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 && ctrl->tagset)
120 queue_work(nvme_wq, &ctrl->scan_work);
124 * Use this function to proceed with scheduling reset_work for a controller
125 * that had previously been set to the resetting state. This is intended for
126 * code paths that can't be interrupted by other reset attempts. A hot removal
127 * may prevent this from succeeding.
129 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
131 if (ctrl->state != NVME_CTRL_RESETTING)
133 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
137 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
139 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
141 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
143 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
147 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
149 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
153 ret = nvme_reset_ctrl(ctrl);
155 flush_work(&ctrl->reset_work);
156 if (ctrl->state != NVME_CTRL_LIVE)
162 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
164 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
166 dev_info(ctrl->device,
167 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
169 flush_work(&ctrl->reset_work);
170 nvme_stop_ctrl(ctrl);
171 nvme_remove_namespaces(ctrl);
172 ctrl->ops->delete_ctrl(ctrl);
173 nvme_uninit_ctrl(ctrl);
177 static void nvme_delete_ctrl_work(struct work_struct *work)
179 struct nvme_ctrl *ctrl =
180 container_of(work, struct nvme_ctrl, delete_work);
182 nvme_do_delete_ctrl(ctrl);
185 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
189 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
193 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
195 static int nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
200 * Keep a reference until nvme_do_delete_ctrl() complete,
201 * since ->delete_ctrl can free the controller.
204 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
207 nvme_do_delete_ctrl(ctrl);
212 static inline bool nvme_ns_has_pi(struct nvme_ns *ns)
214 return ns->pi_type && ns->ms == sizeof(struct t10_pi_tuple);
217 static blk_status_t nvme_error_status(u16 status)
219 switch (status & 0x7ff) {
220 case NVME_SC_SUCCESS:
222 case NVME_SC_CAP_EXCEEDED:
223 return BLK_STS_NOSPC;
224 case NVME_SC_LBA_RANGE:
225 return BLK_STS_TARGET;
226 case NVME_SC_BAD_ATTRIBUTES:
227 case NVME_SC_ONCS_NOT_SUPPORTED:
228 case NVME_SC_INVALID_OPCODE:
229 case NVME_SC_INVALID_FIELD:
230 case NVME_SC_INVALID_NS:
231 return BLK_STS_NOTSUPP;
232 case NVME_SC_WRITE_FAULT:
233 case NVME_SC_READ_ERROR:
234 case NVME_SC_UNWRITTEN_BLOCK:
235 case NVME_SC_ACCESS_DENIED:
236 case NVME_SC_READ_ONLY:
237 case NVME_SC_COMPARE_FAILED:
238 return BLK_STS_MEDIUM;
239 case NVME_SC_GUARD_CHECK:
240 case NVME_SC_APPTAG_CHECK:
241 case NVME_SC_REFTAG_CHECK:
242 case NVME_SC_INVALID_PI:
243 return BLK_STS_PROTECTION;
244 case NVME_SC_RESERVATION_CONFLICT:
245 return BLK_STS_NEXUS;
246 case NVME_SC_HOST_PATH_ERROR:
247 return BLK_STS_TRANSPORT;
249 return BLK_STS_IOERR;
253 static inline bool nvme_req_needs_retry(struct request *req)
255 if (blk_noretry_request(req))
257 if (nvme_req(req)->status & NVME_SC_DNR)
259 if (nvme_req(req)->retries >= nvme_max_retries)
264 static void nvme_retry_req(struct request *req)
266 struct nvme_ns *ns = req->q->queuedata;
267 unsigned long delay = 0;
270 /* The mask and shift result must be <= 3 */
271 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
273 delay = ns->ctrl->crdt[crd - 1] * 100;
275 nvme_req(req)->retries++;
276 blk_mq_requeue_request(req, false);
277 blk_mq_delay_kick_requeue_list(req->q, delay);
280 void nvme_complete_rq(struct request *req)
282 blk_status_t status = nvme_error_status(nvme_req(req)->status);
284 trace_nvme_complete_rq(req);
286 nvme_cleanup_cmd(req);
288 if (nvme_req(req)->ctrl->kas)
289 nvme_req(req)->ctrl->comp_seen = true;
291 if (unlikely(status != BLK_STS_OK && nvme_req_needs_retry(req))) {
292 if ((req->cmd_flags & REQ_NVME_MPATH) &&
293 blk_path_error(status)) {
294 nvme_failover_req(req);
298 if (!blk_queue_dying(req->q)) {
304 nvme_trace_bio_complete(req, status);
305 blk_mq_end_request(req, status);
307 EXPORT_SYMBOL_GPL(nvme_complete_rq);
309 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
311 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
312 "Cancelling I/O %d", req->tag);
314 /* don't abort one completed request */
315 if (blk_mq_request_completed(req))
318 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
319 blk_mq_complete_request(req);
322 EXPORT_SYMBOL_GPL(nvme_cancel_request);
324 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
325 enum nvme_ctrl_state new_state)
327 enum nvme_ctrl_state old_state;
329 bool changed = false;
331 spin_lock_irqsave(&ctrl->lock, flags);
333 old_state = ctrl->state;
338 case NVME_CTRL_RESETTING:
339 case NVME_CTRL_CONNECTING:
346 case NVME_CTRL_RESETTING:
356 case NVME_CTRL_CONNECTING:
359 case NVME_CTRL_RESETTING:
366 case NVME_CTRL_DELETING:
369 case NVME_CTRL_RESETTING:
370 case NVME_CTRL_CONNECTING:
379 case NVME_CTRL_DELETING:
391 ctrl->state = new_state;
392 wake_up_all(&ctrl->state_wq);
395 spin_unlock_irqrestore(&ctrl->lock, flags);
396 if (changed && ctrl->state == NVME_CTRL_LIVE)
397 nvme_kick_requeue_lists(ctrl);
400 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
403 * Returns true for sink states that can't ever transition back to live.
405 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
407 switch (ctrl->state) {
410 case NVME_CTRL_RESETTING:
411 case NVME_CTRL_CONNECTING:
413 case NVME_CTRL_DELETING:
417 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
423 * Waits for the controller state to be resetting, or returns false if it is
424 * not possible to ever transition to that state.
426 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
428 wait_event(ctrl->state_wq,
429 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
430 nvme_state_terminal(ctrl));
431 return ctrl->state == NVME_CTRL_RESETTING;
433 EXPORT_SYMBOL_GPL(nvme_wait_reset);
435 static void nvme_free_ns_head(struct kref *ref)
437 struct nvme_ns_head *head =
438 container_of(ref, struct nvme_ns_head, ref);
440 nvme_mpath_remove_disk(head);
441 ida_simple_remove(&head->subsys->ns_ida, head->instance);
442 list_del_init(&head->entry);
443 cleanup_srcu_struct(&head->srcu);
444 nvme_put_subsystem(head->subsys);
448 static void nvme_put_ns_head(struct nvme_ns_head *head)
450 kref_put(&head->ref, nvme_free_ns_head);
453 static void nvme_free_ns(struct kref *kref)
455 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
458 nvme_nvm_unregister(ns);
461 nvme_put_ns_head(ns->head);
462 nvme_put_ctrl(ns->ctrl);
466 static void nvme_put_ns(struct nvme_ns *ns)
468 kref_put(&ns->kref, nvme_free_ns);
471 static inline void nvme_clear_nvme_request(struct request *req)
473 if (!(req->rq_flags & RQF_DONTPREP)) {
474 nvme_req(req)->retries = 0;
475 nvme_req(req)->flags = 0;
476 req->rq_flags |= RQF_DONTPREP;
480 struct request *nvme_alloc_request(struct request_queue *q,
481 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
483 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
486 if (qid == NVME_QID_ANY) {
487 req = blk_mq_alloc_request(q, op, flags);
489 req = blk_mq_alloc_request_hctx(q, op, flags,
495 req->cmd_flags |= REQ_FAILFAST_DRIVER;
496 nvme_clear_nvme_request(req);
497 nvme_req(req)->cmd = cmd;
501 EXPORT_SYMBOL_GPL(nvme_alloc_request);
503 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
505 struct nvme_command c;
507 memset(&c, 0, sizeof(c));
509 c.directive.opcode = nvme_admin_directive_send;
510 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
511 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
512 c.directive.dtype = NVME_DIR_IDENTIFY;
513 c.directive.tdtype = NVME_DIR_STREAMS;
514 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
516 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
519 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
521 return nvme_toggle_streams(ctrl, false);
524 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
526 return nvme_toggle_streams(ctrl, true);
529 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
530 struct streams_directive_params *s, u32 nsid)
532 struct nvme_command c;
534 memset(&c, 0, sizeof(c));
535 memset(s, 0, sizeof(*s));
537 c.directive.opcode = nvme_admin_directive_recv;
538 c.directive.nsid = cpu_to_le32(nsid);
539 c.directive.numd = cpu_to_le32((sizeof(*s) >> 2) - 1);
540 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
541 c.directive.dtype = NVME_DIR_STREAMS;
543 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
546 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
548 struct streams_directive_params s;
551 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
556 ret = nvme_enable_streams(ctrl);
560 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
564 ctrl->nssa = le16_to_cpu(s.nssa);
565 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
566 dev_info(ctrl->device, "too few streams (%u) available\n",
568 nvme_disable_streams(ctrl);
572 ctrl->nr_streams = min_t(unsigned, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
573 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
578 * Check if 'req' has a write hint associated with it. If it does, assign
579 * a valid namespace stream to the write.
581 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
582 struct request *req, u16 *control,
585 enum rw_hint streamid = req->write_hint;
587 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
591 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
594 *control |= NVME_RW_DTYPE_STREAMS;
595 *dsmgmt |= streamid << 16;
598 if (streamid < ARRAY_SIZE(req->q->write_hints))
599 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
602 static inline void nvme_setup_flush(struct nvme_ns *ns,
603 struct nvme_command *cmnd)
605 cmnd->common.opcode = nvme_cmd_flush;
606 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
609 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
610 struct nvme_command *cmnd)
612 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
613 struct nvme_dsm_range *range;
617 * Some devices do not consider the DSM 'Number of Ranges' field when
618 * determining how much data to DMA. Always allocate memory for maximum
619 * number of segments to prevent device reading beyond end of buffer.
621 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
623 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
626 * If we fail allocation our range, fallback to the controller
627 * discard page. If that's also busy, it's safe to return
628 * busy, as we know we can make progress once that's freed.
