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>
25 #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(struct request *req)
202 switch (nvme_req(req)->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;
230 return BLK_STS_IOERR;
234 static inline bool nvme_req_needs_retry(struct request *req)
236 if (blk_noretry_request(req))
238 if (nvme_req(req)->status & NVME_SC_DNR)
240 if (nvme_req(req)->retries >= nvme_max_retries)
245 static void nvme_retry_req(struct request *req)
247 struct nvme_ns *ns = req->q->queuedata;
248 unsigned long delay = 0;
251 /* The mask and shift result must be <= 3 */
252 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
254 delay = ns->ctrl->crdt[crd - 1] * 100;
256 nvme_req(req)->retries++;
257 blk_mq_requeue_request(req, false);
258 blk_mq_delay_kick_requeue_list(req->q, delay);
261 void nvme_complete_rq(struct request *req)
263 blk_status_t status = nvme_error_status(req);
265 trace_nvme_complete_rq(req);
267 if (nvme_req(req)->ctrl->kas)
268 nvme_req(req)->ctrl->comp_seen = true;
270 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
271 if ((req->cmd_flags & REQ_NVME_MPATH) &&
272 blk_path_error(status)) {
273 nvme_failover_req(req);
277 if (!blk_queue_dying(req->q)) {
282 blk_mq_end_request(req, status);
284 EXPORT_SYMBOL_GPL(nvme_complete_rq);
286 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
288 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
289 "Cancelling I/O %d", req->tag);
291 nvme_req(req)->status = NVME_SC_ABORT_REQ;
292 blk_mq_complete_request_sync(req);
295 EXPORT_SYMBOL_GPL(nvme_cancel_request);
297 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
298 enum nvme_ctrl_state new_state)
300 enum nvme_ctrl_state old_state;
302 bool changed = false;
304 spin_lock_irqsave(&ctrl->lock, flags);
306 old_state = ctrl->state;
308 case NVME_CTRL_ADMIN_ONLY:
310 case NVME_CTRL_CONNECTING:
320 case NVME_CTRL_RESETTING:
321 case NVME_CTRL_CONNECTING:
328 case NVME_CTRL_RESETTING:
332 case NVME_CTRL_ADMIN_ONLY:
339 case NVME_CTRL_CONNECTING:
342 case NVME_CTRL_RESETTING:
349 case NVME_CTRL_DELETING:
352 case NVME_CTRL_ADMIN_ONLY:
353 case NVME_CTRL_RESETTING:
354 case NVME_CTRL_CONNECTING:
363 case NVME_CTRL_DELETING:
375 ctrl->state = new_state;
377 spin_unlock_irqrestore(&ctrl->lock, flags);
378 if (changed && ctrl->state == NVME_CTRL_LIVE)
379 nvme_kick_requeue_lists(ctrl);
382 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
384 static void nvme_free_ns_head(struct kref *ref)
386 struct nvme_ns_head *head =
387 container_of(ref, struct nvme_ns_head, ref);
389 nvme_mpath_remove_disk(head);
390 ida_simple_remove(&head->subsys->ns_ida, head->instance);
391 list_del_init(&head->entry);
392 cleanup_srcu_struct(&head->srcu);
393 nvme_put_subsystem(head->subsys);
397 static void nvme_put_ns_head(struct nvme_ns_head *head)
399 kref_put(&head->ref, nvme_free_ns_head);
402 static void nvme_free_ns(struct kref *kref)
404 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
407 nvme_nvm_unregister(ns);
410 nvme_put_ns_head(ns->head);
411 nvme_put_ctrl(ns->ctrl);
415 static void nvme_put_ns(struct nvme_ns *ns)
417 kref_put(&ns->kref, nvme_free_ns);
420 static inline void nvme_clear_nvme_request(struct request *req)
422 if (!(req->rq_flags & RQF_DONTPREP)) {
423 nvme_req(req)->retries = 0;
424 nvme_req(req)->flags = 0;
425 req->rq_flags |= RQF_DONTPREP;
429 struct request *nvme_alloc_request(struct request_queue *q,
430 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
432 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
435 if (qid == NVME_QID_ANY) {
436 req = blk_mq_alloc_request(q, op, flags);
438 req = blk_mq_alloc_request_hctx(q, op, flags,
444 req->cmd_flags |= REQ_FAILFAST_DRIVER;
445 nvme_clear_nvme_request(req);
446 nvme_req(req)->cmd = cmd;
450 EXPORT_SYMBOL_GPL(nvme_alloc_request);
452 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
454 struct nvme_command c;
456 memset(&c, 0, sizeof(c));
458 c.directive.opcode = nvme_admin_directive_send;
459 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
460 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
461 c.directive.dtype = NVME_DIR_IDENTIFY;
462 c.directive.tdtype = NVME_DIR_STREAMS;
463 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
465 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
468 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
470 return nvme_toggle_streams(ctrl, false);
473 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
475 return nvme_toggle_streams(ctrl, true);
478 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
479 struct streams_directive_params *s, u32 nsid)
481 struct nvme_command c;
483 memset(&c, 0, sizeof(c));
484 memset(s, 0, sizeof(*s));
486 c.directive.opcode = nvme_admin_directive_recv;
487 c.directive.nsid = cpu_to_le32(nsid);
488 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
489 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
490 c.directive.dtype = NVME_DIR_STREAMS;
492 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
495 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
497 struct streams_directive_params s;
500 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
505 ret = nvme_enable_streams(ctrl);
509 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
513 ctrl->nssa = le16_to_cpu(s.nssa);
514 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
515 dev_info(ctrl->device, "too few streams (%u) available\n",
517 nvme_disable_streams(ctrl);
521 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
522 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
527 * Check if 'req' has a write hint associated with it. If it does, assign
528 * a valid namespace stream to the write.
530 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
531 struct request *req, u16 *control,
534 enum rw_hint streamid = req->write_hint;
536 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
540 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
543 *control |= NVME_RW_DTYPE_STREAMS;
544 *dsmgmt |= streamid << 16;
547 if (streamid < ARRAY_SIZE(req->q->write_hints))
548 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
551 static inline void nvme_setup_flush(struct nvme_ns *ns,
552 struct nvme_command *cmnd)
554 cmnd->common.opcode = nvme_cmd_flush;
555 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
558 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
559 struct nvme_command *cmnd)
561 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
562 struct nvme_dsm_range *range;
565 range = kmalloc_array(segments, sizeof(*range),
566 GFP_ATOMIC | __GFP_NOWARN);
569 * If we fail allocation our range, fallback to the controller
570 * discard page. If that's also busy, it's safe to return
571 * busy, as we know we can make progress once that's freed.
573 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
574 return BLK_STS_RESOURCE;
576 range = page_address(ns->ctrl->discard_page);
579 __rq_for_each_bio(bio, req) {
580 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
581 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
584 range[n].cattr = cpu_to_le32(0);
585 range[n].nlb = cpu_to_le32(nlb);
586 range[n].slba = cpu_to_le64(slba);
591 if (WARN_ON_ONCE(n != segments)) {
592 if (virt_to_page(range) == ns->ctrl->discard_page)
593 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
596 return BLK_STS_IOERR;
599 cmnd->dsm.opcode = nvme_cmd_dsm;
600 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
601 cmnd->dsm.nr = cpu_to_le32(segments - 1);
602 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
604 req->special_vec.bv_page = virt_to_page(range);
605 req->special_vec.bv_offset = offset_in_page(range);
606 req->special_vec.bv_len = sizeof(*range) * segments;
607 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
612 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
613 struct request *req, struct nvme_command *cmnd)
615 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
616 return nvme_setup_discard(ns, req, cmnd);
618 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
619 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
620 cmnd->write_zeroes.slba =
621 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
622 cmnd->write_zeroes.length =
623 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
624 cmnd->write_zeroes.control = 0;
628 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
629 struct request *req, struct nvme_command *cmnd)
631 struct nvme_ctrl *ctrl = ns->ctrl;
635 if (req->cmd_flags & REQ_FUA)
636 control |= NVME_RW_FUA;
637 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
638 control |= NVME_RW_LR;
640 if (req->cmd_flags & REQ_RAHEAD)
641 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
643 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
644 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
645 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
646 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
648 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
649 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
653 * If formated with metadata, the block layer always provides a
654 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
655 * we enable the PRACT bit for protection information or set the
656 * namespace capacity to zero to prevent any I/O.
