1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/delay.h>
10 #include <linux/errno.h>
11 #include <linux/hdreg.h>
12 #include <linux/kernel.h>
13 #include <linux/module.h>
14 #include <linux/backing-dev.h>
15 #include <linux/list_sort.h>
16 #include <linux/slab.h>
17 #include <linux/types.h>
19 #include <linux/ptrace.h>
20 #include <linux/nvme_ioctl.h>
21 #include <linux/t10-pi.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
61 module_param(streams, bool, 0644);
62 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
65 * nvme_wq - hosts nvme related works that are not reset or delete
66 * nvme_reset_wq - hosts nvme reset works
67 * nvme_delete_wq - hosts nvme delete works
69 * nvme_wq will host works such are scan, aen handling, fw activation,
70 * keep-alive error recovery, periodic reconnects etc. nvme_reset_wq
71 * runs reset works which also flush works hosted on nvme_wq for
72 * serialization purposes. nvme_delete_wq host controller deletion
73 * works which flush reset works for serialization.
75 struct workqueue_struct *nvme_wq;
76 EXPORT_SYMBOL_GPL(nvme_wq);
78 struct workqueue_struct *nvme_reset_wq;
79 EXPORT_SYMBOL_GPL(nvme_reset_wq);
81 struct workqueue_struct *nvme_delete_wq;
82 EXPORT_SYMBOL_GPL(nvme_delete_wq);
84 static LIST_HEAD(nvme_subsystems);
85 static DEFINE_MUTEX(nvme_subsystems_lock);
87 static DEFINE_IDA(nvme_instance_ida);
88 static dev_t nvme_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_set_queue_dying(struct nvme_ns *ns)
100 * Revalidating a dead namespace sets capacity to 0. This will end
101 * buffered writers dirtying pages that can't be synced.
103 if (!ns->disk || test_and_set_bit(NVME_NS_DEAD, &ns->flags))
105 blk_set_queue_dying(ns->queue);
106 /* Forcibly unquiesce queues to avoid blocking dispatch */
107 blk_mq_unquiesce_queue(ns->queue);
109 * Revalidate after unblocking dispatchers that may be holding bd_butex
111 revalidate_disk(ns->disk);
114 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
117 * Only new queue scan work when admin and IO queues are both alive
119 if (ctrl->state == NVME_CTRL_LIVE)
120 queue_work(nvme_wq, &ctrl->scan_work);
123 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
125 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
127 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
131 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
133 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
137 ret = nvme_reset_ctrl(ctrl);
139 flush_work(&ctrl->reset_work);
140 if (ctrl->state != NVME_CTRL_LIVE &&
141 ctrl->state != NVME_CTRL_ADMIN_ONLY)
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
149 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
151 dev_info(ctrl->device,
152 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
154 flush_work(&ctrl->reset_work);
155 nvme_stop_ctrl(ctrl);
156 nvme_remove_namespaces(ctrl);
157 ctrl->ops->delete_ctrl(ctrl);
158 nvme_uninit_ctrl(ctrl);
162 static void nvme_delete_ctrl_work(struct work_struct *work)
164 struct nvme_ctrl *ctrl =
165 container_of(work, struct nvme_ctrl, delete_work);
167 nvme_do_delete_ctrl(ctrl);
170 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
172 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
174 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
178 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
180 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
185 * Keep a reference until nvme_do_delete_ctrl() complete,
186 * since ->delete_ctrl can free the controller.
189 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
192 nvme_do_delete_ctrl(ctrl);
197 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
199 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
202 static blk_status_t nvme_error_status(u16 status)
204 switch (status & 0x7ff) {
205 case NVME_SC_SUCCESS:
207 case NVME_SC_CAP_EXCEEDED:
208 return BLK_STS_NOSPC;
209 case NVME_SC_LBA_RANGE:
210 return BLK_STS_TARGET;
211 case NVME_SC_BAD_ATTRIBUTES:
212 case NVME_SC_ONCS_NOT_SUPPORTED:
213 case NVME_SC_INVALID_OPCODE:
214 case NVME_SC_INVALID_FIELD:
215 case NVME_SC_INVALID_NS:
216 return BLK_STS_NOTSUPP;
217 case NVME_SC_WRITE_FAULT:
218 case NVME_SC_READ_ERROR:
219 case NVME_SC_UNWRITTEN_BLOCK:
220 case NVME_SC_ACCESS_DENIED:
221 case NVME_SC_READ_ONLY:
222 case NVME_SC_COMPARE_FAILED:
223 return BLK_STS_MEDIUM;
224 case NVME_SC_GUARD_CHECK:
225 case NVME_SC_APPTAG_CHECK:
226 case NVME_SC_REFTAG_CHECK:
227 case NVME_SC_INVALID_PI:
228 return BLK_STS_PROTECTION;
229 case NVME_SC_RESERVATION_CONFLICT:
230 return BLK_STS_NEXUS;
231 case NVME_SC_HOST_PATH_ERROR:
232 return BLK_STS_TRANSPORT;
234 return BLK_STS_IOERR;
238 static inline bool nvme_req_needs_retry(struct request *req)
240 if (blk_noretry_request(req))
242 if (nvme_req(req)->status & NVME_SC_DNR)
244 if (nvme_req(req)->retries >= nvme_max_retries)
249 static void nvme_retry_req(struct request *req)
251 struct nvme_ns *ns = req->q->queuedata;
252 unsigned long delay = 0;
255 /* The mask and shift result must be <= 3 */
256 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
258 delay = ns->ctrl->crdt[crd - 1] * 100;
260 nvme_req(req)->retries++;
261 blk_mq_requeue_request(req, false);
262 blk_mq_delay_kick_requeue_list(req->q, delay);
265 void nvme_complete_rq(struct request *req)
267 blk_status_t status = nvme_error_status(nvme_req(req)->status);
269 trace_nvme_complete_rq(req);
271 if (nvme_req(req)->ctrl->kas)
272 nvme_req(req)->ctrl->comp_seen = true;
274 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
275 if ((req->cmd_flags & REQ_NVME_MPATH) &&
276 blk_path_error(status)) {
277 nvme_failover_req(req);
281 if (!blk_queue_dying(req->q)) {
287 nvme_trace_bio_complete(req, status);
288 blk_mq_end_request(req, status);
290 EXPORT_SYMBOL_GPL(nvme_complete_rq);
292 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
294 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
295 "Cancelling I/O %d", req->tag);
297 /* don't abort one completed request */
298 if (blk_mq_request_completed(req))
301 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
302 blk_mq_complete_request(req);
305 EXPORT_SYMBOL_GPL(nvme_cancel_request);
307 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
308 enum nvme_ctrl_state new_state)
310 enum nvme_ctrl_state old_state;
312 bool changed = false;
314 spin_lock_irqsave(&ctrl->lock, flags);
316 old_state = ctrl->state;
318 case NVME_CTRL_ADMIN_ONLY:
320 case NVME_CTRL_CONNECTING:
330 case NVME_CTRL_RESETTING:
331 case NVME_CTRL_CONNECTING:
338 case NVME_CTRL_RESETTING:
342 case NVME_CTRL_ADMIN_ONLY:
349 case NVME_CTRL_CONNECTING:
352 case NVME_CTRL_RESETTING:
359 case NVME_CTRL_DELETING:
362 case NVME_CTRL_ADMIN_ONLY:
363 case NVME_CTRL_RESETTING:
364 case NVME_CTRL_CONNECTING:
373 case NVME_CTRL_DELETING:
385 ctrl->state = new_state;
387 spin_unlock_irqrestore(&ctrl->lock, flags);
388 if (changed && ctrl->state == NVME_CTRL_LIVE)
389 nvme_kick_requeue_lists(ctrl);
392 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
394 static void nvme_free_ns_head(struct kref *ref)
396 struct nvme_ns_head *head =
397 container_of(ref, struct nvme_ns_head, ref);
399 nvme_mpath_remove_disk(head);
400 ida_simple_remove(&head->subsys->ns_ida, head->instance);
401 list_del_init(&head->entry);
402 cleanup_srcu_struct(&head->srcu);
403 nvme_put_subsystem(head->subsys);
407 static void nvme_put_ns_head(struct nvme_ns_head *head)
409 kref_put(&head->ref, nvme_free_ns_head);
412 static void nvme_free_ns(struct kref *kref)
414 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
417 nvme_nvm_unregister(ns);
420 nvme_put_ns_head(ns->head);
421 nvme_put_ctrl(ns->ctrl);
425 static void nvme_put_ns(struct nvme_ns *ns)
427 kref_put(&ns->kref, nvme_free_ns);
430 static inline void nvme_clear_nvme_request(struct request *req)
432 if (!(req->rq_flags & RQF_DONTPREP)) {
433 nvme_req(req)->retries = 0;
434 nvme_req(req)->flags = 0;
435 req->rq_flags |= RQF_DONTPREP;
439 struct request *nvme_alloc_request(struct request_queue *q,
440 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
442 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
445 if (qid == NVME_QID_ANY) {
446 req = blk_mq_alloc_request(q, op, flags);
448 req = blk_mq_alloc_request_hctx(q, op, flags,
454 req->cmd_flags |= REQ_FAILFAST_DRIVER;
455 nvme_clear_nvme_request(req);
456 nvme_req(req)->cmd = cmd;
460 EXPORT_SYMBOL_GPL(nvme_alloc_request);
462 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
464 struct nvme_command c;
466 memset(&c, 0, sizeof(c));
468 c.directive.opcode = nvme_admin_directive_send;
469 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
470 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
471 c.directive.dtype = NVME_DIR_IDENTIFY;
472 c.directive.tdtype = NVME_DIR_STREAMS;
473 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
475 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
478 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
480 return nvme_toggle_streams(ctrl, false);
483 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
485 return nvme_toggle_streams(ctrl, true);
488 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
489 struct streams_directive_params *s, u32 nsid)
491 struct nvme_command c;
493 memset(&c, 0, sizeof(c));
494 memset(s, 0, sizeof(*s));
496 c.directive.opcode = nvme_admin_directive_recv;
497 c.directive.nsid = cpu_to_le32(nsid);
498 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
499 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
500 c.directive.dtype = NVME_DIR_STREAMS;
502 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
505 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
507 struct streams_directive_params s;
510 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
515 ret = nvme_enable_streams(ctrl);
519 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
523 ctrl->nssa = le16_to_cpu(s.nssa);
524 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
525 dev_info(ctrl->device, "too few streams (%u) available\n",
527 nvme_disable_streams(ctrl);
531 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
532 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
537 * Check if 'req' has a write hint associated with it. If it does, assign
538 * a valid namespace stream to the write.
