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);
1311 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1314 * Revalidate LBA changes prior to unfreezing. This is necessary to
1315 * prevent memory corruption if a logical block size was changed by
1318 if (effects & NVME_CMD_EFFECTS_LBCC)
1319 nvme_update_formats(ctrl);
1320 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1321 nvme_unfreeze(ctrl);
1322 nvme_mpath_unfreeze(ctrl->subsys);
1323 mutex_unlock(&ctrl->subsys->lock);
1324 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1325 mutex_unlock(&ctrl->scan_lock);
1327 if (effects & NVME_CMD_EFFECTS_CCC)
1328 nvme_init_identify(ctrl);
1329 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1330 nvme_queue_scan(ctrl);
1333 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1334 struct nvme_passthru_cmd __user *ucmd)
1336 struct nvme_passthru_cmd cmd;
1337 struct nvme_command c;
1338 unsigned timeout = 0;
1343 if (!capable(CAP_SYS_ADMIN))
1345 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1350 memset(&c, 0, sizeof(c));
1351 c.common.opcode = cmd.opcode;
1352 c.common.flags = cmd.flags;
1353 c.common.nsid = cpu_to_le32(cmd.nsid);
1354 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1355 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1356 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1357 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1358 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1359 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1360 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1361 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1364 timeout = msecs_to_jiffies(cmd.timeout_ms);
1366 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1367 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1368 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1369 (void __user *)(uintptr_t)cmd.metadata,
1370 cmd.metadata_len, 0, &result, timeout);
1371 nvme_passthru_end(ctrl, effects);
1374 if (put_user(result, &ucmd->result))
1381 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1382 struct nvme_passthru_cmd64 __user *ucmd)
1384 struct nvme_passthru_cmd64 cmd;
1385 struct nvme_command c;
1386 unsigned timeout = 0;
1390 if (!capable(CAP_SYS_ADMIN))
1392 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1397 memset(&c, 0, sizeof(c));
1398 c.common.opcode = cmd.opcode;
1399 c.common.flags = cmd.flags;
1400 c.common.nsid = cpu_to_le32(cmd.nsid);
1401 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1402 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1403 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1404 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1405 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1406 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1407 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1408 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1411 timeout = msecs_to_jiffies(cmd.timeout_ms);
1413 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1414 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1415 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1416 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1417 0, &cmd.result, timeout);
1418 nvme_passthru_end(ctrl, effects);
1421 if (put_user(cmd.result, &ucmd->result))
1429 * Issue ioctl requests on the first available path. Note that unlike normal
1430 * block layer requests we will not retry failed request on another controller.
1432 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1433 struct nvme_ns_head **head, int *srcu_idx)
1435 #ifdef CONFIG_NVME_MULTIPATH
1436 if (disk->fops == &nvme_ns_head_ops) {
1439 *head = disk->private_data;
1440 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1441 ns = nvme_find_path(*head);
1443 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1449 return disk->private_data;
1452 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1455 srcu_read_unlock(&head->srcu, idx);
1458 static bool is_ctrl_ioctl(unsigned int cmd)
1460 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1462 if (is_sed_ioctl(cmd))
1467 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1469 struct nvme_ns_head *head,
1472 struct nvme_ctrl *ctrl = ns->ctrl;
1475 nvme_get_ctrl(ns->ctrl);
1476 nvme_put_ns_from_disk(head, srcu_idx);
1479 case NVME_IOCTL_ADMIN_CMD:
1480 ret = nvme_user_cmd(ctrl, NULL, argp);
1482 case NVME_IOCTL_ADMIN64_CMD:
1483 ret = nvme_user_cmd64(ctrl, NULL, argp);
1486 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1489 nvme_put_ctrl(ctrl);
1493 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1494 unsigned int cmd, unsigned long arg)
1496 struct nvme_ns_head *head = NULL;
1497 void __user *argp = (void __user *)arg;
1501 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1503 return -EWOULDBLOCK;
1506 * Handle ioctls that apply to the controller instead of the namespace
1507 * seperately and drop the ns SRCU reference early. This avoids a
1508 * deadlock when deleting namespaces using the passthrough interface.
1510 if (is_ctrl_ioctl(cmd))
1511 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1515 force_successful_syscall_return();
1516 ret = ns->head->ns_id;
1518 case NVME_IOCTL_IO_CMD:
1519 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1521 case NVME_IOCTL_SUBMIT_IO:
1522 ret = nvme_submit_io(ns, argp);
1524 case NVME_IOCTL_IO64_CMD:
1525 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1529 ret = nvme_nvm_ioctl(ns, cmd, arg);
1534 nvme_put_ns_from_disk(head, srcu_idx);
1538 static int nvme_open(struct block_device *bdev, fmode_t mode)
1540 struct nvme_ns *ns = bdev->bd_disk->private_data;
1542 #ifdef CONFIG_NVME_MULTIPATH
1543 /* should never be called due to GENHD_FL_HIDDEN */
1544 if (WARN_ON_ONCE(ns->head->disk))
1547 if (!kref_get_unless_zero(&ns->kref))
1549 if (!try_module_get(ns->ctrl->ops->module))
1560 static void nvme_release(struct gendisk *disk, fmode_t mode)
1562 struct nvme_ns *ns = disk->private_data;
1564 module_put(ns->ctrl->ops->module);
1568 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1570 /* some standard values */
1571 geo->heads = 1 << 6;
1572 geo->sectors = 1 << 5;
1573 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1577 #ifdef CONFIG_BLK_DEV_INTEGRITY
1578 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1580 struct blk_integrity integrity;
1582 memset(&integrity, 0, sizeof(integrity));
1584 case NVME_NS_DPS_PI_TYPE3:
1585 integrity.profile = &t10_pi_type3_crc;
1586 integrity.tag_size = sizeof(u16) + sizeof(u32);
1587 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1589 case NVME_NS_DPS_PI_TYPE1:
1590 case NVME_NS_DPS_PI_TYPE2:
1591 integrity.profile = &t10_pi_type1_crc;
1592 integrity.tag_size = sizeof(u16);
1593 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1596 integrity.profile = NULL;
1599 integrity.tuple_size = ms;
1600 blk_integrity_register(disk, &integrity);
1601 blk_queue_max_integrity_segments(disk->queue, 1);
1604 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1607 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1609 static void nvme_set_chunk_size(struct nvme_ns *ns)
1611 u32 chunk_size = (((u32)ns->noiob) << (ns->lba_shift - 9));
1612 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1615 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1617 struct nvme_ctrl *ctrl = ns->ctrl;
1618 struct request_queue *queue = disk->queue;
1619 u32 size = queue_logical_block_size(queue);
1621 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1622 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1626 if (ctrl->nr_streams && ns->sws && ns->sgs)
1627 size *= ns->sws * ns->sgs;
1629 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1630 NVME_DSM_MAX_RANGES);
1632 queue->limits.discard_alignment = 0;
1633 queue->limits.discard_granularity = size;
1635 /* If discard is already enabled, don't reset queue limits */
1636 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1639 blk_queue_max_discard_sectors(queue, UINT_MAX);
1640 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1642 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1643 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1646 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1649 unsigned short bs = 1 << ns->lba_shift;
1651 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1652 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1655 * Even though NVMe spec explicitly states that MDTS is not
1656 * applicable to the write-zeroes:- "The restriction does not apply to
1657 * commands that do not transfer data between the host and the
1658 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1659 * In order to be more cautious use controller's max_hw_sectors value
1660 * to configure the maximum sectors for the write-zeroes which is
1661 * configured based on the controller's MDTS field in the
1662 * nvme_init_identify() if available.
