3067 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3069 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3071 static ssize_t nvme_sysfs_show_address(struct device *dev,
3072 struct device_attribute *attr,
3075 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3077 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3079 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3081 static struct attribute *nvme_dev_attrs[] = {
3082 &dev_attr_reset_controller.attr,
3083 &dev_attr_rescan_controller.attr,
3084 &dev_attr_model.attr,
3085 &dev_attr_serial.attr,
3086 &dev_attr_firmware_rev.attr,
3087 &dev_attr_cntlid.attr,
3088 &dev_attr_delete_controller.attr,
3089 &dev_attr_transport.attr,
3090 &dev_attr_subsysnqn.attr,
3091 &dev_attr_address.attr,
3092 &dev_attr_state.attr,
3093 &dev_attr_numa_node.attr,
3097 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3098 struct attribute *a, int n)
3100 struct device *dev = container_of(kobj, struct device, kobj);
3101 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3103 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3105 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3111 static struct attribute_group nvme_dev_attrs_group = {
3112 .attrs = nvme_dev_attrs,
3113 .is_visible = nvme_dev_attrs_are_visible,
3116 static const struct attribute_group *nvme_dev_attr_groups[] = {
3117 &nvme_dev_attrs_group,
3121 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3124 struct nvme_ns_head *h;
3126 lockdep_assert_held(&subsys->lock);
3128 list_for_each_entry(h, &subsys->nsheads, entry) {
3129 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3136 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3137 struct nvme_ns_head *new)
3139 struct nvme_ns_head *h;
3141 lockdep_assert_held(&subsys->lock);
3143 list_for_each_entry(h, &subsys->nsheads, entry) {
3144 if (nvme_ns_ids_valid(&new->ids) &&
3145 !list_empty(&h->list) &&
3146 nvme_ns_ids_equal(&new->ids, &h->ids))
3153 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3154 unsigned nsid, struct nvme_id_ns *id)
3156 struct nvme_ns_head *head;
3157 size_t size = sizeof(*head);
3160 #ifdef CONFIG_NVME_MULTIPATH
3161 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3164 head = kzalloc(size, GFP_KERNEL);
3167 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3170 head->instance = ret;
3171 INIT_LIST_HEAD(&head->list);
3172 ret = init_srcu_struct(&head->srcu);
3174 goto out_ida_remove;
3175 head->subsys = ctrl->subsys;
3177 kref_init(&head->ref);
3179 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3181 ret = __nvme_check_ids(ctrl->subsys, head);
3183 dev_err(ctrl->device,
3184 "duplicate IDs for nsid %d\n", nsid);
3185 goto out_cleanup_srcu;
3188 ret = nvme_mpath_alloc_disk(ctrl, head);
3190 goto out_cleanup_srcu;
3192 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3194 kref_get(&ctrl->subsys->ref);
3198 cleanup_srcu_struct(&head->srcu);
3200 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3204 return ERR_PTR(ret);
3207 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3208 struct nvme_id_ns *id)
3210 struct nvme_ctrl *ctrl = ns->ctrl;
3211 bool is_shared = id->nmic & (1 << 0);
3212 struct nvme_ns_head *head = NULL;
3215 mutex_lock(&ctrl->subsys->lock);
3217 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3219 head = nvme_alloc_ns_head(ctrl, nsid, id);
3221 ret = PTR_ERR(head);
3225 struct nvme_ns_ids ids;
3227 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3228 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3229 dev_err(ctrl->device,
3230 "IDs don't match for shared namespace %d\n",
3237 list_add_tail(&ns->siblings, &head->list);
3241 mutex_unlock(&ctrl->subsys->lock);
3245 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3247 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3248 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3250 return nsa->head->ns_id - nsb->head->ns_id;
3253 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3255 struct nvme_ns *ns, *ret = NULL;
3257 down_read(&ctrl->namespaces_rwsem);
3258 list_for_each_entry(ns, &ctrl->namespaces, list) {
3259 if (ns->head->ns_id == nsid) {
3260 if (!kref_get_unless_zero(&ns->kref))
3265 if (ns->head->ns_id > nsid)
3268 up_read(&ctrl->namespaces_rwsem);
3272 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3274 struct streams_directive_params s;
3277 if (!ctrl->nr_streams)
3280 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3284 ns->sws = le32_to_cpu(s.sws);
3285 ns->sgs = le16_to_cpu(s.sgs);