540 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
541 struct request *req, u16 *control,
544 enum rw_hint streamid = req->write_hint;
546 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
550 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
553 *control |= NVME_RW_DTYPE_STREAMS;
554 *dsmgmt |= streamid << 16;
557 if (streamid < ARRAY_SIZE(req->q->write_hints))
558 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
561 static inline void nvme_setup_flush(struct nvme_ns *ns,
562 struct nvme_command *cmnd)
564 cmnd->common.opcode = nvme_cmd_flush;
565 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
568 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
569 struct nvme_command *cmnd)
571 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
572 struct nvme_dsm_range *range;
575 range = kmalloc_array(segments, sizeof(*range),
576 GFP_ATOMIC | __GFP_NOWARN);
579 * If we fail allocation our range, fallback to the controller
580 * discard page. If that's also busy, it's safe to return
581 * busy, as we know we can make progress once that's freed.
583 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
584 return BLK_STS_RESOURCE;
586 range = page_address(ns->ctrl->discard_page);
589 __rq_for_each_bio(bio, req) {
590 u64 slba = nvme_block_nr(ns, bio->bi_iter.bi_sector);
591 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
594 range[n].cattr = cpu_to_le32(0);
595 range[n].nlb = cpu_to_le32(nlb);
596 range[n].slba = cpu_to_le64(slba);
601 if (WARN_ON_ONCE(n != segments)) {
602 if (virt_to_page(range) == ns->ctrl->discard_page)
603 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
606 return BLK_STS_IOERR;
609 cmnd->dsm.opcode = nvme_cmd_dsm;
610 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
611 cmnd->dsm.nr = cpu_to_le32(segments - 1);
612 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
614 req->special_vec.bv_page = virt_to_page(range);
615 req->special_vec.bv_offset = offset_in_page(range);
616 req->special_vec.bv_len = sizeof(*range) * segments;
617 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
622 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
623 struct request *req, struct nvme_command *cmnd)
625 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
626 return nvme_setup_discard(ns, req, cmnd);
628 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
629 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
630 cmnd->write_zeroes.slba =
631 cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
632 cmnd->write_zeroes.length =
633 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
634 cmnd->write_zeroes.control = 0;
638 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
639 struct request *req, struct nvme_command *cmnd)
641 struct nvme_ctrl *ctrl = ns->ctrl;
645 if (req->cmd_flags & REQ_FUA)
646 control |= NVME_RW_FUA;
647 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
648 control |= NVME_RW_LR;
650 if (req->cmd_flags & REQ_RAHEAD)
651 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
653 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
654 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
655 cmnd->rw.slba = cpu_to_le64(nvme_block_nr(ns, blk_rq_pos(req)));
656 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
658 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
659 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
663 * If formated with metadata, the block layer always provides a
664 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
665 * we enable the PRACT bit for protection information or set the
666 * namespace capacity to zero to prevent any I/O.
668 if (!blk_integrity_rq(req)) {
669 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
670 return BLK_STS_NOTSUPP;
671 control |= NVME_RW_PRINFO_PRACT;
674 switch (ns->pi_type) {
675 case NVME_NS_DPS_PI_TYPE3:
676 control |= NVME_RW_PRINFO_PRCHK_GUARD;
678 case NVME_NS_DPS_PI_TYPE1:
679 case NVME_NS_DPS_PI_TYPE2:
680 control |= NVME_RW_PRINFO_PRCHK_GUARD |
681 NVME_RW_PRINFO_PRCHK_REF;
682 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
687 cmnd->rw.control = cpu_to_le16(control);
688 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
692 void nvme_cleanup_cmd(struct request *req)
694 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
695 struct nvme_ns *ns = req->rq_disk->private_data;
696 struct page *page = req->special_vec.bv_page;
698 if (page == ns->ctrl->discard_page)
699 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
701 kfree(page_address(page) + req->special_vec.bv_offset);
704 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
706 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
707 struct nvme_command *cmd)
709 blk_status_t ret = BLK_STS_OK;
711 nvme_clear_nvme_request(req);
713 memset(cmd, 0, sizeof(*cmd));
714 switch (req_op(req)) {
717 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
720 nvme_setup_flush(ns, cmd);
722 case REQ_OP_WRITE_ZEROES:
723 ret = nvme_setup_write_zeroes(ns, req, cmd);
726 ret = nvme_setup_discard(ns, req, cmd);
730 ret = nvme_setup_rw(ns, req, cmd);
734 return BLK_STS_IOERR;
737 cmd->common.command_id = req->tag;
738 trace_nvme_setup_cmd(req, cmd);
741 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
743 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
745 struct completion *waiting = rq->end_io_data;
747 rq->end_io_data = NULL;
751 static void nvme_execute_rq_polled(struct request_queue *q,
752 struct gendisk *bd_disk, struct request *rq, int at_head)
754 DECLARE_COMPLETION_ONSTACK(wait);
756 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
758 rq->cmd_flags |= REQ_HIPRI;
759 rq->end_io_data = &wait;
760 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
762 while (!completion_done(&wait)) {
763 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
769 * Returns 0 on success. If the result is negative, it's a Linux error code;
770 * if the result is positive, it's an NVM Express status code
772 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
773 union nvme_result *result, void *buffer, unsigned bufflen,
774 unsigned timeout, int qid, int at_head,
775 blk_mq_req_flags_t flags, bool poll)
780 req = nvme_alloc_request(q, cmd, flags, qid);
784 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
786 if (buffer && bufflen) {
787 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
793 nvme_execute_rq_polled(req->q, NULL, req, at_head);
795 blk_execute_rq(req->q, NULL, req, at_head);
797 *result = nvme_req(req)->result;
798 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
801 ret = nvme_req(req)->status;
803 blk_mq_free_request(req);
806 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
808 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
809 void *buffer, unsigned bufflen)
811 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
812 NVME_QID_ANY, 0, 0, false);
814 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
816 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
817 unsigned len, u32 seed, bool write)
819 struct bio_integrity_payload *bip;
823 buf = kmalloc(len, GFP_KERNEL);
828 if (write && copy_from_user(buf, ubuf, len))
831 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
837 bip->bip_iter.bi_size = len;
838 bip->bip_iter.bi_sector = seed;
839 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
840 offset_in_page(buf));
850 static int nvme_submit_user_cmd(struct request_queue *q,
851 struct nvme_command *cmd, void __user *ubuffer,
852 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
853 u32 meta_seed, u64 *result, unsigned timeout)
855 bool write = nvme_is_write(cmd);
856 struct nvme_ns *ns = q->queuedata;
857 struct gendisk *disk = ns ? ns->disk : NULL;
859 struct bio *bio = NULL;
863 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
867 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
868 nvme_req(req)->flags |= NVME_REQ_USERCMD;
870 if (ubuffer && bufflen) {
871 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
877 if (disk && meta_buffer && meta_len) {
878 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
884 req->cmd_flags |= REQ_INTEGRITY;
888 blk_execute_rq(req->q, disk, req, 0);
889 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
892 ret = nvme_req(req)->status;
894 *result = le64_to_cpu(nvme_req(req)->result.u64);
895 if (meta && !ret && !write) {
896 if (copy_to_user(meta_buffer, meta, meta_len))
902 blk_rq_unmap_user(bio);
904 blk_mq_free_request(req);
908 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
910 struct nvme_ctrl *ctrl = rq->end_io_data;
912 bool startka = false;
914 blk_mq_free_request(rq);
917 dev_err(ctrl->device,
918 "failed nvme_keep_alive_end_io error=%d\n",
923 ctrl->comp_seen = false;
924 spin_lock_irqsave(&ctrl->lock, flags);
925 if (ctrl->state == NVME_CTRL_LIVE ||
926 ctrl->state == NVME_CTRL_CONNECTING)
928 spin_unlock_irqrestore(&ctrl->lock, flags);
930 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
933 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
937 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
942 rq->timeout = ctrl->kato * HZ;
943 rq->end_io_data = ctrl;
945 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
950 static void nvme_keep_alive_work(struct work_struct *work)
952 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
953 struct nvme_ctrl, ka_work);
954 bool comp_seen = ctrl->comp_seen;
956 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
957 dev_dbg(ctrl->device,
958 "reschedule traffic based keep-alive timer\n");
959 ctrl->comp_seen = false;
960 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
964 if (nvme_keep_alive(ctrl)) {
965 /* allocation failure, reset the controller */
966 dev_err(ctrl->device, "keep-alive failed\n");
967 nvme_reset_ctrl(ctrl);
972 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
974 if (unlikely(ctrl->kato == 0))
977 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
980 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
982 if (unlikely(ctrl->kato == 0))
985 cancel_delayed_work_sync(&ctrl->ka_work);
987 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
989 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
991 struct nvme_command c = { };
994 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
995 c.identify.opcode = nvme_admin_identify;
996 c.identify.cns = NVME_ID_CNS_CTRL;
998 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1002 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1003 sizeof(struct nvme_id_ctrl));
1009 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1010 struct nvme_ns_ids *ids)
1012 struct nvme_command c = { };
1018 c.identify.opcode = nvme_admin_identify;
1019 c.identify.nsid = cpu_to_le32(nsid);
1020 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1022 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1026 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1027 NVME_IDENTIFY_DATA_SIZE);
1031 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1032 struct nvme_ns_id_desc *cur = data + pos;
1037 switch (cur->nidt) {
1038 case NVME_NIDT_EUI64:
1039 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1040 dev_warn(ctrl->device,
1041 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1045 len = NVME_NIDT_EUI64_LEN;
1046 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1048 case NVME_NIDT_NGUID:
1049 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1050 dev_warn(ctrl->device,
1051 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1055 len = NVME_NIDT_NGUID_LEN;
1056 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1058 case NVME_NIDT_UUID:
1059 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1060 dev_warn(ctrl->device,
1061 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1065 len = NVME_NIDT_UUID_LEN;
1066 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1069 /* Skip unknown types */
1074 len += sizeof(*cur);
1081 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1083 struct nvme_command c = { };
1085 c.identify.opcode = nvme_admin_identify;
1086 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1087 c.identify.nsid = cpu_to_le32(nsid);
1088 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1089 NVME_IDENTIFY_DATA_SIZE);
1092 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1093 unsigned nsid, 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);
1116 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1117 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1119 struct nvme_command c;
1120 union nvme_result res;
1123 memset(&c, 0, sizeof(c));
1124 c.features.opcode = op;
1125 c.features.fid = cpu_to_le32(fid);
1126 c.features.dword11 = cpu_to_le32(dword11);
1128 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1129 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1130 if (ret >= 0 && result)
1131 *result = le32_to_cpu(res.u32);
1135 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1136 unsigned int dword11, void *buffer, size_t buflen,
1139 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1142 EXPORT_SYMBOL_GPL(nvme_set_features);
1144 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1145 unsigned int dword11, void *buffer, size_t buflen,
1148 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1151 EXPORT_SYMBOL_GPL(nvme_get_features);
1153 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1155 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1157 int status, nr_io_queues;
1159 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1165 * Degraded controllers might return an error when setting the queue
1166 * count. We still want to be able to bring them online and offer
1167 * access to the admin queue, as that might be only way to fix them up.
