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/compat.h>
10 #include <linux/delay.h>
11 #include <linux/errno.h>
12 #include <linux/hdreg.h>
13 #include <linux/kernel.h>
14 #include <linux/module.h>
15 #include <linux/backing-dev.h>
16 #include <linux/list_sort.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.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 as scan, aen handling, fw activation,
70 * keep-alive, 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 void nvme_put_subsystem(struct nvme_subsystem *subsys);
93 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
96 static void nvme_update_bdev_size(struct gendisk *disk)
98 struct block_device *bdev = bdget_disk(disk, 0);
101 bd_set_nr_sectors(bdev, get_capacity(disk));
107 * Prepare a queue for teardown.
109 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
110 * the capacity to 0 after that to avoid blocking dispatchers that may be
111 * holding bd_butex. This will end buffered writers dirtying pages that can't
114 static void nvme_set_queue_dying(struct nvme_ns *ns)
116 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
119 blk_set_queue_dying(ns->queue);
120 blk_mq_unquiesce_queue(ns->queue);
122 set_capacity(ns->disk, 0);
123 nvme_update_bdev_size(ns->disk);
126 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
129 * Only new queue scan work when admin and IO queues are both alive
131 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
132 queue_work(nvme_wq, &ctrl->scan_work);
136 * Use this function to proceed with scheduling reset_work for a controller
137 * that had previously been set to the resetting state. This is intended for
138 * code paths that can't be interrupted by other reset attempts. A hot removal
139 * may prevent this from succeeding.
141 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
143 if (ctrl->state != NVME_CTRL_RESETTING)
145 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
149 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
151 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
153 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
155 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
159 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
161 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
165 ret = nvme_reset_ctrl(ctrl);
167 flush_work(&ctrl->reset_work);
168 if (ctrl->state != NVME_CTRL_LIVE)
174 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
176 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
178 dev_info(ctrl->device,
179 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
181 flush_work(&ctrl->reset_work);
182 nvme_stop_ctrl(ctrl);
183 nvme_remove_namespaces(ctrl);
184 ctrl->ops->delete_ctrl(ctrl);
185 nvme_uninit_ctrl(ctrl);
188 static void nvme_delete_ctrl_work(struct work_struct *work)
190 struct nvme_ctrl *ctrl =
191 container_of(work, struct nvme_ctrl, delete_work);
193 nvme_do_delete_ctrl(ctrl);
196 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
198 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
200 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
204 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
206 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
209 * Keep a reference until nvme_do_delete_ctrl() complete,
210 * since ->delete_ctrl can free the controller.
213 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
214 nvme_do_delete_ctrl(ctrl);
218 static blk_status_t nvme_error_status(u16 status)
220 switch (status & 0x7ff) {
221 case NVME_SC_SUCCESS:
223 case NVME_SC_CAP_EXCEEDED:
224 return BLK_STS_NOSPC;
225 case NVME_SC_LBA_RANGE:
226 case NVME_SC_CMD_INTERRUPTED:
227 case NVME_SC_NS_NOT_READY:
228 return BLK_STS_TARGET;
229 case NVME_SC_BAD_ATTRIBUTES:
230 case NVME_SC_ONCS_NOT_SUPPORTED:
231 case NVME_SC_INVALID_OPCODE:
232 case NVME_SC_INVALID_FIELD:
233 case NVME_SC_INVALID_NS:
234 return BLK_STS_NOTSUPP;
235 case NVME_SC_WRITE_FAULT:
236 case NVME_SC_READ_ERROR:
237 case NVME_SC_UNWRITTEN_BLOCK:
238 case NVME_SC_ACCESS_DENIED:
239 case NVME_SC_READ_ONLY:
240 case NVME_SC_COMPARE_FAILED:
241 return BLK_STS_MEDIUM;
242 case NVME_SC_GUARD_CHECK:
243 case NVME_SC_APPTAG_CHECK:
244 case NVME_SC_REFTAG_CHECK:
245 case NVME_SC_INVALID_PI:
246 return BLK_STS_PROTECTION;
247 case NVME_SC_RESERVATION_CONFLICT:
248 return BLK_STS_NEXUS;
249 case NVME_SC_HOST_PATH_ERROR:
250 return BLK_STS_TRANSPORT;
252 return BLK_STS_IOERR;
256 static void nvme_retry_req(struct request *req)
258 struct nvme_ns *ns = req->q->queuedata;
259 unsigned long delay = 0;
262 /* The mask and shift result must be <= 3 */
263 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
265 delay = ns->ctrl->crdt[crd - 1] * 100;
267 nvme_req(req)->retries++;
268 blk_mq_requeue_request(req, false);
269 blk_mq_delay_kick_requeue_list(req->q, delay);
272 enum nvme_disposition {
278 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
280 if (likely(nvme_req(req)->status == 0))
283 if (blk_noretry_request(req) ||
284 (nvme_req(req)->status & NVME_SC_DNR) ||
285 nvme_req(req)->retries >= nvme_max_retries)
288 if (req->cmd_flags & REQ_NVME_MPATH) {
289 if (nvme_is_path_error(nvme_req(req)->status) ||
290 blk_queue_dying(req->q))
293 if (blk_queue_dying(req->q))
300 static inline void nvme_end_req(struct request *req)
302 blk_status_t status = nvme_error_status(nvme_req(req)->status);
304 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
305 req_op(req) == REQ_OP_ZONE_APPEND)
306 req->__sector = nvme_lba_to_sect(req->q->queuedata,
307 le64_to_cpu(nvme_req(req)->result.u64));
309 nvme_trace_bio_complete(req, status);
310 blk_mq_end_request(req, status);
313 void nvme_complete_rq(struct request *req)
315 trace_nvme_complete_rq(req);
316 nvme_cleanup_cmd(req);
318 if (nvme_req(req)->ctrl->kas)
319 nvme_req(req)->ctrl->comp_seen = true;
321 switch (nvme_decide_disposition(req)) {
329 nvme_failover_req(req);
333 EXPORT_SYMBOL_GPL(nvme_complete_rq);
335 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
337 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
338 "Cancelling I/O %d", req->tag);
340 /* don't abort one completed request */
341 if (blk_mq_request_completed(req))
344 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
345 blk_mq_complete_request(req);
348 EXPORT_SYMBOL_GPL(nvme_cancel_request);
350 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
351 enum nvme_ctrl_state new_state)
353 enum nvme_ctrl_state old_state;
355 bool changed = false;
357 spin_lock_irqsave(&ctrl->lock, flags);
359 old_state = ctrl->state;
364 case NVME_CTRL_RESETTING:
365 case NVME_CTRL_CONNECTING:
372 case NVME_CTRL_RESETTING:
382 case NVME_CTRL_CONNECTING:
385 case NVME_CTRL_RESETTING:
392 case NVME_CTRL_DELETING:
395 case NVME_CTRL_RESETTING:
396 case NVME_CTRL_CONNECTING:
403 case NVME_CTRL_DELETING_NOIO:
405 case NVME_CTRL_DELETING:
415 case NVME_CTRL_DELETING:
427 ctrl->state = new_state;
428 wake_up_all(&ctrl->state_wq);
431 spin_unlock_irqrestore(&ctrl->lock, flags);
432 if (changed && ctrl->state == NVME_CTRL_LIVE)
433 nvme_kick_requeue_lists(ctrl);
436 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
439 * Returns true for sink states that can't ever transition back to live.
441 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
443 switch (ctrl->state) {
446 case NVME_CTRL_RESETTING:
447 case NVME_CTRL_CONNECTING:
449 case NVME_CTRL_DELETING:
450 case NVME_CTRL_DELETING_NOIO:
454 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
460 * Waits for the controller state to be resetting, or returns false if it is
461 * not possible to ever transition to that state.