3288 unsigned int bs = 1 << ns->lba_shift;
3290 blk_queue_io_min(ns->queue, bs * ns->sws);
3292 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3298 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3301 struct gendisk *disk;
3302 struct nvme_id_ns *id;
3303 char disk_name[DISK_NAME_LEN];
3304 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3306 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3310 ns->queue = blk_mq_init_queue(ctrl->tagset);
3311 if (IS_ERR(ns->queue)) {
3312 ret = PTR_ERR(ns->queue);
3316 if (ctrl->opts && ctrl->opts->data_digest)
3317 ns->queue->backing_dev_info->capabilities
3318 |= BDI_CAP_STABLE_WRITES;
3320 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3321 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3322 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3324 ns->queue->queuedata = ns;
3327 kref_init(&ns->kref);
3328 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3330 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3331 nvme_set_queue_limits(ctrl, ns->queue);
3333 ret = nvme_identify_ns(ctrl, nsid, &id);
3335 goto out_free_queue;
3337 if (id->ncap == 0) {
3342 ret = nvme_init_ns_head(ns, nsid, id);
3345 nvme_setup_streams_ns(ctrl, ns);
3346 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3348 disk = alloc_disk_node(0, node);
3354 disk->fops = &nvme_fops;
3355 disk->private_data = ns;
3356 disk->queue = ns->queue;
3357 disk->flags = flags;
3358 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3361 __nvme_revalidate_disk(disk, id);
3363 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3364 ret = nvme_nvm_register(ns, disk_name, node);
3366 dev_warn(ctrl->device, "LightNVM init failure\n");
3371 down_write(&ctrl->namespaces_rwsem);
3372 list_add_tail(&ns->list, &ctrl->namespaces);
3373 up_write(&ctrl->namespaces_rwsem);
3375 nvme_get_ctrl(ctrl);
3377 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3379 nvme_mpath_add_disk(ns, id);
3380 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3387 mutex_lock(&ctrl->subsys->lock);
3388 list_del_rcu(&ns->siblings);
3389 mutex_unlock(&ctrl->subsys->lock);
3390 nvme_put_ns_head(ns->head);
3394 blk_cleanup_queue(ns->queue);
3398 ret = blk_status_to_errno(nvme_error_status(ret));
3402 static void nvme_ns_remove(struct nvme_ns *ns)
3404 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3407 nvme_fault_inject_fini(&ns->fault_inject);
3409 mutex_lock(&ns->ctrl->subsys->lock);
3410 list_del_rcu(&ns->siblings);
3411 mutex_unlock(&ns->ctrl->subsys->lock);
3412 synchronize_rcu(); /* guarantee not available in head->list */
3413 nvme_mpath_clear_current_path(ns);
3414 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3416 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3417 del_gendisk(ns->disk);
3418 blk_cleanup_queue(ns->queue);
3419 if (blk_get_integrity(ns->disk))
3420 blk_integrity_unregister(ns->disk);
3423 down_write(&ns->ctrl->namespaces_rwsem);
3424 list_del_init(&ns->list);
3425 up_write(&ns->ctrl->namespaces_rwsem);
3427 nvme_mpath_check_last_path(ns);
3431 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3435 ns = nvme_find_get_ns(ctrl, nsid);
3437 if (ns->disk && revalidate_disk(ns->disk))
3441 nvme_alloc_ns(ctrl, nsid);
3444 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3447 struct nvme_ns *ns, *next;
3450 down_write(&ctrl->namespaces_rwsem);
3451 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3452 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3453 list_move_tail(&ns->list, &rm_list);
3455 up_write(&ctrl->namespaces_rwsem);
3457 list_for_each_entry_safe(ns, next, &rm_list, list)
3462 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3466 unsigned i, j, nsid, prev = 0;
3467 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3470 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3474 for (i = 0; i < num_lists; i++) {
3475 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3479 for (j = 0; j < min(nn, 1024U); j++) {
3480 nsid = le32_to_cpu(ns_list[j]);
3484 nvme_validate_ns(ctrl, nsid);
3486 while (++prev < nsid) {
3487 ns = nvme_find_get_ns(ctrl, prev);
3497 nvme_remove_invalid_namespaces(ctrl, prev);
3503 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3507 for (i = 1; i <= nn; i++)
3508 nvme_validate_ns(ctrl, i);
3510 nvme_remove_invalid_namespaces(ctrl, nn);
3513 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3515 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3519 log = kzalloc(log_size, GFP_KERNEL);