630 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
631 return BLK_STS_RESOURCE;
633 range = page_address(ns->ctrl->discard_page);
636 __rq_for_each_bio(bio, req) {
637 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
638 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
641 range[n].cattr = cpu_to_le32(0);
642 range[n].nlb = cpu_to_le32(nlb);
643 range[n].slba = cpu_to_le64(slba);
648 if (WARN_ON_ONCE(n != segments)) {
649 if (virt_to_page(range) == ns->ctrl->discard_page)
650 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
653 return BLK_STS_IOERR;
656 cmnd->dsm.opcode = nvme_cmd_dsm;
657 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
658 cmnd->dsm.nr = cpu_to_le32(segments - 1);
659 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
661 req->special_vec.bv_page = virt_to_page(range);
662 req->special_vec.bv_offset = offset_in_page(range);
663 req->special_vec.bv_len = alloc_size;
664 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
669 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
670 struct request *req, struct nvme_command *cmnd)
672 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
673 return nvme_setup_discard(ns, req, cmnd);
675 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
676 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
677 cmnd->write_zeroes.slba =
678 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
679 cmnd->write_zeroes.length =
680 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
681 cmnd->write_zeroes.control = 0;
685 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
686 struct request *req, struct nvme_command *cmnd)
688 struct nvme_ctrl *ctrl = ns->ctrl;
692 if (req->cmd_flags & REQ_FUA)
693 control |= NVME_RW_FUA;
694 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
695 control |= NVME_RW_LR;
697 if (req->cmd_flags & REQ_RAHEAD)
698 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
700 cmnd->rw.opcode = (rq_data_dir(req) ? nvme_cmd_write : nvme_cmd_read);
701 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
702 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
703 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
705 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
706 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
710 * If formated with metadata, the block layer always provides a
711 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
712 * we enable the PRACT bit for protection information or set the
713 * namespace capacity to zero to prevent any I/O.
715 if (!blk_integrity_rq(req)) {
716 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
717 return BLK_STS_NOTSUPP;
718 control |= NVME_RW_PRINFO_PRACT;
721 switch (ns->pi_type) {
722 case NVME_NS_DPS_PI_TYPE3:
723 control |= NVME_RW_PRINFO_PRCHK_GUARD;
725 case NVME_NS_DPS_PI_TYPE1:
726 case NVME_NS_DPS_PI_TYPE2:
727 control |= NVME_RW_PRINFO_PRCHK_GUARD |
728 NVME_RW_PRINFO_PRCHK_REF;
729 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
734 cmnd->rw.control = cpu_to_le16(control);
735 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
739 void nvme_cleanup_cmd(struct request *req)
741 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
742 struct nvme_ns *ns = req->rq_disk->private_data;
743 struct page *page = req->special_vec.bv_page;
745 if (page == ns->ctrl->discard_page)
746 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
748 kfree(page_address(page) + req->special_vec.bv_offset);
751 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
753 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
754 struct nvme_command *cmd)
756 blk_status_t ret = BLK_STS_OK;
758 nvme_clear_nvme_request(req);
760 memset(cmd, 0, sizeof(*cmd));
761 switch (req_op(req)) {
764 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
767 nvme_setup_flush(ns, cmd);
769 case REQ_OP_WRITE_ZEROES:
770 ret = nvme_setup_write_zeroes(ns, req, cmd);
773 ret = nvme_setup_discard(ns, req, cmd);
777 ret = nvme_setup_rw(ns, req, cmd);
781 return BLK_STS_IOERR;
784 cmd->common.command_id = req->tag;
785 trace_nvme_setup_cmd(req, cmd);
788 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
790 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
792 struct completion *waiting = rq->end_io_data;
794 rq->end_io_data = NULL;
798 static void nvme_execute_rq_polled(struct request_queue *q,
799 struct gendisk *bd_disk, struct request *rq, int at_head)
801 DECLARE_COMPLETION_ONSTACK(wait);
803 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
805 rq->cmd_flags |= REQ_HIPRI;
806 rq->end_io_data = &wait;
807 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
809 while (!completion_done(&wait)) {
810 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
816 * Returns 0 on success. If the result is negative, it's a Linux error code;
817 * if the result is positive, it's an NVM Express status code
819 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
820 union nvme_result *result, void *buffer, unsigned bufflen,
821 unsigned timeout, int qid, int at_head,
822 blk_mq_req_flags_t flags, bool poll)
827 req = nvme_alloc_request(q, cmd, flags, qid);
831 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
833 if (buffer && bufflen) {
834 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
840 nvme_execute_rq_polled(req->q, NULL, req, at_head);
842 blk_execute_rq(req->q, NULL, req, at_head);
844 *result = nvme_req(req)->result;
845 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
848 ret = nvme_req(req)->status;
850 blk_mq_free_request(req);
853 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
855 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
856 void *buffer, unsigned bufflen)
858 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
859 NVME_QID_ANY, 0, 0, false);
861 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
863 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
864 unsigned len, u32 seed, bool write)
866 struct bio_integrity_payload *bip;
870 buf = kmalloc(len, GFP_KERNEL);
875 if (write && copy_from_user(buf, ubuf, len))
878 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
884 bip->bip_iter.bi_size = len;
885 bip->bip_iter.bi_sector = seed;
886 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
887 offset_in_page(buf));
897 static int nvme_submit_user_cmd(struct request_queue *q,
898 struct nvme_command *cmd, void __user *ubuffer,
899 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
900 u32 meta_seed, u64 *result, unsigned timeout)
902 bool write = nvme_is_write(cmd);
903 struct nvme_ns *ns = q->queuedata;
904 struct gendisk *disk = ns ? ns->disk : NULL;
906 struct bio *bio = NULL;
910 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
914 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
915 nvme_req(req)->flags |= NVME_REQ_USERCMD;
917 if (ubuffer && bufflen) {
918 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
924 if (disk && meta_buffer && meta_len) {
925 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
931 req->cmd_flags |= REQ_INTEGRITY;
935 blk_execute_rq(req->q, disk, req, 0);
936 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
939 ret = nvme_req(req)->status;
941 *result = le64_to_cpu(nvme_req(req)->result.u64);
942 if (meta && !ret && !write) {
943 if (copy_to_user(meta_buffer, meta, meta_len))
949 blk_rq_unmap_user(bio);
951 blk_mq_free_request(req);
955 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
957 struct nvme_ctrl *ctrl = rq->end_io_data;
959 bool startka = false;
961 blk_mq_free_request(rq);
964 dev_err(ctrl->device,
965 "failed nvme_keep_alive_end_io error=%d\n",
970 ctrl->comp_seen = false;
971 spin_lock_irqsave(&ctrl->lock, flags);
972 if (ctrl->state == NVME_CTRL_LIVE ||
973 ctrl->state == NVME_CTRL_CONNECTING)
975 spin_unlock_irqrestore(&ctrl->lock, flags);
977 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
980 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
984 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
989 rq->timeout = ctrl->kato * HZ;
990 rq->end_io_data = ctrl;
992 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
997 static void nvme_keep_alive_work(struct work_struct *work)
999 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1000 struct nvme_ctrl, ka_work);
1001 bool comp_seen = ctrl->comp_seen;
1003 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1004 dev_dbg(ctrl->device,
1005 "reschedule traffic based keep-alive timer\n");
1006 ctrl->comp_seen = false;
1007 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1011 if (nvme_keep_alive(ctrl)) {
1012 /* allocation failure, reset the controller */
1013 dev_err(ctrl->device, "keep-alive failed\n");
1014 nvme_reset_ctrl(ctrl);
1019 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1021 if (unlikely(ctrl->kato == 0))
1024 schedule_delayed_work(&ctrl->ka_work, ctrl->kato * HZ);
1027 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1029 if (unlikely(ctrl->kato == 0))
1032 cancel_delayed_work_sync(&ctrl->ka_work);
1034 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1036 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1038 struct nvme_command c = { };
1041 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1042 c.identify.opcode = nvme_admin_identify;
1043 c.identify.cns = NVME_ID_CNS_CTRL;
1045 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1049 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1050 sizeof(struct nvme_id_ctrl));
1056 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1057 struct nvme_ns_ids *ids)
1059 struct nvme_command c = { };
1065 c.identify.opcode = nvme_admin_identify;
1066 c.identify.nsid = cpu_to_le32(nsid);
1067 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1069 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1073 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1074 NVME_IDENTIFY_DATA_SIZE);
1078 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1079 struct nvme_ns_id_desc *cur = data + pos;
1084 switch (cur->nidt) {
1085 case NVME_NIDT_EUI64:
1086 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1087 dev_warn(ctrl->device,
1088 "ctrl returned bogus length: %d for NVME_NIDT_EUI64\n",
1092 len = NVME_NIDT_EUI64_LEN;
1093 memcpy(ids->eui64, data + pos + sizeof(*cur), len);
1095 case NVME_NIDT_NGUID:
1096 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1097 dev_warn(ctrl->device,
1098 "ctrl returned bogus length: %d for NVME_NIDT_NGUID\n",
1102 len = NVME_NIDT_NGUID_LEN;
1103 memcpy(ids->nguid, data + pos + sizeof(*cur), len);
1105 case NVME_NIDT_UUID:
1106 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1107 dev_warn(ctrl->device,
1108 "ctrl returned bogus length: %d for NVME_NIDT_UUID\n",
1112 len = NVME_NIDT_UUID_LEN;
1113 uuid_copy(&ids->uuid, data + pos + sizeof(*cur));
1116 /* Skip unknown types */
1121 len += sizeof(*cur);
1128 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1130 struct nvme_command c = { };
1132 c.identify.opcode = nvme_admin_identify;
1133 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1134 c.identify.nsid = cpu_to_le32(nsid);
1135 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1136 NVME_IDENTIFY_DATA_SIZE);
1139 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1140 unsigned nsid, struct nvme_id_ns **id)
1142 struct nvme_command c = { };
1145 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1146 c.identify.opcode = nvme_admin_identify;
1147 c.identify.nsid = cpu_to_le32(nsid);
1148 c.identify.cns = NVME_ID_CNS_NS;
1150 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1154 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1156 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1163 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1164 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1166 struct nvme_command c;
1167 union nvme_result res;
1170 memset(&c, 0, sizeof(c));
1171 c.features.opcode = op;
1172 c.features.fid = cpu_to_le32(fid);
1173 c.features.dword11 = cpu_to_le32(dword11);
1175 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1176 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1177 if (ret >= 0 && result)
1178 *result = le32_to_cpu(res.u32);
1182 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1183 unsigned int dword11, void *buffer, size_t buflen,
1186 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1189 EXPORT_SYMBOL_GPL(nvme_set_features);
1191 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1192 unsigned int dword11, void *buffer, size_t buflen,
1195 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1198 EXPORT_SYMBOL_GPL(nvme_get_features);
1200 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1202 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1204 int status, nr_io_queues;
1206 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1212 * Degraded controllers might return an error when setting the queue
1213 * count. We still want to be able to bring them online and offer
1214 * access to the admin queue, as that might be only way to fix them up.