463 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
465 wait_event(ctrl->state_wq,
466 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
467 nvme_state_terminal(ctrl));
468 return ctrl->state == NVME_CTRL_RESETTING;
470 EXPORT_SYMBOL_GPL(nvme_wait_reset);
472 static void nvme_free_ns_head(struct kref *ref)
474 struct nvme_ns_head *head =
475 container_of(ref, struct nvme_ns_head, ref);
477 nvme_mpath_remove_disk(head);
478 ida_simple_remove(&head->subsys->ns_ida, head->instance);
479 cleanup_srcu_struct(&head->srcu);
480 nvme_put_subsystem(head->subsys);
484 static void nvme_put_ns_head(struct nvme_ns_head *head)
486 kref_put(&head->ref, nvme_free_ns_head);
489 static void nvme_free_ns(struct kref *kref)
491 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
494 nvme_nvm_unregister(ns);
497 nvme_put_ns_head(ns->head);
498 nvme_put_ctrl(ns->ctrl);
502 void nvme_put_ns(struct nvme_ns *ns)
504 kref_put(&ns->kref, nvme_free_ns);
506 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
508 static inline void nvme_clear_nvme_request(struct request *req)
510 if (!(req->rq_flags & RQF_DONTPREP)) {
511 nvme_req(req)->retries = 0;
512 nvme_req(req)->flags = 0;
513 req->rq_flags |= RQF_DONTPREP;
517 struct request *nvme_alloc_request(struct request_queue *q,
518 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
520 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
523 if (qid == NVME_QID_ANY) {
524 req = blk_mq_alloc_request(q, op, flags);
526 req = blk_mq_alloc_request_hctx(q, op, flags,
532 req->cmd_flags |= REQ_FAILFAST_DRIVER;
533 nvme_clear_nvme_request(req);
534 nvme_req(req)->cmd = cmd;
538 EXPORT_SYMBOL_GPL(nvme_alloc_request);
540 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
542 struct nvme_command c;
544 memset(&c, 0, sizeof(c));
546 c.directive.opcode = nvme_admin_directive_send;
547 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
548 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
549 c.directive.dtype = NVME_DIR_IDENTIFY;
550 c.directive.tdtype = NVME_DIR_STREAMS;
551 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
553 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
556 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
558 return nvme_toggle_streams(ctrl, false);
561 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
563 return nvme_toggle_streams(ctrl, true);
566 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
567 struct streams_directive_params *s, u32 nsid)
569 struct nvme_command c;
571 memset(&c, 0, sizeof(c));
572 memset(s, 0, sizeof(*s));
574 c.directive.opcode = nvme_admin_directive_recv;
575 c.directive.nsid = cpu_to_le32(nsid);
576 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
577 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
578 c.directive.dtype = NVME_DIR_STREAMS;
580 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
583 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
585 struct streams_directive_params s;
588 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
593 ret = nvme_enable_streams(ctrl);
597 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
599 goto out_disable_stream;
601 ctrl->nssa = le16_to_cpu(s.nssa);
602 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
603 dev_info(ctrl->device, "too few streams (%u) available\n",
605 goto out_disable_stream;
608 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
609 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
613 nvme_disable_streams(ctrl);
618 * Check if 'req' has a write hint associated with it. If it does, assign
619 * a valid namespace stream to the write.
621 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
622 struct request *req, u16 *control,
625 enum rw_hint streamid = req->write_hint;
627 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
631 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
634 *control |= NVME_RW_DTYPE_STREAMS;
635 *dsmgmt |= streamid << 16;
638 if (streamid < ARRAY_SIZE(req->q->write_hints))
639 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
642 static void nvme_setup_passthrough(struct request *req,
643 struct nvme_command *cmd)
645 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
646 /* passthru commands should let the driver set the SGL flags */
647 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
650 static inline void nvme_setup_flush(struct nvme_ns *ns,
651 struct nvme_command *cmnd)
653 cmnd->common.opcode = nvme_cmd_flush;
654 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
657 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
658 struct nvme_command *cmnd)
660 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
661 struct nvme_dsm_range *range;
665 * Some devices do not consider the DSM 'Number of Ranges' field when
666 * determining how much data to DMA. Always allocate memory for maximum
667 * number of segments to prevent device reading beyond end of buffer.
669 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
671 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
674 * If we fail allocation our range, fallback to the controller
675 * discard page. If that's also busy, it's safe to return
676 * busy, as we know we can make progress once that's freed.
678 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
679 return BLK_STS_RESOURCE;
681 range = page_address(ns->ctrl->discard_page);
684 __rq_for_each_bio(bio, req) {
685 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
686 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
689 range[n].cattr = cpu_to_le32(0);
690 range[n].nlb = cpu_to_le32(nlb);
691 range[n].slba = cpu_to_le64(slba);
696 if (WARN_ON_ONCE(n != segments)) {
697 if (virt_to_page(range) == ns->ctrl->discard_page)
698 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
701 return BLK_STS_IOERR;
704 cmnd->dsm.opcode = nvme_cmd_dsm;
705 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
706 cmnd->dsm.nr = cpu_to_le32(segments - 1);
707 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
709 req->special_vec.bv_page = virt_to_page(range);
710 req->special_vec.bv_offset = offset_in_page(range);
711 req->special_vec.bv_len = alloc_size;
712 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
717 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
718 struct request *req, struct nvme_command *cmnd)
720 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
721 return nvme_setup_discard(ns, req, cmnd);
723 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
724 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
725 cmnd->write_zeroes.slba =
726 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
727 cmnd->write_zeroes.length =
728 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
729 cmnd->write_zeroes.control = 0;
733 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
734 struct request *req, struct nvme_command *cmnd,
737 struct nvme_ctrl *ctrl = ns->ctrl;
741 if (req->cmd_flags & REQ_FUA)
742 control |= NVME_RW_FUA;
743 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
744 control |= NVME_RW_LR;
746 if (req->cmd_flags & REQ_RAHEAD)
747 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
749 cmnd->rw.opcode = op;
750 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
751 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
752 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
754 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
755 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
759 * If formated with metadata, the block layer always provides a
760 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
761 * we enable the PRACT bit for protection information or set the
762 * namespace capacity to zero to prevent any I/O.
764 if (!blk_integrity_rq(req)) {
765 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
766 return BLK_STS_NOTSUPP;
767 control |= NVME_RW_PRINFO_PRACT;
770 switch (ns->pi_type) {
771 case NVME_NS_DPS_PI_TYPE3:
772 control |= NVME_RW_PRINFO_PRCHK_GUARD;
774 case NVME_NS_DPS_PI_TYPE1:
775 case NVME_NS_DPS_PI_TYPE2:
776 control |= NVME_RW_PRINFO_PRCHK_GUARD |
777 NVME_RW_PRINFO_PRCHK_REF;
778 if (op == nvme_cmd_zone_append)
779 control |= NVME_RW_APPEND_PIREMAP;
780 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
785 cmnd->rw.control = cpu_to_le16(control);
786 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
790 void nvme_cleanup_cmd(struct request *req)
792 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
793 struct nvme_ns *ns = req->rq_disk->private_data;
794 struct page *page = req->special_vec.bv_page;
796 if (page == ns->ctrl->discard_page)
797 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
799 kfree(page_address(page) + req->special_vec.bv_offset);
802 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
804 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
805 struct nvme_command *cmd)
807 blk_status_t ret = BLK_STS_OK;
809 nvme_clear_nvme_request(req);
811 memset(cmd, 0, sizeof(*cmd));
812 switch (req_op(req)) {
815 nvme_setup_passthrough(req, cmd);
818 nvme_setup_flush(ns, cmd);
820 case REQ_OP_ZONE_RESET_ALL:
821 case REQ_OP_ZONE_RESET:
822 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
824 case REQ_OP_ZONE_OPEN:
825 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
827 case REQ_OP_ZONE_CLOSE:
828 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
830 case REQ_OP_ZONE_FINISH:
831 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
833 case REQ_OP_WRITE_ZEROES:
834 ret = nvme_setup_write_zeroes(ns, req, cmd);
837 ret = nvme_setup_discard(ns, req, cmd);
840 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
843 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
845 case REQ_OP_ZONE_APPEND:
846 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
850 return BLK_STS_IOERR;
853 cmd->common.command_id = req->tag;
854 trace_nvme_setup_cmd(req, cmd);
857 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
859 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
861 struct completion *waiting = rq->end_io_data;
863 rq->end_io_data = NULL;
867 static void nvme_execute_rq_polled(struct request_queue *q,
868 struct gendisk *bd_disk, struct request *rq, int at_head)
870 DECLARE_COMPLETION_ONSTACK(wait);
872 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
874 rq->cmd_flags |= REQ_HIPRI;
875 rq->end_io_data = &wait;
876 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
878 while (!completion_done(&wait)) {
879 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
885 * Returns 0 on success. If the result is negative, it's a Linux error code;
886 * if the result is positive, it's an NVM Express status code
888 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
889 union nvme_result *result, void *buffer, unsigned bufflen,
890 unsigned timeout, int qid, int at_head,
891 blk_mq_req_flags_t flags, bool poll)
896 req = nvme_alloc_request(q, cmd, flags, qid);
900 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
902 if (buffer && bufflen) {
903 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
909 nvme_execute_rq_polled(req->q, NULL, req, at_head);
911 blk_execute_rq(req->q, NULL, req, at_head);
913 *result = nvme_req(req)->result;
914 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
917 ret = nvme_req(req)->status;
919 blk_mq_free_request(req);
922 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
924 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
925 void *buffer, unsigned bufflen)
927 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
928 NVME_QID_ANY, 0, 0, false);
930 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
932 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
933 unsigned len, u32 seed, bool write)
935 struct bio_integrity_payload *bip;
939 buf = kmalloc(len, GFP_KERNEL);
944 if (write && copy_from_user(buf, ubuf, len))
947 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
953 bip->bip_iter.bi_size = len;
954 bip->bip_iter.bi_sector = seed;
955 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
956 offset_in_page(buf));
966 static u32 nvme_known_admin_effects(u8 opcode)
969 case nvme_admin_format_nvm:
970 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
971 NVME_CMD_EFFECTS_CSE_MASK;
972 case nvme_admin_sanitize_nvm:
973 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
980 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
985 if (ns->head->effects)
986 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
987 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
988 dev_warn(ctrl->device,
989 "IO command:%02x has unhandled effects:%08x\n",
995 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
996 effects |= nvme_known_admin_effects(opcode);
1000 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1002 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1005 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1008 * For simplicity, IO to all namespaces is quiesced even if the command
1009 * effects say only one namespace is affected.