658 if (!blk_integrity_rq(req)) {
659 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
660 return BLK_STS_NOTSUPP;
661 control |= NVME_RW_PRINFO_PRACT;
662 } else if (req_op(req) == REQ_OP_WRITE) {
663 t10_pi_prepare(req, ns->pi_type);
666 switch (ns->pi_type) {
667 case NVME_NS_DPS_PI_TYPE3:
668 control |= NVME_RW_PRINFO_PRCHK_GUARD;
670 case NVME_NS_DPS_PI_TYPE1:
671 case NVME_NS_DPS_PI_TYPE2:
672 control |= NVME_RW_PRINFO_PRCHK_GUARD |
673 NVME_RW_PRINFO_PRCHK_REF;
674 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
679 cmnd->rw.control = cpu_to_le16(control);
680 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
684 void nvme_cleanup_cmd(struct request *req)
686 if (blk_integrity_rq(req) && req_op(req) == REQ_OP_READ &&
687 nvme_req(req)->status == 0) {
688 struct nvme_ns *ns = req->rq_disk->private_data;
690 t10_pi_complete(req, ns->pi_type,
691 blk_rq_bytes(req) >> ns->lba_shift);
693 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
694 struct nvme_ns *ns = req->rq_disk->private_data;
695 struct page *page = req->special_vec.bv_page;
697 if (page == ns->ctrl->discard_page)
698 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
700 kfree(page_address(page) + req->special_vec.bv_offset);
703 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
705 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
706 struct nvme_command *cmd)
708 blk_status_t ret = BLK_STS_OK;
710 nvme_clear_nvme_request(req);
712 memset(cmd, 0, sizeof(*cmd));
713 switch (req_op(req)) {
716 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
719 nvme_setup_flush(ns, cmd);
721 case REQ_OP_WRITE_ZEROES:
722 ret = nvme_setup_write_zeroes(ns, req, cmd);
725 ret = nvme_setup_discard(ns, req, cmd);
729 ret = nvme_setup_rw(ns, req, cmd);
733 return BLK_STS_IOERR;
736 cmd->common.command_id = req->tag;
737 trace_nvme_setup_cmd(req, cmd);
740 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
742 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
744 struct completion *waiting = rq->end_io_data;
746 rq->end_io_data = NULL;
750 static void nvme_execute_rq_polled(struct request_queue *q,
751 struct gendisk *bd_disk, struct request *rq, int at_head)
753 DECLARE_COMPLETION_ONSTACK(wait);
755 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
757 rq->cmd_flags |= REQ_HIPRI;
758 rq->end_io_data = &wait;
759 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
761 while (!completion_done(&wait)) {
762 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
768 * Returns 0 on success. If the result is negative, it's a Linux error code;
769 * if the result is positive, it's an NVM Express status code
771 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
772 union nvme_result *result, void *buffer, unsigned bufflen,
773 unsigned timeout, int qid, int at_head,
774 blk_mq_req_flags_t flags, bool poll)
779 req = nvme_alloc_request(q, cmd, flags, qid);
783 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
785 if (buffer && bufflen) {
786 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
792 nvme_execute_rq_polled(req->q, NULL, req, at_head);
794 blk_execute_rq(req->q, NULL, req, at_head);
796 *result = nvme_req(req)->result;
797 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
800 ret = nvme_req(req)->status;
802 blk_mq_free_request(req);
805 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
807 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
808 void *buffer, unsigned bufflen)
810 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
811 NVME_QID_ANY, 0, 0, false);
813 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
815 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
816 unsigned len, u32 seed, bool write)
818 struct bio_integrity_payload *bip;
822 buf = kmalloc(len, GFP_KERNEL);
827 if (write && copy_from_user(buf, ubuf, len))
830 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
836 bip->bip_iter.bi_size = len;
837 bip->bip_iter.bi_sector = seed;
838 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
839 offset_in_page(buf));
849 static int nvme_submit_user_cmd(struct request_queue *q,
850 struct nvme_command *cmd, void __user *ubuffer,
851 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
852 u32 meta_seed, u32 *result, unsigned timeout)
854 bool write = nvme_is_write(cmd);
855 struct nvme_ns *ns = q->queuedata;
856 struct gendisk *disk = ns ? ns->disk : NULL;
858 struct bio *bio = NULL;
862 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
866 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
867 nvme_req(req)->flags |= NVME_REQ_USERCMD;
869 if (ubuffer && bufflen) {
870 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
876 if (disk && meta_buffer && meta_len) {
877 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
883 req->cmd_flags |= REQ_INTEGRITY;
887 blk_execute_rq(req->q, disk, req, 0);
888 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
891 ret = nvme_req(req)->status;
893 *result = le32_to_cpu(nvme_req(req)->result.u32);
894 if (meta && !ret && !write) {
895 if (copy_to_user(meta_buffer, meta, meta_len))
901 blk_rq_unmap_user(bio);
903 blk_mq_free_request(req);
907 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
909 struct nvme_ctrl *ctrl = rq->end_io_data;
911 bool startka = false;
913 blk_mq_free_request(rq);
916 dev_err(ctrl->device,
917 "failed nvme_keep_alive_end_io error=%d\n",
922 ctrl->comp_seen = false;
923 spin_lock_irqsave(&ctrl->lock, flags);
924 if (ctrl->state == NVME_CTRL_LIVE ||
925 ctrl->state == NVME_CTRL_CONNECTING)
927 spin_unlock_irqrestore(&ctrl->lock, flags);
929 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
932 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
936 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
941 rq->timeout = ctrl->kato * HZ;
942 rq->end_io_data = ctrl;
944 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
949 static void nvme_keep_alive_work(struct work_struct *work)
951 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
952 struct nvme_ctrl, ka_work);
953 bool comp_seen = ctrl->comp_seen;
955 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
956 dev_dbg(ctrl->device,
957 "reschedule traffic based keep-alive timer\n");
958 ctrl->comp_seen = false;
959 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
963 if (nvme_keep_alive(ctrl)) {
964 /* allocation failure, reset the controller */
965 dev_err(ctrl->device, "keep-alive failed\n");
966 nvme_reset_ctrl(ctrl);
971 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
973 if (unlikely(ctrl->kato == 0))
976 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
979 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
981 if (unlikely(ctrl->kato == 0))
984 cancel_delayed_work_sync(&ctrl->ka_work);
986 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
988 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
990 struct nvme_command c = { };
993 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
994 c.identify.opcode = nvme_admin_identify;
995 c.identify.cns = NVME_ID_CNS_CTRL;
997 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1001 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1002 sizeof(struct nvme_id_ctrl));
1008 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1009 struct nvme_ns_ids *ids)
1011 struct nvme_command c = { };
1017 c.identify.opcode = nvme_admin_identify;
1018 c.identify.nsid = cpu_to_le32(nsid);
1019 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1021 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1025 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1026 NVME_IDENTIFY_DATA_SIZE);
1030 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1031 struct nvme_ns_id_desc *cur = data + pos;
1036 switch (cur->nidt) {
1037 case NVME_NIDT_EUI64:
1038 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1039 dev_warn(ctrl->device,
1040 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1044 len = NVME_NIDT_EUI64_LEN;
1045 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1047 case NVME_NIDT_NGUID:
1048 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1049 dev_warn(ctrl->device,
1050 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1054 len = NVME_NIDT_NGUID_LEN;
1055 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1057 case NVME_NIDT_UUID:
1058 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1059 dev_warn(ctrl->device,
1060 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1064 len = NVME_NIDT_UUID_LEN;
1065 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1068 /* Skip unknown types */
1073 len += sizeof(*cur);
1080 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1082 struct nvme_command c = { };
1084 c.identify.opcode = nvme_admin_identify;
1085 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1086 c.identify.nsid = cpu_to_le32(nsid);
1087 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1088 NVME_IDENTIFY_DATA_SIZE);
1091 static struct nvme_id_ns *nvme_identify_ns(struct nvme_ctrl *ctrl,
1094 struct nvme_id_ns *id;
1095 struct nvme_command c = { };
1098 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1099 c.identify.opcode = nvme_admin_identify;
1100 c.identify.nsid = cpu_to_le32(nsid);
1101 c.identify.cns = NVME_ID_CNS_NS;
1103 id = kmalloc(sizeof(*id), GFP_KERNEL);
1107 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1109 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1117 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1118 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1120 struct nvme_command c;
1121 union nvme_result res;
1124 memset(&c, 0, sizeof(c));
1125 c.features.opcode = op;
1126 c.features.fid = cpu_to_le32(fid);
1127 c.features.dword11 = cpu_to_le32(dword11);
1129 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1130 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1131 if (ret >= 0 && result)
1132 *result = le32_to_cpu(res.u32);
1136 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1137 unsigned int dword11, void *buffer, size_t buflen,
1140 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1143 EXPORT_SYMBOL_GPL(nvme_set_features);
1145 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1146 unsigned int dword11, void *buffer, size_t buflen,
1149 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1152 EXPORT_SYMBOL_GPL(nvme_get_features);
1154 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1156 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1158 int status, nr_io_queues;
1160 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1166 * Degraded controllers might return an error when setting the queue
1167 * count. We still want to be able to bring them online and offer
1168 * access to the admin queue, as that might be only way to fix them up.
1171 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1174 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1175 *count = min(*count, nr_io_queues);
1180 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1182 #define NVME_AEN_SUPPORTED \
1183 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | NVME_AEN_CFG_ANA_CHANGE)
1185 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1187 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1190 if (!supported_aens)
1193 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1196 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1200 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1202 struct nvme_user_io io;
1203 struct nvme_command c;
1204 unsigned length, meta_len;
1205 void __user *metadata;
1207 if (copy_from_user(&io, uio, sizeof(io)))
1212 switch (io.opcode) {
1213 case nvme_cmd_write:
1215 case nvme_cmd_compare:
1221 length = (io.nblocks + 1) << ns->lba_shift;
1222 meta_len = (io.nblocks + 1) * ns->ms;
1223 metadata = (void __user *)(uintptr_t)io.metadata;
1228 } else if (meta_len) {
1229 if ((io.metadata & 3) || !io.metadata)
1233 memset(&c, 0, sizeof(c));
1234 c.rw.opcode = io.opcode;
1235 c.rw.flags = io.flags;
1236 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1237 c.rw.slba = cpu_to_le64(io.slba);
1238 c.rw.length = cpu_to_le16(io.nblocks);
1239 c.rw.control = cpu_to_le16(io.control);
1240 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1241 c.rw.reftag = cpu_to_le32(io.reftag);
1242 c.rw.apptag = cpu_to_le16(io.apptag);
1243 c.rw.appmask = cpu_to_le16(io.appmask);
1245 return nvme_submit_user_cmd(ns->queue, &c,
1246 (void __user *)(uintptr_t)io.addr, length,
1247 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1250 static u32 nvme_known_admin_effects(u8 opcode)
1253 case nvme_admin_format_nvm:
1254 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1255 NVME_CMD_EFFECTS_CSE_MASK;
1256 case nvme_admin_sanitize_nvm:
1257 return NVME_CMD_EFFECTS_CSE_MASK;
1264 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1271 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1272 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1273 dev_warn(ctrl->device,
1274 "IO command:%02x has unhandled effects:%08x\n",
1280 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1281 effects |= nvme_known_admin_effects(opcode);
1284 * For simplicity, IO to all namespaces is quiesced even if the command
1285 * effects say only one namespace is affected.