1217 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1220 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1221 *count = min(*count, nr_io_queues);
1226 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1228 #define NVME_AEN_SUPPORTED \
1229 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1230 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1232 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1234 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1237 if (!supported_aens)
1240 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1243 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1246 queue_work(nvme_wq, &ctrl->async_event_work);
1249 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1251 struct nvme_user_io io;
1252 struct nvme_command c;
1253 unsigned length, meta_len;
1254 void __user *metadata;
1256 if (copy_from_user(&io, uio, sizeof(io)))
1261 switch (io.opcode) {
1262 case nvme_cmd_write:
1264 case nvme_cmd_compare:
1270 length = (io.nblocks + 1) << ns->lba_shift;
1271 meta_len = (io.nblocks + 1) * ns->ms;
1272 metadata = (void __user *)(uintptr_t)io.metadata;
1277 } else if (meta_len) {
1278 if ((io.metadata & 3) || !io.metadata)
1282 memset(&c, 0, sizeof(c));
1283 c.rw.opcode = io.opcode;
1284 c.rw.flags = io.flags;
1285 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1286 c.rw.slba = cpu_to_le64(io.slba);
1287 c.rw.length = cpu_to_le16(io.nblocks);
1288 c.rw.control = cpu_to_le16(io.control);
1289 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1290 c.rw.reftag = cpu_to_le32(io.reftag);
1291 c.rw.apptag = cpu_to_le16(io.apptag);
1292 c.rw.appmask = cpu_to_le16(io.appmask);
1294 return nvme_submit_user_cmd(ns->queue, &c,
1295 (void __user *)(uintptr_t)io.addr, length,
1296 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1299 static u32 nvme_known_admin_effects(u8 opcode)
1302 case nvme_admin_format_nvm:
1303 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
1304 NVME_CMD_EFFECTS_CSE_MASK;
1305 case nvme_admin_sanitize_nvm:
1306 return NVME_CMD_EFFECTS_CSE_MASK;
1313 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1320 effects = le32_to_cpu(ctrl->effects->iocs[opcode]);
1321 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1322 dev_warn(ctrl->device,
1323 "IO command:%02x has unhandled effects:%08x\n",
1329 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1330 effects |= nvme_known_admin_effects(opcode);
1333 * For simplicity, IO to all namespaces is quiesced even if the command
1334 * effects say only one namespace is affected.
1336 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1337 mutex_lock(&ctrl->scan_lock);
1338 mutex_lock(&ctrl->subsys->lock);
1339 nvme_mpath_start_freeze(ctrl->subsys);
1340 nvme_mpath_wait_freeze(ctrl->subsys);
1341 nvme_start_freeze(ctrl);
1342 nvme_wait_freeze(ctrl);
1347 static void nvme_update_formats(struct nvme_ctrl *ctrl)
1351 down_read(&ctrl->namespaces_rwsem);
1352 list_for_each_entry(ns, &ctrl->namespaces, list)
1353 if (ns->disk && nvme_revalidate_disk(ns->disk))
1354 nvme_set_queue_dying(ns);
1355 up_read(&ctrl->namespaces_rwsem);
1358 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1361 * Revalidate LBA changes prior to unfreezing. This is necessary to
1362 * prevent memory corruption if a logical block size was changed by
1365 if (effects & NVME_CMD_EFFECTS_LBCC)
1366 nvme_update_formats(ctrl);
1367 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1368 nvme_unfreeze(ctrl);
1369 nvme_mpath_unfreeze(ctrl->subsys);
1370 mutex_unlock(&ctrl->subsys->lock);
1371 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1372 mutex_unlock(&ctrl->scan_lock);
1374 if (effects & NVME_CMD_EFFECTS_CCC)
1375 nvme_init_identify(ctrl);
1376 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC))
1377 nvme_queue_scan(ctrl);
1380 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1381 struct nvme_passthru_cmd __user *ucmd)
1383 struct nvme_passthru_cmd cmd;
1384 struct nvme_command c;
1385 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,
1417 cmd.metadata_len, 0, &result, timeout);
1418 nvme_passthru_end(ctrl, effects);
1421 if (put_user(result, &ucmd->result))
1428 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1429 struct nvme_passthru_cmd64 __user *ucmd)
1431 struct nvme_passthru_cmd64 cmd;
1432 struct nvme_command c;
1433 unsigned timeout = 0;
1437 if (!capable(CAP_SYS_ADMIN))
1439 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1444 memset(&c, 0, sizeof(c));
1445 c.common.opcode = cmd.opcode;
1446 c.common.flags = cmd.flags;
1447 c.common.nsid = cpu_to_le32(cmd.nsid);
1448 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1449 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1450 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1451 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1452 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1453 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1454 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1455 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1458 timeout = msecs_to_jiffies(cmd.timeout_ms);
1460 effects = nvme_passthru_start(ctrl, ns, cmd.opcode);
1461 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1462 (void __user *)(uintptr_t)cmd.addr, cmd.data_len,
1463 (void __user *)(uintptr_t)cmd.metadata, cmd.metadata_len,
1464 0, &cmd.result, timeout);
1465 nvme_passthru_end(ctrl, effects);
1468 if (put_user(cmd.result, &ucmd->result))
1476 * Issue ioctl requests on the first available path. Note that unlike normal
1477 * block layer requests we will not retry failed request on another controller.
1479 static struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1480 struct nvme_ns_head **head, int *srcu_idx)
1482 #ifdef CONFIG_NVME_MULTIPATH
1483 if (disk->fops == &nvme_ns_head_ops) {
1486 *head = disk->private_data;
1487 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1488 ns = nvme_find_path(*head);
1490 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1496 return disk->private_data;
1499 static void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1502 srcu_read_unlock(&head->srcu, idx);
1505 static bool is_ctrl_ioctl(unsigned int cmd)
1507 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1509 if (is_sed_ioctl(cmd))
1514 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1516 struct nvme_ns_head *head,
1519 struct nvme_ctrl *ctrl = ns->ctrl;
1522 nvme_get_ctrl(ns->ctrl);
1523 nvme_put_ns_from_disk(head, srcu_idx);
1526 case NVME_IOCTL_ADMIN_CMD:
1527 ret = nvme_user_cmd(ctrl, NULL, argp);
1529 case NVME_IOCTL_ADMIN64_CMD:
1530 ret = nvme_user_cmd64(ctrl, NULL, argp);
1533 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1536 nvme_put_ctrl(ctrl);
1540 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1541 unsigned int cmd, unsigned long arg)
1543 struct nvme_ns_head *head = NULL;
1544 void __user *argp = (void __user *)arg;
1548 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1550 return -EWOULDBLOCK;
1553 * Handle ioctls that apply to the controller instead of the namespace
1554 * seperately and drop the ns SRCU reference early. This avoids a
1555 * deadlock when deleting namespaces using the passthrough interface.
1557 if (is_ctrl_ioctl(cmd))
1558 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1562 force_successful_syscall_return();
1563 ret = ns->head->ns_id;
1565 case NVME_IOCTL_IO_CMD:
1566 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1568 case NVME_IOCTL_SUBMIT_IO:
1569 ret = nvme_submit_io(ns, argp);
1571 case NVME_IOCTL_IO64_CMD:
1572 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1576 ret = nvme_nvm_ioctl(ns, cmd, arg);
1581 nvme_put_ns_from_disk(head, srcu_idx);
1585 static int nvme_open(struct block_device *bdev, fmode_t mode)
1587 struct nvme_ns *ns = bdev->bd_disk->private_data;
1589 #ifdef CONFIG_NVME_MULTIPATH
1590 /* should never be called due to GENHD_FL_HIDDEN */
1591 if (WARN_ON_ONCE(ns->head->disk))
1594 if (!kref_get_unless_zero(&ns->kref))
1596 if (!try_module_get(ns->ctrl->ops->module))
1607 static void nvme_release(struct gendisk *disk, fmode_t mode)
1609 struct nvme_ns *ns = disk->private_data;
1611 module_put(ns->ctrl->ops->module);
1615 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1617 /* some standard values */
1618 geo->heads = 1 << 6;
1619 geo->sectors = 1 << 5;
1620 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1624 #ifdef CONFIG_BLK_DEV_INTEGRITY
1625 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1627 struct blk_integrity integrity;
1629 memset(&integrity, 0, sizeof(integrity));
1631 case NVME_NS_DPS_PI_TYPE3:
1632 integrity.profile = &t10_pi_type3_crc;
1633 integrity.tag_size = sizeof(u16) + sizeof(u32);
1634 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1636 case NVME_NS_DPS_PI_TYPE1:
1637 case NVME_NS_DPS_PI_TYPE2:
1638 integrity.profile = &t10_pi_type1_crc;
1639 integrity.tag_size = sizeof(u16);
1640 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1643 integrity.profile = NULL;
1646 integrity.tuple_size = ms;
1647 blk_integrity_register(disk, &integrity);
1648 blk_queue_max_integrity_segments(disk->queue, 1);
1651 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type)
1654 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1656 static void nvme_set_chunk_size(struct nvme_ns *ns)
1658 u32 chunk_size = nvme_lba_to_sect(ns, ns->noiob);
1659 blk_queue_chunk_sectors(ns->queue, rounddown_pow_of_two(chunk_size));
1662 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1664 struct nvme_ctrl *ctrl = ns->ctrl;
1665 struct request_queue *queue = disk->queue;
1666 u32 size = queue_logical_block_size(queue);
1668 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1669 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1673 if (ctrl->nr_streams && ns->sws && ns->sgs)
1674 size *= ns->sws * ns->sgs;
1676 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1677 NVME_DSM_MAX_RANGES);
1679 queue->limits.discard_alignment = 0;
1680 queue->limits.discard_granularity = size;
1682 /* If discard is already enabled, don't reset queue limits */
1683 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1686 blk_queue_max_discard_sectors(queue, UINT_MAX);
1687 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1689 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1690 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1693 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1697 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1698 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1701 * Even though NVMe spec explicitly states that MDTS is not
1702 * applicable to the write-zeroes:- "The restriction does not apply to
1703 * commands that do not transfer data between the host and the
1704 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1705 * In order to be more cautious use controller's max_hw_sectors value
1706 * to configure the maximum sectors for the write-zeroes which is
1707 * configured based on the controller's MDTS field in the
1708 * nvme_init_identify() if available.
1710 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1711 max_blocks = (u64)USHRT_MAX + 1;
1713 max_blocks = ns->ctrl->max_hw_sectors + 1;
1715 blk_queue_max_write_zeroes_sectors(disk->queue,
1716 nvme_lba_to_sect(ns, max_blocks));
1719 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1720 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1724 memset(ids, 0, sizeof(*ids));
1726 if (ctrl->vs >= NVME_VS(1, 1, 0))
1727 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1728 if (ctrl->vs >= NVME_VS(1, 2, 0))
1729 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1730 if (ctrl->vs >= NVME_VS(1, 3, 0)) {
1731 /* Don't treat error as fatal we potentially
1732 * already have a NGUID or EUI-64
1734 ret = nvme_identify_ns_descs(ctrl, nsid, ids);
1736 dev_warn(ctrl->device,
1737 "Identify Descriptors failed (%d)\n", ret);
1744 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1746 return !uuid_is_null(&ids->uuid) ||
1747 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1748 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1751 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1753 return uuid_equal(&a->uuid, &b->uuid) &&
1754 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1755 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0;
1758 static void nvme_update_disk_info(struct gendisk *disk,
1759 struct nvme_ns *ns, struct nvme_id_ns *id)
1761 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1762 unsigned short bs = 1 << ns->lba_shift;
1763 u32 atomic_bs, phys_bs, io_opt;
1765 if (ns->lba_shift > PAGE_SHIFT) {
1766 /* unsupported block size, set capacity to 0 later */
1769 blk_mq_freeze_queue(disk->queue);
1770 blk_integrity_unregister(disk);
1772 if (id->nabo == 0) {
1774 * Bit 1 indicates whether NAWUPF is defined for this namespace
1775 * and whether it should be used instead of AWUPF. If NAWUPF ==
1776 * 0 then AWUPF must be used instead.