1664 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1665 max_sectors = ((u32)(USHRT_MAX + 1) * bs) >> 9;
1667 max_sectors = ((u32)(ns->ctrl->max_hw_sectors + 1) * bs) >> 9;
1669 blk_queue_max_write_zeroes_sectors(disk->queue, max_sectors);
1672 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1673 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1677 memset(ids, 0, sizeof(*ids));
1679 if (ctrl->vs >= NVME_VS(1, 1, 0))
1680 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1681 if (ctrl->vs >= NVME_VS(1, 2, 0))
1682 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1683 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1684 /* Don't treat error as fatal we potentially
1685 * already have a NGUID or EUI-64
1687 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1689 dev_warn(ctrl->device,
1690 "Identify Descriptors failed (%d)\n", ret);
1695 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1697 return !uuid_is_null(&ids->uuid) ||
1698 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1699 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1702 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1704 return uuid_equal(&a->uuid, &b->uuid) &&
1705 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1706 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1709 static void nvme_update_disk_info(struct gendisk *disk,
1710 struct nvme_ns *ns, struct nvme_id_ns *id)
1712 sector_t capacity = le64_to_cpu(id->nsze) << (ns->lba_shift - 9);
1713 unsigned short bs = 1 << ns->lba_shift;
1714 u32 atomic_bs, phys_bs, io_opt;
1716 if (ns->lba_shift > PAGE_SHIFT) {
1717 /* unsupported block size, set capacity to 0 later */
1720 blk_mq_freeze_queue(disk->queue);
1721 blk_integrity_unregister(disk);
1723 if (id->nabo == 0) {
1725 * Bit 1 indicates whether NAWUPF is defined for this namespace
1726 * and whether it should be used instead of AWUPF. If NAWUPF ==
1727 * 0 then AWUPF must be used instead.
1729 if (id->nsfeat & (1 << 1) && id->nawupf)
1730 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1732 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1738 if (id->nsfeat & (1 << 4)) {
1739 /* NPWG = Namespace Preferred Write Granularity */
1740 phys_bs *= 1 + le16_to_cpu(id->npwg);
1741 /* NOWS = Namespace Optimal Write Size */
1742 io_opt *= 1 + le16_to_cpu(id->nows);
1745 blk_queue_logical_block_size(disk->queue, bs);
1747 * Linux filesystems assume writing a single physical block is
1748 * an atomic operation. Hence limit the physical block size to the
1749 * value of the Atomic Write Unit Power Fail parameter.
1751 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1752 blk_queue_io_min(disk->queue, phys_bs);
1753 blk_queue_io_opt(disk->queue, io_opt);
1755 if (ns->ms && !ns->ext &&
1756 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1757 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1758 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1759 ns->lba_shift > PAGE_SHIFT)
1762 set_capacity(disk, capacity);
1764 nvme_config_discard(disk, ns);
1765 nvme_config_write_zeroes(disk, ns);
1767 if (id->nsattr & (1 << 0))
1768 set_disk_ro(disk, true);
1770 set_disk_ro(disk, false);
1772 blk_mq_unfreeze_queue(disk->queue);
1775 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1777 struct nvme_ns *ns = disk->private_data;
1780 * If identify namespace failed, use default 512 byte block size so
1781 * block layer can use before failing read/write for 0 capacity.
1783 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1784 if (ns->lba_shift == 0)
1786 ns->noiob = le16_to_cpu(id->noiob);
1787 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1788 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1789 /* the PI implementation requires metadata equal t10 pi tuple size */
1790 if (ns->ms == sizeof(struct t10_pi_tuple))
1791 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1796 nvme_set_chunk_size(ns);
1797 nvme_update_disk_info(disk, ns, id);
1798 #ifdef CONFIG_NVME_MULTIPATH
1799 if (ns->head->disk) {
1800 nvme_update_disk_info(ns->head->disk, ns, id);
1801 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1802 revalidate_disk(ns->head->disk);
1807 static int nvme_revalidate_disk(struct gendisk *disk)
1809 struct nvme_ns *ns = disk->private_data;
1810 struct nvme_ctrl *ctrl = ns->ctrl;
1811 struct nvme_id_ns *id;
1812 struct nvme_ns_ids ids;
1815 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1816 set_capacity(disk, 0);
1820 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1824 if (id->ncap == 0) {
1829 __nvme_revalidate_disk(disk, id);
1830 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1834 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1835 dev_err(ctrl->device,
1836 "identifiers changed for nsid %d\n", ns->head->ns_id);
1844 * Only fail the function if we got a fatal error back from the
1845 * device, otherwise ignore the error and just move on.
1847 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1850 ret = blk_status_to_errno(nvme_error_status(ret));
1854 static char nvme_pr_type(enum pr_type type)
1857 case PR_WRITE_EXCLUSIVE:
1859 case PR_EXCLUSIVE_ACCESS:
1861 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1863 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1865 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1867 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1874 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1875 u64 key, u64 sa_key, u8 op)
1877 struct nvme_ns_head *head = NULL;
1879 struct nvme_command c;
1881 u8 data[16] = { 0, };
1883 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1885 return -EWOULDBLOCK;
1887 put_unaligned_le64(key, &data[0]);
1888 put_unaligned_le64(sa_key, &data[8]);
1890 memset(&c, 0, sizeof(c));
1891 c.common.opcode = op;
1892 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1893 c.common.cdw10 = cpu_to_le32(cdw10);
1895 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1896 nvme_put_ns_from_disk(head, srcu_idx);
1900 static int nvme_pr_register(struct block_device *bdev, u64 old,
1901 u64 new, unsigned flags)
1905 if (flags & ~PR_FL_IGNORE_KEY)
1908 cdw10 = old ? 2 : 0;
1909 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1910 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1911 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1914 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1915 enum pr_type type, unsigned flags)
1919 if (flags & ~PR_FL_IGNORE_KEY)
1922 cdw10 = nvme_pr_type(type) << 8;
1923 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1924 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1927 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1928 enum pr_type type, bool abort)
1930 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1931 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1934 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1936 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1937 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1940 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1942 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1943 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1946 static const struct pr_ops nvme_pr_ops = {
1947 .pr_register = nvme_pr_register,
1948 .pr_reserve = nvme_pr_reserve,
1949 .pr_release = nvme_pr_release,
1950 .pr_preempt = nvme_pr_preempt,
1951 .pr_clear = nvme_pr_clear,
1954 #ifdef CONFIG_BLK_SED_OPAL
1955 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1958 struct nvme_ctrl *ctrl = data;
1959 struct nvme_command cmd;
1961 memset(&cmd, 0, sizeof(cmd));
1963 cmd.common.opcode = nvme_admin_security_send;
1965 cmd.common.opcode = nvme_admin_security_recv;
1966 cmd.common.nsid = 0;
1967 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1968 cmd.common.cdw11 = cpu_to_le32(len);
1970 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
1971 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
1973 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1974 #endif /* CONFIG_BLK_SED_OPAL */
1976 static const struct block_device_operations nvme_fops = {
1977 .owner = THIS_MODULE,
1978 .ioctl = nvme_ioctl,
1979 .compat_ioctl = nvme_ioctl,
1981 .release = nvme_release,
1982 .getgeo = nvme_getgeo,
1983 .revalidate_disk= nvme_revalidate_disk,
1984 .pr_ops = &nvme_pr_ops,
1987 #ifdef CONFIG_NVME_MULTIPATH
1988 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1990 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1992 if (!kref_get_unless_zero(&head->ref))
1997 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1999 nvme_put_ns_head(disk->private_data);
2002 const struct block_device_operations nvme_ns_head_ops = {
2003 .owner = THIS_MODULE,
2004 .open = nvme_ns_head_open,
2005 .release = nvme_ns_head_release,
2006 .ioctl = nvme_ioctl,
2007 .compat_ioctl = nvme_ioctl,
2008 .getgeo = nvme_getgeo,
2009 .pr_ops = &nvme_pr_ops,
2011 #endif /* CONFIG_NVME_MULTIPATH */
2013 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2015 unsigned long timeout =
2016 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2017 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2020 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2023 if ((csts & NVME_CSTS_RDY) == bit)
2027 if (fatal_signal_pending(current))
2029 if (time_after(jiffies, timeout)) {
2030 dev_err(ctrl->device,
2031 "Device not ready; aborting %s\n", enabled ?