1287 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1288 mutex_lock(&ctrl->scan_lock);
1289 mutex_lock(&ctrl->subsys->lock);
1290 nvme_mpath_start_freeze(ctrl->subsys);
1291 nvme_mpath_wait_freeze(ctrl->subsys);
1292 nvme_start_freeze(ctrl);
1293 nvme_wait_freeze(ctrl);
1298 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1302 down_read(&ctrl->namespaces_rwsem);
1303 list_for_each_entry(ns, &ctrl->namespaces, list)
1304 if (ns->disk && nvme_revalidate_disk(ns->disk))
1305 nvme_set_queue_dying(ns);
1306 up_read(&ctrl->namespaces_rwsem);
1308 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1311 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1314 * Revalidate LBA changes prior to unfreezing. This is necessary to
1315 * prevent memory corruption if a logical block size was changed by
1318 if (effects & NVME_CMD_EFFECTS_LBCC)
1319 nvme_update_formats(ctrl);
1320 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1321 nvme_unfreeze(ctrl);
1322 nvme_mpath_unfreeze(ctrl->subsys);
1323 mutex_unlock(&ctrl->subsys->lock);
1324 mutex_unlock(&ctrl->scan_lock);
1326 if (effects & NVME_CMD_EFFECTS_CCC)
1327 nvme_init_identify(ctrl);
1328 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1329 nvme_queue_scan(ctrl);
1332 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1333 struct nvme_passthru_cmd __user *ucmd)
1335 struct nvme_passthru_cmd cmd;
1336 struct nvme_command c;
1337 unsigned timeout = 0;
1341 if (!capable(CAP_SYS_ADMIN))
1343 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1348 memset(&c, 0, sizeof(c));
1349 c.common.opcode = cmd.opcode;
1350 c.common.flags = cmd.flags;
1351 c.common.nsid = cpu_to_le32(cmd.nsid);
1352 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1353 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1354 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1355 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1356 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1357 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1358 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1359 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1362 timeout = msecs_to_jiffies(cmd.timeout_ms);
1364 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1365 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1366 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1367 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1368 0, &cmd.result, timeout);
1369 nvme_passthru_end(ctrl, effects);
1372 if (put_user(cmd.result, &ucmd->result))
1380 * Issue ioctl requests on the first available path. Note that unlike normal
1381 * block layer requests we will not retry failed request on another controller.
1383 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1384 struct nvme_ns_head **head, int *srcu_idx)
1386 #ifdef CONFIG_NVME_MULTIPATH
1387 if (disk->fops == &nvme_ns_head_ops) {
1390 *head = disk->private_data;
1391 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1392 ns = nvme_find_path(*head);
1394 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1400 return disk->private_data;
1403 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1406 srcu_read_unlock(&head->srcu, idx);
1409 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1410 unsigned int cmd, unsigned long arg)
1412 struct nvme_ns_head *head = NULL;
1413 void __user *argp = (void __user *)arg;
1417 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1419 return -EWOULDBLOCK;
1422 * Handle ioctls that apply to the controller instead of the namespace
1423 * seperately and drop the ns SRCU reference early. This avoids a
1424 * deadlock when deleting namespaces using the passthrough interface.
1426 if (cmd == NVME_IOCTL_ADMIN_CMD || is_sed_ioctl(cmd)) {
1427 struct nvme_ctrl *ctrl = ns->ctrl;
1429 nvme_get_ctrl(ns->ctrl);
1430 nvme_put_ns_from_disk(head, srcu_idx);
1432 if (cmd == NVME_IOCTL_ADMIN_CMD)
1433 ret = nvme_user_cmd(ctrl, NULL, argp);
1435 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1437 nvme_put_ctrl(ctrl);
1443 force_successful_syscall_return();
1444 ret = ns->head->ns_id;
1446 case NVME_IOCTL_IO_CMD:
1447 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1449 case NVME_IOCTL_SUBMIT_IO:
1450 ret = nvme_submit_io(ns, argp);
1454 ret = nvme_nvm_ioctl(ns, cmd, arg);
1459 nvme_put_ns_from_disk(head, srcu_idx);
1463 static int nvme_open(struct block_device *bdev, fmode_t mode)
1465 struct nvme_ns *ns = bdev->bd_disk->private_data;
1467 #ifdef CONFIG_NVME_MULTIPATH
1468 /* should never be called due to GENHD_FL_HIDDEN */
1469 if (WARN_ON_ONCE(ns->head->disk))
1472 if (!kref_get_unless_zero(&ns->kref))
1474 if (!try_module_get(ns->ctrl->ops->module))
1485 static void nvme_release(struct gendisk *disk, fmode_t mode)
1487 struct nvme_ns *ns = disk->private_data;
1489 module_put(ns->ctrl->ops->module);
1493 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1495 /* some standard values */
1496 geo->heads = 1 << 6;
1497 geo->sectors = 1 << 5;
1498 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1502 #ifdef CONFIG_BLK_DEV_INTEGRITY
1503 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1505 struct blk_integrity integrity;
1507 memset(&integrity, 0, sizeof(integrity));
1509 case NVME_NS_DPS_PI_TYPE3:
1510 integrity.profile = &t10_pi_type3_crc;
1511 integrity.tag_size = sizeof(u16) + sizeof(u32);
1512 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1514 case NVME_NS_DPS_PI_TYPE1:
1515 case NVME_NS_DPS_PI_TYPE2:
1516 integrity.profile = &t10_pi_type1_crc;
1517 integrity.tag_size = sizeof(u16);
1518 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1521 integrity.profile = NULL;
1524 integrity.tuple_size = ms;
1525 blk_integrity_register(disk, &integrity);
1526 blk_queue_max_integrity_segments(disk->queue, 1);
1529 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1532 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1534 static void nvme_set_chunk_size(struct nvme_ns *ns)
1536 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1537 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1540 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1542 struct nvme_ctrl *ctrl = ns->ctrl;
1543 struct request_queue *queue = disk->queue;
1544 u32 size = queue_logical_block_size(queue);
1546 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1547 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1551 if (ctrl->nr_streams && ns->sws && ns->sgs)
1552 size *= ns->sws * ns->sgs;
1554 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1555 NVME_DSM_MAX_RANGES);
1557 queue->limits.discard_alignment = 0;
1558 queue->limits.discard_granularity = size;
1560 /* If discard is already enabled, don't reset queue limits */
1561 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1564 blk_queue_max_discard_sectors(queue, UINT_MAX);
1565 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1567 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1568 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1571 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1574 unsigned short bs = 1 << ns->lba_shift;
1576 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1577 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1580 * Even though NVMe spec explicitly states that MDTS is not
1581 * applicable to the write-zeroes:- "The restriction does not apply to
1582 * commands that do not transfer data between the host and the
1583 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1584 * In order to be more cautious use controller's max_hw_sectors value
1585 * to configure the maximum sectors for the write-zeroes which is
1586 * configured based on the controller's MDTS field in the
1587 * nvme_init_identify() if available.
1589 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1590 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1592 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1594 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1597 static void nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1598 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1600 memset(ids, 0, sizeof(*ids));
1602 if (ctrl->vs >= NVME_VS(1, 1, 0))
1603 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1604 if (ctrl->vs >= NVME_VS(1, 2, 0))
1605 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1606 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1607 /* Don't treat error as fatal we potentially
1608 * already have a NGUID or EUI-64
1610 if (nvme_identify_ns_descs(ctrl, nsid, ids))
1611 dev_warn(ctrl->device,
1612 "%s: Identify Descriptors failed\n", __func__);
1616 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1618 return !uuid_is_null(&ids->uuid) ||
1619 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1620 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1623 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1625 return uuid_equal(&a->uuid, &b->uuid) &&
1626 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1627 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1630 static void nvme_update_disk_info(struct gendisk *disk,
1631 struct nvme_ns *ns, struct nvme_id_ns *id)
1633 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1634 unsigned short bs = 1 << ns->lba_shift;
1635 u32 atomic_bs, phys_bs, io_opt;
1637 if (ns->lba_shift > PAGE_SHIFT) {
1638 /* unsupported block size, set capacity to 0 later */
1641 blk_mq_freeze_queue(disk->queue);
1642 blk_integrity_unregister(disk);
1644 if (id->nabo == 0) {
1646 * Bit 1 indicates whether NAWUPF is defined for this namespace
1647 * and whether it should be used instead of AWUPF. If NAWUPF ==
1648 * 0 then AWUPF must be used instead.
1650 if (id->nsfeat & (1 << 1) && id->nawupf)
1651 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1653 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1659 if (id->nsfeat & (1 << 4)) {
1660 /* NPWG = Namespace Preferred Write Granularity */
1661 phys_bs *= 1 + le16_to_cpu(id->npwg);
1662 /* NOWS = Namespace Optimal Write Size */
1663 io_opt *= 1 + le16_to_cpu(id->nows);
1666 blk_queue_logical_block_size(disk->queue, bs);
1668 * Linux filesystems assume writing a single physical block is
1669 * an atomic operation. Hence limit the physical block size to the
1670 * value of the Atomic Write Unit Power Fail parameter.
1672 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1673 blk_queue_io_min(disk->queue, phys_bs);
1674 blk_queue_io_opt(disk->queue, io_opt);
1676 if (ns->ms && !ns->ext &&
1677 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1678 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1679 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1680 ns->lba_shift > PAGE_SHIFT)
1683 set_capacity(disk, capacity);
1685 nvme_config_discard(disk, ns);
1686 nvme_config_write_zeroes(disk, ns);
1688 if (id->nsattr & (1 << 0))
1689 set_disk_ro(disk, true);
1691 set_disk_ro(disk, false);
1693 blk_mq_unfreeze_queue(disk->queue);
1696 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1698 struct nvme_ns *ns = disk->private_data;
1701 * If identify namespace failed, use default 512 byte block size so
1702 * block layer can use before failing read/write for 0 capacity.