1170 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1173 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1174 *count = min(*count, nr_io_queues);
1179 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1181 #define NVME_AEN_SUPPORTED \
1182 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1183 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_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",
1199 queue_work(nvme_wq, &ctrl->async_event_work);
1202 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1204 struct nvme_user_io io;
1205 struct nvme_command c;
1206 unsigned length, meta_len;
1207 void __user *metadata;
1209 if (copy_from_user(&io, uio, sizeof(io)))
1214 switch (io.opcode) {
1215 case nvme_cmd_write:
1217 case nvme_cmd_compare:
1223 length = (io.nblocks + 1) << ns->lba_shift;
1224 meta_len = (io.nblocks + 1) * ns->ms;
1225 metadata = (void __user *)(uintptr_t)io.metadata;
1230 } else if (meta_len) {
1231 if ((io.metadata & 3) || !io.metadata)
1235 memset(&c, 0, sizeof(c));
1236 c.rw.opcode = io.opcode;
1237 c.rw.flags = io.flags;
1238 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1239 c.rw.slba = cpu_to_le64(io.slba);
1240 c.rw.length = cpu_to_le16(io.nblocks);
1241 c.rw.control = cpu_to_le16(io.control);
1242 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1243 c.rw.reftag = cpu_to_le32(io.reftag);
1244 c.rw.apptag = cpu_to_le16(io.apptag);
1245 c.rw.appmask = cpu_to_le16(io.appmask);
1247 return nvme_submit_user_cmd(ns->queue, &c,
1248 (void __user *)(uintptr_t)io.addr, length,
1249 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1252 static u32 nvme_known_admin_effects(u8 opcode)
1255 case nvme_admin_format_nvm:
1256 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1257 NVME_CMD_EFFECTS_CSE_MASK;
1258 case nvme_admin_sanitize_nvm:
1259 return NVME_CMD_EFFECTS_CSE_MASK;
1266 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1273 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1274 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1275 dev_warn(ctrl->device,
1276 "IO command:%02x has unhandled effects:%08x\n",
1282 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1283 effects |= nvme_known_admin_effects(opcode);
1286 * For simplicity, IO to all namespaces is quiesced even if the command
1287 * effects say only one namespace is affected.
1289 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1290 mutex_lock(&ctrl->scan_lock);
1291 mutex_lock(&ctrl->subsys->lock);
1292 nvme_mpath_start_freeze(ctrl->subsys);
1293 nvme_mpath_wait_freeze(ctrl->subsys);
1294 nvme_start_freeze(ctrl);
1295 nvme_wait_freeze(ctrl);
1300 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1304 down_read(&ctrl->namespaces_rwsem);
1305 list_for_each_entry(ns, &ctrl->namespaces, list)
1306 if (ns->disk && nvme_revalidate_disk(ns->disk))
1307 nvme_set_queue_dying(ns);
1308 up_read(&ctrl->namespaces_rwsem);
1310 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1313 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1316 * Revalidate LBA changes prior to unfreezing. This is necessary to
1317 * prevent memory corruption if a logical block size was changed by
1320 if (effects & NVME_CMD_EFFECTS_LBCC)
1321 nvme_update_formats(ctrl);
1322 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1323 nvme_unfreeze(ctrl);
1324 nvme_mpath_unfreeze(ctrl->subsys);
1325 mutex_unlock(&ctrl->subsys->lock);
1326 mutex_unlock(&ctrl->scan_lock);
1328 if (effects & NVME_CMD_EFFECTS_CCC)
1329 nvme_init_identify(ctrl);
1330 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1331 nvme_queue_scan(ctrl);
1334 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1335 struct nvme_passthru_cmd __user *ucmd)
1337 struct nvme_passthru_cmd cmd;
1338 struct nvme_command c;
1339 unsigned timeout = 0;
1344 if (!capable(CAP_SYS_ADMIN))
1346 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1351 memset(&c, 0, sizeof(c));
1352 c.common.opcode = cmd.opcode;
1353 c.common.flags = cmd.flags;
1354 c.common.nsid = cpu_to_le32(cmd.nsid);
1355 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1356 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1357 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1358 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1359 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1360 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1361 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1362 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1365 timeout = msecs_to_jiffies(cmd.timeout_ms);
1367 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1368 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1369 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1370 (void __user *)(uintptr_t)cmd.metadata,
1371 cmd.metadata_len, 0, &result, timeout);
1372 nvme_passthru_end(ctrl, effects);
1375 if (put_user(result, &ucmd->result))
1382 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1383 struct nvme_passthru_cmd64 __user *ucmd)
1385 struct nvme_passthru_cmd64 cmd;
1386 struct nvme_command c;
1387 unsigned timeout = 0;
1391 if (!capable(CAP_SYS_ADMIN))
1393 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1398 memset(&c, 0, sizeof(c));
1399 c.common.opcode = cmd.opcode;
1400 c.common.flags = cmd.flags;
1401 c.common.nsid = cpu_to_le32(cmd.nsid);
1402 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1403 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1404 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1405 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1406 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1407 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1408 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1409 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1412 timeout = msecs_to_jiffies(cmd.timeout_ms);
1414 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1415 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1416 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1417 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1418 0, &cmd.result, timeout);
1419 nvme_passthru_end(ctrl, effects);
1422 if (put_user(cmd.result, &ucmd->result))
1430 * Issue ioctl requests on the first available path. Note that unlike normal
1431 * block layer requests we will not retry failed request on another controller.
1433 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1434 struct nvme_ns_head **head, int *srcu_idx)
1436 #ifdef CONFIG_NVME_MULTIPATH
1437 if (disk->fops == &nvme_ns_head_ops) {
1440 *head = disk->private_data;
1441 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1442 ns = nvme_find_path(*head);
1444 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1450 return disk->private_data;
1453 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1456 srcu_read_unlock(&head->srcu, idx);
1459 static bool is_ctrl_ioctl(unsigned int cmd)
1461 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1463 if (is_sed_ioctl(cmd))
1468 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1470 struct nvme_ns_head *head,
1473 struct nvme_ctrl *ctrl = ns->ctrl;
1476 nvme_get_ctrl(ns->ctrl);
1477 nvme_put_ns_from_disk(head, srcu_idx);
1480 case NVME_IOCTL_ADMIN_CMD:
1481 ret = nvme_user_cmd(ctrl, NULL, argp);
1483 case NVME_IOCTL_ADMIN64_CMD:
1484 ret = nvme_user_cmd64(ctrl, NULL, argp);
1487 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1490 nvme_put_ctrl(ctrl);
1494 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1495 unsigned int cmd, unsigned long arg)
1497 struct nvme_ns_head *head = NULL;
1498 void __user *argp = (void __user *)arg;
1502 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1504 return -EWOULDBLOCK;
1507 * Handle ioctls that apply to the controller instead of the namespace
1508 * seperately and drop the ns SRCU reference early. This avoids a
1509 * deadlock when deleting namespaces using the passthrough interface.
1511 if (is_ctrl_ioctl(cmd))
1512 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1516 force_successful_syscall_return();
1517 ret = ns->head->ns_id;
1519 case NVME_IOCTL_IO_CMD:
1520 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1522 case NVME_IOCTL_SUBMIT_IO:
1523 ret = nvme_submit_io(ns, argp);
1525 case NVME_IOCTL_IO64_CMD:
1526 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1530 ret = nvme_nvm_ioctl(ns, cmd, arg);
1535 nvme_put_ns_from_disk(head, srcu_idx);
1539 static int nvme_open(struct block_device *bdev, fmode_t mode)
1541 struct nvme_ns *ns = bdev->bd_disk->private_data;
1543 #ifdef CONFIG_NVME_MULTIPATH
1544 /* should never be called due to GENHD_FL_HIDDEN */
1545 if (WARN_ON_ONCE(ns->head->disk))
1548 if (!kref_get_unless_zero(&ns->kref))
1550 if (!try_module_get(ns->ctrl->ops->module))
1561 static void nvme_release(struct gendisk *disk, fmode_t mode)
1563 struct nvme_ns *ns = disk->private_data;
1565 module_put(ns->ctrl->ops->module);
1569 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1571 /* some standard values */
1572 geo->heads = 1 << 6;
1573 geo->sectors = 1 << 5;
1574 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1578 #ifdef CONFIG_BLK_DEV_INTEGRITY
1579 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1581 struct blk_integrity integrity;
1583 memset(&integrity, 0, sizeof(integrity));
1585 case NVME_NS_DPS_PI_TYPE3:
1586 integrity.profile = &t10_pi_type3_crc;
1587 integrity.tag_size = sizeof(u16) + sizeof(u32);
1588 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1590 case NVME_NS_DPS_PI_TYPE1:
1591 case NVME_NS_DPS_PI_TYPE2:
1592 integrity.profile = &t10_pi_type1_crc;
1593 integrity.tag_size = sizeof(u16);
1594 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1597 integrity.profile = NULL;
1600 integrity.tuple_size = ms;
1601 blk_integrity_register(disk, &integrity);
1602 blk_queue_max_integrity_segments(disk->queue, 1);
1605 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1608 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1610 static void nvme_set_chunk_size(struct nvme_ns *ns)
1612 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1613 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1616 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1618 struct nvme_ctrl *ctrl = ns->ctrl;
1619 struct request_queue *queue = disk->queue;
1620 u32 size = queue_logical_block_size(queue);
1622 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1623 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1627 if (ctrl->nr_streams && ns->sws && ns->sgs)
1628 size *= ns->sws * ns->sgs;
1630 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1631 NVME_DSM_MAX_RANGES);
1633 queue->limits.discard_alignment = 0;
1634 queue->limits.discard_granularity = size;
1636 /* If discard is already enabled, don't reset queue limits */
1637 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1640 blk_queue_max_discard_sectors(queue, UINT_MAX);
1641 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1643 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1644 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1647 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1650 unsigned short bs = 1 << ns->lba_shift;
1652 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1653 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1656 * Even though NVMe spec explicitly states that MDTS is not
1657 * applicable to the write-zeroes:- "The restriction does not apply to
1658 * commands that do not transfer data between the host and the
1659 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1660 * In order to be more cautious use controller's max_hw_sectors value
1661 * to configure the maximum sectors for the write-zeroes which is
1662 * configured based on the controller's MDTS field in the
1663 * nvme_init_identify() if available.