2032 "initialisation" : "reset");
2041 * If the device has been passed off to us in an enabled state, just clear
2042 * the enabled bit. The spec says we should set the 'shutdown notification
2043 * bits', but doing so may cause the device to complete commands to the
2044 * admin queue ... and we don't know what memory that might be pointing at!
2046 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2050 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2051 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2053 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2057 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2058 msleep(NVME_QUIRK_DELAY_AMOUNT);
2060 return nvme_wait_ready(ctrl, ctrl->cap, false);
2062 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2064 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2067 * Default to a 4K page size, with the intention to update this
2068 * path in the future to accomodate architectures with differing
2069 * kernel and IO page sizes.
2071 unsigned dev_page_min, page_shift = 12;
2074 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2076 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2079 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2081 if (page_shift < dev_page_min) {
2082 dev_err(ctrl->device,
2083 "Minimum device page size %u too large for host (%u)\n",
2084 1 << dev_page_min, 1 << page_shift);
2088 ctrl->page_size = 1 << page_shift;
2090 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2091 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2092 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2093 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2094 ctrl->ctrl_config |= NVME_CC_ENABLE;
2096 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2099 return nvme_wait_ready(ctrl, ctrl->cap, true);
2101 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2103 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2105 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2109 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2110 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2112 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2116 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2117 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2121 if (fatal_signal_pending(current))
2123 if (time_after(jiffies, timeout)) {
2124 dev_err(ctrl->device,
2125 "Device shutdown incomplete; abort shutdown\n");
2132 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2134 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2135 struct request_queue *q)
2139 if (ctrl->max_hw_sectors) {
2141 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2143 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2144 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2145 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2147 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2148 is_power_of_2(ctrl->max_hw_sectors))
2149 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2150 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2151 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2153 blk_queue_write_cache(q, vwc, vwc);
2156 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2161 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2164 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2165 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2168 dev_warn_once(ctrl->device,
2169 "could not set timestamp (%d)\n", ret);
2173 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2175 struct nvme_feat_host_behavior *host;
2178 /* Don't bother enabling the feature if retry delay is not reported */
2182 host = kzalloc(sizeof(*host), GFP_KERNEL);
2186 host->acre = NVME_ENABLE_ACRE;
2187 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2188 host, sizeof(*host), NULL);
2193 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2196 * APST (Autonomous Power State Transition) lets us program a
2197 * table of power state transitions that the controller will
2198 * perform automatically. We configure it with a simple
2199 * heuristic: we are willing to spend at most 2% of the time
2200 * transitioning between power states. Therefore, when running
2201 * in any given state, we will enter the next lower-power
2202 * non-operational state after waiting 50 * (enlat + exlat)
2203 * microseconds, as long as that state's exit latency is under
2204 * the requested maximum latency.
2206 * We will not autonomously enter any non-operational state for
2207 * which the total latency exceeds ps_max_latency_us. Users
2208 * can set ps_max_latency_us to zero to turn off APST.
2212 struct nvme_feat_auto_pst *table;
2218 * If APST isn't supported or if we haven't been initialized yet,
2219 * then don't do anything.
2224 if (ctrl->npss > 31) {
2225 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2229 table = kzalloc(sizeof(*table), GFP_KERNEL);
2233 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2234 /* Turn off APST. */
2236 dev_dbg(ctrl->device, "APST disabled\n");
2238 __le64 target = cpu_to_le64(0);
2242 * Walk through all states from lowest- to highest-power.
2243 * According to the spec, lower-numbered states use more
2244 * power. NPSS, despite the name, is the index of the
2245 * lowest-power state, not the number of states.
2247 for (state = (int)ctrl->npss; state >= 0; state--) {
2248 u64 total_latency_us, exit_latency_us, transition_ms;
2251 table->entries[state] = target;
2254 * Don't allow transitions to the deepest state
2255 * if it's quirked off.
2257 if (state == ctrl->npss &&
2258 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2262 * Is this state a useful non-operational state for
2263 * higher-power states to autonomously transition to?
2265 if (!(ctrl->psd[state].flags &
2266 NVME_PS_FLAGS_NON_OP_STATE))
2270 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2271 if (exit_latency_us > ctrl->ps_max_latency_us)
2276 le32_to_cpu(ctrl->psd[state].entry_lat);
2279 * This state is good. Use it as the APST idle
2280 * target for higher power states.
2282 transition_ms = total_latency_us + 19;
2283 do_div(transition_ms, 20);
2284 if (transition_ms > (1 << 24) - 1)
2285 transition_ms = (1 << 24) - 1;
2287 target = cpu_to_le64((state << 3) |
2288 (transition_ms << 8));
2293 if (total_latency_us > max_lat_us)
2294 max_lat_us = total_latency_us;
2300 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2302 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2303 max_ps, max_lat_us, (int)sizeof(*table), table);
2307 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2308 table, sizeof(*table), NULL);
2310 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2316 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2318 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2322 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2323 case PM_QOS_LATENCY_ANY:
2331 if (ctrl->ps_max_latency_us != latency) {
2332 ctrl->ps_max_latency_us = latency;
2333 nvme_configure_apst(ctrl);
2337 struct nvme_core_quirk_entry {
2339 * NVMe model and firmware strings are padded with spaces. For
2340 * simplicity, strings in the quirk table are padded with NULLs
2346 unsigned long quirks;
2349 static const struct nvme_core_quirk_entry core_quirks[] = {
2352 * This Toshiba device seems to die using any APST states. See:
2353 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2356 .mn = "THNSF5256GPUK TOSHIBA",
2357 .quirks = NVME_QUIRK_NO_APST,
2361 * This LiteON CL1-3D*-Q11 firmware version has a race
2362 * condition associated with actions related to suspend to idle
2363 * LiteON has resolved the problem in future firmware
2367 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2371 * This Kingston E8FK11.T firmware version has no interrupt
2372 * after resume with actions related to suspend to idle
2373 * https://bugzilla.kernel.org/show_bug.cgi?id=204887
2377 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2381 /* match is null-terminated but idstr is space-padded. */
2382 static bool string_matches(const char *idstr, const char *match, size_t len)
2389 matchlen = strlen(match);
2390 WARN_ON_ONCE(matchlen > len);
2392 if (memcmp(idstr, match, matchlen))
2395 for (; matchlen < len; matchlen++)
2396 if (idstr[matchlen] != ' ')
2402 static bool quirk_matches(const struct nvme_id_ctrl *id,
2403 const struct nvme_core_quirk_entry *q)
2405 return q->vid == le16_to_cpu(id->vid) &&
2406 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2407 string_matches(id->fr, q->fr, sizeof(id->fr));
2410 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2411 struct nvme_id_ctrl *id)
2416 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2417 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2418 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2419 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2423 if (ctrl->vs >= NVME_VS(1, 2, 1))
2424 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2427 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2428 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2429 "nqn.2014.08.org.nvmexpress:%04x%04x",
2430 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2431 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2432 off += sizeof(id->sn);
2433 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2434 off += sizeof(id->mn);
2435 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2438 static void nvme_release_subsystem(struct device *dev)
2440 struct nvme_subsystem *subsys =
2441 container_of(dev, struct nvme_subsystem, dev);
2443 if (subsys->instance >= 0)
2444 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2448 static void nvme_destroy_subsystem(struct kref *ref)
2450 struct nvme_subsystem *subsys =
2451 container_of(ref, struct nvme_subsystem, ref);
2453 mutex_lock(&nvme_subsystems_lock);
2454 list_del(&subsys->entry);
2455 mutex_unlock(&nvme_subsystems_lock);
2457 ida_destroy(&subsys->ns_ida);
2458 device_del(&subsys->dev);
2459 put_device(&subsys->dev);
2462 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2464 kref_put(&subsys->ref, nvme_destroy_subsystem);
2467 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2469 struct nvme_subsystem *subsys;
2471 lockdep_assert_held(&nvme_subsystems_lock);
2474 * Fail matches for discovery subsystems. This results
2475 * in each discovery controller bound to a unique subsystem.