1704 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1705 if (ns->lba_shift == 0)
1707 ns->noiob = le16_to_cpu(id->noiob);
1708 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1709 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1710 /* the PI implementation requires metadata equal t10 pi tuple size */
1711 if (ns->ms == sizeof(struct t10_pi_tuple))
1712 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1717 nvme_set_chunk_size(ns);
1718 nvme_update_disk_info(disk, ns, id);
1719 #ifdef CONFIG_NVME_MULTIPATH
1720 if (ns->head->disk) {
1721 nvme_update_disk_info(ns->head->disk, ns, id);
1722 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1723 revalidate_disk(ns->head->disk);
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 id = nvme_identify_ns(ctrl, ns->head->ns_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);
1763 static char nvme_pr_type(enum pr_type type)
1766 case PR_WRITE_EXCLUSIVE:
1768 case PR_EXCLUSIVE_ACCESS:
1770 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1772 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1774 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1776 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1783 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1784 u64 key, u64 sa_key, u8 op)
1786 struct nvme_ns_head *head = NULL;
1788 struct nvme_command c;
1790 u8 data[16] = { 0, };
1792 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1794 return -EWOULDBLOCK;
1796 put_unaligned_le64(key, &data[0]);
1797 put_unaligned_le64(sa_key, &data[8]);
1799 memset(&c, 0, sizeof(c));
1800 c.common.opcode = op;
1801 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1802 c.common.cdw10 = cpu_to_le32(cdw10);
1804 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1805 nvme_put_ns_from_disk(head, srcu_idx);
1809 static int nvme_pr_register(struct block_device *bdev, u64 old,
1810 u64 new, unsigned flags)
1814 if (flags & ~PR_FL_IGNORE_KEY)
1817 cdw10 = old ? 2 : 0;
1818 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1819 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1820 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1823 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1824 enum pr_type type, unsigned flags)
1828 if (flags & ~PR_FL_IGNORE_KEY)
1831 cdw10 = nvme_pr_type(type) << 8;
1832 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1833 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1836 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1837 enum pr_type type, bool abort)
1839 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1840 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1843 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1845 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1846 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1849 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1851 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1852 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1855 static const struct pr_ops nvme_pr_ops = {
1856 .pr_register = nvme_pr_register,
1857 .pr_reserve = nvme_pr_reserve,
1858 .pr_release = nvme_pr_release,
1859 .pr_preempt = nvme_pr_preempt,
1860 .pr_clear = nvme_pr_clear,
1863 #ifdef CONFIG_BLK_SED_OPAL
1864 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1867 struct nvme_ctrl *ctrl = data;
1868 struct nvme_command cmd;
1870 memset(&cmd, 0, sizeof(cmd));
1872 cmd.common.opcode = nvme_admin_security_send;
1874 cmd.common.opcode = nvme_admin_security_recv;
1875 cmd.common.nsid = 0;
1876 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1877 cmd.common.cdw11 = cpu_to_le32(len);
1879 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1880 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1882 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1883 #endif /* CONFIG_BLK_SED_OPAL */
1885 static const struct block_device_operations nvme_fops = {
1886 .owner = THIS_MODULE,
1887 .ioctl = nvme_ioctl,
1888 .compat_ioctl = nvme_ioctl,
1890 .release = nvme_release,
1891 .getgeo = nvme_getgeo,
1892 .revalidate_disk= nvme_revalidate_disk,
1893 .pr_ops = &nvme_pr_ops,
1896 #ifdef CONFIG_NVME_MULTIPATH
1897 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1899 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1901 if (!kref_get_unless_zero(&head->ref))
1906 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1908 nvme_put_ns_head(disk->private_data);
1911 const struct block_device_operations nvme_ns_head_ops = {
1912 .owner = THIS_MODULE,
1913 .open = nvme_ns_head_open,
1914 .release = nvme_ns_head_release,
1915 .ioctl = nvme_ioctl,
1916 .compat_ioctl = nvme_ioctl,
1917 .getgeo = nvme_getgeo,
1918 .pr_ops = &nvme_pr_ops,
1920 #endif /* CONFIG_NVME_MULTIPATH */
1922 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
1924 unsigned long timeout =
1925 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
1926 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
1929 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
1932 if ((csts & NVME_CSTS_RDY) == bit)
1936 if (fatal_signal_pending(current))
1938 if (time_after(jiffies, timeout)) {
1939 dev_err(ctrl->device,
1940 "Device not ready; aborting %s\n", enabled ?
1941 "initialisation" : "reset");
1950 * If the device has been passed off to us in an enabled state, just clear
1951 * the enabled bit. The spec says we should set the 'shutdown notification
1952 * bits', but doing so may cause the device to complete commands to the
1953 * admin queue ... and we don't know what memory that might be pointing at!
1955 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1959 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
1960 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
1962 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
1966 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
1967 msleep(NVME_QUIRK_DELAY_AMOUNT);
1969 return nvme_wait_ready(ctrl, cap, false);
1971 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
1973 int nvme_enable_ctrl(struct nvme_ctrl *ctrl, u64 cap)
1976 * Default to a 4K page size, with the intention to update this
1977 * path in the future to accomodate architectures with differing
1978 * kernel and IO page sizes.
1980 unsigned dev_page_min = NVME_CAP_MPSMIN(cap) + 12, page_shift = 12;
1983 if (page_shift < dev_page_min) {
1984 dev_err(ctrl->device,
1985 "Minimum device page size %u too large for host (%u)\n",
1986 1 << dev_page_min, 1 << page_shift);
1990 ctrl->page_size = 1 << page_shift;
1992 ctrl->ctrl_config = NVME_CC_CSS_NVM;
1993 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
1994 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
1995 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
1996 ctrl->ctrl_config |= NVME_CC_ENABLE;
1998 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2001 return nvme_wait_ready(ctrl, cap, true);
2003 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2005 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2007 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2011 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2012 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2014 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2018 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2019 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2023 if (fatal_signal_pending(current))
2025 if (time_after(jiffies, timeout)) {
2026 dev_err(ctrl->device,
2027 "Device shutdown incomplete; abort shutdown\n");
2034 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2036 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2037 struct request_queue *q)
2041 if (ctrl->max_hw_sectors) {
2043 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2045 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2046 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2047 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2049 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2050 is_power_of_2(ctrl->max_hw_sectors))
2051 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2052 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2053 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2055 blk_queue_write_cache(q, vwc, vwc);
2058 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2063 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2066 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2067 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2070 dev_warn_once(ctrl->device,
2071 "could not set timestamp (%d)\n", ret);
2075 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2077 struct nvme_feat_host_behavior *host;
2080 /* Don't bother enabling the feature if retry delay is not reported */
2084 host = kzalloc(sizeof(*host), GFP_KERNEL);
2088 host->acre = NVME_ENABLE_ACRE;
2089 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2090 host, sizeof(*host), NULL);
2095 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2098 * APST (Autonomous Power State Transition) lets us program a
2099 * table of power state transitions that the controller will
2100 * perform automatically. We configure it with a simple
2101 * heuristic: we are willing to spend at most 2% of the time
2102 * transitioning between power states. Therefore, when running
2103 * in any given state, we will enter the next lower-power
2104 * non-operational state after waiting 50 * (enlat + exlat)
2105 * microseconds, as long as that state's exit latency is under
2106 * the requested maximum latency.
2108 * We will not autonomously enter any non-operational state for
2109 * which the total latency exceeds ps_max_latency_us. Users
2110 * can set ps_max_latency_us to zero to turn off APST.
2114 struct nvme_feat_auto_pst *table;
2120 * If APST isn't supported or if we haven't been initialized yet,
2121 * then don't do anything.
2126 if (ctrl->npss > 31) {
2127 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2131 table = kzalloc(sizeof(*table), GFP_KERNEL);
2135 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2136 /* Turn off APST. */
2138 dev_dbg(ctrl->device, "APST disabled\n");
2140 __le64 target = cpu_to_le64(0);
2144 * Walk through all states from lowest- to highest-power.
2145 * According to the spec, lower-numbered states use more
2146 * power. NPSS, despite the name, is the index of the
2147 * lowest-power state, not the number of states.
2149 for (state = (int)ctrl->npss; state >= 0; state--) {
2150 u64 total_latency_us, exit_latency_us, transition_ms;
2153 table->entries[state] = target;
2156 * Don't allow transitions to the deepest state
2157 * if it's quirked off.
2159 if (state == ctrl->npss &&
2160 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2164 * Is this state a useful non-operational state for
2165 * higher-power states to autonomously transition to?
2167 if (!(ctrl->psd[state].flags &
2168 NVME_PS_FLAGS_NON_OP_STATE))
2172 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2173 if (exit_latency_us > ctrl->ps_max_latency_us)
2178 le32_to_cpu(ctrl->psd[state].entry_lat);
2181 * This state is good. Use it as the APST idle
2182 * target for higher power states.