1778 if (id->nsfeat & (1 << 1) && id->nawupf)
1779 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1781 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1787 if (id->nsfeat & (1 << 4)) {
1788 /* NPWG = Namespace Preferred Write Granularity */
1789 phys_bs *= 1 + le16_to_cpu(id->npwg);
1790 /* NOWS = Namespace Optimal Write Size */
1791 io_opt *= 1 + le16_to_cpu(id->nows);
1794 blk_queue_logical_block_size(disk->queue, bs);
1796 * Linux filesystems assume writing a single physical block is
1797 * an atomic operation. Hence limit the physical block size to the
1798 * value of the Atomic Write Unit Power Fail parameter.
1800 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1801 blk_queue_io_min(disk->queue, phys_bs);
1802 blk_queue_io_opt(disk->queue, io_opt);
1804 if (ns->ms && !ns->ext &&
1805 (ns->ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1806 nvme_init_integrity(disk, ns->ms, ns->pi_type);
1807 if ((ns->ms && !nvme_ns_has_pi(ns) && !blk_get_integrity(disk)) ||
1808 ns->lba_shift > PAGE_SHIFT)
1811 set_capacity(disk, capacity);
1813 nvme_config_discard(disk, ns);
1814 nvme_config_write_zeroes(disk, ns);
1816 if (id->nsattr & (1 << 0))
1817 set_disk_ro(disk, true);
1819 set_disk_ro(disk, false);
1821 blk_mq_unfreeze_queue(disk->queue);
1824 static void __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
1826 struct nvme_ns *ns = disk->private_data;
1829 * If identify namespace failed, use default 512 byte block size so
1830 * block layer can use before failing read/write for 0 capacity.
1832 ns->lba_shift = id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ds;
1833 if (ns->lba_shift == 0)
1835 ns->noiob = le16_to_cpu(id->noiob);
1836 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1837 ns->ext = ns->ms && (id->flbas & NVME_NS_FLBAS_META_EXT);
1838 /* the PI implementation requires metadata equal t10 pi tuple size */
1839 if (ns->ms == sizeof(struct t10_pi_tuple))
1840 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1845 nvme_set_chunk_size(ns);
1846 nvme_update_disk_info(disk, ns, id);
1847 #ifdef CONFIG_NVME_MULTIPATH
1848 if (ns->head->disk) {
1849 nvme_update_disk_info(ns->head->disk, ns, id);
1850 blk_queue_stack_limits(ns->head->disk->queue, ns->queue);
1851 revalidate_disk(ns->head->disk);
1856 static int nvme_revalidate_disk(struct gendisk *disk)
1858 struct nvme_ns *ns = disk->private_data;
1859 struct nvme_ctrl *ctrl = ns->ctrl;
1860 struct nvme_id_ns *id;
1861 struct nvme_ns_ids ids;
1864 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
1865 set_capacity(disk, 0);
1869 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
1873 if (id->ncap == 0) {
1878 __nvme_revalidate_disk(disk, id);
1879 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
1883 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
1884 dev_err(ctrl->device,
1885 "identifiers changed for nsid %d\n", ns->head->ns_id);
1893 * Only fail the function if we got a fatal error back from the
1894 * device, otherwise ignore the error and just move on.
1896 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
1899 ret = blk_status_to_errno(nvme_error_status(ret));
1903 static char nvme_pr_type(enum pr_type type)
1906 case PR_WRITE_EXCLUSIVE:
1908 case PR_EXCLUSIVE_ACCESS:
1910 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1912 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1914 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1916 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1923 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1924 u64 key, u64 sa_key, u8 op)
1926 struct nvme_ns_head *head = NULL;
1928 struct nvme_command c;
1930 u8 data[16] = { 0, };
1932 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1934 return -EWOULDBLOCK;
1936 put_unaligned_le64(key, &data[0]);
1937 put_unaligned_le64(sa_key, &data[8]);
1939 memset(&c, 0, sizeof(c));
1940 c.common.opcode = op;
1941 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1942 c.common.cdw10 = cpu_to_le32(cdw10);
1944 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1945 nvme_put_ns_from_disk(head, srcu_idx);
1949 static int nvme_pr_register(struct block_device *bdev, u64 old,
1950 u64 new, unsigned flags)
1954 if (flags & ~PR_FL_IGNORE_KEY)
1957 cdw10 = old ? 2 : 0;
1958 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1959 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1960 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1963 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1964 enum pr_type type, unsigned flags)
1968 if (flags & ~PR_FL_IGNORE_KEY)
1971 cdw10 = nvme_pr_type(type) << 8;
1972 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1973 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1976 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1977 enum pr_type type, bool abort)
1979 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1980 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1983 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1985 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1986 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1989 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1991 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1992 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1995 static const struct pr_ops nvme_pr_ops = {
1996 .pr_register = nvme_pr_register,
1997 .pr_reserve = nvme_pr_reserve,
1998 .pr_release = nvme_pr_release,
1999 .pr_preempt = nvme_pr_preempt,
2000 .pr_clear = nvme_pr_clear,
2003 #ifdef CONFIG_BLK_SED_OPAL
2004 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2007 struct nvme_ctrl *ctrl = data;
2008 struct nvme_command cmd;
2010 memset(&cmd, 0, sizeof(cmd));
2012 cmd.common.opcode = nvme_admin_security_send;
2014 cmd.common.opcode = nvme_admin_security_recv;
2015 cmd.common.nsid = 0;
2016 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2017 cmd.common.cdw11 = cpu_to_le32(len);
2019 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2020 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2022 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2023 #endif /* CONFIG_BLK_SED_OPAL */
2025 static const struct block_device_operations nvme_fops = {
2026 .owner = THIS_MODULE,
2027 .ioctl = nvme_ioctl,
2028 .compat_ioctl = nvme_ioctl,
2030 .release = nvme_release,
2031 .getgeo = nvme_getgeo,
2032 .revalidate_disk= nvme_revalidate_disk,
2033 .pr_ops = &nvme_pr_ops,
2036 #ifdef CONFIG_NVME_MULTIPATH
2037 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2039 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2041 if (!kref_get_unless_zero(&head->ref))
2046 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2048 nvme_put_ns_head(disk->private_data);
2051 const struct block_device_operations nvme_ns_head_ops = {
2052 .owner = THIS_MODULE,
2053 .open = nvme_ns_head_open,
2054 .release = nvme_ns_head_release,
2055 .ioctl = nvme_ioctl,
2056 .compat_ioctl = nvme_ioctl,
2057 .getgeo = nvme_getgeo,
2058 .pr_ops = &nvme_pr_ops,
2060 #endif /* CONFIG_NVME_MULTIPATH */
2062 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2064 unsigned long timeout =
2065 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2066 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2069 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2072 if ((csts & NVME_CSTS_RDY) == bit)
2076 if (fatal_signal_pending(current))
2078 if (time_after(jiffies, timeout)) {
2079 dev_err(ctrl->device,
2080 "Device not ready; aborting %s\n", enabled ?
2081 "initialisation" : "reset");
2090 * If the device has been passed off to us in an enabled state, just clear
2091 * the enabled bit. The spec says we should set the 'shutdown notification
2092 * bits', but doing so may cause the device to complete commands to the
2093 * admin queue ... and we don't know what memory that might be pointing at!
2095 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2099 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2100 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2102 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2106 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2107 msleep(NVME_QUIRK_DELAY_AMOUNT);
2109 return nvme_wait_ready(ctrl, ctrl->cap, false);
2111 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2113 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2116 * Default to a 4K page size, with the intention to update this
2117 * path in the future to accomodate architectures with differing
2118 * kernel and IO page sizes.
2120 unsigned dev_page_min, page_shift = 12;
2123 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2125 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2128 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2130 if (page_shift < dev_page_min) {
2131 dev_err(ctrl->device,
2132 "Minimum device page size %u too large for host (%u)\n",
2133 1 << dev_page_min, 1 << page_shift);
2137 ctrl->page_size = 1 << page_shift;
2139 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2140 ctrl->ctrl_config |= (page_shift - 12) << NVME_CC_MPS_SHIFT;
2141 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2142 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2143 ctrl->ctrl_config |= NVME_CC_ENABLE;
2145 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2148 return nvme_wait_ready(ctrl, ctrl->cap, true);
2150 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2152 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2154 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2158 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2159 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2161 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2165 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2166 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2170 if (fatal_signal_pending(current))
2172 if (time_after(jiffies, timeout)) {
2173 dev_err(ctrl->device,
2174 "Device shutdown incomplete; abort shutdown\n");
2181 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2183 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2184 struct request_queue *q)
2188 if (ctrl->max_hw_sectors) {
2190 (ctrl->max_hw_sectors / (ctrl->page_size >> 9)) + 1;
2192 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2193 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2194 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2196 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2197 is_power_of_2(ctrl->max_hw_sectors))
2198 blk_queue_chunk_sectors(q, ctrl->max_hw_sectors);
2199 blk_queue_virt_boundary(q, ctrl->page_size - 1);
2200 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2202 blk_queue_write_cache(q, vwc, vwc);
2205 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2210 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2213 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2214 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2217 dev_warn_once(ctrl->device,
2218 "could not set timestamp (%d)\n", ret);
2222 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2224 struct nvme_feat_host_behavior *host;
2227 /* Don't bother enabling the feature if retry delay is not reported */
2231 host = kzalloc(sizeof(*host), GFP_KERNEL);
2235 host->acre = NVME_ENABLE_ACRE;
2236 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2237 host, sizeof(*host), NULL);
2242 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2245 * APST (Autonomous Power State Transition) lets us program a
2246 * table of power state transitions that the controller will
2247 * perform automatically. We configure it with a simple
2248 * heuristic: we are willing to spend at most 2% of the time
2249 * transitioning between power states. Therefore, when running
2250 * in any given state, we will enter the next lower-power
2251 * non-operational state after waiting 50 * (enlat + exlat)
2252 * microseconds, as long as that state's exit latency is under
2253 * the requested maximum latency.
2255 * We will not autonomously enter any non-operational state for
2256 * which the total latency exceeds ps_max_latency_us. Users
2257 * can set ps_max_latency_us to zero to turn off APST.
2261 struct nvme_feat_auto_pst *table;
2267 * If APST isn't supported or if we haven't been initialized yet,
2268 * then don't do anything.