1011 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1012 mutex_lock(&ctrl->scan_lock);
1013 mutex_lock(&ctrl->subsys->lock);
1014 nvme_mpath_start_freeze(ctrl->subsys);
1015 nvme_mpath_wait_freeze(ctrl->subsys);
1016 nvme_start_freeze(ctrl);
1017 nvme_wait_freeze(ctrl);
1022 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1024 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1025 nvme_unfreeze(ctrl);
1026 nvme_mpath_unfreeze(ctrl->subsys);
1027 mutex_unlock(&ctrl->subsys->lock);
1028 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1029 mutex_unlock(&ctrl->scan_lock);
1031 if (effects & NVME_CMD_EFFECTS_CCC)
1032 nvme_init_identify(ctrl);
1033 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1034 nvme_queue_scan(ctrl);
1035 flush_work(&ctrl->scan_work);
1039 void nvme_execute_passthru_rq(struct request *rq)
1041 struct nvme_command *cmd = nvme_req(rq)->cmd;
1042 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1043 struct nvme_ns *ns = rq->q->queuedata;
1044 struct gendisk *disk = ns ? ns->disk : NULL;
1047 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1048 blk_execute_rq(rq->q, disk, rq, 0);
1049 nvme_passthru_end(ctrl, effects);
1051 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1053 static int nvme_submit_user_cmd(struct request_queue *q,
1054 struct nvme_command *cmd, void __user *ubuffer,
1055 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1056 u32 meta_seed, u64 *result, unsigned timeout)
1058 bool write = nvme_is_write(cmd);
1059 struct nvme_ns *ns = q->queuedata;
1060 struct gendisk *disk = ns ? ns->disk : NULL;
1061 struct request *req;
1062 struct bio *bio = NULL;
1066 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
1068 return PTR_ERR(req);
1070 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
1071 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1073 if (ubuffer && bufflen) {
1074 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1079 bio->bi_disk = disk;
1080 if (disk && meta_buffer && meta_len) {
1081 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1084 ret = PTR_ERR(meta);
1087 req->cmd_flags |= REQ_INTEGRITY;
1091 nvme_execute_passthru_rq(req);
1092 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1095 ret = nvme_req(req)->status;
1097 *result = le64_to_cpu(nvme_req(req)->result.u64);
1098 if (meta && !ret && !write) {
1099 if (copy_to_user(meta_buffer, meta, meta_len))
1105 blk_rq_unmap_user(bio);
1107 blk_mq_free_request(req);
1111 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1113 struct nvme_ctrl *ctrl = rq->end_io_data;
1114 unsigned long flags;
1115 bool startka = false;
1117 blk_mq_free_request(rq);
1120 dev_err(ctrl->device,
1121 "failed nvme_keep_alive_end_io error=%d\n",
1126 ctrl->comp_seen = false;
1127 spin_lock_irqsave(&ctrl->lock, flags);
1128 if (ctrl->state == NVME_CTRL_LIVE ||
1129 ctrl->state == NVME_CTRL_CONNECTING)
1131 spin_unlock_irqrestore(&ctrl->lock, flags);
1133 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1136 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1140 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
1145 rq->timeout = ctrl->kato * HZ;
1146 rq->end_io_data = ctrl;
1148 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1153 static void nvme_keep_alive_work(struct work_struct *work)
1155 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1156 struct nvme_ctrl, ka_work);
1157 bool comp_seen = ctrl->comp_seen;
1159 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1160 dev_dbg(ctrl->device,
1161 "reschedule traffic based keep-alive timer\n");
1162 ctrl->comp_seen = false;
1163 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1167 if (nvme_keep_alive(ctrl)) {
1168 /* allocation failure, reset the controller */
1169 dev_err(ctrl->device, "keep-alive failed\n");
1170 nvme_reset_ctrl(ctrl);
1175 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1177 if (unlikely(ctrl->kato == 0))
1180 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1183 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1185 if (unlikely(ctrl->kato == 0))
1188 cancel_delayed_work_sync(&ctrl->ka_work);
1190 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1193 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1194 * flag, thus sending any new CNS opcodes has a big chance of not working.
1195 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1196 * (but not for any later version).
1198 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1200 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1201 return ctrl->vs < NVME_VS(1, 2, 0);
1202 return ctrl->vs < NVME_VS(1, 1, 0);
1205 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1207 struct nvme_command c = { };
1210 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1211 c.identify.opcode = nvme_admin_identify;
1212 c.identify.cns = NVME_ID_CNS_CTRL;
1214 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1218 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1219 sizeof(struct nvme_id_ctrl));
1225 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1227 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1230 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1231 struct nvme_ns_id_desc *cur, bool *csi_seen)
1233 const char *warn_str = "ctrl returned bogus length:";
1236 switch (cur->nidt) {
1237 case NVME_NIDT_EUI64:
1238 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1239 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1240 warn_str, cur->nidl);
1243 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1244 return NVME_NIDT_EUI64_LEN;
1245 case NVME_NIDT_NGUID:
1246 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1247 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1248 warn_str, cur->nidl);
1251 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1252 return NVME_NIDT_NGUID_LEN;
1253 case NVME_NIDT_UUID:
1254 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1255 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1256 warn_str, cur->nidl);
1259 uuid_copy(&ids->uuid, data + sizeof(*cur));
1260 return NVME_NIDT_UUID_LEN;
1262 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1263 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1264 warn_str, cur->nidl);
1267 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1269 return NVME_NIDT_CSI_LEN;
1271 /* Skip unknown types */
1276 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1277 struct nvme_ns_ids *ids)
1279 struct nvme_command c = { };
1280 bool csi_seen = false;
1281 int status, pos, len;
1284 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1287 c.identify.opcode = nvme_admin_identify;
1288 c.identify.nsid = cpu_to_le32(nsid);
1289 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1291 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1295 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1296 NVME_IDENTIFY_DATA_SIZE);
1298 dev_warn(ctrl->device,
1299 "Identify Descriptors failed (%d)\n", status);
1303 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1304 struct nvme_ns_id_desc *cur = data + pos;
1309 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1313 len += sizeof(*cur);
1316 if (nvme_multi_css(ctrl) && !csi_seen) {
1317 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1327 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1329 struct nvme_command c = { };
1331 c.identify.opcode = nvme_admin_identify;
1332 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1333 c.identify.nsid = cpu_to_le32(nsid);
1334 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1335 NVME_IDENTIFY_DATA_SIZE);
1338 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1339 unsigned nsid, struct nvme_id_ns **id)
1341 struct nvme_command c = { };
1344 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1345 c.identify.opcode = nvme_admin_identify;
1346 c.identify.nsid = cpu_to_le32(nsid);
1347 c.identify.cns = NVME_ID_CNS_NS;
1349 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1353 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1355 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1360 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1369 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1370 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1372 union nvme_result res = { 0 };
1373 struct nvme_command c;
1376 memset(&c, 0, sizeof(c));
1377 c.features.opcode = op;
1378 c.features.fid = cpu_to_le32(fid);
1379 c.features.dword11 = cpu_to_le32(dword11);
1381 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1382 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1383 if (ret >= 0 && result)
1384 *result = le32_to_cpu(res.u32);
1388 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1389 unsigned int dword11, void *buffer, size_t buflen,
1392 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1395 EXPORT_SYMBOL_GPL(nvme_set_features);
1397 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1398 unsigned int dword11, void *buffer, size_t buflen,
1401 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1404 EXPORT_SYMBOL_GPL(nvme_get_features);
1406 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1408 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1410 int status, nr_io_queues;
1412 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1418 * Degraded controllers might return an error when setting the queue
1419 * count. We still want to be able to bring them online and offer
1420 * access to the admin queue, as that might be only way to fix them up.