2184 transition_ms = total_latency_us + 19;
2185 do_div(transition_ms, 20);
2186 if (transition_ms > (1 << 24) - 1)
2187 transition_ms = (1 << 24) - 1;
2189 target = cpu_to_le64((state << 3) |
2190 (transition_ms << 8));
2195 if (total_latency_us > max_lat_us)
2196 max_lat_us = total_latency_us;
2202 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2204 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2205 max_ps, max_lat_us, (int)sizeof(*table), table);
2209 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2210 table, sizeof(*table), NULL);
2212 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2218 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2220 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2224 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2225 case PM_QOS_LATENCY_ANY:
2233 if (ctrl->ps_max_latency_us != latency) {
2234 ctrl->ps_max_latency_us = latency;
2235 nvme_configure_apst(ctrl);
2239 struct nvme_core_quirk_entry {
2241 * NVMe model and firmware strings are padded with spaces. For
2242 * simplicity, strings in the quirk table are padded with NULLs
2248 unsigned long quirks;
2251 static const struct nvme_core_quirk_entry core_quirks[] = {
2254 * This Toshiba device seems to die using any APST states. See:
2255 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2258 .mn = "THNSF5256GPUK TOSHIBA",
2259 .quirks = NVME_QUIRK_NO_APST,
2263 /* match is null-terminated but idstr is space-padded. */
2264 static bool string_matches(const char *idstr, const char *match, size_t len)
2271 matchlen = strlen(match);
2272 WARN_ON_ONCE(matchlen > len);
2274 if (memcmp(idstr, match, matchlen))
2277 for (; matchlen < len; matchlen++)
2278 if (idstr[matchlen] != ' ')
2284 static bool quirk_matches(const struct nvme_id_ctrl *id,
2285 const struct nvme_core_quirk_entry *q)
2287 return q->vid == le16_to_cpu(id->vid) &&
2288 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2289 string_matches(id->fr, q->fr, sizeof(id->fr));
2292 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2293 struct nvme_id_ctrl *id)
2298 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2299 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2300 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2301 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2305 if (ctrl->vs >= NVME_VS(1, 2, 1))
2306 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2309 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2310 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2311 "nqn.2014.08.org.nvmexpress:%04x%04x",
2312 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2313 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2314 off += sizeof(id->sn);
2315 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2316 off += sizeof(id->mn);
2317 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2320 static void nvme_release_subsystem(struct device *dev)
2322 struct nvme_subsystem *subsys =
2323 container_of(dev, struct nvme_subsystem, dev);
2325 ida_simple_remove(&nvme_subsystems_ida, subsys->instance);
2329 static void nvme_destroy_subsystem(struct kref *ref)
2331 struct nvme_subsystem *subsys =
2332 container_of(ref, struct nvme_subsystem, ref);
2334 mutex_lock(&nvme_subsystems_lock);
2335 list_del(&subsys->entry);
2336 mutex_unlock(&nvme_subsystems_lock);
2338 ida_destroy(&subsys->ns_ida);
2339 device_del(&subsys->dev);
2340 put_device(&subsys->dev);
2343 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2345 kref_put(&subsys->ref, nvme_destroy_subsystem);
2348 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2350 struct nvme_subsystem *subsys;
2352 lockdep_assert_held(&nvme_subsystems_lock);
2354 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2355 if (strcmp(subsys->subnqn, subsysnqn))
2357 if (!kref_get_unless_zero(&subsys->ref))
2365 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2366 struct device_attribute subsys_attr_##_name = \
2367 __ATTR(_name, _mode, _show, NULL)
2369 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2370 struct device_attribute *attr,
2373 struct nvme_subsystem *subsys =
2374 container_of(dev, struct nvme_subsystem, dev);
2376 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2378 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2380 #define nvme_subsys_show_str_function(field) \
2381 static ssize_t subsys_##field##_show(struct device *dev, \
2382 struct device_attribute *attr, char *buf) \
2384 struct nvme_subsystem *subsys = \
2385 container_of(dev, struct nvme_subsystem, dev); \
2386 return sprintf(buf, "%.*s\n", \
2387 (int)sizeof(subsys->field), subsys->field); \
2389 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2391 nvme_subsys_show_str_function(model);
2392 nvme_subsys_show_str_function(serial);
2393 nvme_subsys_show_str_function(firmware_rev);
2395 static struct attribute *nvme_subsys_attrs[] = {
2396 &subsys_attr_model.attr,
2397 &subsys_attr_serial.attr,
2398 &subsys_attr_firmware_rev.attr,
2399 &subsys_attr_subsysnqn.attr,
2400 #ifdef CONFIG_NVME_MULTIPATH
2401 &subsys_attr_iopolicy.attr,
2406 static struct attribute_group nvme_subsys_attrs_group = {
2407 .attrs = nvme_subsys_attrs,
2410 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2411 &nvme_subsys_attrs_group,
2415 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2416 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2418 struct nvme_ctrl *tmp;
2420 lockdep_assert_held(&nvme_subsystems_lock);
2422 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2423 if (tmp->state == NVME_CTRL_DELETING ||
2424 tmp->state == NVME_CTRL_DEAD)
2427 if (tmp->cntlid == ctrl->cntlid) {
2428 dev_err(ctrl->device,
2429 "Duplicate cntlid %u with %s, rejecting\n",
2430 ctrl->cntlid, dev_name(tmp->device));
2434 if ((id->cmic & (1 << 1)) ||
2435 (ctrl->opts && ctrl->opts->discovery_nqn))
2438 dev_err(ctrl->device,
2439 "Subsystem does not support multiple controllers\n");
2446 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2448 struct nvme_subsystem *subsys, *found;
2451 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2454 ret = ida_simple_get(&nvme_subsystems_ida, 0, 0, GFP_KERNEL);
2459 subsys->instance = ret;
2460 mutex_init(&subsys->lock);
2461 kref_init(&subsys->ref);
2462 INIT_LIST_HEAD(&subsys->ctrls);
2463 INIT_LIST_HEAD(&subsys->nsheads);
2464 nvme_init_subnqn(subsys, ctrl, id);
2465 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2466 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2467 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2468 subsys->vendor_id = le16_to_cpu(id->vid);
2469 subsys->cmic = id->cmic;
2470 subsys->awupf = le16_to_cpu(id->awupf);
2471 #ifdef CONFIG_NVME_MULTIPATH
2472 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2475 subsys->dev.class = nvme_subsys_class;
2476 subsys->dev.release = nvme_release_subsystem;
2477 subsys->dev.groups = nvme_subsys_attrs_groups;
2478 dev_set_name(&subsys->dev, "nvme-subsys%d", subsys->instance);
2479 device_initialize(&subsys->dev);
2481 mutex_lock(&nvme_subsystems_lock);
2482 found = __nvme_find_get_subsystem(subsys->subnqn);
2484 put_device(&subsys->dev);
2487 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2489 goto out_put_subsystem;
2492 ret = device_add(&subsys->dev);
2494 dev_err(ctrl->device,
2495 "failed to register subsystem device.\n");
2496 put_device(&subsys->dev);
2499 ida_init(&subsys->ns_ida);
2500 list_add_tail(&subsys->entry, &nvme_subsystems);
2503 if (sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2504 dev_name(ctrl->device))) {
2505 dev_err(ctrl->device,
2506 "failed to create sysfs link from subsystem.\n");
2507 goto out_put_subsystem;
2510 ctrl->subsys = subsys;
2511 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2512 mutex_unlock(&nvme_subsystems_lock);
2516 nvme_put_subsystem(subsys);
2518 mutex_unlock(&nvme_subsystems_lock);
2522 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2523 void *log, size_t size, u64 offset)
2525 struct nvme_command c = { };
2526 unsigned long dwlen = size / 4 - 1;
2528 c.get_log_page.opcode = nvme_admin_get_log_page;
2529 c.get_log_page.nsid = cpu_to_le32(nsid);
2530 c.get_log_page.lid = log_page;
2531 c.get_log_page.lsp = lsp;
2532 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2533 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2534 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2535 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2537 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2540 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2545 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2550 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2551 ctrl->effects, sizeof(*ctrl->effects), 0);
2553 kfree(ctrl->effects);
2554 ctrl->effects = NULL;
2560 * Initialize the cached copies of the Identify data and various controller
2561 * register in our nvme_ctrl structure. This should be called as soon as
2562 * the admin queue is fully up and running.
2564 int nvme_init_identify(struct nvme_ctrl *ctrl)
2566 struct nvme_id_ctrl *id;
2568 int ret, page_shift;
2570 bool prev_apst_enabled;
2572 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2574 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2578 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &cap);
2580 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2583 page_shift = NVME_CAP_MPSMIN(cap) + 12;
2585 if (ctrl->vs >= NVME_VS(1, 1, 0))
2586 ctrl->subsystem = NVME_CAP_NSSRC(cap);
2588 ret = nvme_identify_ctrl(ctrl, &id);
2590 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2594 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2595 ret = nvme_get_effects_log(ctrl);
2600 if (!ctrl->identified) {
2603 ret = nvme_init_subsystem(ctrl, id);
2608 * Check for quirks. Quirk can depend on firmware version,
2609 * so, in principle, the set of quirks present can change
2610 * across a reset. As a possible future enhancement, we
2611 * could re-scan for quirks every time we reinitialize
2612 * the device, but we'd have to make sure that the driver
2613 * behaves intelligently if the quirks change.