2273 if (ctrl->npss > 31) {
2274 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2278 table = kzalloc(sizeof(*table), GFP_KERNEL);
2282 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2283 /* Turn off APST. */
2285 dev_dbg(ctrl->device, "APST disabled\n");
2287 __le64 target = cpu_to_le64(0);
2291 * Walk through all states from lowest- to highest-power.
2292 * According to the spec, lower-numbered states use more
2293 * power. NPSS, despite the name, is the index of the
2294 * lowest-power state, not the number of states.
2296 for (state = (int)ctrl->npss; state >= 0; state--) {
2297 u64 total_latency_us, exit_latency_us, transition_ms;
2300 table->entries[state] = target;
2303 * Don't allow transitions to the deepest state
2304 * if it's quirked off.
2306 if (state == ctrl->npss &&
2307 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2311 * Is this state a useful non-operational state for
2312 * higher-power states to autonomously transition to?
2314 if (!(ctrl->psd[state].flags &
2315 NVME_PS_FLAGS_NON_OP_STATE))
2319 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2320 if (exit_latency_us > ctrl->ps_max_latency_us)
2325 le32_to_cpu(ctrl->psd[state].entry_lat);
2328 * This state is good. Use it as the APST idle
2329 * target for higher power states.
2331 transition_ms = total_latency_us + 19;
2332 do_div(transition_ms, 20);
2333 if (transition_ms > (1 << 24) - 1)
2334 transition_ms = (1 << 24) - 1;
2336 target = cpu_to_le64((state << 3) |
2337 (transition_ms << 8));
2342 if (total_latency_us > max_lat_us)
2343 max_lat_us = total_latency_us;
2349 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2351 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2352 max_ps, max_lat_us, (int)sizeof(*table), table);
2356 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2357 table, sizeof(*table), NULL);
2359 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2365 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2367 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2371 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2372 case PM_QOS_LATENCY_ANY:
2380 if (ctrl->ps_max_latency_us != latency) {
2381 ctrl->ps_max_latency_us = latency;
2382 nvme_configure_apst(ctrl);
2386 struct nvme_core_quirk_entry {
2388 * NVMe model and firmware strings are padded with spaces. For
2389 * simplicity, strings in the quirk table are padded with NULLs
2395 unsigned long quirks;
2398 static const struct nvme_core_quirk_entry core_quirks[] = {
2401 * This Toshiba device seems to die using any APST states. See:
2402 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2405 .mn = "THNSF5256GPUK TOSHIBA",
2406 .quirks = NVME_QUIRK_NO_APST,
2410 * This LiteON CL1-3D*-Q11 firmware version has a race
2411 * condition associated with actions related to suspend to idle
2412 * LiteON has resolved the problem in future firmware
2416 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2420 /* match is null-terminated but idstr is space-padded. */
2421 static bool string_matches(const char *idstr, const char *match, size_t len)
2428 matchlen = strlen(match);
2429 WARN_ON_ONCE(matchlen > len);
2431 if (memcmp(idstr, match, matchlen))
2434 for (; matchlen < len; matchlen++)
2435 if (idstr[matchlen] != ' ')
2441 static bool quirk_matches(const struct nvme_id_ctrl *id,
2442 const struct nvme_core_quirk_entry *q)
2444 return q->vid == le16_to_cpu(id->vid) &&
2445 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2446 string_matches(id->fr, q->fr, sizeof(id->fr));
2449 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2450 struct nvme_id_ctrl *id)
2455 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2456 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2457 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2458 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2462 if (ctrl->vs >= NVME_VS(1, 2, 1))
2463 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2466 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2467 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2468 "nqn.2014.08.org.nvmexpress:%04x%04x",
2469 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2470 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2471 off += sizeof(id->sn);
2472 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2473 off += sizeof(id->mn);
2474 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2477 static void nvme_release_subsystem(struct device *dev)
2479 struct nvme_subsystem *subsys =
2480 container_of(dev, struct nvme_subsystem, dev);
2482 if (subsys->instance >= 0)
2483 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2487 static void nvme_destroy_subsystem(struct kref *ref)
2489 struct nvme_subsystem *subsys =
2490 container_of(ref, struct nvme_subsystem, ref);
2492 mutex_lock(&nvme_subsystems_lock);
2493 list_del(&subsys->entry);
2494 mutex_unlock(&nvme_subsystems_lock);
2496 ida_destroy(&subsys->ns_ida);
2497 device_del(&subsys->dev);
2498 put_device(&subsys->dev);
2501 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2503 kref_put(&subsys->ref, nvme_destroy_subsystem);
2506 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2508 struct nvme_subsystem *subsys;
2510 lockdep_assert_held(&nvme_subsystems_lock);
2513 * Fail matches for discovery subsystems. This results
2514 * in each discovery controller bound to a unique subsystem.
2515 * This avoids issues with validating controller values
2516 * that can only be true when there is a single unique subsystem.
2517 * There may be multiple and completely independent entities
2518 * that provide discovery controllers.
2520 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2523 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2524 if (strcmp(subsys->subnqn, subsysnqn))
2526 if (!kref_get_unless_zero(&subsys->ref))
2534 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2535 struct device_attribute subsys_attr_##_name = \
2536 __ATTR(_name, _mode, _show, NULL)
2538 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2539 struct device_attribute *attr,
2542 struct nvme_subsystem *subsys =
2543 container_of(dev, struct nvme_subsystem, dev);
2545 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2547 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2549 #define nvme_subsys_show_str_function(field) \
2550 static ssize_t subsys_##field##_show(struct device *dev, \
2551 struct device_attribute *attr, char *buf) \
2553 struct nvme_subsystem *subsys = \
2554 container_of(dev, struct nvme_subsystem, dev); \
2555 return sprintf(buf, "%.*s\n", \
2556 (int)sizeof(subsys->field), subsys->field); \
2558 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2560 nvme_subsys_show_str_function(model);
2561 nvme_subsys_show_str_function(serial);
2562 nvme_subsys_show_str_function(firmware_rev);
2564 static struct attribute *nvme_subsys_attrs[] = {
2565 &subsys_attr_model.attr,
2566 &subsys_attr_serial.attr,
2567 &subsys_attr_firmware_rev.attr,
2568 &subsys_attr_subsysnqn.attr,
2569 #ifdef CONFIG_NVME_MULTIPATH
2570 &subsys_attr_iopolicy.attr,
2575 static struct attribute_group nvme_subsys_attrs_group = {
2576 .attrs = nvme_subsys_attrs,
2579 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2580 &nvme_subsys_attrs_group,
2584 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2585 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2587 struct nvme_ctrl *tmp;
2589 lockdep_assert_held(&nvme_subsystems_lock);
2591 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2592 if (tmp->state == NVME_CTRL_DELETING ||
2593 tmp->state == NVME_CTRL_DEAD)
2596 if (tmp->cntlid == ctrl->cntlid) {
2597 dev_err(ctrl->device,
2598 "Duplicate cntlid %u with %s, rejecting\n",
2599 ctrl->cntlid, dev_name(tmp->device));
2603 if ((id->cmic & (1 << 1)) ||
2604 (ctrl->opts && ctrl->opts->discovery_nqn))
2607 dev_err(ctrl->device,
2608 "Subsystem does not support multiple controllers\n");
2615 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2617 struct nvme_subsystem *subsys, *found;
2620 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2624 subsys->instance = -1;
2625 mutex_init(&subsys->lock);
2626 kref_init(&subsys->ref);
2627 INIT_LIST_HEAD(&subsys->ctrls);
2628 INIT_LIST_HEAD(&subsys->nsheads);
2629 nvme_init_subnqn(subsys, ctrl, id);
2630 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2631 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2632 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2633 subsys->vendor_id = le16_to_cpu(id->vid);
2634 subsys->cmic = id->cmic;
2635 subsys->awupf = le16_to_cpu(id->awupf);
2636 #ifdef CONFIG_NVME_MULTIPATH
2637 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2640 subsys->dev.class = nvme_subsys_class;
2641 subsys->dev.release = nvme_release_subsystem;
2642 subsys->dev.groups = nvme_subsys_attrs_groups;
2643 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2644 device_initialize(&subsys->dev);
2646 mutex_lock(&nvme_subsystems_lock);
2647 found = __nvme_find_get_subsystem(subsys->subnqn);
2649 put_device(&subsys->dev);
2652 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2654 goto out_put_subsystem;
2657 ret = device_add(&subsys->dev);
2659 dev_err(ctrl->device,
2660 "failed to register subsystem device.\n");
2661 put_device(&subsys->dev);
2664 ida_init(&subsys->ns_ida);
2665 list_add_tail(&subsys->entry, &nvme_subsystems);
2668 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2669 dev_name(ctrl->device));
2671 dev_err(ctrl->device,
2672 "failed to create sysfs link from subsystem.\n");
2673 goto out_put_subsystem;
2677 subsys->instance = ctrl->instance;
2678 ctrl->subsys = subsys;
2679 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2680 mutex_unlock(&nvme_subsystems_lock);
2684 nvme_put_subsystem(subsys);
2686 mutex_unlock(&nvme_subsystems_lock);
2690 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp,
2691 void *log, size_t size, u64 offset)
2693 struct nvme_command c = { };
2694 unsigned long dwlen = size / 4 - 1;
2696 c.get_log_page.opcode = nvme_admin_get_log_page;
2697 c.get_log_page.nsid = cpu_to_le32(nsid);
2698 c.get_log_page.lid = log_page;
2699 c.get_log_page.lsp = lsp;
2700 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2701 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2702 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2703 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2705 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2708 static int nvme_get_effects_log(struct nvme_ctrl *ctrl)
2713 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
2718 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0,
2719 ctrl->effects, sizeof(*ctrl->effects), 0);
2721 kfree(ctrl->effects);
2722 ctrl->effects = NULL;
2728 * Initialize the cached copies of the Identify data and various controller
2729 * register in our nvme_ctrl structure. This should be called as soon as
2730 * the admin queue is fully up and running.
2732 int nvme_init_identify(struct nvme_ctrl *ctrl)
2734 struct nvme_id_ctrl *id;
2735 int ret, page_shift;
2737 bool prev_apst_enabled;
2739 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2741 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2744 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2745 ctrl->sqsize = min_t(int, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2747 if (ctrl->vs >= NVME_VS(1, 1, 0))
2748 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2750 ret = nvme_identify_ctrl(ctrl, &id);
2752 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2756 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2757 ret = nvme_get_effects_log(ctrl);
2762 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2763 ctrl->cntlid = le16_to_cpu(id->cntlid);
2765 if (!ctrl->identified) {
2768 ret = nvme_init_subsystem(ctrl, id);
2773 * Check for quirks. Quirk can depend on firmware version,
2774 * so, in principle, the set of quirks present can change
2775 * across a reset. As a possible future enhancement, we
2776 * could re-scan for quirks every time we reinitialize
2777 * the device, but we'd have to make sure that the driver
2778 * behaves intelligently if the quirks change.