2476 * This avoids issues with validating controller values
2477 * that can only be true when there is a single unique subsystem.
2478 * There may be multiple and completely independent entities
2479 * that provide discovery controllers.
2481 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2484 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2485 if (strcmp(subsys->subnqn, subsysnqn))
2487 if (!kref_get_unless_zero(&subsys->ref))
2495 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2496 struct device_attribute subsys_attr_##_name = \
2497 __ATTR(_name, _mode, _show, NULL)
2499 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2500 struct device_attribute *attr,
2503 struct nvme_subsystem *subsys =
2504 container_of(dev, struct nvme_subsystem, dev);
2506 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2508 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2510 #define nvme_subsys_show_str_function(field) \
2511 static ssize_t subsys_##field##_show(struct device *dev, \
2512 struct device_attribute *attr, char *buf) \
2514 struct nvme_subsystem *subsys = \
2515 container_of(dev, struct nvme_subsystem, dev); \
2516 return sprintf(buf, "%.*s\n", \
2517 (int)sizeof(subsys->field), subsys->field); \
2519 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2521 nvme_subsys_show_str_function(model);
2522 nvme_subsys_show_str_function(serial);
2523 nvme_subsys_show_str_function(firmware_rev);
2525 static struct attribute *nvme_subsys_attrs[] = {
2526 &subsys_attr_model.attr,
2527 &subsys_attr_serial.attr,
2528 &subsys_attr_firmware_rev.attr,
2529 &subsys_attr_subsysnqn.attr,
2530 #ifdef CONFIG_NVME_MULTIPATH
2531 &subsys_attr_iopolicy.attr,
2536 static struct attribute_group nvme_subsys_attrs_group = {
2537 .attrs = nvme_subsys_attrs,
2540 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2541 &nvme_subsys_attrs_group,
2545 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2546 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2548 struct nvme_ctrl *tmp;
2550 lockdep_assert_held(&nvme_subsystems_lock);
2552 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2553 if (tmp->state == NVME_CTRL_DELETING ||
2554 tmp->state == NVME_CTRL_DEAD)
2557 if (tmp->cntlid == ctrl->cntlid) {
2558 dev_err(ctrl->device,
2559 "Duplicate cntlid %u with %s, rejecting\n",
2560 ctrl->cntlid, dev_name(tmp->device));
2564 if ((id->cmic & (1 << 1)) ||
2565 (ctrl->opts && ctrl->opts->discovery_nqn))
2568 dev_err(ctrl->device,
2569 "Subsystem does not support multiple controllers\n");
2576 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2578 struct nvme_subsystem *subsys, *found;
2581 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2585 subsys->instance = -1;
2586 mutex_init(&subsys->lock);
2587 kref_init(&subsys->ref);
2588 INIT_LIST_HEAD(&subsys->ctrls);
2589 INIT_LIST_HEAD(&subsys->nsheads);
2590 nvme_init_subnqn(subsys, ctrl, id);
2591 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2592 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2593 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2594 subsys->vendor_id = le16_to_cpu(id->vid);
2595 subsys->cmic = id->cmic;
2596 subsys->awupf = le16_to_cpu(id->awupf);
2597 #ifdef CONFIG_NVME_MULTIPATH
2598 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2601 subsys->dev.class = nvme_subsys_class;
2602 subsys->dev.release = nvme_release_subsystem;
2603 subsys->dev.groups = nvme_subsys_attrs_groups;
2604 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2605 device_initialize(&subsys->dev);
2607 mutex_lock(&nvme_subsystems_lock);
2608 found = __nvme_find_get_subsystem(subsys->subnqn);
2610 put_device(&subsys->dev);
2613 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2615 goto out_put_subsystem;
2618 ret = device_add(&subsys->dev);
2620 dev_err(ctrl->device,
2621 "failed to register subsystem device.\n");
2622 put_device(&subsys->dev);
2625 ida_init(&subsys->ns_ida);
2626 list_add_tail(&subsys->entry, &nvme_subsystems);
2629 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2630 dev_name(ctrl->device));
2632 dev_err(ctrl->device,
2633 "failed to create sysfs link from subsystem.\n");
2634 goto out_put_subsystem;
2638 subsys->instance = ctrl->instance;
2639 ctrl->subsys = subsys;
2640 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2641 mutex_unlock(&nvme_subsystems_lock);
2645 nvme_put_subsystem(subsys);
2647 mutex_unlock(&nvme_subsystems_lock);
2651 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2652 void *log, size_t size, u64 offset)
2654 struct nvme_command c = { };
2655 unsigned long dwlen = size / 4 - 1;
2657 c.get_log_page.opcode = nvme_admin_get_log_page;
2658 c.get_log_page.nsid = cpu_to_le32(nsid);
2659 c.get_log_page.lid = log_page;
2660 c.get_log_page.lsp = lsp;
2661 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2662 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2663 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2664 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2666 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2669 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2674 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2679 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2680 ctrl->effects, sizeof(*ctrl->effects), 0);
2682 kfree(ctrl->effects);
2683 ctrl->effects = NULL;
2689 * Initialize the cached copies of the Identify data and various controller
2690 * register in our nvme_ctrl structure. This should be called as soon as
2691 * the admin queue is fully up and running.
2693 int nvme_init_identify(struct nvme_ctrl *ctrl)
2695 struct nvme_id_ctrl *id;
2696 int ret, page_shift;
2698 bool prev_apst_enabled;
2700 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2702 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2705 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2706 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2708 if (ctrl->vs >= NVME_VS(1, 1, 0))
2709 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2711 ret = nvme_identify_ctrl(ctrl, &id);
2713 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2717 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2718 ret = nvme_get_effects_log(ctrl);
2723 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2724 ctrl->cntlid = le16_to_cpu(id->cntlid);
2726 if (!ctrl->identified) {
2729 ret = nvme_init_subsystem(ctrl, id);
2734 * Check for quirks. Quirk can depend on firmware version,
2735 * so, in principle, the set of quirks present can change
2736 * across a reset. As a possible future enhancement, we
2737 * could re-scan for quirks every time we reinitialize
2738 * the device, but we'd have to make sure that the driver
2739 * behaves intelligently if the quirks change.