2615 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2616 if (quirk_matches(id, &core_quirks[i]))
2617 ctrl->quirks |= core_quirks[i].quirks;
2621 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2622 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2623 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2626 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2627 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2628 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2630 ctrl->oacs = le16_to_cpu(id->oacs);
2631 ctrl->oncs = le16_to_cpu(id->oncs);
2632 ctrl->mtfa = le16_to_cpu(id->mtfa);
2633 ctrl->oaes = le32_to_cpu(id->oaes);
2634 atomic_set(&ctrl->abort_limit, id->acl + 1);
2635 ctrl->vwc = id->vwc;
2637 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2639 max_hw_sectors = UINT_MAX;
2640 ctrl->max_hw_sectors =
2641 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2643 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2644 ctrl->sgls = le32_to_cpu(id->sgls);
2645 ctrl->kas = le16_to_cpu(id->kas);
2646 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2647 ctrl->ctratt = le32_to_cpu(id->ctratt);
2651 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2653 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2654 shutdown_timeout, 60);
2656 if (ctrl->shutdown_timeout != shutdown_timeout)
2657 dev_info(ctrl->device,
2658 "Shutdown timeout set to %u seconds\n",
2659 ctrl->shutdown_timeout);
2661 ctrl->shutdown_timeout = shutdown_timeout;
2663 ctrl->npss = id->npss;
2664 ctrl->apsta = id->apsta;
2665 prev_apst_enabled = ctrl->apst_enabled;
2666 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2667 if (force_apst && id->apsta) {
2668 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2669 ctrl->apst_enabled = true;
2671 ctrl->apst_enabled = false;
2674 ctrl->apst_enabled = id->apsta;
2676 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2678 if (ctrl->ops->flags & NVME_F_FABRICS) {
2679 ctrl->icdoff = le16_to_cpu(id->icdoff);
2680 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2681 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2682 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2685 * In fabrics we need to verify the cntlid matches the
2688 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2693 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2694 dev_err(ctrl->device,
2695 "keep-alive support is mandatory for fabrics\n");
2700 ctrl->cntlid = le16_to_cpu(id->cntlid);
2701 ctrl->hmpre = le32_to_cpu(id->hmpre);
2702 ctrl->hmmin = le32_to_cpu(id->hmmin);
2703 ctrl->hmminds = le32_to_cpu(id->hmminds);
2704 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2707 ret = nvme_mpath_init(ctrl, id);
2713 if (ctrl->apst_enabled && !prev_apst_enabled)
2714 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2715 else if (!ctrl->apst_enabled && prev_apst_enabled)
2716 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2718 ret = nvme_configure_apst(ctrl);
2722 ret = nvme_configure_timestamp(ctrl);
2726 ret = nvme_configure_directives(ctrl);
2730 ret = nvme_configure_acre(ctrl);
2734 ctrl->identified = true;
2742 EXPORT_SYMBOL_GPL(nvme_init_identify);
2744 static int nvme_dev_open(struct inode *inode, struct file *file)
2746 struct nvme_ctrl *ctrl =
2747 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2749 switch (ctrl->state) {
2750 case NVME_CTRL_LIVE:
2751 case NVME_CTRL_ADMIN_ONLY:
2754 return -EWOULDBLOCK;
2757 file->private_data = ctrl;
2761 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2766 down_read(&ctrl->namespaces_rwsem);
2767 if (list_empty(&ctrl->namespaces)) {
2772 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2773 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2774 dev_warn(ctrl->device,
2775 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2780 dev_warn(ctrl->device,
2781 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2782 kref_get(&ns->kref);
2783 up_read(&ctrl->namespaces_rwsem);
2785 ret = nvme_user_cmd(ctrl, ns, argp);
2790 up_read(&ctrl->namespaces_rwsem);
2794 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2797 struct nvme_ctrl *ctrl = file->private_data;
2798 void __user *argp = (void __user *)arg;
2801 case NVME_IOCTL_ADMIN_CMD:
2802 return nvme_user_cmd(ctrl, NULL, argp);
2803 case NVME_IOCTL_IO_CMD:
2804 return nvme_dev_user_cmd(ctrl, argp);
2805 case NVME_IOCTL_RESET:
2806 dev_warn(ctrl->device, "resetting controller\n");
2807 return nvme_reset_ctrl_sync(ctrl);
2808 case NVME_IOCTL_SUBSYS_RESET:
2809 return nvme_reset_subsystem(ctrl);
2810 case NVME_IOCTL_RESCAN:
2811 nvme_queue_scan(ctrl);
2818 static const struct file_operations nvme_dev_fops = {
2819 .owner = THIS_MODULE,
2820 .open = nvme_dev_open,
2821 .unlocked_ioctl = nvme_dev_ioctl,
2822 .compat_ioctl = nvme_dev_ioctl,
2825 static ssize_t nvme_sysfs_reset(struct device *dev,
2826 struct device_attribute *attr, const char *buf,
2829 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2832 ret = nvme_reset_ctrl_sync(ctrl);
2837 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2839 static ssize_t nvme_sysfs_rescan(struct device *dev,
2840 struct device_attribute *attr, const char *buf,
2843 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2845 nvme_queue_scan(ctrl);
2848 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2850 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2852 struct gendisk *disk = dev_to_disk(dev);
2854 if (disk->fops == &nvme_fops)
2855 return nvme_get_ns_from_dev(dev)->head;
2857 return disk->private_data;
2860 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2863 struct nvme_ns_head *head = dev_to_ns_head(dev);
2864 struct nvme_ns_ids *ids = &head->ids;
2865 struct nvme_subsystem *subsys = head->subsys;
2866 int serial_len = sizeof(subsys->serial);
2867 int model_len = sizeof(subsys->model);
2869 if (!uuid_is_null(&ids->uuid))
2870 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2872 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2873 return sprintf(buf, "eui.%16phN\n", ids->nguid);
2875 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2876 return sprintf(buf, "eui.%8phN\n", ids->eui64);
2878 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
2879 subsys->serial[serial_len - 1] == '\0'))
2881 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
2882 subsys->model[model_len - 1] == '\0'))
2885 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
2886 serial_len, subsys->serial, model_len, subsys->model,
2889 static DEVICE_ATTR_RO(wwid);
2891 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
2894 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
2896 static DEVICE_ATTR_RO(nguid);
2898 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
2901 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2903 /* For backward compatibility expose the NGUID to userspace if
2904 * we have no UUID set
2906 if (uuid_is_null(&ids->uuid)) {
2907 printk_ratelimited(KERN_WARNING
2908 "No UUID available providing old NGUID\n");
2909 return sprintf(buf, "%pU\n", ids->nguid);
2911 return sprintf(buf, "%pU\n", &ids->uuid);
2913 static DEVICE_ATTR_RO(uuid);
2915 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
2918 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
2920 static DEVICE_ATTR_RO(eui);
2922 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
2925 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
2927 static DEVICE_ATTR_RO(nsid);
2929 static struct attribute *nvme_ns_id_attrs[] = {
2930 &dev_attr_wwid.attr,
2931 &dev_attr_uuid.attr,
2932 &dev_attr_nguid.attr,
2934 &dev_attr_nsid.attr,
2935 #ifdef CONFIG_NVME_MULTIPATH
2936 &dev_attr_ana_grpid.attr,
2937 &dev_attr_ana_state.attr,
2942 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
2943 struct attribute *a, int n)
2945 struct device *dev = container_of(kobj, struct device, kobj);
2946 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
2948 if (a == &dev_attr_uuid.attr) {
2949 if (uuid_is_null(&ids->uuid) &&
2950 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2953 if (a == &dev_attr_nguid.attr) {
2954 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
2957 if (a == &dev_attr_eui.attr) {
2958 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
2961 #ifdef CONFIG_NVME_MULTIPATH
2962 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
2963 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
2965 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
2972 static const struct attribute_group nvme_ns_id_attr_group = {
2973 .attrs = nvme_ns_id_attrs,
2974 .is_visible = nvme_ns_id_attrs_are_visible,
2977 const struct attribute_group *nvme_ns_id_attr_groups[] = {
2978 &nvme_ns_id_attr_group,
2980 &nvme_nvm_attr_group,
2985 #define nvme_show_str_function(field) \
2986 static ssize_t field##_show(struct device *dev, \
2987 struct device_attribute *attr, char *buf) \
2989 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
2990 return sprintf(buf, "%.*s\n", \
2991 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
2993 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
2995 nvme_show_str_function(model);
2996 nvme_show_str_function(serial);
2997 nvme_show_str_function(firmware_rev);
2999 #define nvme_show_int_function(field) \
3000 static ssize_t field##_show(struct device *dev, \
3001 struct device_attribute *attr, char *buf) \
3003 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3004 return sprintf(buf, "%d\n", ctrl->field); \
3006 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3008 nvme_show_int_function(cntlid);
3009 nvme_show_int_function(numa_node);
3011 static ssize_t nvme_sysfs_delete(struct device *dev,
3012 struct device_attribute *attr, const char *buf,
3015 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3017 if (device_remove_file_self(dev, attr))
3018 nvme_delete_ctrl_sync(ctrl);
3021 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3023 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3024 struct device_attribute *attr,
3027 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3029 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3031 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3033 static ssize_t nvme_sysfs_show_state(struct device *dev,
3034 struct device_attribute *attr,
3037 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3038 static const char *const state_name[] = {
3039 [NVME_CTRL_NEW] = "new",
3040 [NVME_CTRL_LIVE] = "live",
3041 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3042 [NVME_CTRL_RESETTING] = "resetting",
3043 [NVME_CTRL_CONNECTING] = "connecting",
3044 [NVME_CTRL_DELETING] = "deleting",
3045 [NVME_CTRL_DEAD] = "dead",
3048 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3049 state_name[ctrl->state])
3050 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3052 return sprintf(buf, "unknown state\n");
3055 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3057 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3058 struct device_attribute *attr,