2780 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2781 if (quirk_matches(id, &core_quirks[i]))
2782 ctrl->quirks |= core_quirks[i].quirks;
2786 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2787 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2788 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2791 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2792 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2793 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2795 ctrl->oacs = le16_to_cpu(id->oacs);
2796 ctrl->oncs = le16_to_cpu(id->oncs);
2797 ctrl->mtfa = le16_to_cpu(id->mtfa);
2798 ctrl->oaes = le32_to_cpu(id->oaes);
2799 ctrl->wctemp = le16_to_cpu(id->wctemp);
2800 ctrl->cctemp = le16_to_cpu(id->cctemp);
2802 atomic_set(&ctrl->abort_limit, id->acl + 1);
2803 ctrl->vwc = id->vwc;
2805 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
2807 max_hw_sectors = UINT_MAX;
2808 ctrl->max_hw_sectors =
2809 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2811 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2812 ctrl->sgls = le32_to_cpu(id->sgls);
2813 ctrl->kas = le16_to_cpu(id->kas);
2814 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2815 ctrl->ctratt = le32_to_cpu(id->ctratt);
2819 u32 transition_time = le32_to_cpu(id->rtd3e) / 1000000;
2821 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2822 shutdown_timeout, 60);
2824 if (ctrl->shutdown_timeout != shutdown_timeout)
2825 dev_info(ctrl->device,
2826 "Shutdown timeout set to %u seconds\n",
2827 ctrl->shutdown_timeout);
2829 ctrl->shutdown_timeout = shutdown_timeout;
2831 ctrl->npss = id->npss;
2832 ctrl->apsta = id->apsta;
2833 prev_apst_enabled = ctrl->apst_enabled;
2834 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2835 if (force_apst && id->apsta) {
2836 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2837 ctrl->apst_enabled = true;
2839 ctrl->apst_enabled = false;
2842 ctrl->apst_enabled = id->apsta;
2844 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2846 if (ctrl->ops->flags & NVME_F_FABRICS) {
2847 ctrl->icdoff = le16_to_cpu(id->icdoff);
2848 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2849 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2850 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2853 * In fabrics we need to verify the cntlid matches the
2856 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2857 dev_err(ctrl->device,
2858 "Mismatching cntlid: Connect %u vs Identify "
2860 ctrl->cntlid, le16_to_cpu(id->cntlid));
2865 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
2866 dev_err(ctrl->device,
2867 "keep-alive support is mandatory for fabrics\n");
2872 ctrl->hmpre = le32_to_cpu(id->hmpre);
2873 ctrl->hmmin = le32_to_cpu(id->hmmin);
2874 ctrl->hmminds = le32_to_cpu(id->hmminds);
2875 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2878 ret = nvme_mpath_init(ctrl, id);
2884 if (ctrl->apst_enabled && !prev_apst_enabled)
2885 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2886 else if (!ctrl->apst_enabled && prev_apst_enabled)
2887 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2889 ret = nvme_configure_apst(ctrl);
2893 ret = nvme_configure_timestamp(ctrl);
2897 ret = nvme_configure_directives(ctrl);
2901 ret = nvme_configure_acre(ctrl);
2905 if (!ctrl->identified)
2906 nvme_hwmon_init(ctrl);
2908 ctrl->identified = true;
2916 EXPORT_SYMBOL_GPL(nvme_init_identify);
2918 static int nvme_dev_open(struct inode *inode, struct file *file)
2920 struct nvme_ctrl *ctrl =
2921 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2923 switch (ctrl->state) {
2924 case NVME_CTRL_LIVE:
2927 return -EWOULDBLOCK;
2930 file->private_data = ctrl;
2934 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
2939 down_read(&ctrl->namespaces_rwsem);
2940 if (list_empty(&ctrl->namespaces)) {
2945 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
2946 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
2947 dev_warn(ctrl->device,
2948 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
2953 dev_warn(ctrl->device,
2954 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
2955 kref_get(&ns->kref);
2956 up_read(&ctrl->namespaces_rwsem);
2958 ret = nvme_user_cmd(ctrl, ns, argp);
2963 up_read(&ctrl->namespaces_rwsem);
2967 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
2970 struct nvme_ctrl *ctrl = file->private_data;
2971 void __user *argp = (void __user *)arg;
2974 case NVME_IOCTL_ADMIN_CMD:
2975 return nvme_user_cmd(ctrl, NULL, argp);
2976 case NVME_IOCTL_ADMIN64_CMD:
2977 return nvme_user_cmd64(ctrl, NULL, argp);
2978 case NVME_IOCTL_IO_CMD:
2979 return nvme_dev_user_cmd(ctrl, argp);
2980 case NVME_IOCTL_RESET:
2981 dev_warn(ctrl->device, "resetting controller\n");
2982 return nvme_reset_ctrl_sync(ctrl);
2983 case NVME_IOCTL_SUBSYS_RESET:
2984 return nvme_reset_subsystem(ctrl);
2985 case NVME_IOCTL_RESCAN:
2986 nvme_queue_scan(ctrl);
2993 static const struct file_operations nvme_dev_fops = {
2994 .owner = THIS_MODULE,
2995 .open = nvme_dev_open,
2996 .unlocked_ioctl = nvme_dev_ioctl,
2997 .compat_ioctl = compat_ptr_ioctl,
3000 static ssize_t nvme_sysfs_reset(struct device *dev,
3001 struct device_attribute *attr, const char *buf,
3004 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3007 ret = nvme_reset_ctrl_sync(ctrl);
3012 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3014 static ssize_t nvme_sysfs_rescan(struct device *dev,
3015 struct device_attribute *attr, const char *buf,
3018 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3020 nvme_queue_scan(ctrl);
3023 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3025 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3027 struct gendisk *disk = dev_to_disk(dev);
3029 if (disk->fops == &nvme_fops)
3030 return nvme_get_ns_from_dev(dev)->head;
3032 return disk->private_data;
3035 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3038 struct nvme_ns_head *head = dev_to_ns_head(dev);
3039 struct nvme_ns_ids *ids = &head->ids;
3040 struct nvme_subsystem *subsys = head->subsys;
3041 int serial_len = sizeof(subsys->serial);
3042 int model_len = sizeof(subsys->model);
3044 if (!uuid_is_null(&ids->uuid))
3045 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3047 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3048 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3050 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3051 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3053 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3054 subsys->serial[serial_len - 1] == '\0'))
3056 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3057 subsys->model[model_len - 1] == '\0'))
3060 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3061 serial_len, subsys->serial, model_len, subsys->model,
3064 static DEVICE_ATTR_RO(wwid);
3066 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3069 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3071 static DEVICE_ATTR_RO(nguid);
3073 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3076 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3078 /* For backward compatibility expose the NGUID to userspace if
3079 * we have no UUID set
3081 if (uuid_is_null(&ids->uuid)) {
3082 printk_ratelimited(KERN_WARNING
3083 "No UUID available providing old NGUID\n");
3084 return sprintf(buf, "%pU\n", ids->nguid);
3086 return sprintf(buf, "%pU\n", &ids->uuid);
3088 static DEVICE_ATTR_RO(uuid);
3090 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3093 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3095 static DEVICE_ATTR_RO(eui);
3097 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3100 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3102 static DEVICE_ATTR_RO(nsid);
3104 static struct attribute *nvme_ns_id_attrs[] = {
3105 &dev_attr_wwid.attr,
3106 &dev_attr_uuid.attr,
3107 &dev_attr_nguid.attr,
3109 &dev_attr_nsid.attr,
3110 #ifdef CONFIG_NVME_MULTIPATH
3111 &dev_attr_ana_grpid.attr,
3112 &dev_attr_ana_state.attr,
3117 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3118 struct attribute *a, int n)
3120 struct device *dev = container_of(kobj, struct device, kobj);
3121 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3123 if (a == &dev_attr_uuid.attr) {
3124 if (uuid_is_null(&ids->uuid) &&
3125 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3128 if (a == &dev_attr_nguid.attr) {
3129 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3132 if (a == &dev_attr_eui.attr) {
3133 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3136 #ifdef CONFIG_NVME_MULTIPATH
3137 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3138 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3140 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3147 static const struct attribute_group nvme_ns_id_attr_group = {
3148 .attrs = nvme_ns_id_attrs,
3149 .is_visible = nvme_ns_id_attrs_are_visible,
3152 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3153 &nvme_ns_id_attr_group,
3155 &nvme_nvm_attr_group,
3160 #define nvme_show_str_function(field) \
3161 static ssize_t field##_show(struct device *dev, \
3162 struct device_attribute *attr, char *buf) \
3164 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3165 return sprintf(buf, "%.*s\n", \
3166 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3168 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3170 nvme_show_str_function(model);
3171 nvme_show_str_function(serial);
3172 nvme_show_str_function(firmware_rev);
3174 #define nvme_show_int_function(field) \
3175 static ssize_t field##_show(struct device *dev, \
3176 struct device_attribute *attr, char *buf) \
3178 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3179 return sprintf(buf, "%d\n", ctrl->field); \
3181 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3183 nvme_show_int_function(cntlid);
3184 nvme_show_int_function(numa_node);
3185 nvme_show_int_function(queue_count);
3186 nvme_show_int_function(sqsize);
3188 static ssize_t nvme_sysfs_delete(struct device *dev,
3189 struct device_attribute *attr, const char *buf,
3192 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3194 if (device_remove_file_self(dev, attr))
3195 nvme_delete_ctrl_sync(ctrl);
3198 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3200 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3201 struct device_attribute *attr,
3204 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3206 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3208 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3210 static ssize_t nvme_sysfs_show_state(struct device *dev,
3211 struct device_attribute *attr,
3214 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3215 static const char *const state_name[] = {
3216 [NVME_CTRL_NEW] = "new",
3217 [NVME_CTRL_LIVE] = "live",
3218 [NVME_CTRL_RESETTING] = "resetting",
3219 [NVME_CTRL_CONNECTING] = "connecting",
3220 [NVME_CTRL_DELETING] = "deleting",
3221 [NVME_CTRL_DEAD] = "dead",
3224 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3225 state_name[ctrl->state])
3226 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3228 return sprintf(buf, "unknown state\n");
3231 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3233 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3234 struct device_attribute *attr,
3237 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3239 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3241 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3243 static ssize_t nvme_sysfs_show_address(struct device *dev,