1423 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1426 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1427 *count = min(*count, nr_io_queues);
1432 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1434 #define NVME_AEN_SUPPORTED \
1435 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1436 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1438 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1440 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1443 if (!supported_aens)
1446 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1449 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1452 queue_work(nvme_wq, &ctrl->async_event_work);
1456 * Convert integer values from ioctl structures to user pointers, silently
1457 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1460 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1462 if (in_compat_syscall())
1463 ptrval = (compat_uptr_t)ptrval;
1464 return (void __user *)ptrval;
1467 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1469 struct nvme_user_io io;
1470 struct nvme_command c;
1471 unsigned length, meta_len;
1472 void __user *metadata;
1474 if (copy_from_user(&io, uio, sizeof(io)))
1479 switch (io.opcode) {
1480 case nvme_cmd_write:
1482 case nvme_cmd_compare:
1488 length = (io.nblocks + 1) << ns->lba_shift;
1489 meta_len = (io.nblocks + 1) * ns->ms;
1490 metadata = nvme_to_user_ptr(io.metadata);
1492 if (ns->features & NVME_NS_EXT_LBAS) {
1495 } else if (meta_len) {
1496 if ((io.metadata & 3) || !io.metadata)
1500 memset(&c, 0, sizeof(c));
1501 c.rw.opcode = io.opcode;
1502 c.rw.flags = io.flags;
1503 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1504 c.rw.slba = cpu_to_le64(io.slba);
1505 c.rw.length = cpu_to_le16(io.nblocks);
1506 c.rw.control = cpu_to_le16(io.control);
1507 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1508 c.rw.reftag = cpu_to_le32(io.reftag);
1509 c.rw.apptag = cpu_to_le16(io.apptag);
1510 c.rw.appmask = cpu_to_le16(io.appmask);
1512 return nvme_submit_user_cmd(ns->queue, &c,
1513 nvme_to_user_ptr(io.addr), length,
1514 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1517 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1518 struct nvme_passthru_cmd __user *ucmd)
1520 struct nvme_passthru_cmd cmd;
1521 struct nvme_command c;
1522 unsigned timeout = 0;
1526 if (!capable(CAP_SYS_ADMIN))
1528 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1533 memset(&c, 0, sizeof(c));
1534 c.common.opcode = cmd.opcode;
1535 c.common.flags = cmd.flags;
1536 c.common.nsid = cpu_to_le32(cmd.nsid);
1537 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1538 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1539 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1540 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1541 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1542 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1543 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1544 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1547 timeout = msecs_to_jiffies(cmd.timeout_ms);
1549 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1550 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1551 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1552 0, &result, timeout);
1555 if (put_user(result, &ucmd->result))
1562 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1563 struct nvme_passthru_cmd64 __user *ucmd)
1565 struct nvme_passthru_cmd64 cmd;
1566 struct nvme_command c;
1567 unsigned timeout = 0;
1570 if (!capable(CAP_SYS_ADMIN))
1572 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1577 memset(&c, 0, sizeof(c));
1578 c.common.opcode = cmd.opcode;
1579 c.common.flags = cmd.flags;
1580 c.common.nsid = cpu_to_le32(cmd.nsid);
1581 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1582 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1583 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1584 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1585 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1586 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1587 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1588 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1591 timeout = msecs_to_jiffies(cmd.timeout_ms);
1593 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1594 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1595 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1596 0, &cmd.result, timeout);
1599 if (put_user(cmd.result, &ucmd->result))
1607 * Issue ioctl requests on the first available path. Note that unlike normal
1608 * block layer requests we will not retry failed request on another controller.
1610 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1611 struct nvme_ns_head **head, int *srcu_idx)
1613 #ifdef CONFIG_NVME_MULTIPATH
1614 if (disk->fops == &nvme_ns_head_ops) {
1617 *head = disk->private_data;
1618 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1619 ns = nvme_find_path(*head);
1621 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1627 return disk->private_data;
1630 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1633 srcu_read_unlock(&head->srcu, idx);
1636 static bool is_ctrl_ioctl(unsigned int cmd)
1638 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1640 if (is_sed_ioctl(cmd))
1645 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1647 struct nvme_ns_head *head,
1650 struct nvme_ctrl *ctrl = ns->ctrl;
1653 nvme_get_ctrl(ns->ctrl);
1654 nvme_put_ns_from_disk(head, srcu_idx);
1657 case NVME_IOCTL_ADMIN_CMD:
1658 ret = nvme_user_cmd(ctrl, NULL, argp);
1660 case NVME_IOCTL_ADMIN64_CMD:
1661 ret = nvme_user_cmd64(ctrl, NULL, argp);
1664 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1667 nvme_put_ctrl(ctrl);
1671 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1672 unsigned int cmd, unsigned long arg)
1674 struct nvme_ns_head *head = NULL;
1675 void __user *argp = (void __user *)arg;
1679 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1681 return -EWOULDBLOCK;
1684 * Handle ioctls that apply to the controller instead of the namespace
1685 * seperately and drop the ns SRCU reference early. This avoids a
1686 * deadlock when deleting namespaces using the passthrough interface.
1688 if (is_ctrl_ioctl(cmd))
1689 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1693 force_successful_syscall_return();
1694 ret = ns->head->ns_id;
1696 case NVME_IOCTL_IO_CMD:
1697 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1699 case NVME_IOCTL_SUBMIT_IO:
1700 ret = nvme_submit_io(ns, argp);
1702 case NVME_IOCTL_IO64_CMD:
1703 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1707 ret = nvme_nvm_ioctl(ns, cmd, arg);
1712 nvme_put_ns_from_disk(head, srcu_idx);
1716 #ifdef CONFIG_COMPAT
1717 struct nvme_user_io32 {
1730 } __attribute__((__packed__));
1732 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1734 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1735 unsigned int cmd, unsigned long arg)
1738 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1739 * between 32 bit programs and 64 bit kernel.
1740 * The cause is that the results of sizeof(struct nvme_user_io),
1741 * which is used to define NVME_IOCTL_SUBMIT_IO,
1742 * are not same between 32 bit compiler and 64 bit compiler.
1743 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1744 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1745 * Other IOCTL numbers are same between 32 bit and 64 bit.
1746 * So there is nothing to do regarding to other IOCTL numbers.
1748 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1749 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1751 return nvme_ioctl(bdev, mode, cmd, arg);
1754 #define nvme_compat_ioctl NULL
1755 #endif /* CONFIG_COMPAT */
1757 static int nvme_open(struct block_device *bdev, fmode_t mode)
1759 struct nvme_ns *ns = bdev->bd_disk->private_data;
1761 #ifdef CONFIG_NVME_MULTIPATH
1762 /* should never be called due to GENHD_FL_HIDDEN */
1763 if (WARN_ON_ONCE(ns->head->disk))
1766 if (!kref_get_unless_zero(&ns->kref))
1768 if (!try_module_get(ns->ctrl->ops->module))
1779 static void nvme_release(struct gendisk *disk, fmode_t mode)
1781 struct nvme_ns *ns = disk->private_data;
1783 module_put(ns->ctrl->ops->module);
1787 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1789 /* some standard values */
1790 geo->heads = 1 << 6;
1791 geo->sectors = 1 << 5;
1792 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1796 #ifdef CONFIG_BLK_DEV_INTEGRITY
1797 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1798 u32 max_integrity_segments)
1800 struct blk_integrity integrity;
1802 memset(&integrity, 0, sizeof(integrity));
1804 case NVME_NS_DPS_PI_TYPE3:
1805 integrity.profile = &t10_pi_type3_crc;
1806 integrity.tag_size = sizeof(u16) + sizeof(u32);
1807 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1809 case NVME_NS_DPS_PI_TYPE1:
1810 case NVME_NS_DPS_PI_TYPE2:
1811 integrity.profile = &t10_pi_type1_crc;
1812 integrity.tag_size = sizeof(u16);
1813 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1816 integrity.profile = NULL;
1819 integrity.tuple_size = ms;
1820 blk_integrity_register(disk, &integrity);
1821 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1824 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1825 u32 max_integrity_segments)
1828 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1830 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1832 struct nvme_ctrl *ctrl = ns->ctrl;
1833 struct request_queue *queue = disk->queue;
1834 u32 size = queue_logical_block_size(queue);
1836 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1837 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1841 if (ctrl->nr_streams && ns->sws && ns->sgs)
1842 size *= ns->sws * ns->sgs;
1844 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1845 NVME_DSM_MAX_RANGES);
1847 queue->limits.discard_alignment = 0;
1848 queue->limits.discard_granularity = size;
1850 /* If discard is already enabled, don't reset queue limits */
1851 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1854 blk_queue_max_discard_sectors(queue, UINT_MAX);
1855 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1857 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1858 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1861 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1865 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1866 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1869 * Even though NVMe spec explicitly states that MDTS is not
1870 * applicable to the write-zeroes:- "The restriction does not apply to
1871 * commands that do not transfer data between the host and the
1872 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1873 * In order to be more cautious use controller's max_hw_sectors value
1874 * to configure the maximum sectors for the write-zeroes which is
1875 * configured based on the controller's MDTS field in the
1876 * nvme_init_identify() if available.