3061 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3063 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3065 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3067 static ssize_t nvme_sysfs_show_address(struct device *dev,
3068 struct device_attribute *attr,
3071 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3073 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3075 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3077 static struct attribute *nvme_dev_attrs[] = {
3078 &dev_attr_reset_controller.attr,
3079 &dev_attr_rescan_controller.attr,
3080 &dev_attr_model.attr,
3081 &dev_attr_serial.attr,
3082 &dev_attr_firmware_rev.attr,
3083 &dev_attr_cntlid.attr,
3084 &dev_attr_delete_controller.attr,
3085 &dev_attr_transport.attr,
3086 &dev_attr_subsysnqn.attr,
3087 &dev_attr_address.attr,
3088 &dev_attr_state.attr,
3089 &dev_attr_numa_node.attr,
3093 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3094 struct attribute *a, int n)
3096 struct device *dev = container_of(kobj, struct device, kobj);
3097 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3099 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3101 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3107 static struct attribute_group nvme_dev_attrs_group = {
3108 .attrs = nvme_dev_attrs,
3109 .is_visible = nvme_dev_attrs_are_visible,
3112 static const struct attribute_group *nvme_dev_attr_groups[] = {
3113 &nvme_dev_attrs_group,
3117 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3120 struct nvme_ns_head *h;
3122 lockdep_assert_held(&subsys->lock);
3124 list_for_each_entry(h, &subsys->nsheads, entry) {
3125 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3132 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3133 struct nvme_ns_head *new)
3135 struct nvme_ns_head *h;
3137 lockdep_assert_held(&subsys->lock);
3139 list_for_each_entry(h, &subsys->nsheads, entry) {
3140 if (nvme_ns_ids_valid(&new->ids) &&
3141 !list_empty(&h->list) &&
3142 nvme_ns_ids_equal(&new->ids, &h->ids))
3149 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3150 unsigned nsid, struct nvme_id_ns *id)
3152 struct nvme_ns_head *head;
3153 size_t size = sizeof(*head);
3156 #ifdef CONFIG_NVME_MULTIPATH
3157 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3160 head = kzalloc(size, GFP_KERNEL);
3163 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3166 head->instance = ret;
3167 INIT_LIST_HEAD(&head->list);
3168 ret = init_srcu_struct(&head->srcu);
3170 goto out_ida_remove;
3171 head->subsys = ctrl->subsys;
3173 kref_init(&head->ref);
3175 nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3177 ret = __nvme_check_ids(ctrl->subsys, head);
3179 dev_err(ctrl->device,
3180 "duplicate IDs for nsid %d\n", nsid);
3181 goto out_cleanup_srcu;
3184 ret = nvme_mpath_alloc_disk(ctrl, head);
3186 goto out_cleanup_srcu;
3188 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3190 kref_get(&ctrl->subsys->ref);
3194 cleanup_srcu_struct(&head->srcu);
3196 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3200 return ERR_PTR(ret);
3203 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3204 struct nvme_id_ns *id)
3206 struct nvme_ctrl *ctrl = ns->ctrl;
3207 bool is_shared = id->nmic & (1 << 0);
3208 struct nvme_ns_head *head = NULL;
3211 mutex_lock(&ctrl->subsys->lock);
3213 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3215 head = nvme_alloc_ns_head(ctrl, nsid, id);
3217 ret = PTR_ERR(head);
3221 struct nvme_ns_ids ids;
3223 nvme_report_ns_ids(ctrl, nsid, id, &ids);
3224 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3225 dev_err(ctrl->device,
3226 "IDs don't match for shared namespace %d\n",
3233 list_add_tail(&ns->siblings, &head->list);
3237 mutex_unlock(&ctrl->subsys->lock);
3241 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3243 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3244 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3246 return nsa->head->ns_id - nsb->head->ns_id;
3249 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3251 struct nvme_ns *ns, *ret = NULL;
3253 down_read(&ctrl->namespaces_rwsem);
3254 list_for_each_entry(ns, &ctrl->namespaces, list) {
3255 if (ns->head->ns_id == nsid) {
3256 if (!kref_get_unless_zero(&ns->kref))
3261 if (ns->head->ns_id > nsid)
3264 up_read(&ctrl->namespaces_rwsem);
3268 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3270 struct streams_directive_params s;
3273 if (!ctrl->nr_streams)
3276 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3280 ns->sws = le32_to_cpu(s.sws);
3281 ns->sgs = le16_to_cpu(s.sgs);
3284 unsigned int bs = 1 << ns->lba_shift;
3286 blk_queue_io_min(ns->queue, bs * ns->sws);
3288 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3294 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3297 struct gendisk *disk;
3298 struct nvme_id_ns *id;
3299 char disk_name[DISK_NAME_LEN];
3300 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3302 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3306 ns->queue = blk_mq_init_queue(ctrl->tagset);
3307 if (IS_ERR(ns->queue)) {
3308 ret = PTR_ERR(ns->queue);
3312 if (ctrl->opts && ctrl->opts->data_digest)
3313 ns->queue->backing_dev_info->capabilities
3314 |= BDI_CAP_STABLE_WRITES;
3316 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3317 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3318 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3320 ns->queue->queuedata = ns;
3323 kref_init(&ns->kref);
3324 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3326 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3327 nvme_set_queue_limits(ctrl, ns->queue);
3329 id = nvme_identify_ns(ctrl, nsid);
3332 goto out_free_queue;
3335 if (id->ncap == 0) {
3340 ret = nvme_init_ns_head(ns, nsid, id);
3343 nvme_setup_streams_ns(ctrl, ns);
3344 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3346 disk = alloc_disk_node(0, node);
3352 disk->fops = &nvme_fops;
3353 disk->private_data = ns;
3354 disk->queue = ns->queue;
3355 disk->flags = flags;
3356 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3359 __nvme_revalidate_disk(disk, id);
3361 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3362 ret = nvme_nvm_register(ns, disk_name, node);
3364 dev_warn(ctrl->device, "LightNVM init failure\n");
3369 down_write(&ctrl->namespaces_rwsem);
3370 list_add_tail(&ns->list, &ctrl->namespaces);
3371 up_write(&ctrl->namespaces_rwsem);
3373 nvme_get_ctrl(ctrl);
3375 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3377 nvme_mpath_add_disk(ns, id);
3378 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3385 mutex_lock(&ctrl->subsys->lock);
3386 list_del_rcu(&ns->siblings);
3387 mutex_unlock(&ctrl->subsys->lock);
3388 nvme_put_ns_head(ns->head);
3392 blk_cleanup_queue(ns->queue);
3398 static void nvme_ns_remove(struct nvme_ns *ns)
3400 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3403 nvme_fault_inject_fini(&ns->fault_inject);
3405 mutex_lock(&ns->ctrl->subsys->lock);
3406 list_del_rcu(&ns->siblings);
3407 mutex_unlock(&ns->ctrl->subsys->lock);
3408 synchronize_rcu(); /* guarantee not available in head->list */
3409 nvme_mpath_clear_current_path(ns);
3410 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3412 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3413 del_gendisk(ns->disk);
3414 blk_cleanup_queue(ns->queue);
3415 if (blk_get_integrity(ns->disk))
3416 blk_integrity_unregister(ns->disk);
3419 down_write(&ns->ctrl->namespaces_rwsem);
3420 list_del_init(&ns->list);
3421 up_write(&ns->ctrl->namespaces_rwsem);
3423 nvme_mpath_check_last_path(ns);
3427 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3431 ns = nvme_find_get_ns(ctrl, nsid);
3433 if (ns->disk && revalidate_disk(ns->disk))
3437 nvme_alloc_ns(ctrl, nsid);
3440 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3443 struct nvme_ns *ns, *next;
3446 down_write(&ctrl->namespaces_rwsem);
3447 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3448 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3449 list_move_tail(&ns->list, &rm_list);
3451 up_write(&ctrl->namespaces_rwsem);
3453 list_for_each_entry_safe(ns, next, &rm_list, list)
3458 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3462 unsigned i, j, nsid, prev = 0;
3463 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3466 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3470 for (i = 0; i < num_lists; i++) {
3471 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3475 for (j = 0; j < min(nn, 1024U); j++) {
3476 nsid = le32_to_cpu(ns_list[j]);
3480 nvme_validate_ns(ctrl, nsid);
3482 while (++prev < nsid) {
3483 ns = nvme_find_get_ns(ctrl, prev);
3493 nvme_remove_invalid_namespaces(ctrl, prev);
3499 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3503 for (i = 1; i <= nn; i++)
3504 nvme_validate_ns(ctrl, i);
3506 nvme_remove_invalid_namespaces(ctrl, nn);
3509 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3511 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3515 log = kzalloc(log_size, GFP_KERNEL);
3520 * We need to read the log to clear the AEN, but we don't want to rely
3521 * on it for the changed namespace information as userspace could have
3522 * raced with us in reading the log page, which could cause us to miss
3525 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3528 dev_warn(ctrl->device,
3529 "reading changed ns log failed: %d\n", error);
3534 static void nvme_scan_work(struct work_struct *work)
3536 struct nvme_ctrl *ctrl =
3537 container_of(work, struct nvme_ctrl, scan_work);
3538 struct nvme_id_ctrl *id;
3541 if (ctrl->state != NVME_CTRL_LIVE)
3544 WARN_ON_ONCE(!ctrl->tagset);
3546 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3547 dev_info(ctrl->device, "rescanning namespaces.\n");
3548 nvme_clear_changed_ns_log(ctrl);
3551 if (nvme_identify_ctrl(ctrl, &id))
3554 mutex_lock(&ctrl->scan_lock);
3555 nn = le32_to_cpu(id->nn);
3556 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3557 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3558 if (!nvme_scan_ns_list(ctrl, nn))
3561 nvme_scan_ns_sequential(ctrl, nn);
3563 mutex_unlock(&ctrl->scan_lock);
3565 down_write(&ctrl->namespaces_rwsem);
3566 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3567 up_write(&ctrl->namespaces_rwsem);
3571 * This function iterates the namespace list unlocked to allow recovery from
3572 * controller failure. It is up to the caller to ensure the namespace list is
3573 * not modified by scan work while this function is executing.
3575 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3577 struct nvme_ns *ns, *next;
3580 /* prevent racing with ns scanning */
3581 flush_work(&ctrl->scan_work);
3584 * The dead states indicates the controller was not gracefully
3585 * disconnected. In that case, we won't be able to flush any data while
3586 * removing the namespaces' disks; fail all the queues now to avoid
3587 * potentially having to clean up the failed sync later.