3244 struct device_attribute *attr,
3247 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3249 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3251 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3253 static struct attribute *nvme_dev_attrs[] = {
3254 &dev_attr_reset_controller.attr,
3255 &dev_attr_rescan_controller.attr,
3256 &dev_attr_model.attr,
3257 &dev_attr_serial.attr,
3258 &dev_attr_firmware_rev.attr,
3259 &dev_attr_cntlid.attr,
3260 &dev_attr_delete_controller.attr,
3261 &dev_attr_transport.attr,
3262 &dev_attr_subsysnqn.attr,
3263 &dev_attr_address.attr,
3264 &dev_attr_state.attr,
3265 &dev_attr_numa_node.attr,
3266 &dev_attr_queue_count.attr,
3267 &dev_attr_sqsize.attr,
3271 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3272 struct attribute *a, int n)
3274 struct device *dev = container_of(kobj, struct device, kobj);
3275 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3277 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3279 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3285 static struct attribute_group nvme_dev_attrs_group = {
3286 .attrs = nvme_dev_attrs,
3287 .is_visible = nvme_dev_attrs_are_visible,
3290 static const struct attribute_group *nvme_dev_attr_groups[] = {
3291 &nvme_dev_attrs_group,
3295 static struct nvme_ns_head *__nvme_find_ns_head(struct nvme_subsystem *subsys,
3298 struct nvme_ns_head *h;
3300 lockdep_assert_held(&subsys->lock);
3302 list_for_each_entry(h, &subsys->nsheads, entry) {
3303 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3310 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3311 struct nvme_ns_head *new)
3313 struct nvme_ns_head *h;
3315 lockdep_assert_held(&subsys->lock);
3317 list_for_each_entry(h, &subsys->nsheads, entry) {
3318 if (nvme_ns_ids_valid(&new->ids) &&
3319 !list_empty(&h->list) &&
3320 nvme_ns_ids_equal(&new->ids, &h->ids))
3327 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3328 unsigned nsid, struct nvme_id_ns *id)
3330 struct nvme_ns_head *head;
3331 size_t size = sizeof(*head);
3334 #ifdef CONFIG_NVME_MULTIPATH
3335 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3338 head = kzalloc(size, GFP_KERNEL);
3341 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3344 head->instance = ret;
3345 INIT_LIST_HEAD(&head->list);
3346 ret = init_srcu_struct(&head->srcu);
3348 goto out_ida_remove;
3349 head->subsys = ctrl->subsys;
3351 kref_init(&head->ref);
3353 ret = nvme_report_ns_ids(ctrl, nsid, id, &head->ids);
3355 goto out_cleanup_srcu;
3357 ret = __nvme_check_ids(ctrl->subsys, head);
3359 dev_err(ctrl->device,
3360 "duplicate IDs for nsid %d\n", nsid);
3361 goto out_cleanup_srcu;
3364 ret = nvme_mpath_alloc_disk(ctrl, head);
3366 goto out_cleanup_srcu;
3368 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3370 kref_get(&ctrl->subsys->ref);
3374 cleanup_srcu_struct(&head->srcu);
3376 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3381 ret = blk_status_to_errno(nvme_error_status(ret));
3382 return ERR_PTR(ret);
3385 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3386 struct nvme_id_ns *id)
3388 struct nvme_ctrl *ctrl = ns->ctrl;
3389 bool is_shared = id->nmic & (1 << 0);
3390 struct nvme_ns_head *head = NULL;
3393 mutex_lock(&ctrl->subsys->lock);
3395 head = __nvme_find_ns_head(ctrl->subsys, nsid);
3397 head = nvme_alloc_ns_head(ctrl, nsid, id);
3399 ret = PTR_ERR(head);
3403 struct nvme_ns_ids ids;
3405 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3409 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3410 dev_err(ctrl->device,
3411 "IDs don't match for shared namespace %d\n",
3418 list_add_tail(&ns->siblings, &head->list);
3422 mutex_unlock(&ctrl->subsys->lock);
3424 ret = blk_status_to_errno(nvme_error_status(ret));
3428 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3430 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3431 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3433 return nsa->head->ns_id - nsb->head->ns_id;
3436 static struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3438 struct nvme_ns *ns, *ret = NULL;
3440 down_read(&ctrl->namespaces_rwsem);
3441 list_for_each_entry(ns, &ctrl->namespaces, list) {
3442 if (ns->head->ns_id == nsid) {
3443 if (!kref_get_unless_zero(&ns->kref))
3448 if (ns->head->ns_id > nsid)
3451 up_read(&ctrl->namespaces_rwsem);
3455 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns)
3457 struct streams_directive_params s;
3460 if (!ctrl->nr_streams)
3463 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
3467 ns->sws = le32_to_cpu(s.sws);
3468 ns->sgs = le16_to_cpu(s.sgs);
3471 unsigned int bs = 1 << ns->lba_shift;
3473 blk_queue_io_min(ns->queue, bs * ns->sws);
3475 blk_queue_io_opt(ns->queue, bs * ns->sws * ns->sgs);
3481 static int nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3484 struct gendisk *disk;
3485 struct nvme_id_ns *id;
3486 char disk_name[DISK_NAME_LEN];
3487 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3489 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3493 ns->queue = blk_mq_init_queue(ctrl->tagset);
3494 if (IS_ERR(ns->queue)) {
3495 ret = PTR_ERR(ns->queue);
3499 if (ctrl->opts && ctrl->opts->data_digest)
3500 ns->queue->backing_dev_info->capabilities
3501 |= BDI_CAP_STABLE_WRITES;
3503 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3504 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3505 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3507 ns->queue->queuedata = ns;
3510 kref_init(&ns->kref);
3511 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3513 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3514 nvme_set_queue_limits(ctrl, ns->queue);
3516 ret = nvme_identify_ns(ctrl, nsid, &id);
3518 goto out_free_queue;
3520 if (id->ncap == 0) {
3525 ret = nvme_init_ns_head(ns, nsid, id);
3528 nvme_setup_streams_ns(ctrl, ns);
3529 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3531 disk = alloc_disk_node(0, node);
3537 disk->fops = &nvme_fops;
3538 disk->private_data = ns;
3539 disk->queue = ns->queue;
3540 disk->flags = flags;
3541 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3544 __nvme_revalidate_disk(disk, id);
3546 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3547 ret = nvme_nvm_register(ns, disk_name, node);
3549 dev_warn(ctrl->device, "LightNVM init failure\n");
3554 down_write(&ctrl->namespaces_rwsem);
3555 list_add_tail(&ns->list, &ctrl->namespaces);
3556 up_write(&ctrl->namespaces_rwsem);
3558 nvme_get_ctrl(ctrl);
3560 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3562 nvme_mpath_add_disk(ns, id);
3563 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3570 mutex_lock(&ctrl->subsys->lock);
3571 list_del_rcu(&ns->siblings);
3572 mutex_unlock(&ctrl->subsys->lock);
3573 nvme_put_ns_head(ns->head);
3577 blk_cleanup_queue(ns->queue);
3581 ret = blk_status_to_errno(nvme_error_status(ret));
3585 static void nvme_ns_remove(struct nvme_ns *ns)
3587 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3590 nvme_fault_inject_fini(&ns->fault_inject);
3592 mutex_lock(&ns->ctrl->subsys->lock);
3593 list_del_rcu(&ns->siblings);
3594 mutex_unlock(&ns->ctrl->subsys->lock);
3595 synchronize_rcu(); /* guarantee not available in head->list */
3596 nvme_mpath_clear_current_path(ns);
3597 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3599 if (ns->disk && ns->disk->flags & GENHD_FL_UP) {
3600 del_gendisk(ns->disk);
3601 blk_cleanup_queue(ns->queue);
3602 if (blk_get_integrity(ns->disk))
3603 blk_integrity_unregister(ns->disk);
3606 down_write(&ns->ctrl->namespaces_rwsem);
3607 list_del_init(&ns->list);
3608 up_write(&ns->ctrl->namespaces_rwsem);
3610 nvme_mpath_check_last_path(ns);
3614 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3618 ns = nvme_find_get_ns(ctrl, nsid);
3620 if (ns->disk && revalidate_disk(ns->disk))
3624 nvme_alloc_ns(ctrl, nsid);
3627 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3630 struct nvme_ns *ns, *next;
3633 down_write(&ctrl->namespaces_rwsem);
3634 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3635 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3636 list_move_tail(&ns->list, &rm_list);
3638 up_write(&ctrl->namespaces_rwsem);
3640 list_for_each_entry_safe(ns, next, &rm_list, list)
3645 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl, unsigned nn)
3649 unsigned i, j, nsid, prev = 0;
3650 unsigned num_lists = DIV_ROUND_UP_ULL((u64)nn, 1024);
3653 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3657 for (i = 0; i < num_lists; i++) {
3658 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
3662 for (j = 0; j < min(nn, 1024U); j++) {
3663 nsid = le32_to_cpu(ns_list[j]);
3667 nvme_validate_ns(ctrl, nsid);
3669 while (++prev < nsid) {
3670 ns = nvme_find_get_ns(ctrl, prev);
3680 nvme_remove_invalid_namespaces(ctrl, prev);
3686 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl, unsigned nn)
3690 for (i = 1; i <= nn; i++)
3691 nvme_validate_ns(ctrl, i);
3693 nvme_remove_invalid_namespaces(ctrl, nn);
3696 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3698 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3702 log = kzalloc(log_size, GFP_KERNEL);
3707 * We need to read the log to clear the AEN, but we don't want to rely
3708 * on it for the changed namespace information as userspace could have
3709 * raced with us in reading the log page, which could cause us to miss
3712 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0, log,
3715 dev_warn(ctrl->device,
3716 "reading changed ns log failed: %d\n", error);
3721 static void nvme_scan_work(struct work_struct *work)
3723 struct nvme_ctrl *ctrl =
3724 container_of(work, struct nvme_ctrl, scan_work);
3725 struct nvme_id_ctrl *id;
3728 /* No tagset on a live ctrl means IO queues could not created */
3729 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3732 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3733 dev_info(ctrl->device, "rescanning namespaces.\n");
3734 nvme_clear_changed_ns_log(ctrl);
3737 if (nvme_identify_ctrl(ctrl, &id))
3740 mutex_lock(&ctrl->scan_lock);
3741 nn = le32_to_cpu(id->nn);
3742 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
3743 !(ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)) {
3744 if (!nvme_scan_ns_list(ctrl, nn))
3747 nvme_scan_ns_sequential(ctrl, nn);
3749 mutex_unlock(&ctrl->scan_lock);
3751 down_write(&ctrl->namespaces_rwsem);
3752 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3753 up_write(&ctrl->namespaces_rwsem);
3757 * This function iterates the namespace list unlocked to allow recovery from
3758 * controller failure. It is up to the caller to ensure the namespace list is
3759 * not modified by scan work while this function is executing.
3761 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3763 struct nvme_ns *ns, *next;
3767 * make sure to requeue I/O to all namespaces as these
3768 * might result from the scan itself and must complete
3769 * for the scan_work to make progress
3771 nvme_mpath_clear_ctrl_paths(ctrl);
3773 /* prevent racing with ns scanning */
3774 flush_work(&ctrl->scan_work);
3777 * The dead states indicates the controller was not gracefully
3778 * disconnected. In that case, we won't be able to flush any data while
3779 * removing the namespaces' disks; fail all the queues now to avoid
3780 * potentially having to clean up the failed sync later.