1665 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1666 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1668 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1670 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1673 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1674 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1678 memset(ids, 0, sizeof(*ids));
1680 if (ctrl->vs >= NVME_VS(1, 1, 0))
1681 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1682 if (ctrl->vs >= NVME_VS(1, 2, 0))
1683 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1684 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1685 /* Don't treat error as fatal we potentially
1686 * already have a NGUID or EUI-64
1688 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1690 dev_warn(ctrl->device,
1691 "Identify Descriptors failed (%d)\n", ret);
1696 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1698 return !uuid_is_null(&ids->uuid) ||
1699 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1700 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1703 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1705 return uuid_equal(&a->uuid, &b->uuid) &&
1706 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1707 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1710 static void nvme_update_disk_info(struct gendisk *disk,
1711 struct nvme_ns *ns, struct nvme_id_ns *id)
1713 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1714 unsigned short bs = 1 << ns->lba_shift;
1715 u32 atomic_bs, phys_bs, io_opt;
1717 if (ns->lba_shift > PAGE_SHIFT) {
1718 /* unsupported block size, set capacity to 0 later */
1721 blk_mq_freeze_queue(disk->queue);
1722 blk_integrity_unregister(disk);
1724 if (id->nabo == 0) {
1726 * Bit 1 indicates whether NAWUPF is defined for this namespace
1727 * and whether it should be used instead of AWUPF. If NAWUPF ==
1728 * 0 then AWUPF must be used instead.
1730 if (id->nsfeat & (1 << 1) && id->nawupf)
1731 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1733 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1739 if (id->nsfeat & (1 << 4)) {
1740 /* NPWG = Namespace Preferred Write Granularity */
1741 phys_bs *= 1 + le16_to_cpu(id->npwg);
1742 /* NOWS = Namespace Optimal Write Size */
1743 io_opt *= 1 + le16_to_cpu(id->nows);
1746 blk_queue_logical_block_size(disk->queue, bs);
1748 * Linux filesystems assume writing a single physical block is
1749 * an atomic operation. Hence limit the physical block size to the
1750 * value of the Atomic Write Unit Power Fail parameter.
1752 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1753 blk_queue_io_min(disk->queue, phys_bs);
1754 blk_queue_io_opt(disk->queue, io_opt);
1756 if (ns->ms && !ns->ext &&
1757 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1758 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1759 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1760 ns->lba_shift > PAGE_SHIFT)
1763 set_capacity(disk, capacity);
1765 nvme_config_discard(disk, ns);
1766 nvme_config_write_zeroes(disk, ns);
1768 if (id->nsattr & (1 << 0))
1769 set_disk_ro(disk, true);
1771 set_disk_ro(disk, false);
1773 blk_mq_unfreeze_queue(disk->queue);
1776 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1778 struct nvme_ns *ns = disk->private_data;
1781 * If identify namespace failed, use default 512 byte block size so
1782 * block layer can use before failing read/write for 0 capacity.
1784 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1785 if (ns->lba_shift == 0)
1787 ns->noiob = le16_to_cpu(id->noiob);
1788 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1789 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1790 /* the PI implementation requires metadata equal t10 pi tuple size */
1791 if (ns->ms == sizeof(struct t10_pi_tuple))
1792 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1797 nvme_set_chunk_size(ns);
1798 nvme_update_disk_info(disk, ns, id);
1799 #ifdef CONFIG_NVME_MULTIPATH
1800 if (ns->head->disk) {
1801 nvme_update_disk_info(ns->head->disk, ns, id);
1802 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1803 revalidate_disk(ns->head->disk);
1808 static int nvme_revalidate_disk(struct gendisk *disk)
1810 struct nvme_ns *ns = disk->private_data;
1811 struct nvme_ctrl *ctrl = ns->ctrl;
1812 struct nvme_id_ns *id;
1813 struct nvme_ns_ids ids;
1816 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1817 set_capacity(disk, 0);
1821 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1825 if (id->ncap == 0) {
1830 __nvme_revalidate_disk(disk, id);
1831 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1835 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1836 dev_err(ctrl->device,
1837 "identifiers changed for nsid %d\n", ns->head->ns_id);
1845 * Only fail the function if we got a fatal error back from the
1846 * device, otherwise ignore the error and just move on.
1848 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1851 ret = blk_status_to_errno(nvme_error_status(ret));
1855 static char nvme_pr_type(enum pr_type type)
1858 case PR_WRITE_EXCLUSIVE:
1860 case PR_EXCLUSIVE_ACCESS:
1862 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1864 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1866 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1868 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1875 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1876 u64 key, u64 sa_key, u8 op)
1878 struct nvme_ns_head *head = NULL;
1880 struct nvme_command c;
1882 u8 data[16] = { 0, };
1884 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1886 return -EWOULDBLOCK;
1888 put_unaligned_le64(key, &data[0]);
1889 put_unaligned_le64(sa_key, &data[8]);
1891 memset(&c, 0, sizeof(c));
1892 c.common.opcode = op;
1893 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1894 c.common.cdw10 = cpu_to_le32(cdw10);
1896 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1897 nvme_put_ns_from_disk(head, srcu_idx);
1901 static int nvme_pr_register(struct block_device *bdev, u64 old,
1902 u64 new, unsigned flags)
1906 if (flags & ~PR_FL_IGNORE_KEY)
1909 cdw10 = old ? 2 : 0;
1910 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1911 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1912 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1915 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1916 enum pr_type type, unsigned flags)
1920 if (flags & ~PR_FL_IGNORE_KEY)
1923 cdw10 = nvme_pr_type(type) << 8;
1924 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1925 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1928 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1929 enum pr_type type, bool abort)
1931 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1932 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1935 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1937 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1938 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1941 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1943 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1944 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1947 static const struct pr_ops nvme_pr_ops = {
1948 .pr_register = nvme_pr_register,
1949 .pr_reserve = nvme_pr_reserve,
1950 .pr_release = nvme_pr_release,
1951 .pr_preempt = nvme_pr_preempt,
1952 .pr_clear = nvme_pr_clear,
1955 #ifdef CONFIG_BLK_SED_OPAL
1956 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1959 struct nvme_ctrl *ctrl = data;
1960 struct nvme_command cmd;
1962 memset(&cmd, 0, sizeof(cmd));
1964 cmd.common.opcode = nvme_admin_security_send;
1966 cmd.common.opcode = nvme_admin_security_recv;
1967 cmd.common.nsid = 0;
1968 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1969 cmd.common.cdw11 = cpu_to_le32(len);
1971 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1972 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1974 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1975 #endif /* CONFIG_BLK_SED_OPAL */
1977 static const struct block_device_operations nvme_fops = {
1978 .owner = THIS_MODULE,
1979 .ioctl = nvme_ioctl,
1980 .compat_ioctl = nvme_ioctl,
1982 .release = nvme_release,
1983 .getgeo = nvme_getgeo,
1984 .revalidate_disk= nvme_revalidate_disk,
1985 .pr_ops = &nvme_pr_ops,
1988 #ifdef CONFIG_NVME_MULTIPATH
1989 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1991 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1993 if (!kref_get_unless_zero(&head->ref))
1998 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2000 nvme_put_ns_head(disk->private_data);
2003 const struct block_device_operations nvme_ns_head_ops = {
2004 .owner = THIS_MODULE,
2005 .open = nvme_ns_head_open,
2006 .release = nvme_ns_head_release,
2007 .ioctl = nvme_ioctl,
2008 .compat_ioctl = nvme_ioctl,
2009 .getgeo = nvme_getgeo,
2010 .pr_ops = &nvme_pr_ops,
2012 #endif /* CONFIG_NVME_MULTIPATH */
2014 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2016 unsigned long timeout =
2017 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2018 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2021 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2024 if ((csts & NVME_CSTS_RDY) == bit)
2028 if (fatal_signal_pending(current))
2030 if (time_after(jiffies, timeout)) {
2031 dev_err(ctrl->device,
2032 "Device not ready; aborting %s\n", enabled ?
2033 "initialisation" : "reset");
2042 * If the device has been passed off to us in an enabled state, just clear
2043 * the enabled bit. The spec says we should set the 'shutdown notification
2044 * bits', but doing so may cause the device to complete commands to the
2045 * admin queue ... and we don't know what memory that might be pointing at!
2047 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2051 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2052 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2054 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2058 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2059 msleep(NVME_QUIRK_DELAY_AMOUNT);
2061 return nvme_wait_ready(ctrl, ctrl->cap, false);
2063 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2065 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2068 * Default to a 4K page size, with the intention to update this
2069 * path in the future to accomodate architectures with differing
2070 * kernel and IO page sizes.
2072 unsigned dev_page_min, page_shift = 12;
2075 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2077 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2080 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2082 if (page_shift < dev_page_min) {
2083 dev_err(ctrl->device,
2084 "Minimum device page size %u too large for host (%u)\n",
2085 1 << dev_page_min, 1 << page_shift);
2089 ctrl->page_size = 1 << page_shift;
2091 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2092 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2093 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2094 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2095 ctrl->ctrl_config |= NVME_CC_ENABLE;
2097 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2100 return nvme_wait_ready(ctrl, ctrl->cap, true);
2102 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2104 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2106 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2110 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2111 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2113 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2117 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2118 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2122 if (fatal_signal_pending(current))
2124 if (time_after(jiffies, timeout)) {
2125 dev_err(ctrl->device,
2126 "Device shutdown incomplete; abort shutdown\n");
2133 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2135 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2136 struct request_queue *q)
2140 if (ctrl->max_hw_sectors) {
2142 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2144 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2145 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2146 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2148 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2149 is_power_of_2(ctrl->max_hw_sectors))
2150 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2151 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2152 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2154 blk_queue_write_cache(q, vwc, vwc);
2157 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2162 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2165 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2166 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2169 dev_warn_once(ctrl->device,
2170 "could not set timestamp (%d)\n", ret);
2174 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2176 struct nvme_feat_host_behavior *host;
2179 /* Don't bother enabling the feature if retry delay is not reported */
2183 host = kzalloc(sizeof(*host), GFP_KERNEL);
2187 host->acre = NVME_ENABLE_ACRE;
2188 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2189 host, sizeof(*host), NULL);
2194 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2197 * APST (Autonomous Power State Transition) lets us program a
2198 * table of power state transitions that the controller will
2199 * perform automatically. We configure it with a simple
2200 * heuristic: we are willing to spend at most 2% of the time
2201 * transitioning between power states. Therefore, when running
2202 * in any given state, we will enter the next lower-power
2203 * non-operational state after waiting 50 * (enlat + exlat)
2204 * microseconds, as long as that state's exit latency is under
2205 * the requested maximum latency.
2207 * We will not autonomously enter any non-operational state for
2208 * which the total latency exceeds ps_max_latency_us. Users
2209 * can set ps_max_latency_us to zero to turn off APST.