3524 * We need to read the log to clear the AEN, but we don't want to rely
3525 * on it for the changed namespace information as userspace could have
3526 * raced with us in reading the log page, which could cause us to miss
3529 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3532 dev_warn(ctrl->device,
3533 "reading changed ns log failed: %d\n", error);
3538 static void nvme_scan_work(struct work_struct *work)
3540 struct nvme_ctrl *ctrl =
3541 container_of(work, struct nvme_ctrl, scan_work);
3542 struct nvme_id_ctrl *id;
3545 if (ctrl->state != NVME_CTRL_LIVE)
3548 WARN_ON_ONCE(!ctrl->tagset);
3550 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3551 dev_info(ctrl->device, "rescanning namespaces.\n");
3552 nvme_clear_changed_ns_log(ctrl);
3555 if (nvme_identify_ctrl(ctrl, &id))
3558 mutex_lock(&ctrl->scan_lock);
3559 nn = le32_to_cpu(id->nn);
3560 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3561 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3562 if (!nvme_scan_ns_list(ctrl, nn))
3565 nvme_scan_ns_sequential(ctrl, nn);
3567 mutex_unlock(&ctrl->scan_lock);
3569 down_write(&ctrl->namespaces_rwsem);
3570 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3571 up_write(&ctrl->namespaces_rwsem);
3575 * This function iterates the namespace list unlocked to allow recovery from
3576 * controller failure. It is up to the caller to ensure the namespace list is
3577 * not modified by scan work while this function is executing.
3579 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3581 struct nvme_ns *ns, *next;
3584 /* prevent racing with ns scanning */
3585 flush_work(&ctrl->scan_work);
3588 * The dead states indicates the controller was not gracefully
3589 * disconnected. In that case, we won't be able to flush any data while
3590 * removing the namespaces' disks; fail all the queues now to avoid
3591 * potentially having to clean up the failed sync later.
3593 if (ctrl->state == NVME_CTRL_DEAD)
3594 nvme_kill_queues(ctrl);
3596 down_write(&ctrl->namespaces_rwsem);
3597 list_splice_init(&ctrl->namespaces, &ns_list);
3598 up_write(&ctrl->namespaces_rwsem);
3600 list_for_each_entry_safe(ns, next, &ns_list, list)
3603 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3605 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3607 char *envp[2] = { NULL, NULL };
3608 u32 aen_result = ctrl->aen_result;
3610 ctrl->aen_result = 0;
3614 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3617 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3621 static void nvme_async_event_work(struct work_struct *work)
3623 struct nvme_ctrl *ctrl =
3624 container_of(work, struct nvme_ctrl, async_event_work);
3626 nvme_aen_uevent(ctrl);
3627 ctrl->ops->submit_async_event(ctrl);
3630 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3635 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3641 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3644 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3646 struct nvme_fw_slot_info_log *log;
3648 log = kmalloc(sizeof(*log), GFP_KERNEL);
3652 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3654 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3658 static void nvme_fw_act_work(struct work_struct *work)
3660 struct nvme_ctrl *ctrl = container_of(work,
3661 struct nvme_ctrl, fw_act_work);
3662 unsigned long fw_act_timeout;
3665 fw_act_timeout = jiffies +
3666 msecs_to_jiffies(ctrl->mtfa * 100);
3668 fw_act_timeout = jiffies +
3669 msecs_to_jiffies(admin_timeout * 1000);
3671 nvme_stop_queues(ctrl);
3672 while (nvme_ctrl_pp_status(ctrl)) {
3673 if (time_after(jiffies, fw_act_timeout)) {
3674 dev_warn(ctrl->device,
3675 "Fw activation timeout, reset controller\n");
3676 nvme_reset_ctrl(ctrl);
3682 if (ctrl->state != NVME_CTRL_LIVE)
3685 nvme_start_queues(ctrl);
3686 /* read FW slot information to clear the AER */
3687 nvme_get_fw_slot_info(ctrl);
3690 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3692 u32 aer_notice_type = (result & 0xff00) >> 8;
3694 trace_nvme_async_event(ctrl, aer_notice_type);
3696 switch (aer_notice_type) {
3697 case NVME_AER_NOTICE_NS_CHANGED:
3698 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3699 nvme_queue_scan(ctrl);
3701 case NVME_AER_NOTICE_FW_ACT_STARTING:
3702 queue_work(nvme_wq, &ctrl->fw_act_work);
3704 #ifdef CONFIG_NVME_MULTIPATH
3705 case NVME_AER_NOTICE_ANA:
3706 if (!ctrl->ana_log_buf)
3708 queue_work(nvme_wq, &ctrl->ana_work);
3712 dev_warn(ctrl->device, "async event result %08x\n", result);
3716 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3717 volatile union nvme_result *res)