2741 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2742 if (quirk_matches(id, &core_quirks[i]))
2743 ctrl->quirks |= core_quirks[i].quirks;
2747 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2748 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2749 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2752 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2753 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2754 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2756 ctrl->oacs = le16_to_cpu(id->oacs);
2757 ctrl->oncs = le16_to_cpu(id->oncs);
2758 ctrl->mtfa = le16_to_cpu(id->mtfa);
2759 ctrl->oaes = le32_to_cpu(id->oaes);
2760 atomic_set(&ctrl->abort_limit, id->acl + 1);
2761 ctrl->vwc = id->vwc;
2763 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2765 max_hw_sectors = UINT_MAX;
2766 ctrl->max_hw_sectors =
2767 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2769 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2770 ctrl->sgls = le32_to_cpu(id->sgls);
2771 ctrl->kas = le16_to_cpu(id->kas);
2772 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2773 ctrl->ctratt = le32_to_cpu(id->ctratt);
2777 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2779 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2780 shutdown_timeout, 60);
2782 if (ctrl->shutdown_timeout != shutdown_timeout)
2783 dev_info(ctrl->device,
2784 "Shutdown timeout set to %u seconds\n",
2785 ctrl->shutdown_timeout);
2787 ctrl->shutdown_timeout = shutdown_timeout;
2789 ctrl->npss = id->npss;
2790 ctrl->apsta = id->apsta;
2791 prev_apst_enabled = ctrl->apst_enabled;
2792 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2793 if (force_apst && id->apsta) {
2794 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2795 ctrl->apst_enabled = true;
2797 ctrl->apst_enabled = false;
2800 ctrl->apst_enabled = id->apsta;
2802 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2804 if (ctrl->ops->flags & NVME_F_FABRICS) {
2805 ctrl->icdoff = le16_to_cpu(id->icdoff);
2806 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2807 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2808 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2811 * In fabrics we need to verify the cntlid matches the
2814 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2819 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2820 dev_err(ctrl->device,
2821 "keep-alive support is mandatory for fabrics\n");
2826 ctrl->hmpre = le32_to_cpu(id->hmpre);
2827 ctrl->hmmin = le32_to_cpu(id->hmmin);
2828 ctrl->hmminds = le32_to_cpu(id->hmminds);
2829 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2832 ret = nvme_mpath_init(ctrl, id);
2838 if (ctrl->apst_enabled && !prev_apst_enabled)
2839 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2840 else if (!ctrl->apst_enabled && prev_apst_enabled)
2841 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2843 ret = nvme_configure_apst(ctrl);
2847 ret = nvme_configure_timestamp(ctrl);
2851 ret = nvme_configure_directives(ctrl);
2855 ret = nvme_configure_acre(ctrl);
2859 ctrl->identified = true;
2867 EXPORT_SYMBOL_GPL(nvme_init_identify);
2869 static int nvme_dev_open(struct inode *inode, struct file *file)
2871 struct nvme_ctrl *ctrl =
2872 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2874 switch (ctrl->state) {
2875 case NVME_CTRL_LIVE:
2876 case NVME_CTRL_ADMIN_ONLY:
2879 return -EWOULDBLOCK;
2882 file->private_data = ctrl;
2886 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2891 down_read(&ctrl->namespaces_rwsem);
2892 if (list_empty(&ctrl->namespaces)) {
2897 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2898 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2899 dev_warn(ctrl->device,
2900 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2905 dev_warn(ctrl->device,
2906 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2907 kref_get(&ns->kref);
2908 up_read(&ctrl->namespaces_rwsem);
2910 ret = nvme_user_cmd(ctrl, ns, argp);
2915 up_read(&ctrl->namespaces_rwsem);
2919 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2922 struct nvme_ctrl *ctrl = file->private_data;
2923 void __user *argp = (void __user *)arg;
2926 case NVME_IOCTL_ADMIN_CMD:
2927 return nvme_user_cmd(ctrl, NULL, argp);
2928 case NVME_IOCTL_ADMIN64_CMD:
2929 return nvme_user_cmd64(ctrl, NULL, argp);
2930 case NVME_IOCTL_IO_CMD:
2931 return nvme_dev_user_cmd(ctrl, argp);
2932 case NVME_IOCTL_RESET:
2933 dev_warn(ctrl->device, "resetting controller\n");
2934 return nvme_reset_ctrl_sync(ctrl);
2935 case NVME_IOCTL_SUBSYS_RESET:
2936 return nvme_reset_subsystem(ctrl);
2937 case NVME_IOCTL_RESCAN:
2938 nvme_queue_scan(ctrl);
2945 static const struct file_operations nvme_dev_fops = {
2946 .owner = THIS_MODULE,
2947 .open = nvme_dev_open,
2948 .unlocked_ioctl = nvme_dev_ioctl,
2949 .compat_ioctl = nvme_dev_ioctl,
2952 static ssize_t nvme_sysfs_reset(struct device *dev,
2953 struct device_attribute *attr, const char *buf,
2956 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2959 ret = nvme_reset_ctrl_sync(ctrl);
2964 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2966 static ssize_t nvme_sysfs_rescan(struct device *dev,
2967 struct device_attribute *attr, const char *buf,
2970 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2972 nvme_queue_scan(ctrl);
2975 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
2977 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
2979 struct gendisk *disk = dev_to_disk(dev);
2981 if (disk->fops == &nvme_fops)
2982 return nvme_get_ns_from_dev(dev)->head;
2984 return disk->private_data;
2987 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
2990 struct nvme_ns_head *head = dev_to_ns_head(dev);
2991 struct nvme_ns_ids *ids = &head->ids;
2992 struct nvme_subsystem *subsys = head->subsys;
2993 int serial_len = sizeof(subsys->serial);
2994 int model_len = sizeof(subsys->model);
2996 if (!uuid_is_null(&ids->uuid))
2997 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
2999 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3000 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3002 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3003 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3005 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3006 subsys->serial[serial_len - 1] == '\0'))
3008 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3009 subsys->model[model_len - 1] == '\0'))
3012 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3013 serial_len, subsys->serial, model_len, subsys->model,
3016 static DEVICE_ATTR_RO(wwid);
3018 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3021 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3023 static DEVICE_ATTR_RO(nguid);
3025 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3028 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3030 /* For backward compatibility expose the NGUID to userspace if
3031 * we have no UUID set
3033 if (uuid_is_null(&ids->uuid)) {
3034 printk_ratelimited(KERN_WARNING
3035 "No UUID available providing old NGUID\n");
3036 return sprintf(buf, "%pU\n", ids->nguid);
3038 return sprintf(buf, "%pU\n", &ids->uuid);
3040 static DEVICE_ATTR_RO(uuid);
3042 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3045 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3047 static DEVICE_ATTR_RO(eui);
3049 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3052 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3054 static DEVICE_ATTR_RO(nsid);
3056 static struct attribute *nvme_ns_id_attrs[] = {
3057 &dev_attr_wwid.attr,
3058 &dev_attr_uuid.attr,
3059 &dev_attr_nguid.attr,
3061 &dev_attr_nsid.attr,
3062 #ifdef CONFIG_NVME_MULTIPATH
3063 &dev_attr_ana_grpid.attr,
3064 &dev_attr_ana_state.attr,
3069 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3070 struct attribute *a, int n)
3072 struct device *dev = container_of(kobj, struct device, kobj);
3073 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3075 if (a == &dev_attr_uuid.attr) {
3076 if (uuid_is_null(&ids->uuid) &&
3077 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3080 if (a == &dev_attr_nguid.attr) {
3081 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3084 if (a == &dev_attr_eui.attr) {
3085 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3088 #ifdef CONFIG_NVME_MULTIPATH
3089 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3090 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3092 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3099 static const struct attribute_group nvme_ns_id_attr_group = {
3100 .attrs = nvme_ns_id_attrs,
3101 .is_visible = nvme_ns_id_attrs_are_visible,
3104 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3105 &nvme_ns_id_attr_group,
3107 &nvme_nvm_attr_group,
3112 #define nvme_show_str_function(field) \
3113 static ssize_t field##_show(struct device *dev, \