1878 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1879 max_blocks = (u64)USHRT_MAX + 1;
1881 max_blocks = ns->ctrl->max_hw_sectors + 1;
1883 blk_queue_max_write_zeroes_sectors(disk->queue,
1884 nvme_lba_to_sect(ns, max_blocks));
1887 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1888 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1890 memset(ids, 0, sizeof(*ids));
1892 if (ctrl->vs >= NVME_VS(1, 1, 0))
1893 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1894 if (ctrl->vs >= NVME_VS(1, 2, 0))
1895 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1896 if (ctrl->vs >= NVME_VS(1, 3, 0) || nvme_multi_css(ctrl))
1897 return nvme_identify_ns_descs(ctrl, nsid, ids);
1901 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1903 return !uuid_is_null(&ids->uuid) ||
1904 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1905 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1908 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1910 return uuid_equal(&a->uuid, &b->uuid) &&
1911 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1912 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1916 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1917 u32 *phys_bs, u32 *io_opt)
1919 struct streams_directive_params s;
1922 if (!ctrl->nr_streams)
1925 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1929 ns->sws = le32_to_cpu(s.sws);
1930 ns->sgs = le16_to_cpu(s.sgs);
1933 *phys_bs = ns->sws * (1 << ns->lba_shift);
1935 *io_opt = *phys_bs * ns->sgs;
1941 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1943 struct nvme_ctrl *ctrl = ns->ctrl;
1946 * The PI implementation requires the metadata size to be equal to the
1947 * t10 pi tuple size.
1949 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1950 if (ns->ms == sizeof(struct t10_pi_tuple))
1951 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1955 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1956 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1958 if (ctrl->ops->flags & NVME_F_FABRICS) {
1960 * The NVMe over Fabrics specification only supports metadata as
1961 * part of the extended data LBA. We rely on HCA/HBA support to
1962 * remap the separate metadata buffer from the block layer.
1964 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1966 if (ctrl->max_integrity_segments)
1968 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1971 * For PCIe controllers, we can't easily remap the separate
1972 * metadata buffer from the block layer and thus require a
1973 * separate metadata buffer for block layer metadata/PI support.
1974 * We allow extended LBAs for the passthrough interface, though.
1976 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1977 ns->features |= NVME_NS_EXT_LBAS;
1979 ns->features |= NVME_NS_METADATA_SUPPORTED;
1985 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1986 struct request_queue *q)
1990 if (ctrl->max_hw_sectors) {
1992 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1994 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1995 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1996 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1998 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1999 blk_queue_dma_alignment(q, 7);
2000 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2002 blk_queue_write_cache(q, vwc, vwc);
2005 static void nvme_update_disk_info(struct gendisk *disk,
2006 struct nvme_ns *ns, struct nvme_id_ns *id)
2008 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2009 unsigned short bs = 1 << ns->lba_shift;
2010 u32 atomic_bs, phys_bs, io_opt = 0;
2013 * The block layer can't support LBA sizes larger than the page size
2014 * yet, so catch this early and don't allow block I/O.
2016 if (ns->lba_shift > PAGE_SHIFT) {
2021 blk_integrity_unregister(disk);
2023 atomic_bs = phys_bs = bs;
2024 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2025 if (id->nabo == 0) {
2027 * Bit 1 indicates whether NAWUPF is defined for this namespace
2028 * and whether it should be used instead of AWUPF. If NAWUPF ==
2029 * 0 then AWUPF must be used instead.
2031 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2032 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2034 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2037 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2038 /* NPWG = Namespace Preferred Write Granularity */
2039 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2040 /* NOWS = Namespace Optimal Write Size */
2041 io_opt = bs * (1 + le16_to_cpu(id->nows));
2044 blk_queue_logical_block_size(disk->queue, bs);
2046 * Linux filesystems assume writing a single physical block is
2047 * an atomic operation. Hence limit the physical block size to the
2048 * value of the Atomic Write Unit Power Fail parameter.
2050 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2051 blk_queue_io_min(disk->queue, phys_bs);
2052 blk_queue_io_opt(disk->queue, io_opt);
2055 * Register a metadata profile for PI, or the plain non-integrity NVMe
2056 * metadata masquerading as Type 0 if supported, otherwise reject block
2057 * I/O to namespaces with metadata except when the namespace supports
2058 * PI, as it can strip/insert in that case.
2061 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2062 (ns->features & NVME_NS_METADATA_SUPPORTED))
2063 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2064 ns->ctrl->max_integrity_segments);
2065 else if (!nvme_ns_has_pi(ns))
2069 set_capacity_revalidate_and_notify(disk, capacity, false);
2071 nvme_config_discard(disk, ns);
2072 nvme_config_write_zeroes(disk, ns);
2074 if (id->nsattr & NVME_NS_ATTR_RO)
2075 set_disk_ro(disk, true);
2077 set_disk_ro(disk, false);
2080 static inline bool nvme_first_scan(struct gendisk *disk)
2082 /* nvme_alloc_ns() scans the disk prior to adding it */
2083 return !(disk->flags & GENHD_FL_UP);
2086 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2088 struct nvme_ctrl *ctrl = ns->ctrl;
2091 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2092 is_power_of_2(ctrl->max_hw_sectors))
2093 iob = ctrl->max_hw_sectors;
2095 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2100 if (!is_power_of_2(iob)) {
2101 if (nvme_first_scan(ns->disk))
2102 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2103 ns->disk->disk_name, iob);
2107 if (blk_queue_is_zoned(ns->disk->queue)) {
2108 if (nvme_first_scan(ns->disk))
2109 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2110 ns->disk->disk_name);
2114 blk_queue_chunk_sectors(ns->queue, iob);
2117 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2119 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2120 struct nvme_ctrl *ctrl = ns->ctrl;
2123 blk_mq_freeze_queue(ns->disk->queue);
2124 ns->lba_shift = id->lbaf[lbaf].ds;
2125 nvme_set_queue_limits(ctrl, ns->queue);
2127 switch (ns->head->ids.csi) {
2131 ret = nvme_update_zone_info(ns, lbaf);
2133 dev_warn(ctrl->device,
2134 "failed to add zoned namespace:%u ret:%d\n",
2135 ns->head->ns_id, ret);
2140 dev_warn(ctrl->device, "unknown csi:%u ns:%u\n",
2141 ns->head->ids.csi, ns->head->ns_id);
2146 ret = nvme_configure_metadata(ns, id);
2149 nvme_set_chunk_sectors(ns, id);
2150 nvme_update_disk_info(ns->disk, ns, id);
2151 blk_mq_unfreeze_queue(ns->disk->queue);
2153 if (blk_queue_is_zoned(ns->queue)) {
2154 ret = nvme_revalidate_zones(ns);
2159 #ifdef CONFIG_NVME_MULTIPATH
2160 if (ns->head->disk) {
2161 blk_mq_freeze_queue(ns->head->disk->queue);
2162 nvme_update_disk_info(ns->head->disk, ns, id);
2163 blk_stack_limits(&ns->head->disk->queue->limits,
2164 &ns->queue->limits, 0);
2165 blk_queue_update_readahead(ns->head->disk->queue);
2166 nvme_update_bdev_size(ns->head->disk);
2167 blk_mq_unfreeze_queue(ns->head->disk->queue);
2173 blk_mq_unfreeze_queue(ns->disk->queue);
2177 static int nvme_validate_ns(struct nvme_ns *ns)
2179 struct nvme_ctrl *ctrl = ns->ctrl;
2180 struct nvme_id_ns *id;
2181 struct nvme_ns_ids ids;
2184 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
2185 set_capacity(ns->disk, 0);
2189 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
2193 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
2197 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
2198 dev_err(ctrl->device,
2199 "identifiers changed for nsid %d\n", ns->head->ns_id);
2204 ret = nvme_update_ns_info(ns, id);
2209 * Only fail the function if we got a fatal error back from the
2210 * device, otherwise ignore the error and just move on.