3589 if (ctrl->state == NVME_CTRL_DEAD)
3590 nvme_kill_queues(ctrl);
3592 down_write(&ctrl->namespaces_rwsem);
3593 list_splice_init(&ctrl->namespaces, &ns_list);
3594 up_write(&ctrl->namespaces_rwsem);
3596 list_for_each_entry_safe(ns, next, &ns_list, list)
3599 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3601 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3603 char *envp[2] = { NULL, NULL };
3604 u32 aen_result = ctrl->aen_result;
3606 ctrl->aen_result = 0;
3610 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3613 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3617 static void nvme_async_event_work(struct work_struct *work)
3619 struct nvme_ctrl *ctrl =
3620 container_of(work, struct nvme_ctrl, async_event_work);
3622 nvme_aen_uevent(ctrl);
3623 ctrl->ops->submit_async_event(ctrl);
3626 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3631 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3637 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3640 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3642 struct nvme_fw_slot_info_log *log;
3644 log = kmalloc(sizeof(*log), GFP_KERNEL);
3648 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3650 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3654 static void nvme_fw_act_work(struct work_struct *work)
3656 struct nvme_ctrl *ctrl = container_of(work,
3657 struct nvme_ctrl, fw_act_work);
3658 unsigned long fw_act_timeout;
3661 fw_act_timeout = jiffies +
3662 msecs_to_jiffies(ctrl->mtfa * 100);
3664 fw_act_timeout = jiffies +
3665 msecs_to_jiffies(admin_timeout * 1000);
3667 nvme_stop_queues(ctrl);
3668 while (nvme_ctrl_pp_status(ctrl)) {
3669 if (time_after(jiffies, fw_act_timeout)) {
3670 dev_warn(ctrl->device,
3671 "Fw activation timeout, reset controller\n");
3672 nvme_reset_ctrl(ctrl);
3678 if (ctrl->state != NVME_CTRL_LIVE)
3681 nvme_start_queues(ctrl);
3682 /* read FW slot information to clear the AER */
3683 nvme_get_fw_slot_info(ctrl);
3686 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3688 u32 aer_notice_type = (result & 0xff00) >> 8;
3690 trace_nvme_async_event(ctrl, aer_notice_type);
3692 switch (aer_notice_type) {
3693 case NVME_AER_NOTICE_NS_CHANGED:
3694 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3695 nvme_queue_scan(ctrl);
3697 case NVME_AER_NOTICE_FW_ACT_STARTING:
3698 queue_work(nvme_wq, &ctrl->fw_act_work);
3700 #ifdef CONFIG_NVME_MULTIPATH
3701 case NVME_AER_NOTICE_ANA:
3702 if (!ctrl->ana_log_buf)
3704 queue_work(nvme_wq, &ctrl->ana_work);
3708 dev_warn(ctrl->device, "async event result %08x\n", result);
3712 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3713 volatile union nvme_result *res)
3715 u32 result = le32_to_cpu(res->u32);
3716 u32 aer_type = result & 0x07;
3718 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3722 case NVME_AER_NOTICE:
3723 nvme_handle_aen_notice(ctrl, result);
3725 case NVME_AER_ERROR:
3726 case NVME_AER_SMART:
3729 trace_nvme_async_event(ctrl, aer_type);
3730 ctrl->aen_result = result;
3735 queue_work(nvme_wq, &ctrl->async_event_work);
3737 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3739 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3741 nvme_mpath_stop(ctrl);
3742 nvme_stop_keep_alive(ctrl);
3743 flush_work(&ctrl->async_event_work);
3744 cancel_work_sync(&ctrl->fw_act_work);
3746 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3748 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3751 nvme_start_keep_alive(ctrl);
3753 if (ctrl->queue_count > 1) {
3754 nvme_queue_scan(ctrl);
3755 nvme_enable_aen(ctrl);
3756 queue_work(nvme_wq, &ctrl->async_event_work);
3757 nvme_start_queues(ctrl);
3760 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3762 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3764 nvme_fault_inject_fini(&ctrl->fault_inject);
3765 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3766 cdev_device_del(&ctrl->cdev, ctrl->device);
3768 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3770 static void nvme_free_ctrl(struct device *dev)
3772 struct nvme_ctrl *ctrl =
3773 container_of(dev, struct nvme_ctrl, ctrl_device);
3774 struct nvme_subsystem *subsys = ctrl->subsys;
3776 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3777 kfree(ctrl->effects);
3778 nvme_mpath_uninit(ctrl);
3779 __free_page(ctrl->discard_page);
3782 mutex_lock(&nvme_subsystems_lock);
3783 list_del(&ctrl->subsys_entry);
3784 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3785 mutex_unlock(&nvme_subsystems_lock);
3788 ctrl->ops->free_ctrl(ctrl);
3791 nvme_put_subsystem(subsys);
3795 * Initialize a NVMe controller structures. This needs to be called during
3796 * earliest initialization so that we have the initialized structured around
3799 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3800 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3804 ctrl->state = NVME_CTRL_NEW;
3805 spin_lock_init(&ctrl->lock);
3806 mutex_init(&ctrl->scan_lock);
3807 INIT_LIST_HEAD(&ctrl->namespaces);
3808 init_rwsem(&ctrl->namespaces_rwsem);
3811 ctrl->quirks = quirks;
3812 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3813 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3814 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3815 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3817 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3818 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3819 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
3821 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
3823 ctrl->discard_page = alloc_page(GFP_KERNEL);
3824 if (!ctrl->discard_page) {
3829 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
3832 ctrl->instance = ret;
3834 device_initialize(&ctrl->ctrl_device);
3835 ctrl->device = &ctrl->ctrl_device;
3836 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
3837 ctrl->device->class = nvme_class;
3838 ctrl->device->parent = ctrl->dev;
3839 ctrl->device->groups = nvme_dev_attr_groups;
3840 ctrl->device->release = nvme_free_ctrl;
3841 dev_set_drvdata(ctrl->device, ctrl);
3842 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
3844 goto out_release_instance;
3846 cdev_init(&ctrl->cdev, &nvme_dev_fops);
3847 ctrl->cdev.owner = ops->module;
3848 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
3853 * Initialize latency tolerance controls. The sysfs files won't
3854 * be visible to userspace unless the device actually supports APST.
3856 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
3857 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
3858 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
3860 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
3864 kfree_const(ctrl->device->kobj.name);
3865 out_release_instance:
3866 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3868 if (ctrl->discard_page)
3869 __free_page(ctrl->discard_page);
3872 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
3875 * nvme_kill_queues(): Ends all namespace queues
3876 * @ctrl: the dead controller that needs to end
3878 * Call this function when the driver determines it is unable to get the
3879 * controller in a state capable of servicing IO.
3881 void nvme_kill_queues(struct nvme_ctrl *ctrl)
3885 down_read(&ctrl->namespaces_rwsem);
3887 /* Forcibly unquiesce queues to avoid blocking dispatch */
3888 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
3889 blk_mq_unquiesce_queue(ctrl->admin_q);
3891 list_for_each_entry(ns, &ctrl->namespaces, list)
3892 nvme_set_queue_dying(ns);
3894 up_read(&ctrl->namespaces_rwsem);
3896 EXPORT_SYMBOL_GPL(nvme_kill_queues);
3898 void nvme_unfreeze(struct nvme_ctrl *ctrl)
3902 down_read(&ctrl->namespaces_rwsem);
3903 list_for_each_entry(ns, &ctrl->namespaces, list)
3904 blk_mq_unfreeze_queue(ns->queue);
3905 up_read(&ctrl->namespaces_rwsem);
3907 EXPORT_SYMBOL_GPL(nvme_unfreeze);
3909 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
3913 down_read(&ctrl->namespaces_rwsem);
3914 list_for_each_entry(ns, &ctrl->namespaces, list) {
3915 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
3919 up_read(&ctrl->namespaces_rwsem);
3921 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
3923 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
3927 down_read(&ctrl->namespaces_rwsem);
3928 list_for_each_entry(ns, &ctrl->namespaces, list)
3929 blk_mq_freeze_queue_wait(ns->queue);
3930 up_read(&ctrl->namespaces_rwsem);
3932 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
3934 void nvme_start_freeze(struct nvme_ctrl *ctrl)
3938 down_read(&ctrl->namespaces_rwsem);
3939 list_for_each_entry(ns, &ctrl->namespaces, list)
3940 blk_freeze_queue_start(ns->queue);
3941 up_read(&ctrl->namespaces_rwsem);
3943 EXPORT_SYMBOL_GPL(nvme_start_freeze);
3945 void nvme_stop_queues(struct nvme_ctrl *ctrl)
3949 down_read(&ctrl->namespaces_rwsem);
3950 list_for_each_entry(ns, &ctrl->namespaces, list)
3951 blk_mq_quiesce_queue(ns->queue);
3952 up_read(&ctrl->namespaces_rwsem);
3954 EXPORT_SYMBOL_GPL(nvme_stop_queues);
3956 void nvme_start_queues(struct nvme_ctrl *ctrl)
3960 down_read(&ctrl->namespaces_rwsem);
3961 list_for_each_entry(ns, &ctrl->namespaces, list)
3962 blk_mq_unquiesce_queue(ns->queue);
3963 up_read(&ctrl->namespaces_rwsem);
3965 EXPORT_SYMBOL_GPL(nvme_start_queues);
3968 void nvme_sync_queues(struct nvme_ctrl *ctrl)
3972 down_read(&ctrl->namespaces_rwsem);
3973 list_for_each_entry(ns, &ctrl->namespaces, list)
3974 blk_sync_queue(ns->queue);
3975 up_read(&ctrl->namespaces_rwsem);
3977 EXPORT_SYMBOL_GPL(nvme_sync_queues);
3980 * Check we didn't inadvertently grow the command structure sizes:
3982 static inline void _nvme_check_size(void)
3984 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
3985 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
3986 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
3987 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
3988 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
3989 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
3990 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
3991 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
3992 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
3993 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
3994 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
3995 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
3996 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
3997 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
3998 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
3999 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4000 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4004 static int __init nvme_core_init(void)
4006 int result = -ENOMEM;
4010 nvme_wq = alloc_workqueue("nvme-wq",
4011 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4015 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4016 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4020 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4021 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4022 if (!nvme_delete_wq)
4023 goto destroy_reset_wq;
4025 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4027 goto destroy_delete_wq;
4029 nvme_class = class_create(THIS_MODULE, "nvme");
4030 if (IS_ERR(nvme_class)) {
4031 result = PTR_ERR(nvme_class);
4032 goto unregister_chrdev;
4035 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4036 if (IS_ERR(nvme_subsys_class)) {
4037 result = PTR_ERR(nvme_subsys_class);
4043 class_destroy(nvme_class);
4045 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4047 destroy_workqueue(nvme_delete_wq);
4049 destroy_workqueue(nvme_reset_wq);
4051 destroy_workqueue(nvme_wq);
4056 static void __exit nvme_core_exit(void)
4058 ida_destroy(&nvme_subsystems_ida);
4059 class_destroy(nvme_subsys_class);
4060 class_destroy(nvme_class);
4061 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4062 destroy_workqueue(nvme_delete_wq);
4063 destroy_workqueue(nvme_reset_wq);
4064 destroy_workqueue(nvme_wq);
4067 MODULE_LICENSE("GPL");
4068 MODULE_VERSION("1.0");
4069 module_init(nvme_core_init);
4070 module_exit(nvme_core_exit);