3782 if (ctrl->state == NVME_CTRL_DEAD)
3783 nvme_kill_queues(ctrl);
3785 down_write(&ctrl->namespaces_rwsem);
3786 list_splice_init(&ctrl->namespaces, &ns_list);
3787 up_write(&ctrl->namespaces_rwsem);
3789 list_for_each_entry_safe(ns, next, &ns_list, list)
3792 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3794 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3796 struct nvme_ctrl *ctrl =
3797 container_of(dev, struct nvme_ctrl, ctrl_device);
3798 struct nvmf_ctrl_options *opts = ctrl->opts;
3801 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3806 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3810 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3811 opts->trsvcid ?: "none");
3815 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3816 opts->host_traddr ?: "none");
3821 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3823 char *envp[2] = { NULL, NULL };
3824 u32 aen_result = ctrl->aen_result;
3826 ctrl->aen_result = 0;
3830 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3833 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3837 static void nvme_async_event_work(struct work_struct *work)
3839 struct nvme_ctrl *ctrl =
3840 container_of(work, struct nvme_ctrl, async_event_work);
3842 nvme_aen_uevent(ctrl);
3843 ctrl->ops->submit_async_event(ctrl);
3846 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
3851 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
3857 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
3860 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
3862 struct nvme_fw_slot_info_log *log;
3864 log = kmalloc(sizeof(*log), GFP_KERNEL);
3868 if (nvme_get_log(ctrl, NVME_NSID_ALL, 0, NVME_LOG_FW_SLOT, log,
3870 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
3874 static void nvme_fw_act_work(struct work_struct *work)
3876 struct nvme_ctrl *ctrl = container_of(work,
3877 struct nvme_ctrl, fw_act_work);
3878 unsigned long fw_act_timeout;
3881 fw_act_timeout = jiffies +
3882 msecs_to_jiffies(ctrl->mtfa * 100);
3884 fw_act_timeout = jiffies +
3885 msecs_to_jiffies(admin_timeout * 1000);
3887 nvme_stop_queues(ctrl);
3888 while (nvme_ctrl_pp_status(ctrl)) {
3889 if (time_after(jiffies, fw_act_timeout)) {
3890 dev_warn(ctrl->device,
3891 "Fw activation timeout, reset controller\n");
3892 nvme_try_sched_reset(ctrl);
3898 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
3901 nvme_start_queues(ctrl);
3902 /* read FW slot information to clear the AER */
3903 nvme_get_fw_slot_info(ctrl);
3906 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
3908 u32 aer_notice_type = (result & 0xff00) >> 8;
3910 trace_nvme_async_event(ctrl, aer_notice_type);
3912 switch (aer_notice_type) {
3913 case NVME_AER_NOTICE_NS_CHANGED:
3914 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
3915 nvme_queue_scan(ctrl);
3917 case NVME_AER_NOTICE_FW_ACT_STARTING:
3919 * We are (ab)using the RESETTING state to prevent subsequent
3920 * recovery actions from interfering with the controller's
3921 * firmware activation.
3923 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
3924 queue_work(nvme_wq, &ctrl->fw_act_work);
3926 #ifdef CONFIG_NVME_MULTIPATH
3927 case NVME_AER_NOTICE_ANA:
3928 if (!ctrl->ana_log_buf)
3930 queue_work(nvme_wq, &ctrl->ana_work);
3933 case NVME_AER_NOTICE_DISC_CHANGED:
3934 ctrl->aen_result = result;
3937 dev_warn(ctrl->device, "async event result %08x\n", result);
3941 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
3942 volatile union nvme_result *res)
3944 u32 result = le32_to_cpu(res->u32);
3945 u32 aer_type = result & 0x07;
3947 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
3951 case NVME_AER_NOTICE:
3952 nvme_handle_aen_notice(ctrl, result);
3954 case NVME_AER_ERROR:
3955 case NVME_AER_SMART:
3958 trace_nvme_async_event(ctrl, aer_type);
3959 ctrl->aen_result = result;
3964 queue_work(nvme_wq, &ctrl->async_event_work);
3966 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
3968 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
3970 nvme_mpath_stop(ctrl);
3971 nvme_stop_keep_alive(ctrl);
3972 flush_work(&ctrl->async_event_work);
3973 cancel_work_sync(&ctrl->fw_act_work);
3975 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
3977 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
3980 nvme_start_keep_alive(ctrl);
3982 nvme_enable_aen(ctrl);
3984 if (ctrl->queue_count > 1) {
3985 nvme_queue_scan(ctrl);
3986 nvme_start_queues(ctrl);
3989 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
3991 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
3993 nvme_fault_inject_fini(&ctrl->fault_inject);
3994 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3995 cdev_device_del(&ctrl->cdev, ctrl->device);
3997 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
3999 static void nvme_free_ctrl(struct device *dev)
4001 struct nvme_ctrl *ctrl =
4002 container_of(dev, struct nvme_ctrl, ctrl_device);
4003 struct nvme_subsystem *subsys = ctrl->subsys;
4005 if (subsys && ctrl->instance != subsys->instance)
4006 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4008 kfree(ctrl->effects);
4009 nvme_mpath_uninit(ctrl);
4010 __free_page(ctrl->discard_page);
4013 mutex_lock(&nvme_subsystems_lock);
4014 list_del(&ctrl->subsys_entry);
4015 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4016 mutex_unlock(&nvme_subsystems_lock);
4019 ctrl->ops->free_ctrl(ctrl);
4022 nvme_put_subsystem(subsys);
4026 * Initialize a NVMe controller structures. This needs to be called during
4027 * earliest initialization so that we have the initialized structured around
4030 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4031 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4035 ctrl->state = NVME_CTRL_NEW;
4036 spin_lock_init(&ctrl->lock);
4037 mutex_init(&ctrl->scan_lock);
4038 INIT_LIST_HEAD(&ctrl->namespaces);
4039 init_rwsem(&ctrl->namespaces_rwsem);
4042 ctrl->quirks = quirks;
4043 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4044 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4045 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4046 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4047 init_waitqueue_head(&ctrl->state_wq);
4049 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4050 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4051 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4053 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4055 ctrl->discard_page = alloc_page(GFP_KERNEL);
4056 if (!ctrl->discard_page) {
4061 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4064 ctrl->instance = ret;
4066 device_initialize(&ctrl->ctrl_device);
4067 ctrl->device = &ctrl->ctrl_device;
4068 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4069 ctrl->device->class = nvme_class;
4070 ctrl->device->parent = ctrl->dev;
4071 ctrl->device->groups = nvme_dev_attr_groups;
4072 ctrl->device->release = nvme_free_ctrl;
4073 dev_set_drvdata(ctrl->device, ctrl);
4074 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4076 goto out_release_instance;
4078 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4079 ctrl->cdev.owner = ops->module;
4080 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4085 * Initialize latency tolerance controls. The sysfs files won't
4086 * be visible to userspace unless the device actually supports APST.
4088 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4089 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4090 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4092 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4096 kfree_const(ctrl->device->kobj.name);
4097 out_release_instance:
4098 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4100 if (ctrl->discard_page)
4101 __free_page(ctrl->discard_page);
4104 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4107 * nvme_kill_queues(): Ends all namespace queues
4108 * @ctrl: the dead controller that needs to end
4110 * Call this function when the driver determines it is unable to get the
4111 * controller in a state capable of servicing IO.
4113 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4117 down_read(&ctrl->namespaces_rwsem);
4119 /* Forcibly unquiesce queues to avoid blocking dispatch */
4120 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4121 blk_mq_unquiesce_queue(ctrl->admin_q);
4123 list_for_each_entry(ns, &ctrl->namespaces, list)
4124 nvme_set_queue_dying(ns);
4126 up_read(&ctrl->namespaces_rwsem);
4128 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4130 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4134 down_read(&ctrl->namespaces_rwsem);
4135 list_for_each_entry(ns, &ctrl->namespaces, list)
4136 blk_mq_unfreeze_queue(ns->queue);
4137 up_read(&ctrl->namespaces_rwsem);
4139 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4141 void nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4145 down_read(&ctrl->namespaces_rwsem);
4146 list_for_each_entry(ns, &ctrl->namespaces, list) {
4147 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4151 up_read(&ctrl->namespaces_rwsem);
4153 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4155 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4159 down_read(&ctrl->namespaces_rwsem);
4160 list_for_each_entry(ns, &ctrl->namespaces, list)
4161 blk_mq_freeze_queue_wait(ns->queue);
4162 up_read(&ctrl->namespaces_rwsem);
4164 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4166 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4170 down_read(&ctrl->namespaces_rwsem);
4171 list_for_each_entry(ns, &ctrl->namespaces, list)
4172 blk_freeze_queue_start(ns->queue);
4173 up_read(&ctrl->namespaces_rwsem);
4175 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4177 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4181 down_read(&ctrl->namespaces_rwsem);
4182 list_for_each_entry(ns, &ctrl->namespaces, list)
4183 blk_mq_quiesce_queue(ns->queue);
4184 up_read(&ctrl->namespaces_rwsem);
4186 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4188 void nvme_start_queues(struct nvme_ctrl *ctrl)
4192 down_read(&ctrl->namespaces_rwsem);
4193 list_for_each_entry(ns, &ctrl->namespaces, list)
4194 blk_mq_unquiesce_queue(ns->queue);
4195 up_read(&ctrl->namespaces_rwsem);
4197 EXPORT_SYMBOL_GPL(nvme_start_queues);
4200 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4204 down_read(&ctrl->namespaces_rwsem);
4205 list_for_each_entry(ns, &ctrl->namespaces, list)
4206 blk_sync_queue(ns->queue);
4207 up_read(&ctrl->namespaces_rwsem);
4210 blk_sync_queue(ctrl->admin_q);
4212 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4215 * Check we didn't inadvertently grow the command structure sizes:
4217 static inline void _nvme_check_size(void)
4219 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4220 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4221 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4222 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4223 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4224 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4225 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4226 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4227 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4228 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4229 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4230 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4231 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4232 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4233 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4234 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4235 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4239 static int __init nvme_core_init(void)
4241 int result = -ENOMEM;
4245 nvme_wq = alloc_workqueue("nvme-wq",
4246 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4250 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4251 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4255 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4256 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4257 if (!nvme_delete_wq)
4258 goto destroy_reset_wq;
4260 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4262 goto destroy_delete_wq;
4264 nvme_class = class_create(THIS_MODULE, "nvme");
4265 if (IS_ERR(nvme_class)) {
4266 result = PTR_ERR(nvme_class);
4267 goto unregister_chrdev;
4269 nvme_class->dev_uevent = nvme_class_uevent;
4271 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4272 if (IS_ERR(nvme_subsys_class)) {
4273 result = PTR_ERR(nvme_subsys_class);
4279 class_destroy(nvme_class);
4281 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4283 destroy_workqueue(nvme_delete_wq);
4285 destroy_workqueue(nvme_reset_wq);
4287 destroy_workqueue(nvme_wq);
4292 static void __exit nvme_core_exit(void)
4294 class_destroy(nvme_subsys_class);
4295 class_destroy(nvme_class);
4296 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4297 destroy_workqueue(nvme_delete_wq);
4298 destroy_workqueue(nvme_reset_wq);
4299 destroy_workqueue(nvme_wq);
4302 MODULE_LICENSE("GPL");
4303 MODULE_VERSION("1.0");
4304 module_init(nvme_core_init);
4305 module_exit(nvme_core_exit);