2212 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
2215 ret = blk_status_to_errno(nvme_error_status(ret));
2219 static char nvme_pr_type(enum pr_type type)
2222 case PR_WRITE_EXCLUSIVE:
2224 case PR_EXCLUSIVE_ACCESS:
2226 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2228 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2230 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2232 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2239 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2240 u64 key, u64 sa_key, u8 op)
2242 struct nvme_ns_head *head = NULL;
2244 struct nvme_command c;
2246 u8 data[16] = { 0, };
2248 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2250 return -EWOULDBLOCK;
2252 put_unaligned_le64(key, &data[0]);
2253 put_unaligned_le64(sa_key, &data[8]);
2255 memset(&c, 0, sizeof(c));
2256 c.common.opcode = op;
2257 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2258 c.common.cdw10 = cpu_to_le32(cdw10);
2260 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2261 nvme_put_ns_from_disk(head, srcu_idx);
2265 static int nvme_pr_register(struct block_device *bdev, u64 old,
2266 u64 new, unsigned flags)
2270 if (flags & ~PR_FL_IGNORE_KEY)
2273 cdw10 = old ? 2 : 0;
2274 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2275 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2276 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2279 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2280 enum pr_type type, unsigned flags)
2284 if (flags & ~PR_FL_IGNORE_KEY)
2287 cdw10 = nvme_pr_type(type) << 8;
2288 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2289 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2292 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2293 enum pr_type type, bool abort)
2295 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2296 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2299 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2301 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2302 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2305 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2307 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2308 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2311 static const struct pr_ops nvme_pr_ops = {
2312 .pr_register = nvme_pr_register,
2313 .pr_reserve = nvme_pr_reserve,
2314 .pr_release = nvme_pr_release,
2315 .pr_preempt = nvme_pr_preempt,
2316 .pr_clear = nvme_pr_clear,
2319 #ifdef CONFIG_BLK_SED_OPAL
2320 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2323 struct nvme_ctrl *ctrl = data;
2324 struct nvme_command cmd;
2326 memset(&cmd, 0, sizeof(cmd));
2328 cmd.common.opcode = nvme_admin_security_send;
2330 cmd.common.opcode = nvme_admin_security_recv;
2331 cmd.common.nsid = 0;
2332 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2333 cmd.common.cdw11 = cpu_to_le32(len);
2335 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2336 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2338 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2339 #endif /* CONFIG_BLK_SED_OPAL */
2341 static const struct block_device_operations nvme_fops = {
2342 .owner = THIS_MODULE,
2343 .ioctl = nvme_ioctl,
2344 .compat_ioctl = nvme_compat_ioctl,
2346 .release = nvme_release,
2347 .getgeo = nvme_getgeo,
2348 .report_zones = nvme_report_zones,
2349 .pr_ops = &nvme_pr_ops,
2352 #ifdef CONFIG_NVME_MULTIPATH
2353 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2355 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2357 if (!kref_get_unless_zero(&head->ref))
2362 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2364 nvme_put_ns_head(disk->private_data);
2367 const struct block_device_operations nvme_ns_head_ops = {
2368 .owner = THIS_MODULE,
2369 .submit_bio = nvme_ns_head_submit_bio,
2370 .open = nvme_ns_head_open,
2371 .release = nvme_ns_head_release,
2372 .ioctl = nvme_ioctl,
2373 .compat_ioctl = nvme_compat_ioctl,
2374 .getgeo = nvme_getgeo,
2375 .report_zones = nvme_report_zones,
2376 .pr_ops = &nvme_pr_ops,
2378 #endif /* CONFIG_NVME_MULTIPATH */
2380 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2382 unsigned long timeout =
2383 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2384 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2387 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2390 if ((csts & NVME_CSTS_RDY) == bit)
2393 usleep_range(1000, 2000);
2394 if (fatal_signal_pending(current))
2396 if (time_after(jiffies, timeout)) {
2397 dev_err(ctrl->device,
2398 "Device not ready; aborting %s, CSTS=0x%x\n",
2399 enabled ? "initialisation" : "reset", csts);
2408 * If the device has been passed off to us in an enabled state, just clear
2409 * the enabled bit. The spec says we should set the 'shutdown notification
2410 * bits', but doing so may cause the device to complete commands to the
2411 * admin queue ... and we don't know what memory that might be pointing at!
2413 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2417 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2418 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2420 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2424 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2425 msleep(NVME_QUIRK_DELAY_AMOUNT);
2427 return nvme_wait_ready(ctrl, ctrl->cap, false);
2429 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2431 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2433 unsigned dev_page_min;
2436 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2438 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2441 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2443 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2444 dev_err(ctrl->device,
2445 "Minimum device page size %u too large for host (%u)\n",
2446 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2450 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2451 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2453 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2454 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2455 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2456 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2457 ctrl->ctrl_config |= NVME_CC_ENABLE;
2459 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2462 return nvme_wait_ready(ctrl, ctrl->cap, true);
2464 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2466 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2468 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2472 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2473 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2475 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2479 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2480 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2484 if (fatal_signal_pending(current))
2486 if (time_after(jiffies, timeout)) {
2487 dev_err(ctrl->device,
2488 "Device shutdown incomplete; abort shutdown\n");
2495 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2497 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2502 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2505 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2506 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2509 dev_warn_once(ctrl->device,
2510 "could not set timestamp (%d)\n", ret);
2514 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2516 struct nvme_feat_host_behavior *host;
2519 /* Don't bother enabling the feature if retry delay is not reported */
2523 host = kzalloc(sizeof(*host), GFP_KERNEL);
2527 host->acre = NVME_ENABLE_ACRE;
2528 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2529 host, sizeof(*host), NULL);
2534 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2537 * APST (Autonomous Power State Transition) lets us program a
2538 * table of power state transitions that the controller will
2539 * perform automatically. We configure it with a simple
2540 * heuristic: we are willing to spend at most 2% of the time
2541 * transitioning between power states. Therefore, when running
2542 * in any given state, we will enter the next lower-power
2543 * non-operational state after waiting 50 * (enlat + exlat)
2544 * microseconds, as long as that state's exit latency is under
2545 * the requested maximum latency.
2547 * We will not autonomously enter any non-operational state for
2548 * which the total latency exceeds ps_max_latency_us. Users
2549 * can set ps_max_latency_us to zero to turn off APST.
2553 struct nvme_feat_auto_pst *table;
2559 * If APST isn't supported or if we haven't been initialized yet,
2560 * then don't do anything.
2565 if (ctrl->npss > 31) {
2566 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2570 table = kzalloc(sizeof(*table), GFP_KERNEL);
2574 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2575 /* Turn off APST. */
2577 dev_dbg(ctrl->device, "APST disabled\n");
2579 __le64 target = cpu_to_le64(0);
2583 * Walk through all states from lowest- to highest-power.
2584 * According to the spec, lower-numbered states use more
2585 * power. NPSS, despite the name, is the index of the
2586 * lowest-power state, not the number of states.
2588 for (state = (int)ctrl->npss; state >= 0; state--) {
2589 u64 total_latency_us, exit_latency_us, transition_ms;
2592 table->entries[state] = target;
2595 * Don't allow transitions to the deepest state
2596 * if it's quirked off.
2598 if (state == ctrl->npss &&
2599 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2603 * Is this state a useful non-operational state for
2604 * higher-power states to autonomously transition to?
2606 if (!(ctrl->psd[state].flags &
2607 NVME_PS_FLAGS_NON_OP_STATE))
2611 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2612 if (exit_latency_us > ctrl->ps_max_latency_us)
2617 le32_to_cpu(ctrl->psd[state].entry_lat);
2620 * This state is good. Use it as the APST idle
2621 * target for higher power states.