2213 struct nvme_feat_auto_pst *table;
2219 * If APST isn't supported or if we haven't been initialized yet,
2220 * then don't do anything.
2225 if (ctrl->npss > 31) {
2226 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2230 table = kzalloc(sizeof(*table), GFP_KERNEL);
2234 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2235 /* Turn off APST. */
2237 dev_dbg(ctrl->device, "APST disabled\n");
2239 __le64 target = cpu_to_le64(0);
2243 * Walk through all states from lowest- to highest-power.
2244 * According to the spec, lower-numbered states use more
2245 * power. NPSS, despite the name, is the index of the
2246 * lowest-power state, not the number of states.
2248 for (state = (int)ctrl->npss; state >= 0; state--) {
2249 u64 total_latency_us, exit_latency_us, transition_ms;
2252 table->entries[state] = target;
2255 * Don't allow transitions to the deepest state
2256 * if it's quirked off.
2258 if (state == ctrl->npss &&
2259 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2263 * Is this state a useful non-operational state for
2264 * higher-power states to autonomously transition to?
2266 if (!(ctrl->psd[state].flags &
2267 NVME_PS_FLAGS_NON_OP_STATE))
2271 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2272 if (exit_latency_us > ctrl->ps_max_latency_us)
2277 le32_to_cpu(ctrl->psd[state].entry_lat);
2280 * This state is good. Use it as the APST idle
2281 * target for higher power states.
2283 transition_ms = total_latency_us + 19;
2284 do_div(transition_ms, 20);
2285 if (transition_ms > (1 << 24) - 1)
2286 transition_ms = (1 << 24) - 1;
2288 target = cpu_to_le64((state << 3) |
2289 (transition_ms << 8));
2294 if (total_latency_us > max_lat_us)
2295 max_lat_us = total_latency_us;
2301 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2303 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2304 max_ps, max_lat_us, (int)sizeof(*table), table);
2308 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2309 table, sizeof(*table), NULL);
2311 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2317 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2319 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2323 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2324 case PM_QOS_LATENCY_ANY:
2332 if (ctrl->ps_max_latency_us != latency) {
2333 ctrl->ps_max_latency_us = latency;
2334 nvme_configure_apst(ctrl);
2338 struct nvme_core_quirk_entry {
2340 * NVMe model and firmware strings are padded with spaces. For
2341 * simplicity, strings in the quirk table are padded with NULLs
2347 unsigned long quirks;
2350 static const struct nvme_core_quirk_entry core_quirks[] = {
2353 * This Toshiba device seems to die using any APST states. See:
2354 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2357 .mn = "THNSF5256GPUK TOSHIBA",
2358 .quirks = NVME_QUIRK_NO_APST,
2362 * This LiteON CL1-3D*-Q11 firmware version has a race
2363 * condition associated with actions related to suspend to idle
2364 * LiteON has resolved the problem in future firmware
2368 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2372 * This Kingston E8FK11.T firmware version has no interrupt
2373 * after resume with actions related to suspend to idle
2374 * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2378 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2382 /* match is null-terminated but idstr is space-padded. */
2383 static bool string_matches(const char *idstr, const char *match, size_t len)
2390 matchlen = strlen(match);
2391 WARN_ON_ONCE(matchlen > len);
2393 if (memcmp(idstr, match, matchlen))
2396 for (; matchlen < len; matchlen++)
2397 if (idstr[matchlen] != ' ')
2403 static bool quirk_matches(const struct nvme_id_ctrl *id,
2404 const struct nvme_core_quirk_entry *q)
2406 return q->vid == le16_to_cpu(id->vid) &&
2407 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2408 string_matches(id->fr, q->fr, sizeof(id->fr));
2411 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2412 struct nvme_id_ctrl *id)
2417 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2418 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2419 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2420 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2424 if (ctrl->vs >= NVME_VS(1, 2, 1))
2425 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2428 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2429 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2430 "nqn.2014.08.org.nvmexpress:%04x%04x",
2431 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2432 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2433 off += sizeof(id->sn);
2434 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2435 off += sizeof(id->mn);
2436 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2439 static void nvme_release_subsystem(struct device *dev)
2441 struct nvme_subsystem *subsys =
2442 container_of(dev, struct nvme_subsystem, dev);
2444 if (subsys->instance >= 0)
2445 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2449 static void nvme_destroy_subsystem(struct kref *ref)
2451 struct nvme_subsystem *subsys =
2452 container_of(ref, struct nvme_subsystem, ref);
2454 mutex_lock(&nvme_subsystems_lock);
2455 list_del(&subsys->entry);
2456 mutex_unlock(&nvme_subsystems_lock);
2458 ida_destroy(&subsys->ns_ida);
2459 device_del(&subsys->dev);
2460 put_device(&subsys->dev);
2463 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2465 kref_put(&subsys->ref, nvme_destroy_subsystem);
2468 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2470 struct nvme_subsystem *subsys;
2472 lockdep_assert_held(&nvme_subsystems_lock);
2475 * Fail matches for discovery subsystems. This results
2476 * in each discovery controller bound to a unique subsystem.
2477 * This avoids issues with validating controller values
2478 * that can only be true when there is a single unique subsystem.
2479 * There may be multiple and completely independent entities
2480 * that provide discovery controllers.
2482 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2485 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2486 if (strcmp(subsys->subnqn, subsysnqn))
2488 if (!kref_get_unless_zero(&subsys->ref))
2496 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2497 struct device_attribute subsys_attr_##_name = \
2498 __ATTR(_name, _mode, _show, NULL)
2500 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2501 struct device_attribute *attr,
2504 struct nvme_subsystem *subsys =
2505 container_of(dev, struct nvme_subsystem, dev);
2507 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2509 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2511 #define nvme_subsys_show_str_function(field) \
2512 static ssize_t subsys_##field##_show(struct device *dev, \
2513 struct device_attribute *attr, char *buf) \
2515 struct nvme_subsystem *subsys = \
2516 container_of(dev, struct nvme_subsystem, dev); \
2517 return sprintf(buf, "%.*s\n", \
2518 (int)sizeof(subsys->field), subsys->field); \
2520 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2522 nvme_subsys_show_str_function(model);
2523 nvme_subsys_show_str_function(serial);
2524 nvme_subsys_show_str_function(firmware_rev);
2526 static struct attribute *nvme_subsys_attrs[] = {
2527 &subsys_attr_model.attr,
2528 &subsys_attr_serial.attr,
2529 &subsys_attr_firmware_rev.attr,
2530 &subsys_attr_subsysnqn.attr,
2531 #ifdef CONFIG_NVME_MULTIPATH
2532 &subsys_attr_iopolicy.attr,
2537 static struct attribute_group nvme_subsys_attrs_group = {
2538 .attrs = nvme_subsys_attrs,
2541 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2542 &nvme_subsys_attrs_group,
2546 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2547 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2549 struct nvme_ctrl *tmp;
2551 lockdep_assert_held(&nvme_subsystems_lock);
2553 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2554 if (tmp->state == NVME_CTRL_DELETING ||
2555 tmp->state == NVME_CTRL_DEAD)
2558 if (tmp->cntlid == ctrl->cntlid) {
2559 dev_err(ctrl->device,
2560 "Duplicate cntlid %u with %s, rejecting\n",
2561 ctrl->cntlid, dev_name(tmp->device));
2565 if ((id->cmic & (1 << 1)) ||
2566 (ctrl->opts && ctrl->opts->discovery_nqn))
2569 dev_err(ctrl->device,
2570 "Subsystem does not support multiple controllers\n");
2577 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2579 struct nvme_subsystem *subsys, *found;
2582 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2586 subsys->instance = -1;
2587 mutex_init(&subsys->lock);
2588 kref_init(&subsys->ref);
2589 INIT_LIST_HEAD(&subsys->ctrls);
2590 INIT_LIST_HEAD(&subsys->nsheads);
2591 nvme_init_subnqn(subsys, ctrl, id);
2592 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2593 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2594 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2595 subsys->vendor_id = le16_to_cpu(id->vid);
2596 subsys->cmic = id->cmic;
2597 subsys->awupf = le16_to_cpu(id->awupf);
2598 #ifdef CONFIG_NVME_MULTIPATH
2599 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2602 subsys->dev.class = nvme_subsys_class;
2603 subsys->dev.release = nvme_release_subsystem;
2604 subsys->dev.groups = nvme_subsys_attrs_groups;
2605 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2606 device_initialize(&subsys->dev);
2608 mutex_lock(&nvme_subsystems_lock);
2609 found = __nvme_find_get_subsystem(subsys->subnqn);
2611 put_device(&subsys->dev);
2614 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2616 goto out_put_subsystem;
2619 ret = device_add(&subsys->dev);
2621 dev_err(ctrl->device,
2622 "failed to register subsystem device.\n");
2623 put_device(&subsys->dev);
2626 ida_init(&subsys->ns_ida);
2627 list_add_tail(&subsys->entry, &nvme_subsystems);
2630 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2631 dev_name(ctrl->device));
2633 dev_err(ctrl->device,
2634 "failed to create sysfs link from subsystem.\n");
2635 goto out_put_subsystem;
2639 subsys->instance = ctrl->instance;
2640 ctrl->subsys = subsys;
2641 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2642 mutex_unlock(&nvme_subsystems_lock);
2646 nvme_put_subsystem(subsys);
2648 mutex_unlock(&nvme_subsystems_lock);
2652 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2653 void *log, size_t size, u64 offset)
2655 struct nvme_command c = { };
2656 unsigned long dwlen = size / 4 - 1;
2658 c.get_log_page.opcode = nvme_admin_get_log_page;
2659 c.get_log_page.nsid = cpu_to_le32(nsid);
2660 c.get_log_page.lid = log_page;
2661 c.get_log_page.lsp = lsp;
2662 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2663 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2664 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2665 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2667 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2670 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2675 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2680 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2681 ctrl->effects, sizeof(*ctrl->effects), 0);
2683 kfree(ctrl->effects);
2684 ctrl->effects = NULL;
2690 * Initialize the cached copies of the Identify data and various controller
2691 * register in our nvme_ctrl structure. This should be called as soon as
2692 * the admin queue is fully up and running.
2694 int nvme_init_identify(struct nvme_ctrl *ctrl)
2696 struct nvme_id_ctrl *id;
2697 int ret, page_shift;
2699 bool prev_apst_enabled;
2701 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2703 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2706 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2707 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2709 if (ctrl->vs >= NVME_VS(1, 1, 0))
2710 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2712 ret = nvme_identify_ctrl(ctrl, &id);
2714 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2718 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2719 ret = nvme_get_effects_log(ctrl);
2724 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2725 ctrl->cntlid = le16_to_cpu(id->cntlid);
2727 if (!ctrl->identified) {
2730 ret = nvme_init_subsystem(ctrl, id);
2735 * Check for quirks. Quirk can depend on firmware version,
2736 * so, in principle, the set of quirks present can change
2737 * across a reset. As a possible future enhancement, we
2738 * could re-scan for quirks every time we reinitialize
2739 * the device, but we'd have to make sure that the driver
2740 * behaves intelligently if the quirks change.