3114 struct device_attribute *attr, char *buf) \
3116 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3117 return sprintf(buf, "%.*s\n", \
3118 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3120 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3122 nvme_show_str_function(model);
3123 nvme_show_str_function(serial);
3124 nvme_show_str_function(firmware_rev);
3126 #define nvme_show_int_function(field) \
3127 static ssize_t field##_show(struct device *dev, \
3128 struct device_attribute *attr, char *buf) \
3130 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3131 return sprintf(buf, "%d\n", ctrl->field); \
3133 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3135 nvme_show_int_function(cntlid);
3136 nvme_show_int_function(numa_node);
3137 nvme_show_int_function(queue_count);
3138 nvme_show_int_function(sqsize);
3140 static ssize_t nvme_sysfs_delete(struct device *dev,
3141 struct device_attribute *attr, const char *buf,
3144 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3146 if (device_remove_file_self(dev, attr))
3147 nvme_delete_ctrl_sync(ctrl);
3150 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3152 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3153 struct device_attribute *attr,
3156 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3158 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3160 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3162 static ssize_t nvme_sysfs_show_state(struct device *dev,
3163 struct device_attribute *attr,
3166 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3167 static const char *const state_name[] = {
3168 [NVME_CTRL_NEW] = "new",
3169 [NVME_CTRL_LIVE] = "live",
3170 [NVME_CTRL_ADMIN_ONLY] = "only-admin",
3171 [NVME_CTRL_RESETTING] = "resetting",
3172 [NVME_CTRL_CONNECTING] = "connecting",
3173 [NVME_CTRL_DELETING] = "deleting",
3174 [NVME_CTRL_DEAD] = "dead",
3177 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3178 state_name[ctrl->state])
3179 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3181 return sprintf(buf, "unknown state\n");
3184 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3186 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3187 struct device_attribute *attr,
3190 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3192 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3194 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3196 static ssize_t nvme_sysfs_show_address(struct device *dev,
3197 struct device_attribute *attr,
3200 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3202 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3204 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3206 static struct attribute *nvme_dev_attrs[] = {
3207 &dev_attr_reset_controller.attr,
3208 &dev_attr_rescan_controller.attr,
3209 &dev_attr_model.attr,
3210 &dev_attr_serial.attr,
3211 &dev_attr_firmware_rev.attr,
3212 &dev_attr_cntlid.attr,
3213 &dev_attr_delete_controller.attr,
3214 &dev_attr_transport.attr,
3215 &dev_attr_subsysnqn.attr,
3216 &dev_attr_address.attr,
3217 &dev_attr_state.attr,
3218 &dev_attr_numa_node.attr,
3219 &dev_attr_queue_count.attr,
3220 &dev_attr_sqsize.attr,
3224 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3225 struct attribute *a, int n)
3227 struct device *dev = container_of(kobj, struct device, kobj);
3228 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3230 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3232 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3238 static struct attribute_group nvme_dev_attrs_group = {
3239 .attrs = nvme_dev_attrs,
3240 .is_visible = nvme_dev_attrs_are_visible,
3243 static const struct attribute_group *nvme_dev_attr_groups[] = {
3244 &nvme_dev_attrs_group,
3248 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3251 struct nvme_ns_head *h;
3253 lockdep_assert_held(&subsys->lock);
3255 list_for_each_entry(h, &subsys->nsheads, entry) {
3256 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3263 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3264 struct nvme_ns_head *new)
3266 struct nvme_ns_head *h;
3268 lockdep_assert_held(&subsys->lock);
3270 list_for_each_entry(h, &subsys->nsheads, entry) {
3271 if (nvme_ns_ids_valid(&new->ids) &&
3272 !list_empty(&h->list) &&
3273 nvme_ns_ids_equal(&new->ids, &h->ids))
3280 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3281 unsigned nsid, struct nvme_id_ns *id)
3283 struct nvme_ns_head *head;
3284 size_t size = sizeof(*head);
3287 #ifdef CONFIG_NVME_MULTIPATH
3288 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3291 head = kzalloc(size, GFP_KERNEL);
3294 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3297 head->instance = ret;
3298 INIT_LIST_HEAD(&head->list);
3299 ret = init_srcu_struct(&head->srcu);
3301 goto out_ida_remove;
3302 head->subsys = ctrl->subsys;
3304 kref_init(&head->ref);
3306 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3308 goto out_cleanup_srcu;
3310 ret = __nvme_check_ids(ctrl->subsys, head);
3312 dev_err(ctrl->device,
3313 "duplicate IDs for nsid %d\n", nsid);
3314 goto out_cleanup_srcu;
3317 ret = nvme_mpath_alloc_disk(ctrl, head);
3319 goto out_cleanup_srcu;
3321 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3323 kref_get(&ctrl->subsys->ref);
3327 cleanup_srcu_struct(&head->srcu);
3329 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3334 ret = blk_status_to_errno(nvme_error_status(ret));
3335 return ERR_PTR(ret);
3338 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3339 struct nvme_id_ns *id)
3341 struct nvme_ctrl *ctrl = ns->ctrl;
3342 bool is_shared = id->nmic & (1 << 0);
3343 struct nvme_ns_head *head = NULL;
3346 mutex_lock(&ctrl->subsys->lock);
3348 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3350 head = nvme_alloc_ns_head(ctrl, nsid, id);
3352 ret = PTR_ERR(head);
3356 struct nvme_ns_ids ids;
3358 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3362 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3363 dev_err(ctrl->device,
3364 "IDs don't match for shared namespace %d\n",
3371 list_add_tail(&ns->siblings, &head->list);
3375 mutex_unlock(&ctrl->subsys->lock);
3377 ret = blk_status_to_errno(nvme_error_status(ret));
3381 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3383 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3384 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3386 return nsa->head->ns_id - nsb->head->ns_id;
3389 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3391 struct nvme_ns *ns, *ret = NULL;
3393 down_read(&ctrl->namespaces_rwsem);
3394 list_for_each_entry(ns, &ctrl->namespaces, list) {
3395 if (ns->head->ns_id == nsid) {
3396 if (!kref_get_unless_zero(&ns->kref))
3401 if (ns->head->ns_id > nsid)
3404 up_read(&ctrl->namespaces_rwsem);
3408 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3410 struct streams_directive_params s;
3413 if (!ctrl->nr_streams)
3416 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3420 ns->sws = le32_to_cpu(s.sws);
3421 ns->sgs = le16_to_cpu(s.sgs);
3424 unsigned int bs = 1 << ns->lba_shift;
3426 blk_queue_io_min(ns->queue, bs * ns->sws);
3428 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3434 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3437 struct gendisk *disk;
3438 struct nvme_id_ns *id;
3439 char disk_name[DISK_NAME_LEN];
3440 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3442 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3446 ns->queue = blk_mq_init_queue(ctrl->tagset);
3447 if (IS_ERR(ns->queue)) {
3448 ret = PTR_ERR(ns->queue);
3452 if (ctrl->opts && ctrl->opts->data_digest)
3453 ns->queue->backing_dev_info->capabilities
3454 |= BDI_CAP_STABLE_WRITES;
3456 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3457 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3458 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3460 ns->queue->queuedata = ns;
3463 kref_init(&ns->kref);
3464 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3466 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3467 nvme_set_queue_limits(ctrl, ns->queue);
3469 ret = nvme_identify_ns(ctrl, nsid, &id);
3471 goto out_free_queue;
3473 if (id->ncap == 0) {
3478 ret = nvme_init_ns_head(ns, nsid, id);
3481 nvme_setup_streams_ns(ctrl, ns);
3482 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3484 disk = alloc_disk_node(0, node);
3490 disk->fops = &nvme_fops;
3491 disk->private_data = ns;
3492 disk->queue = ns->queue;
3493 disk->flags = flags;
3494 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3497 __nvme_revalidate_disk(disk, id);
3499 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3500 ret = nvme_nvm_register(ns, disk_name, node);
3502 dev_warn(ctrl->device, "LightNVM init failure\n");
3507 down_write(&ctrl->namespaces_rwsem);
3508 list_add_tail(&ns->list, &ctrl->namespaces);