2623 transition_ms = total_latency_us + 19;
2624 do_div(transition_ms, 20);
2625 if (transition_ms > (1 << 24) - 1)
2626 transition_ms = (1 << 24) - 1;
2628 target = cpu_to_le64((state << 3) |
2629 (transition_ms << 8));
2634 if (total_latency_us > max_lat_us)
2635 max_lat_us = total_latency_us;
2641 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2643 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2644 max_ps, max_lat_us, (int)sizeof(*table), table);
2648 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2649 table, sizeof(*table), NULL);
2651 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2657 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2659 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2663 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2664 case PM_QOS_LATENCY_ANY:
2672 if (ctrl->ps_max_latency_us != latency) {
2673 ctrl->ps_max_latency_us = latency;
2674 nvme_configure_apst(ctrl);
2678 struct nvme_core_quirk_entry {
2680 * NVMe model and firmware strings are padded with spaces. For
2681 * simplicity, strings in the quirk table are padded with NULLs
2687 unsigned long quirks;
2690 static const struct nvme_core_quirk_entry core_quirks[] = {
2693 * This Toshiba device seems to die using any APST states. See:
2694 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2697 .mn = "THNSF5256GPUK TOSHIBA",
2698 .quirks = NVME_QUIRK_NO_APST,
2702 * This LiteON CL1-3D*-Q11 firmware version has a race
2703 * condition associated with actions related to suspend to idle
2704 * LiteON has resolved the problem in future firmware
2708 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2712 /* match is null-terminated but idstr is space-padded. */
2713 static bool string_matches(const char *idstr, const char *match, size_t len)
2720 matchlen = strlen(match);
2721 WARN_ON_ONCE(matchlen > len);
2723 if (memcmp(idstr, match, matchlen))
2726 for (; matchlen < len; matchlen++)
2727 if (idstr[matchlen] != ' ')
2733 static bool quirk_matches(const struct nvme_id_ctrl *id,
2734 const struct nvme_core_quirk_entry *q)
2736 return q->vid == le16_to_cpu(id->vid) &&
2737 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2738 string_matches(id->fr, q->fr, sizeof(id->fr));
2741 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2742 struct nvme_id_ctrl *id)
2747 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2748 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2749 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2750 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2754 if (ctrl->vs >= NVME_VS(1, 2, 1))
2755 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2758 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2759 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2760 "nqn.2014.08.org.nvmexpress:%04x%04x",
2761 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2762 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2763 off += sizeof(id->sn);
2764 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2765 off += sizeof(id->mn);
2766 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2769 static void nvme_release_subsystem(struct device *dev)
2771 struct nvme_subsystem *subsys =
2772 container_of(dev, struct nvme_subsystem, dev);
2774 if (subsys->instance >= 0)
2775 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2779 static void nvme_destroy_subsystem(struct kref *ref)
2781 struct nvme_subsystem *subsys =
2782 container_of(ref, struct nvme_subsystem, ref);
2784 mutex_lock(&nvme_subsystems_lock);
2785 list_del(&subsys->entry);
2786 mutex_unlock(&nvme_subsystems_lock);
2788 ida_destroy(&subsys->ns_ida);
2789 device_del(&subsys->dev);
2790 put_device(&subsys->dev);
2793 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2795 kref_put(&subsys->ref, nvme_destroy_subsystem);
2798 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2800 struct nvme_subsystem *subsys;
2802 lockdep_assert_held(&nvme_subsystems_lock);
2805 * Fail matches for discovery subsystems. This results
2806 * in each discovery controller bound to a unique subsystem.
2807 * This avoids issues with validating controller values
2808 * that can only be true when there is a single unique subsystem.
2809 * There may be multiple and completely independent entities
2810 * that provide discovery controllers.
2812 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2815 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2816 if (strcmp(subsys->subnqn, subsysnqn))
2818 if (!kref_get_unless_zero(&subsys->ref))
2826 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2827 struct device_attribute subsys_attr_##_name = \
2828 __ATTR(_name, _mode, _show, NULL)
2830 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2831 struct device_attribute *attr,
2834 struct nvme_subsystem *subsys =
2835 container_of(dev, struct nvme_subsystem, dev);
2837 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2839 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2841 #define nvme_subsys_show_str_function(field) \
2842 static ssize_t subsys_##field##_show(struct device *dev, \
2843 struct device_attribute *attr, char *buf) \
2845 struct nvme_subsystem *subsys = \
2846 container_of(dev, struct nvme_subsystem, dev); \
2847 return sprintf(buf, "%.*s\n", \
2848 (int)sizeof(subsys->field), subsys->field); \
2850 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2852 nvme_subsys_show_str_function(model);
2853 nvme_subsys_show_str_function(serial);
2854 nvme_subsys_show_str_function(firmware_rev);
2856 static struct attribute *nvme_subsys_attrs[] = {
2857 &subsys_attr_model.attr,
2858 &subsys_attr_serial.attr,
2859 &subsys_attr_firmware_rev.attr,
2860 &subsys_attr_subsysnqn.attr,
2861 #ifdef CONFIG_NVME_MULTIPATH
2862 &subsys_attr_iopolicy.attr,
2867 static struct attribute_group nvme_subsys_attrs_group = {
2868 .attrs = nvme_subsys_attrs,
2871 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2872 &nvme_subsys_attrs_group,
2876 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2877 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2879 struct nvme_ctrl *tmp;
2881 lockdep_assert_held(&nvme_subsystems_lock);
2883 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2884 if (nvme_state_terminal(tmp))
2887 if (tmp->cntlid == ctrl->cntlid) {
2888 dev_err(ctrl->device,
2889 "Duplicate cntlid %u with %s, rejecting\n",
2890 ctrl->cntlid, dev_name(tmp->device));
2894 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2895 (ctrl->opts && ctrl->opts->discovery_nqn))
2898 dev_err(ctrl->device,
2899 "Subsystem does not support multiple controllers\n");
2906 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2908 struct nvme_subsystem *subsys, *found;
2911 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2915 subsys->instance = -1;
2916 mutex_init(&subsys->lock);
2917 kref_init(&subsys->ref);
2918 INIT_LIST_HEAD(&subsys->ctrls);
2919 INIT_LIST_HEAD(&subsys->nsheads);
2920 nvme_init_subnqn(subsys, ctrl, id);
2921 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2922 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2923 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2924 subsys->vendor_id = le16_to_cpu(id->vid);
2925 subsys->cmic = id->cmic;
2926 subsys->awupf = le16_to_cpu(id->awupf);
2927 #ifdef CONFIG_NVME_MULTIPATH
2928 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2931 subsys->dev.class = nvme_subsys_class;
2932 subsys->dev.release = nvme_release_subsystem;
2933 subsys->dev.groups = nvme_subsys_attrs_groups;
2934 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2935 device_initialize(&subsys->dev);
2937 mutex_lock(&nvme_subsystems_lock);
2938 found = __nvme_find_get_subsystem(subsys->subnqn);
2940 put_device(&subsys->dev);
2943 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2945 goto out_put_subsystem;
2948 ret = device_add(&subsys->dev);
2950 dev_err(ctrl->device,
2951 "failed to register subsystem device.\n");
2952 put_device(&subsys->dev);
2955 ida_init(&subsys->ns_ida);
2956 list_add_tail(&subsys->entry, &nvme_subsystems);
2959 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2960 dev_name(ctrl->device));
2962 dev_err(ctrl->device,
2963 "failed to create sysfs link from subsystem.\n");
2964 goto out_put_subsystem;
2968 subsys->instance = ctrl->instance;
2969 ctrl->subsys = subsys;
2970 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2971 mutex_unlock(&nvme_subsystems_lock);
2975 nvme_put_subsystem(subsys);
2977 mutex_unlock(&nvme_subsystems_lock);
2981 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2982 void *log, size_t size, u64 offset)
2984 struct nvme_command c = { };
2985 u32 dwlen = nvme_bytes_to_numd(size);
2987 c.get_log_page.opcode = nvme_admin_get_log_page;
2988 c.get_log_page.nsid = cpu_to_le32(nsid);
2989 c.get_log_page.lid = log_page;
2990 c.get_log_page.lsp = lsp;
2991 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2992 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2993 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2994 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2995 c.get_log_page.csi = csi;
2997 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3000 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3001 struct nvme_effects_log **log)
3003 struct nvme_cel *cel = xa_load(&ctrl->cels, csi);
3009 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3013 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0, csi,
3014 &cel->log, sizeof(cel->log), 0);
3021 xa_store(&ctrl->cels, cel->csi, cel, GFP_KERNEL);
3028 * Initialize the cached copies of the Identify data and various controller
3029 * register in our nvme_ctrl structure. This should be called as soon as
3030 * the admin queue is fully up and running.
3032 int nvme_init_identify(struct nvme_ctrl *ctrl)
3034 struct nvme_id_ctrl *id;
3035 int ret, page_shift;
3037 bool prev_apst_enabled;
3039 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3041 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3044 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3045 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3047 if (ctrl->vs >= NVME_VS(1, 1, 0))
3048 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3050 ret = nvme_identify_ctrl(ctrl, &id);
3052 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3056 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3057 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3062 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3063 ctrl->cntlid = le16_to_cpu(id->cntlid);
3065 if (!ctrl->identified) {
3068 ret = nvme_init_subsystem(ctrl, id);
3073 * Check for quirks. Quirk can depend on firmware version,
3074 * so, in principle, the set of quirks present can change
3075 * across a reset. As a possible future enhancement, we
3076 * could re-scan for quirks every time we reinitialize
3077 * the device, but we'd have to make sure that the driver
3078 * behaves intelligently if the quirks change.