2742 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2743 if (quirk_matches(id, &core_quirks[i]))
2744 ctrl->quirks |= core_quirks[i].quirks;
2748 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2749 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2750 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2753 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2754 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2755 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2757 ctrl->oacs = le16_to_cpu(id->oacs);
2758 ctrl->oncs = le16_to_cpu(id->oncs);
2759 ctrl->mtfa = le16_to_cpu(id->mtfa);
2760 ctrl->oaes = le32_to_cpu(id->oaes);
2761 atomic_set(&ctrl->abort_limit, id->acl + 1);
2762 ctrl->vwc = id->vwc;
2764 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2766 max_hw_sectors = UINT_MAX;
2767 ctrl->max_hw_sectors =
2768 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2770 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2771 ctrl->sgls = le32_to_cpu(id->sgls);
2772 ctrl->kas = le16_to_cpu(id->kas);
2773 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2774 ctrl->ctratt = le32_to_cpu(id->ctratt);
2778 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2780 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2781 shutdown_timeout, 60);
2783 if (ctrl->shutdown_timeout != shutdown_timeout)
2784 dev_info(ctrl->device,
2785 "Shutdown timeout set to %u seconds\n",
2786 ctrl->shutdown_timeout);
2788 ctrl->shutdown_timeout = shutdown_timeout;
2790 ctrl->npss = id->npss;
2791 ctrl->apsta = id->apsta;
2792 prev_apst_enabled = ctrl->apst_enabled;
2793 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2794 if (force_apst && id->apsta) {
2795 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2796 ctrl->apst_enabled = true;
2798 ctrl->apst_enabled = false;
2801 ctrl->apst_enabled = id->apsta;
2803 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2805 if (ctrl->ops->flags & NVME_F_FABRICS) {
2806 ctrl->icdoff = le16_to_cpu(id->icdoff);
2807 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2808 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2809 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2812 * In fabrics we need to verify the cntlid matches the
2815 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2820 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2821 dev_err(ctrl->device,
2822 "keep-alive support is mandatory for fabrics\n");
2827 ctrl->hmpre = le32_to_cpu(id->hmpre);
2828 ctrl->hmmin = le32_to_cpu(id->hmmin);
2829 ctrl->hmminds = le32_to_cpu(id->hmminds);
2830 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2833 ret = nvme_mpath_init(ctrl, id);
2839 if (ctrl->apst_enabled && !prev_apst_enabled)
2840 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2841 else if (!ctrl->apst_enabled && prev_apst_enabled)
2842 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2844 ret = nvme_configure_apst(ctrl);
2848 ret = nvme_configure_timestamp(ctrl);
2852 ret = nvme_configure_directives(ctrl);
2856 ret = nvme_configure_acre(ctrl);
2860 ctrl->identified = true;
2868 EXPORT_SYMBOL_GPL(nvme_init_identify);
2870 static int nvme_dev_open(struct inode *inode, struct file *file)
2872 struct nvme_ctrl *ctrl =
2873 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2875 switch (ctrl->state) {
2876 case NVME_CTRL_LIVE:
2877 case NVME_CTRL_ADMIN_ONLY:
2880 return -EWOULDBLOCK;
2883 file->private_data = ctrl;
2887 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2892 down_read(&ctrl->namespaces_rwsem);
2893 if (list_empty(&ctrl->namespaces)) {
2898 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2899 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2900 dev_warn(ctrl->device,
2901 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2906 dev_warn(ctrl->device,
2907 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2908 kref_get(&ns->kref);
2909 up_read(&ctrl->namespaces_rwsem);
2911 ret = nvme_user_cmd(ctrl, ns, argp);
2916 up_read(&ctrl->namespaces_rwsem);
2920 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2923 struct nvme_ctrl *ctrl = file->private_data;
2924 void __user *argp = (void __user *)arg;
2927 case NVME_IOCTL_ADMIN_CMD:
2928 return nvme_user_cmd(ctrl, NULL, argp);
2929 case NVME_IOCTL_ADMIN64_CMD:
2930 return nvme_user_cmd64(ctrl, NULL, argp);
2931 case NVME_IOCTL_IO_CMD:
2932 return nvme_dev_user_cmd(ctrl, argp);
2933 case NVME_IOCTL_RESET:
2934 dev_warn(ctrl->device, "resetting controller\n");
2935 return nvme_reset_ctrl_sync(ctrl);
2936 case NVME_IOCTL_SUBSYS_RESET:
2937 return nvme_reset_subsystem(ctrl);
2938 case NVME_IOCTL_RESCAN:
2939 nvme_queue_scan(ctrl);
2946 static const struct file_operations nvme_dev_fops = {
2947 .owner = THIS_MODULE,
2948 .open = nvme_dev_open,
2949 .unlocked_ioctl = nvme_dev_ioctl,
2950 .compat_ioctl = nvme_dev_ioctl,
2953 static ssize_t nvme_sysfs_reset(struct device *dev,
2954 struct device_attribute *attr, const char *buf,
2957 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2960 ret = nvme_reset_ctrl_sync(ctrl);
2965 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2967 static ssize_t nvme_sysfs_rescan(struct device *dev,
2968 struct device_attribute *attr, const char *buf,
2971 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2973 nvme_queue_scan(ctrl);
2976 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2978 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2980 struct gendisk *disk = dev_to_disk(dev);
2982 if (disk->fops == &nvme_fops)
2983 return nvme_get_ns_from_dev(dev)->head;
2985 return disk->private_data;
2988 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2991 struct nvme_ns_head *head = dev_to_ns_head(dev);
2992 struct nvme_ns_ids *ids = &head->ids;
2993 struct nvme_subsystem *subsys = head->subsys;
2994 int serial_len = sizeof(subsys->serial);
2995 int model_len = sizeof(subsys->model);
2997 if (!uuid_is_null(&ids->uuid))
2998 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3000 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3001 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3003 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3004 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3006 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3007 subsys->serial[serial_len - 1] == '\0'))
3009 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3010 subsys->model[model_len - 1] == '\0'))
3013 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3014 serial_len, subsys->serial, model_len, subsys->model,
3017 static DEVICE_ATTR_RO(wwid);
3019 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3022 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3024 static DEVICE_ATTR_RO(nguid);
3026 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3029 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3031 /* For backward compatibility expose the NGUID to userspace if
3032 * we have no UUID set
3034 if (uuid_is_null(&ids->uuid)) {
3035 printk_ratelimited(KERN_WARNING
3036 "No UUID available providing old NGUID\n");
3037 return sprintf(buf, "%pU\n", ids->nguid);
3039 return sprintf(buf, "%pU\n", &ids->uuid);
3041 static DEVICE_ATTR_RO(uuid);
3043 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3046 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3048 static DEVICE_ATTR_RO(eui);
3050 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3053 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3055 static DEVICE_ATTR_RO(nsid);
3057 static struct attribute *nvme_ns_id_attrs[] = {
3058 &dev_attr_wwid.attr,
3059 &dev_attr_uuid.attr,
3060 &dev_attr_nguid.attr,
3062 &dev_attr_nsid.attr,
3063 #ifdef CONFIG_NVME_MULTIPATH
3064 &dev_attr_ana_grpid.attr,
3065 &dev_attr_ana_state.attr,
3070 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3071 struct attribute *a, int n)
3073 struct device *dev = container_of(kobj, struct device, kobj);
3074 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3076 if (a == &dev_attr_uuid.attr) {
3077 if (uuid_is_null(&ids->uuid) &&
3078 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3081 if (a == &dev_attr_nguid.attr) {
3082 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3085 if (a == &dev_attr_eui.attr) {
3086 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3089 #ifdef CONFIG_NVME_MULTIPATH
3090 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3091 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3093 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3100 static const struct attribute_group nvme_ns_id_attr_group = {
3101 .attrs = nvme_ns_id_attrs,
3102 .is_visible = nvme_ns_id_attrs_are_visible,
3105 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3106 &nvme_ns_id_attr_group,
3108 &nvme_nvm_attr_group,
3113 #define nvme_show_str_function(field) \
3114 static ssize_t field##_show(struct device *dev, \
3115 struct device_attribute *attr, char *buf) \
3117 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3118 return sprintf(buf, "%.*s\n", \
3119 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3121 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3123 nvme_show_str_function(model);
3124 nvme_show_str_function(serial);
3125 nvme_show_str_function(firmware_rev);
3127 #define nvme_show_int_function(field) \
3128 static ssize_t field##_show(struct device *dev, \
3129 struct device_attribute *attr, char *buf) \
3131 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3132 return sprintf(buf, "%d\n", ctrl->field); \
3134 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3136 nvme_show_int_function(cntlid);
3137 nvme_show_int_function(numa_node);
3138 nvme_show_int_function(queue_count);
3139 nvme_show_int_function(sqsize);
3141 static ssize_t nvme_sysfs_delete(struct device *dev,
3142 struct device_attribute *attr, const char *buf,
3145 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3147 if (device_remove_file_self(dev, attr))
3148 nvme_delete_ctrl_sync(ctrl);
3151 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3153 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3154 struct device_attribute *attr,
3157 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3159 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3161 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3163 static ssize_t nvme_sysfs_show_state(struct device *dev,
3164 struct device_attribute *attr,
3167 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3168 static const char *const state_name[] = {
3169 [NVME_CTRL_NEW] = "new",
3170 [NVME_CTRL_LIVE] = "live",
3171 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3172 [NVME_CTRL_RESETTING] = "resetting",
3173 [NVME_CTRL_CONNECTING] = "connecting",
3174 [NVME_CTRL_DELETING] = "deleting",
3175 [NVME_CTRL_DEAD] = "dead",
3178 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3179 state_name[ctrl->state])
3180 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3182 return sprintf(buf, "unknown state\n");
3185 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3187 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3188 struct device_attribute *attr,
3191 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3193 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3195 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3197 static ssize_t nvme_sysfs_show_address(struct device *dev,