3509 up_write(&ctrl->namespaces_rwsem);
3511 nvme_get_ctrl(ctrl);
3513 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3515 nvme_mpath_add_disk(ns, id);
3516 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3523 mutex_lock(&ctrl->subsys->lock);
3524 list_del_rcu(&ns->siblings);
3525 mutex_unlock(&ctrl->subsys->lock);
3526 nvme_put_ns_head(ns->head);
3530 blk_cleanup_queue(ns->queue);
3534 ret = blk_status_to_errno(nvme_error_status(ret));
3538 static void nvme_ns_remove(struct nvme_ns *ns)
3540 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3543 nvme_fault_inject_fini(&ns->fault_inject);
3545 mutex_lock(&ns->ctrl->subsys->lock);
3546 list_del_rcu(&ns->siblings);
3547 mutex_unlock(&ns->ctrl->subsys->lock);
3548 synchronize_rcu(); /* guarantee not available in head->list */
3549 nvme_mpath_clear_current_path(ns);
3550 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3552 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3553 del_gendisk(ns->disk);
3554 blk_cleanup_queue(ns->queue);
3555 if (blk_get_integrity(ns->disk))
3556 blk_integrity_unregister(ns->disk);
3559 down_write(&ns->ctrl->namespaces_rwsem);
3560 list_del_init(&ns->list);
3561 up_write(&ns->ctrl->namespaces_rwsem);
3563 nvme_mpath_check_last_path(ns);
3567 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3571 ns = nvme_find_get_ns(ctrl, nsid);
3573 if (ns->disk && revalidate_disk(ns->disk))
3577 nvme_alloc_ns(ctrl, nsid);
3580 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3583 struct nvme_ns *ns, *next;
3586 down_write(&ctrl->namespaces_rwsem);
3587 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3588 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3589 list_move_tail(&ns->list, &rm_list);
3591 up_write(&ctrl->namespaces_rwsem);
3593 list_for_each_entry_safe(ns, next, &rm_list, list)
3598 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3602 unsigned i, j, nsid, prev = 0;
3603 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3606 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3610 for (i = 0; i < num_lists; i++) {
3611 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3615 for (j = 0; j < min(nn, 1024U); j++) {
3616 nsid = le32_to_cpu(ns_list[j]);
3620 nvme_validate_ns(ctrl, nsid);
3622 while (++prev < nsid) {
3623 ns = nvme_find_get_ns(ctrl, prev);
3633 nvme_remove_invalid_namespaces(ctrl, prev);
3639 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3643 for (i = 1; i <= nn; i++)
3644 nvme_validate_ns(ctrl, i);
3646 nvme_remove_invalid_namespaces(ctrl, nn);
3649 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3651 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3655 log = kzalloc(log_size, GFP_KERNEL);
3660 * We need to read the log to clear the AEN, but we don't want to rely
3661 * on it for the changed namespace information as userspace could have
3662 * raced with us in reading the log page, which could cause us to miss
3665 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3668 dev_warn(ctrl->device,
3669 "reading changed ns log failed: %d\n", error);
3674 static void nvme_scan_work(struct work_struct *work)
3676 struct nvme_ctrl *ctrl =
3677 container_of(work, struct nvme_ctrl, scan_work);
3678 struct nvme_id_ctrl *id;
3681 if (ctrl->state != NVME_CTRL_LIVE)
3684 WARN_ON_ONCE(!ctrl->tagset);
3686 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3687 dev_info(ctrl->device, "rescanning namespaces.\n");
3688 nvme_clear_changed_ns_log(ctrl);
3691 if (nvme_identify_ctrl(ctrl, &id))
3694 mutex_lock(&ctrl->scan_lock);
3695 nn = le32_to_cpu(id->nn);
3696 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3697 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3698 if (!nvme_scan_ns_list(ctrl, nn))
3701 nvme_scan_ns_sequential(ctrl, nn);
3703 mutex_unlock(&ctrl->scan_lock);
3705 down_write(&ctrl->namespaces_rwsem);
3706 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3707 up_write(&ctrl->namespaces_rwsem);
3711 * This function iterates the namespace list unlocked to allow recovery from
3712 * controller failure. It is up to the caller to ensure the namespace list is
3713 * not modified by scan work while this function is executing.
3715 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3717 struct nvme_ns *ns, *next;
3721 * make sure to requeue I/O to all namespaces as these
3722 * might result from the scan itself and must complete
3723 * for the scan_work to make progress
3725 nvme_mpath_clear_ctrl_paths(ctrl);
3727 /* prevent racing with ns scanning */
3728 flush_work(&ctrl->scan_work);
3731 * The dead states indicates the controller was not gracefully
3732 * disconnected. In that case, we won't be able to flush any data while
3733 * removing the namespaces' disks; fail all the queues now to avoid
3734 * potentially having to clean up the failed sync later.
3736 if (ctrl->state == NVME_CTRL_DEAD)
3737 nvme_kill_queues(ctrl);
3739 down_write(&ctrl->namespaces_rwsem);
3740 list_splice_init(&ctrl->namespaces, &ns_list);
3741 up_write(&ctrl->namespaces_rwsem);
3743 list_for_each_entry_safe(ns, next, &ns_list, list)
3746 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3748 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3750 struct nvme_ctrl *ctrl =
3751 container_of(dev, struct nvme_ctrl, ctrl_device);
3752 struct nvmf_ctrl_options *opts = ctrl->opts;
3755 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3760 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3764 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3765 opts->trsvcid ?: "none");
3769 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3770 opts->host_traddr ?: "none");
3775 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3777 char *envp[2] = { NULL, NULL };
3778 u32 aen_result = ctrl->aen_result;
3780 ctrl->aen_result = 0;
3784 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3787 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3791 static void nvme_async_event_work(struct work_struct *work)
3793 struct nvme_ctrl *ctrl =
3794 container_of(work, struct nvme_ctrl, async_event_work);
3796 nvme_aen_uevent(ctrl);
3797 ctrl->ops->submit_async_event(ctrl);
3800 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3805 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3811 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3814 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3816 struct nvme_fw_slot_info_log *log;
3818 log = kmalloc(sizeof(*log), GFP_KERNEL);
3822 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3824 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3828 static void nvme_fw_act_work(struct work_struct *work)
3830 struct nvme_ctrl *ctrl = container_of(work,
3831 struct nvme_ctrl, fw_act_work);
3832 unsigned long fw_act_timeout;
3835 fw_act_timeout = jiffies +
3836 msecs_to_jiffies(ctrl->mtfa * 100);
3838 fw_act_timeout = jiffies +
3839 msecs_to_jiffies(admin_timeout * 1000);
3841 nvme_stop_queues(ctrl);
3842 while (nvme_ctrl_pp_status(ctrl)) {
3843 if (time_after(jiffies, fw_act_timeout)) {
3844 dev_warn(ctrl->device,
3845 "Fw activation timeout, reset controller\n");
3846 nvme_reset_ctrl(ctrl);
3852 if (ctrl->state != NVME_CTRL_LIVE)
3855 nvme_start_queues(ctrl);
3856 /* read FW slot information to clear the AER */
3857 nvme_get_fw_slot_info(ctrl);
3860 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3862 u32 aer_notice_type = (result & 0xff00) >> 8;
3864 trace_nvme_async_event(ctrl, aer_notice_type);
3866 switch (aer_notice_type) {
3867 case NVME_AER_NOTICE_NS_CHANGED:
3868 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3869 nvme_queue_scan(ctrl);
3871 case NVME_AER_NOTICE_FW_ACT_STARTING:
3872 queue_work(nvme_wq, &ctrl->fw_act_work);
3874 #ifdef CONFIG_NVME_MULTIPATH
3875 case NVME_AER_NOTICE_ANA:
3876 if (!ctrl->ana_log_buf)
3878 queue_work(nvme_wq, &ctrl->ana_work);
3881 case NVME_AER_NOTICE_DISC_CHANGED:
3882 ctrl->aen_result = result;
3885 dev_warn(ctrl->device, "async event result %08x\n", result);
3889 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3890 volatile union nvme_result *res)
3892 u32 result = le32_to_cpu(res->u32);
3893 u32 aer_type = result & 0x07;
3895 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3899 case NVME_AER_NOTICE:
3900 nvme_handle_aen_notice(ctrl, result);
3902 case NVME_AER_ERROR:
3903 case NVME_AER_SMART:
3906 trace_nvme_async_event(ctrl, aer_type);
3907 ctrl->aen_result = result;
3912 queue_work(nvme_wq, &ctrl->async_event_work);
3914 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3916 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3918 nvme_mpath_stop(ctrl);
3919 nvme_stop_keep_alive(ctrl);
3920 flush_work(&ctrl->async_event_work);
3921 cancel_work_sync(&ctrl->fw_act_work);
3923 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3925 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3928 nvme_start_keep_alive(ctrl);