3719 u32 result = le32_to_cpu(res->u32);
3720 u32 aer_type = result & 0x07;
3722 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3726 case NVME_AER_NOTICE:
3727 nvme_handle_aen_notice(ctrl, result);
3729 case NVME_AER_ERROR:
3730 case NVME_AER_SMART:
3733 trace_nvme_async_event(ctrl, aer_type);
3734 ctrl->aen_result = result;
3739 queue_work(nvme_wq, &ctrl->async_event_work);
3741 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3743 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3745 nvme_mpath_stop(ctrl);
3746 nvme_stop_keep_alive(ctrl);
3747 flush_work(&ctrl->async_event_work);
3748 cancel_work_sync(&ctrl->fw_act_work);
3750 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3752 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3755 nvme_start_keep_alive(ctrl);
3757 if (ctrl->queue_count > 1) {
3758 nvme_queue_scan(ctrl);
3759 nvme_enable_aen(ctrl);
3760 queue_work(nvme_wq, &ctrl->async_event_work);
3761 nvme_start_queues(ctrl);
3764 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3766 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3768 nvme_fault_inject_fini(&ctrl->fault_inject);
3769 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3770 cdev_device_del(&ctrl->cdev, ctrl->device);
3772 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3774 static void nvme_free_ctrl(struct device *dev)
3776 struct nvme_ctrl *ctrl =
3777 container_of(dev, struct nvme_ctrl, ctrl_device);
3778 struct nvme_subsystem *subsys = ctrl->subsys;
3780 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3781 kfree(ctrl->effects);
3782 nvme_mpath_uninit(ctrl);
3783 __free_page(ctrl->discard_page);
3786 mutex_lock(&nvme_subsystems_lock);
3787 list_del(&ctrl->subsys_entry);
3788 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3789 mutex_unlock(&nvme_subsystems_lock);
3792 ctrl->ops->free_ctrl(ctrl);
3795 nvme_put_subsystem(subsys);
3799 * Initialize a NVMe controller structures. This needs to be called during
3800 * earliest initialization so that we have the initialized structured around
3803 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3804 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3808 ctrl->state = NVME_CTRL_NEW;
3809 spin_lock_init(&ctrl->lock);
3810 mutex_init(&ctrl->scan_lock);
3811 INIT_LIST_HEAD(&ctrl->namespaces);
3812 init_rwsem(&ctrl->namespaces_rwsem);
3815 ctrl->quirks = quirks;
3816 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3817 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3818 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3819 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3821 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3822 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3823 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3825 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3827 ctrl->discard_page = alloc_page(GFP_KERNEL);
3828 if (!ctrl->discard_page) {
3833 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3836 ctrl->instance = ret;
3838 device_initialize(&ctrl->ctrl_device);
3839 ctrl->device = &ctrl->ctrl_device;
3840 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3841 ctrl->device->class = nvme_class;
3842 ctrl->device->parent = ctrl->dev;
3843 ctrl->device->groups = nvme_dev_attr_groups;
3844 ctrl->device->release = nvme_free_ctrl;
3845 dev_set_drvdata(ctrl->device, ctrl);
3846 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3848 goto out_release_instance;
3850 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3851 ctrl->cdev.owner = ops->module;
3852 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3857 * Initialize latency tolerance controls. The sysfs files won't
3858 * be visible to userspace unless the device actually supports APST.
3860 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3861 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3862 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3864 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3868 kfree_const(ctrl->device->kobj.name);
3869 out_release_instance:
3870 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3872 if (ctrl->discard_page)
3873 __free_page(ctrl->discard_page);
3876 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3879 * nvme_kill_queues(): Ends all namespace queues
3880 * @ctrl: the dead controller that needs to end
3882 * Call this function when the driver determines it is unable to get the
3883 * controller in a state capable of servicing IO.