3198 struct device_attribute *attr,
3201 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3203 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3205 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3207 static struct attribute *nvme_dev_attrs[] = {
3208 &dev_attr_reset_controller.attr,
3209 &dev_attr_rescan_controller.attr,
3210 &dev_attr_model.attr,
3211 &dev_attr_serial.attr,
3212 &dev_attr_firmware_rev.attr,
3213 &dev_attr_cntlid.attr,
3214 &dev_attr_delete_controller.attr,
3215 &dev_attr_transport.attr,
3216 &dev_attr_subsysnqn.attr,
3217 &dev_attr_address.attr,
3218 &dev_attr_state.attr,
3219 &dev_attr_numa_node.attr,
3220 &dev_attr_queue_count.attr,
3221 &dev_attr_sqsize.attr,
3225 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3226 struct attribute *a, int n)
3228 struct device *dev = container_of(kobj, struct device, kobj);
3229 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3231 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3233 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3239 static struct attribute_group nvme_dev_attrs_group = {
3240 .attrs = nvme_dev_attrs,
3241 .is_visible = nvme_dev_attrs_are_visible,
3244 static const struct attribute_group *nvme_dev_attr_groups[] = {
3245 &nvme_dev_attrs_group,
3249 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3252 struct nvme_ns_head *h;
3254 lockdep_assert_held(&subsys->lock);
3256 list_for_each_entry(h, &subsys->nsheads, entry) {
3257 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3264 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3265 struct nvme_ns_head *new)
3267 struct nvme_ns_head *h;
3269 lockdep_assert_held(&subsys->lock);
3271 list_for_each_entry(h, &subsys->nsheads, entry) {
3272 if (nvme_ns_ids_valid(&new->ids) &&
3273 !list_empty(&h->list) &&
3274 nvme_ns_ids_equal(&new->ids, &h->ids))
3281 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3282 unsigned nsid, struct nvme_id_ns *id)
3284 struct nvme_ns_head *head;
3285 size_t size = sizeof(*head);
3288 #ifdef CONFIG_NVME_MULTIPATH
3289 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3292 head = kzalloc(size, GFP_KERNEL);
3295 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3298 head->instance = ret;
3299 INIT_LIST_HEAD(&head->list);
3300 ret = init_srcu_struct(&head->srcu);
3302 goto out_ida_remove;
3303 head->subsys = ctrl->subsys;
3305 kref_init(&head->ref);
3307 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3309 goto out_cleanup_srcu;
3311 ret = __nvme_check_ids(ctrl->subsys, head);
3313 dev_err(ctrl->device,
3314 "duplicate IDs for nsid %d\n", nsid);
3315 goto out_cleanup_srcu;
3318 ret = nvme_mpath_alloc_disk(ctrl, head);
3320 goto out_cleanup_srcu;
3322 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3324 kref_get(&ctrl->subsys->ref);
3328 cleanup_srcu_struct(&head->srcu);
3330 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3335 ret = blk_status_to_errno(nvme_error_status(ret));
3336 return ERR_PTR(ret);
3339 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3340 struct nvme_id_ns *id)
3342 struct nvme_ctrl *ctrl = ns->ctrl;
3343 bool is_shared = id->nmic & (1 << 0);
3344 struct nvme_ns_head *head = NULL;
3347 mutex_lock(&ctrl->subsys->lock);
3349 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3351 head = nvme_alloc_ns_head(ctrl, nsid, id);
3353 ret = PTR_ERR(head);
3357 struct nvme_ns_ids ids;
3359 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3363 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3364 dev_err(ctrl->device,
3365 "IDs don't match for shared namespace %d\n",
3372 list_add_tail(&ns->siblings, &head->list);
3376 mutex_unlock(&ctrl->subsys->lock);
3378 ret = blk_status_to_errno(nvme_error_status(ret));
3382 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3384 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3385 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3387 return nsa->head->ns_id - nsb->head->ns_id;
3390 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3392 struct nvme_ns *ns, *ret = NULL;
3394 down_read(&ctrl->namespaces_rwsem);
3395 list_for_each_entry(ns, &ctrl->namespaces, list) {
3396 if (ns->head->ns_id == nsid) {
3397 if (!kref_get_unless_zero(&ns->kref))
3402 if (ns->head->ns_id > nsid)
3405 up_read(&ctrl->namespaces_rwsem);
3409 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3411 struct streams_directive_params s;
3414 if (!ctrl->nr_streams)
3417 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3421 ns->sws = le32_to_cpu(s.sws);
3422 ns->sgs = le16_to_cpu(s.sgs);
3425 unsigned int bs = 1 << ns->lba_shift;
3427 blk_queue_io_min(ns->queue, bs * ns->sws);
3429 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3435 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3438 struct gendisk *disk;
3439 struct nvme_id_ns *id;
3440 char disk_name[DISK_NAME_LEN];
3441 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3443 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3447 ns->queue = blk_mq_init_queue(ctrl->tagset);
3448 if (IS_ERR(ns->queue)) {
3449 ret = PTR_ERR(ns->queue);
3453 if (ctrl->opts && ctrl->opts->data_digest)
3454 ns->queue->backing_dev_info->capabilities
3455 |= BDI_CAP_STABLE_WRITES;
3457 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3458 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3459 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3461 ns->queue->queuedata = ns;
3464 kref_init(&ns->kref);
3465 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3467 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3468 nvme_set_queue_limits(ctrl, ns->queue);
3470 ret = nvme_identify_ns(ctrl, nsid, &id);
3472 goto out_free_queue;
3474 if (id->ncap == 0) {
3479 ret = nvme_init_ns_head(ns, nsid, id);
3482 nvme_setup_streams_ns(ctrl, ns);
3483 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3485 disk = alloc_disk_node(0, node);
3491 disk->fops = &nvme_fops;
3492 disk->private_data = ns;
3493 disk->queue = ns->queue;
3494 disk->flags = flags;
3495 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3498 __nvme_revalidate_disk(disk, id);
3500 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3501 ret = nvme_nvm_register(ns, disk_name, node);
3503 dev_warn(ctrl->device, "LightNVM init failure\n");
3508 down_write(&ctrl->namespaces_rwsem);
3509 list_add_tail(&ns->list, &ctrl->namespaces);
3510 up_write(&ctrl->namespaces_rwsem);
3512 nvme_get_ctrl(ctrl);
3514 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3516 nvme_mpath_add_disk(ns, id);
3517 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3524 mutex_lock(&ctrl->subsys->lock);
3525 list_del_rcu(&ns->siblings);
3526 mutex_unlock(&ctrl->subsys->lock);
3527 nvme_put_ns_head(ns->head);
3531 blk_cleanup_queue(ns->queue);
3535 ret = blk_status_to_errno(nvme_error_status(ret));
3539 static void nvme_ns_remove(struct nvme_ns *ns)
3541 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3544 nvme_fault_inject_fini(&ns->fault_inject);
3546 mutex_lock(&ns->ctrl->subsys->lock);
3547 list_del_rcu(&ns->siblings);
3548 mutex_unlock(&ns->ctrl->subsys->lock);
3549 synchronize_rcu(); /* guarantee not available in head->list */
3550 nvme_mpath_clear_current_path(ns);
3551 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3553 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3554 del_gendisk(ns->disk);
3555 blk_cleanup_queue(ns->queue);
3556 if (blk_get_integrity(ns->disk))
3557 blk_integrity_unregister(ns->disk);
3560 down_write(&ns->ctrl->namespaces_rwsem);
3561 list_del_init(&ns->list);
3562 up_write(&ns->ctrl->namespaces_rwsem);
3564 nvme_mpath_check_last_path(ns);
3568 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3572 ns = nvme_find_get_ns(ctrl, nsid);
3574 if (ns->disk && revalidate_disk(ns->disk))
3578 nvme_alloc_ns(ctrl, nsid);
3581 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3584 struct nvme_ns *ns, *next;
3587 down_write(&ctrl->namespaces_rwsem);
3588 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3589 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3590 list_move_tail(&ns->list, &rm_list);
3592 up_write(&ctrl->namespaces_rwsem);
3594 list_for_each_entry_safe(ns, next, &rm_list, list)
3599 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3603 unsigned i, j, nsid, prev = 0;
3604 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3607 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3611 for (i = 0; i < num_lists; i++) {
3612 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3616 for (j = 0; j < min(nn, 1024U); j++) {
3617 nsid = le32_to_cpu(ns_list[j]);
3621 nvme_validate_ns(ctrl, nsid);
3623 while (++prev < nsid) {
3624 ns = nvme_find_get_ns(ctrl, prev);
3634 nvme_remove_invalid_namespaces(ctrl, prev);
3640 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3644 for (i = 1; i <= nn; i++)
3645 nvme_validate_ns(ctrl, i);
3647 nvme_remove_invalid_namespaces(ctrl, nn);
3650 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3652 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3656 log = kzalloc(log_size, GFP_KERNEL);
3661 * We need to read the log to clear the AEN, but we don't want to rely
3662 * on it for the changed namespace information as userspace could have
3663 * raced with us in reading the log page, which could cause us to miss
3666 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3669 dev_warn(ctrl->device,
3670 "reading changed ns log failed: %d\n", error);
3675 static void nvme_scan_work(struct work_struct *work)
3677 struct nvme_ctrl *ctrl =
3678 container_of(work, struct nvme_ctrl, scan_work);
3679 struct nvme_id_ctrl *id;
3682 if (ctrl->state != NVME_CTRL_LIVE)
3685 WARN_ON_ONCE(!ctrl->tagset);
3687 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3688 dev_info(ctrl->device, "rescanning namespaces.\n");
3689 nvme_clear_changed_ns_log(ctrl);
3692 if (nvme_identify_ctrl(ctrl, &id))
3695 mutex_lock(&ctrl->scan_lock);
3696 nn = le32_to_cpu(id->nn);
3697 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3698 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3699 if (!nvme_scan_ns_list(ctrl, nn))
3702 nvme_scan_ns_sequential(ctrl, nn);
3704 mutex_unlock(&ctrl->scan_lock);
3706 down_write(&ctrl->namespaces_rwsem);
3707 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3708 up_write(&ctrl->namespaces_rwsem);
3712 * This function iterates the namespace list unlocked to allow recovery from
3713 * controller failure. It is up to the caller to ensure the namespace list is
3714 * not modified by scan work while this function is executing.
3716 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3718 struct nvme_ns *ns, *next;
3722 * make sure to requeue I/O to all namespaces as these
3723 * might result from the scan itself and must complete
3724 * for the scan_work to make progress
3726 nvme_mpath_clear_ctrl_paths(ctrl);
3728 /* prevent racing with ns scanning */
3729 flush_work(&ctrl->scan_work);
3732 * The dead states indicates the controller was not gracefully
3733 * disconnected. In that case, we won't be able to flush any data while
3734 * removing the namespaces' disks; fail all the queues now to avoid
3735 * potentially having to clean up the failed sync later.