3930 nvme_enable_aen(ctrl);
3932 if (ctrl->queue_count > 1) {
3933 nvme_queue_scan(ctrl);
3934 nvme_start_queues(ctrl);
3937 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3939 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3941 nvme_fault_inject_fini(&ctrl->fault_inject);
3942 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3943 cdev_device_del(&ctrl->cdev, ctrl->device);
3945 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3947 static void nvme_free_ctrl(struct device *dev)
3949 struct nvme_ctrl *ctrl =
3950 container_of(dev, struct nvme_ctrl, ctrl_device);
3951 struct nvme_subsystem *subsys = ctrl->subsys;
3953 if (subsys && ctrl->instance != subsys->instance)
3954 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
3956 kfree(ctrl->effects);
3957 nvme_mpath_uninit(ctrl);
3958 __free_page(ctrl->discard_page);
3961 mutex_lock(&nvme_subsystems_lock);
3962 list_del(&ctrl->subsys_entry);
3963 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
3964 mutex_unlock(&nvme_subsystems_lock);
3967 ctrl->ops->free_ctrl(ctrl);
3970 nvme_put_subsystem(subsys);
3974 * Initialize a NVMe controller structures. This needs to be called during
3975 * earliest initialization so that we have the initialized structured around
3978 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
3979 const struct nvme_ctrl_ops *ops, unsigned long quirks)
3983 ctrl->state = NVME_CTRL_NEW;
3984 spin_lock_init(&ctrl->lock);
3985 mutex_init(&ctrl->scan_lock);
3986 INIT_LIST_HEAD(&ctrl->namespaces);
3987 init_rwsem(&ctrl->namespaces_rwsem);
3990 ctrl->quirks = quirks;
3991 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
3992 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
3993 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
3994 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
3996 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
3997 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
3998 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4000 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4002 ctrl->discard_page = alloc_page(GFP_KERNEL);
4003 if (!ctrl->discard_page) {
4008 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4011 ctrl->instance = ret;
4013 device_initialize(&ctrl->ctrl_device);
4014 ctrl->device = &ctrl->ctrl_device;
4015 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4016 ctrl->device->class = nvme_class;
4017 ctrl->device->parent = ctrl->dev;
4018 ctrl->device->groups = nvme_dev_attr_groups;
4019 ctrl->device->release = nvme_free_ctrl;
4020 dev_set_drvdata(ctrl->device, ctrl);
4021 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4023 goto out_release_instance;
4025 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4026 ctrl->cdev.owner = ops->module;
4027 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4032 * Initialize latency tolerance controls. The sysfs files won't
4033 * be visible to userspace unless the device actually supports APST.
4035 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4036 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4037 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4039 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4043 kfree_const(ctrl->device->kobj.name);
4044 out_release_instance:
4045 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4047 if (ctrl->discard_page)
4048 __free_page(ctrl->discard_page);
4051 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4054 * nvme_kill_queues(): Ends all namespace queues
4055 * @ctrl: the dead controller that needs to end
4057 * Call this function when the driver determines it is unable to get the
4058 * controller in a state capable of servicing IO.
4060 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4064 down_read(&ctrl->namespaces_rwsem);
4066 /* Forcibly unquiesce queues to avoid blocking dispatch */
4067 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4068 blk_mq_unquiesce_queue(ctrl->admin_q);
4070 list_for_each_entry(ns, &ctrl->namespaces, list)
4071 nvme_set_queue_dying(ns);
4073 up_read(&ctrl->namespaces_rwsem);
4075 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4077 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4081 down_read(&ctrl->namespaces_rwsem);
4082 list_for_each_entry(ns, &ctrl->namespaces, list)
4083 blk_mq_unfreeze_queue(ns->queue);
4084 up_read(&ctrl->namespaces_rwsem);
4086 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4088 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4092 down_read(&ctrl->namespaces_rwsem);
4093 list_for_each_entry(ns, &ctrl->namespaces, list) {
4094 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4098 up_read(&ctrl->namespaces_rwsem);
4100 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4102 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4106 down_read(&ctrl->namespaces_rwsem);
4107 list_for_each_entry(ns, &ctrl->namespaces, list)
4108 blk_mq_freeze_queue_wait(ns->queue);
4109 up_read(&ctrl->namespaces_rwsem);
4111 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4113 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4117 down_read(&ctrl->namespaces_rwsem);
4118 list_for_each_entry(ns, &ctrl->namespaces, list)
4119 blk_freeze_queue_start(ns->queue);
4120 up_read(&ctrl->namespaces_rwsem);
4122 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4124 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4128 down_read(&ctrl->namespaces_rwsem);
4129 list_for_each_entry(ns, &ctrl->namespaces, list)
4130 blk_mq_quiesce_queue(ns->queue);
4131 up_read(&ctrl->namespaces_rwsem);
4133 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4135 void nvme_start_queues(struct nvme_ctrl *ctrl)
4139 down_read(&ctrl->namespaces_rwsem);
4140 list_for_each_entry(ns, &ctrl->namespaces, list)
4141 blk_mq_unquiesce_queue(ns->queue);
4142 up_read(&ctrl->namespaces_rwsem);
4144 EXPORT_SYMBOL_GPL(nvme_start_queues);
4147 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4151 down_read(&ctrl->namespaces_rwsem);
4152 list_for_each_entry(ns, &ctrl->namespaces, list)
4153 blk_sync_queue(ns->queue);
4154 up_read(&ctrl->namespaces_rwsem);
4157 blk_sync_queue(ctrl->admin_q);
4159 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4162 * Check we didn't inadvertently grow the command structure sizes:
4164 static inline void _nvme_check_size(void)
4166 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4167 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4168 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4169 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4170 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4171 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4172 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4173 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4174 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4175 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4176 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4177 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4178 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4179 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4180 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4181 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4182 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4186 static int __init nvme_core_init(void)
4188 int result = -ENOMEM;
4192 nvme_wq = alloc_workqueue("nvme-wq",
4193 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4197 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4198 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4202 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4203 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4204 if (!nvme_delete_wq)
4205 goto destroy_reset_wq;
4207 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4209 goto destroy_delete_wq;
4211 nvme_class = class_create(THIS_MODULE, "nvme");
4212 if (IS_ERR(nvme_class)) {
4213 result = PTR_ERR(nvme_class);
4214 goto unregister_chrdev;
4216 nvme_class->dev_uevent = nvme_class_uevent;
4218 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4219 if (IS_ERR(nvme_subsys_class)) {
4220 result = PTR_ERR(nvme_subsys_class);
4226 class_destroy(nvme_class);
4228 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4230 destroy_workqueue(nvme_delete_wq);
4232 destroy_workqueue(nvme_reset_wq);
4234 destroy_workqueue(nvme_wq);
4239 static void __exit nvme_core_exit(void)
4241 class_destroy(nvme_subsys_class);
4242 class_destroy(nvme_class);
4243 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4244 destroy_workqueue(nvme_delete_wq);
4245 destroy_workqueue(nvme_reset_wq);
4246 destroy_workqueue(nvme_wq);
4249 MODULE_LICENSE("GPL");
4250 MODULE_VERSION("1.0");
4251 module_init(nvme_core_init);
4252 module_exit(nvme_core_exit);