3885 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3889 down_read(&ctrl->namespaces_rwsem);
3891 /* Forcibly unquiesce queues to avoid blocking dispatch */
3892 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3893 blk_mq_unquiesce_queue(ctrl->admin_q);
3895 list_for_each_entry(ns, &ctrl->namespaces, list)
3896 nvme_set_queue_dying(ns);
3898 up_read(&ctrl->namespaces_rwsem);
3900 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3902 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3906 down_read(&ctrl->namespaces_rwsem);
3907 list_for_each_entry(ns, &ctrl->namespaces, list)
3908 blk_mq_unfreeze_queue(ns->queue);
3909 up_read(&ctrl->namespaces_rwsem);
3911 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3913 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3917 down_read(&ctrl->namespaces_rwsem);
3918 list_for_each_entry(ns, &ctrl->namespaces, list) {
3919 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3923 up_read(&ctrl->namespaces_rwsem);
3925 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3927 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3931 down_read(&ctrl->namespaces_rwsem);
3932 list_for_each_entry(ns, &ctrl->namespaces, list)
3933 blk_mq_freeze_queue_wait(ns->queue);
3934 up_read(&ctrl->namespaces_rwsem);
3936 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3938 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3942 down_read(&ctrl->namespaces_rwsem);
3943 list_for_each_entry(ns, &ctrl->namespaces, list)
3944 blk_freeze_queue_start(ns->queue);
3945 up_read(&ctrl->namespaces_rwsem);
3947 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3949 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3953 down_read(&ctrl->namespaces_rwsem);
3954 list_for_each_entry(ns, &ctrl->namespaces, list)
3955 blk_mq_quiesce_queue(ns->queue);
3956 up_read(&ctrl->namespaces_rwsem);
3958 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3960 void nvme_start_queues(struct nvme_ctrl *ctrl)
3964 down_read(&ctrl->namespaces_rwsem);
3965 list_for_each_entry(ns, &ctrl->namespaces, list)
3966 blk_mq_unquiesce_queue(ns->queue);
3967 up_read(&ctrl->namespaces_rwsem);
3969 EXPORT_SYMBOL_GPL(nvme_start_queues);
3972 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3976 down_read(&ctrl->namespaces_rwsem);
3977 list_for_each_entry(ns, &ctrl->namespaces, list)
3978 blk_sync_queue(ns->queue);
3979 up_read(&ctrl->namespaces_rwsem);
3981 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3984 * Check we didn't inadvertently grow the command structure sizes:
3986 static inline void _nvme_check_size(void)
3988 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3989 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3990 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3991 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3992 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3993 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3994 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3995 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3996 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3997 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3998 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3999 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4000 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4001 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4002 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4003 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4004 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4008 static int __init nvme_core_init(void)
4010 int result = -ENOMEM;
4014 nvme_wq = alloc_workqueue("nvme-wq",
4015 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4019 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4020 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4024 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4025 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4026 if (!nvme_delete_wq)
4027 goto destroy_reset_wq;
4029 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4031 goto destroy_delete_wq;
4033 nvme_class = class_create(THIS_MODULE, "nvme");
4034 if (IS_ERR(nvme_class)) {
4035 result = PTR_ERR(nvme_class);
4036 goto unregister_chrdev;
4039 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4040 if (IS_ERR(nvme_subsys_class)) {
4041 result = PTR_ERR(nvme_subsys_class);
4047 class_destroy(nvme_class);
4049 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4051 destroy_workqueue(nvme_delete_wq);
4053 destroy_workqueue(nvme_reset_wq);
4055 destroy_workqueue(nvme_wq);
4060 static void __exit nvme_core_exit(void)
4062 ida_destroy(&nvme_subsystems_ida);
4063 class_destroy(nvme_subsys_class);
4064 class_destroy(nvme_class);
4065 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4066 destroy_workqueue(nvme_delete_wq);
4067 destroy_workqueue(nvme_reset_wq);
4068 destroy_workqueue(nvme_wq);
4071 MODULE_LICENSE("GPL");
4072 MODULE_VERSION("1.0");
4073 module_init(nvme_core_init);
4074 module_exit(nvme_core_exit);