3737 if (ctrl->state == NVME_CTRL_DEAD)
3738 nvme_kill_queues(ctrl);
3740 down_write(&ctrl->namespaces_rwsem);
3741 list_splice_init(&ctrl->namespaces, &ns_list);
3742 up_write(&ctrl->namespaces_rwsem);
3744 list_for_each_entry_safe(ns, next, &ns_list, list)
3747 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3749 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3751 struct nvme_ctrl *ctrl =
3752 container_of(dev, struct nvme_ctrl, ctrl_device);
3753 struct nvmf_ctrl_options *opts = ctrl->opts;
3756 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3761 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3765 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3766 opts->trsvcid ?: "none");
3770 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3771 opts->host_traddr ?: "none");
3776 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3778 char *envp[2] = { NULL, NULL };
3779 u32 aen_result = ctrl->aen_result;
3781 ctrl->aen_result = 0;
3785 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3788 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3792 static void nvme_async_event_work(struct work_struct *work)
3794 struct nvme_ctrl *ctrl =
3795 container_of(work, struct nvme_ctrl, async_event_work);
3797 nvme_aen_uevent(ctrl);
3798 ctrl->ops->submit_async_event(ctrl);
3801 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3806 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3812 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3815 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3817 struct nvme_fw_slot_info_log *log;
3819 log = kmalloc(sizeof(*log), GFP_KERNEL);
3823 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3825 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3829 static void nvme_fw_act_work(struct work_struct *work)
3831 struct nvme_ctrl *ctrl = container_of(work,
3832 struct nvme_ctrl, fw_act_work);
3833 unsigned long fw_act_timeout;
3836 fw_act_timeout = jiffies +
3837 msecs_to_jiffies(ctrl->mtfa * 100);
3839 fw_act_timeout = jiffies +
3840 msecs_to_jiffies(admin_timeout * 1000);
3842 nvme_stop_queues(ctrl);
3843 while (nvme_ctrl_pp_status(ctrl)) {
3844 if (time_after(jiffies, fw_act_timeout)) {
3845 dev_warn(ctrl->device,
3846 "Fw activation timeout, reset controller\n");
3847 nvme_reset_ctrl(ctrl);
3853 if (ctrl->state != NVME_CTRL_LIVE)
3856 nvme_start_queues(ctrl);
3857 /* read FW slot information to clear the AER */
3858 nvme_get_fw_slot_info(ctrl);
3861 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3863 u32 aer_notice_type = (result & 0xff00) >> 8;
3865 trace_nvme_async_event(ctrl, aer_notice_type);
3867 switch (aer_notice_type) {
3868 case NVME_AER_NOTICE_NS_CHANGED:
3869 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3870 nvme_queue_scan(ctrl);
3872 case NVME_AER_NOTICE_FW_ACT_STARTING:
3873 queue_work(nvme_wq, &ctrl->fw_act_work);
3875 #ifdef CONFIG_NVME_MULTIPATH
3876 case NVME_AER_NOTICE_ANA:
3877 if (!ctrl->ana_log_buf)
3879 queue_work(nvme_wq, &ctrl->ana_work);
3882 case NVME_AER_NOTICE_DISC_CHANGED:
3883 ctrl->aen_result = result;
3886 dev_warn(ctrl->device, "async event result %08x\n", result);
3890 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3891 volatile union nvme_result *res)
3893 u32 result = le32_to_cpu(res->u32);
3894 u32 aer_type = result & 0x07;
3896 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3900 case NVME_AER_NOTICE:
3901 nvme_handle_aen_notice(ctrl, result);
3903 case NVME_AER_ERROR:
3904 case NVME_AER_SMART:
3907 trace_nvme_async_event(ctrl, aer_type);
3908 ctrl->aen_result = result;
3913 queue_work(nvme_wq, &ctrl->async_event_work);
3915 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3917 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3919 nvme_mpath_stop(ctrl);
3920 nvme_stop_keep_alive(ctrl);
3921 flush_work(&ctrl->async_event_work);
3922 cancel_work_sync(&ctrl->fw_act_work);
3924 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3926 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3929 nvme_start_keep_alive(ctrl);
3931 nvme_enable_aen(ctrl);
3933 if (ctrl->queue_count > 1) {
3934 nvme_queue_scan(ctrl);
3935 nvme_start_queues(ctrl);
3938 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3940 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3942 nvme_fault_inject_fini(&ctrl->fault_inject);
3943 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3944 cdev_device_del(&ctrl->cdev, ctrl->device);
3946 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3948 static void nvme_free_ctrl(struct device *dev)
3950 struct nvme_ctrl *ctrl =
3951 container_of(dev, struct nvme_ctrl, ctrl_device);
3952 struct nvme_subsystem *subsys = ctrl->subsys;
3954 if (subsys && ctrl->instance != subsys->instance)
3955 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3957 kfree(ctrl->effects);
3958 nvme_mpath_uninit(ctrl);
3959 __free_page(ctrl->discard_page);
3962 mutex_lock(&nvme_subsystems_lock);
3963 list_del(&ctrl->subsys_entry);
3964 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3965 mutex_unlock(&nvme_subsystems_lock);
3968 ctrl->ops->free_ctrl(ctrl);
3971 nvme_put_subsystem(subsys);
3975 * Initialize a NVMe controller structures. This needs to be called during
3976 * earliest initialization so that we have the initialized structured around
3979 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3980 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3984 ctrl->state = NVME_CTRL_NEW;
3985 spin_lock_init(&ctrl->lock);
3986 mutex_init(&ctrl->scan_lock);
3987 INIT_LIST_HEAD(&ctrl->namespaces);
3988 init_rwsem(&ctrl->namespaces_rwsem);
3991 ctrl->quirks = quirks;
3992 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3993 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3994 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3995 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3997 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3998 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3999 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4001 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4003 ctrl->discard_page = alloc_page(GFP_KERNEL);
4004 if (!ctrl->discard_page) {
4009 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4012 ctrl->instance = ret;
4014 device_initialize(&ctrl->ctrl_device);
4015 ctrl->device = &ctrl->ctrl_device;
4016 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4017 ctrl->device->class = nvme_class;
4018 ctrl->device->parent = ctrl->dev;
4019 ctrl->device->groups = nvme_dev_attr_groups;
4020 ctrl->device->release = nvme_free_ctrl;
4021 dev_set_drvdata(ctrl->device, ctrl);
4022 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4024 goto out_release_instance;
4026 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4027 ctrl->cdev.owner = ops->module;
4028 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4033 * Initialize latency tolerance controls. The sysfs files won't
4034 * be visible to userspace unless the device actually supports APST.
4036 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4037 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4038 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4040 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4044 kfree_const(ctrl->device->kobj.name);
4045 out_release_instance:
4046 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4048 if (ctrl->discard_page)
4049 __free_page(ctrl->discard_page);
4052 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4055 * nvme_kill_queues(): Ends all namespace queues
4056 * @ctrl: the dead controller that needs to end
4058 * Call this function when the driver determines it is unable to get the
4059 * controller in a state capable of servicing IO.
4061 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4065 down_read(&ctrl->namespaces_rwsem);
4067 /* Forcibly unquiesce queues to avoid blocking dispatch */
4068 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4069 blk_mq_unquiesce_queue(ctrl->admin_q);
4071 list_for_each_entry(ns, &ctrl->namespaces, list)
4072 nvme_set_queue_dying(ns);
4074 up_read(&ctrl->namespaces_rwsem);
4076 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4078 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4082 down_read(&ctrl->namespaces_rwsem);
4083 list_for_each_entry(ns, &ctrl->namespaces, list)
4084 blk_mq_unfreeze_queue(ns->queue);
4085 up_read(&ctrl->namespaces_rwsem);
4087 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4089 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4093 down_read(&ctrl->namespaces_rwsem);
4094 list_for_each_entry(ns, &ctrl->namespaces, list) {
4095 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4099 up_read(&ctrl->namespaces_rwsem);
4101 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4103 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4107 down_read(&ctrl->namespaces_rwsem);
4108 list_for_each_entry(ns, &ctrl->namespaces, list)
4109 blk_mq_freeze_queue_wait(ns->queue);
4110 up_read(&ctrl->namespaces_rwsem);
4112 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4114 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4118 down_read(&ctrl->namespaces_rwsem);
4119 list_for_each_entry(ns, &ctrl->namespaces, list)
4120 blk_freeze_queue_start(ns->queue);
4121 up_read(&ctrl->namespaces_rwsem);
4123 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4125 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4129 down_read(&ctrl->namespaces_rwsem);
4130 list_for_each_entry(ns, &ctrl->namespaces, list)
4131 blk_mq_quiesce_queue(ns->queue);
4132 up_read(&ctrl->namespaces_rwsem);
4134 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4136 void nvme_start_queues(struct nvme_ctrl *ctrl)
4140 down_read(&ctrl->namespaces_rwsem);
4141 list_for_each_entry(ns, &ctrl->namespaces, list)
4142 blk_mq_unquiesce_queue(ns->queue);
4143 up_read(&ctrl->namespaces_rwsem);
4145 EXPORT_SYMBOL_GPL(nvme_start_queues);
4148 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4152 down_read(&ctrl->namespaces_rwsem);
4153 list_for_each_entry(ns, &ctrl->namespaces, list)
4154 blk_sync_queue(ns->queue);
4155 up_read(&ctrl->namespaces_rwsem);
4158 blk_sync_queue(ctrl->admin_q);
4160 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4163 * Check we didn't inadvertently grow the command structure sizes:
4165 static inline void _nvme_check_size(void)
4167 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4168 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4169 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4170 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4171 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4172 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4173 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4174 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4175 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4176 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4177 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4178 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4179 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4180 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4181 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4182 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4183 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4187 static int __init nvme_core_init(void)
4189 int result = -ENOMEM;
4193 nvme_wq = alloc_workqueue("nvme-wq",
4194 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4198 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4199 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4203 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4204 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4205 if (!nvme_delete_wq)
4206 goto destroy_reset_wq;
4208 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4210 goto destroy_delete_wq;
4212 nvme_class = class_create(THIS_MODULE, "nvme");
4213 if (IS_ERR(nvme_class)) {
4214 result = PTR_ERR(nvme_class);
4215 goto unregister_chrdev;
4217 nvme_class->dev_uevent = nvme_class_uevent;
4219 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4220 if (IS_ERR(nvme_subsys_class)) {
4221 result = PTR_ERR(nvme_subsys_class);
4227 class_destroy(nvme_class);
4229 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4231 destroy_workqueue(nvme_delete_wq);
4233 destroy_workqueue(nvme_reset_wq);
4235 destroy_workqueue(nvme_wq);
4240 static void __exit nvme_core_exit(void)
4242 class_destroy(nvme_subsys_class);
4243 class_destroy(nvme_class);
4244 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4245 destroy_workqueue(nvme_delete_wq);
4246 destroy_workqueue(nvme_reset_wq);
4247 destroy_workqueue(nvme_wq);
4250 MODULE_LICENSE("GPL");
4251 MODULE_VERSION("1.0");
4252 module_init(nvme_core_init);
4253 module_exit(nvme_core_exit);