3080 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3081 if (quirk_matches(id, &core_quirks[i]))
3082 ctrl->quirks |= core_quirks[i].quirks;
3086 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3087 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3088 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3091 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3092 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3093 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3095 ctrl->oacs = le16_to_cpu(id->oacs);
3096 ctrl->oncs = le16_to_cpu(id->oncs);
3097 ctrl->mtfa = le16_to_cpu(id->mtfa);
3098 ctrl->oaes = le32_to_cpu(id->oaes);
3099 ctrl->wctemp = le16_to_cpu(id->wctemp);
3100 ctrl->cctemp = le16_to_cpu(id->cctemp);
3102 atomic_set(&ctrl->abort_limit, id->acl + 1);
3103 ctrl->vwc = id->vwc;
3105 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3107 max_hw_sectors = UINT_MAX;
3108 ctrl->max_hw_sectors =
3109 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3111 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3112 ctrl->sgls = le32_to_cpu(id->sgls);
3113 ctrl->kas = le16_to_cpu(id->kas);
3114 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3115 ctrl->ctratt = le32_to_cpu(id->ctratt);
3119 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3121 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3122 shutdown_timeout, 60);
3124 if (ctrl->shutdown_timeout != shutdown_timeout)
3125 dev_info(ctrl->device,
3126 "Shutdown timeout set to %u seconds\n",
3127 ctrl->shutdown_timeout);
3129 ctrl->shutdown_timeout = shutdown_timeout;
3131 ctrl->npss = id->npss;
3132 ctrl->apsta = id->apsta;
3133 prev_apst_enabled = ctrl->apst_enabled;
3134 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3135 if (force_apst && id->apsta) {
3136 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3137 ctrl->apst_enabled = true;
3139 ctrl->apst_enabled = false;
3142 ctrl->apst_enabled = id->apsta;
3144 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3146 if (ctrl->ops->flags & NVME_F_FABRICS) {
3147 ctrl->icdoff = le16_to_cpu(id->icdoff);
3148 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3149 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3150 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3153 * In fabrics we need to verify the cntlid matches the
3156 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3157 dev_err(ctrl->device,
3158 "Mismatching cntlid: Connect %u vs Identify "
3160 ctrl->cntlid, le16_to_cpu(id->cntlid));
3165 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
3166 dev_err(ctrl->device,
3167 "keep-alive support is mandatory for fabrics\n");
3172 ctrl->hmpre = le32_to_cpu(id->hmpre);
3173 ctrl->hmmin = le32_to_cpu(id->hmmin);
3174 ctrl->hmminds = le32_to_cpu(id->hmminds);
3175 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3178 ret = nvme_mpath_init(ctrl, id);
3184 if (ctrl->apst_enabled && !prev_apst_enabled)
3185 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3186 else if (!ctrl->apst_enabled && prev_apst_enabled)
3187 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3189 ret = nvme_configure_apst(ctrl);
3193 ret = nvme_configure_timestamp(ctrl);
3197 ret = nvme_configure_directives(ctrl);
3201 ret = nvme_configure_acre(ctrl);
3205 if (!ctrl->identified)
3206 nvme_hwmon_init(ctrl);
3208 ctrl->identified = true;
3216 EXPORT_SYMBOL_GPL(nvme_init_identify);
3218 static int nvme_dev_open(struct inode *inode, struct file *file)
3220 struct nvme_ctrl *ctrl =
3221 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3223 switch (ctrl->state) {
3224 case NVME_CTRL_LIVE:
3227 return -EWOULDBLOCK;
3230 nvme_get_ctrl(ctrl);
3231 if (!try_module_get(ctrl->ops->module))
3234 file->private_data = ctrl;
3238 static int nvme_dev_release(struct inode *inode, struct file *file)
3240 struct nvme_ctrl *ctrl =
3241 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3243 module_put(ctrl->ops->module);
3244 nvme_put_ctrl(ctrl);
3248 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3253 down_read(&ctrl->namespaces_rwsem);
3254 if (list_empty(&ctrl->namespaces)) {
3259 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3260 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3261 dev_warn(ctrl->device,
3262 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3267 dev_warn(ctrl->device,
3268 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3269 kref_get(&ns->kref);
3270 up_read(&ctrl->namespaces_rwsem);
3272 ret = nvme_user_cmd(ctrl, ns, argp);
3277 up_read(&ctrl->namespaces_rwsem);
3281 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3284 struct nvme_ctrl *ctrl = file->private_data;
3285 void __user *argp = (void __user *)arg;
3288 case NVME_IOCTL_ADMIN_CMD:
3289 return nvme_user_cmd(ctrl, NULL, argp);
3290 case NVME_IOCTL_ADMIN64_CMD:
3291 return nvme_user_cmd64(ctrl, NULL, argp);
3292 case NVME_IOCTL_IO_CMD:
3293 return nvme_dev_user_cmd(ctrl, argp);
3294 case NVME_IOCTL_RESET:
3295 dev_warn(ctrl->device, "resetting controller\n");
3296 return nvme_reset_ctrl_sync(ctrl);
3297 case NVME_IOCTL_SUBSYS_RESET:
3298 return nvme_reset_subsystem(ctrl);
3299 case NVME_IOCTL_RESCAN:
3300 nvme_queue_scan(ctrl);
3307 static const struct file_operations nvme_dev_fops = {
3308 .owner = THIS_MODULE,
3309 .open = nvme_dev_open,
3310 .release = nvme_dev_release,
3311 .unlocked_ioctl = nvme_dev_ioctl,
3312 .compat_ioctl = compat_ptr_ioctl,
3315 static ssize_t nvme_sysfs_reset(struct device *dev,
3316 struct device_attribute *attr, const char *buf,
3319 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3322 ret = nvme_reset_ctrl_sync(ctrl);
3327 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3329 static ssize_t nvme_sysfs_rescan(struct device *dev,
3330 struct device_attribute *attr, const char *buf,
3333 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3335 nvme_queue_scan(ctrl);
3338 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3340 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3342 struct gendisk *disk = dev_to_disk(dev);
3344 if (disk->fops == &nvme_fops)
3345 return nvme_get_ns_from_dev(dev)->head;
3347 return disk->private_data;
3350 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3353 struct nvme_ns_head *head = dev_to_ns_head(dev);
3354 struct nvme_ns_ids *ids = &head->ids;
3355 struct nvme_subsystem *subsys = head->subsys;
3356 int serial_len = sizeof(subsys->serial);
3357 int model_len = sizeof(subsys->model);
3359 if (!uuid_is_null(&ids->uuid))
3360 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3362 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3363 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3365 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3366 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3368 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3369 subsys->serial[serial_len - 1] == '\0'))
3371 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3372 subsys->model[model_len - 1] == '\0'))
3375 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3376 serial_len, subsys->serial, model_len, subsys->model,
3379 static DEVICE_ATTR_RO(wwid);
3381 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3384 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3386 static DEVICE_ATTR_RO(nguid);
3388 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3391 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3393 /* For backward compatibility expose the NGUID to userspace if
3394 * we have no UUID set
3396 if (uuid_is_null(&ids->uuid)) {
3397 printk_ratelimited(KERN_WARNING
3398 "No UUID available providing old NGUID\n");
3399 return sprintf(buf, "%pU\n", ids->nguid);
3401 return sprintf(buf, "%pU\n", &ids->uuid);
3403 static DEVICE_ATTR_RO(uuid);
3405 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3408 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3410 static DEVICE_ATTR_RO(eui);
3412 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3415 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3417 static DEVICE_ATTR_RO(nsid);
3419 static struct attribute *nvme_ns_id_attrs[] = {
3420 &dev_attr_wwid.attr,
3421 &dev_attr_uuid.attr,
3422 &dev_attr_nguid.attr,
3424 &dev_attr_nsid.attr,
3425 #ifdef CONFIG_NVME_MULTIPATH
3426 &dev_attr_ana_grpid.attr,
3427 &dev_attr_ana_state.attr,
3432 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3433 struct attribute *a, int n)
3435 struct device *dev = container_of(kobj, struct device, kobj);
3436 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3438 if (a == &dev_attr_uuid.attr) {
3439 if (uuid_is_null(&ids->uuid) &&
3440 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3443 if (a == &dev_attr_nguid.attr) {
3444 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3447 if (a == &dev_attr_eui.attr) {
3448 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3451 #ifdef CONFIG_NVME_MULTIPATH
3452 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3453 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3455 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3462 static const struct attribute_group nvme_ns_id_attr_group = {
3463 .attrs = nvme_ns_id_attrs,
3464 .is_visible = nvme_ns_id_attrs_are_visible,
3467 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3468 &nvme_ns_id_attr_group,
3470 &nvme_nvm_attr_group,
3475 #define nvme_show_str_function(field) \
3476 static ssize_t field##_show(struct device *dev, \
3477 struct device_attribute *attr, char *buf) \
3479 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3480 return sprintf(buf, "%.*s\n", \
3481 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3483 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3485 nvme_show_str_function(model);
3486 nvme_show_str_function(serial);
3487 nvme_show_str_function(firmware_rev);
3489 #define nvme_show_int_function(field) \
3490 static ssize_t field##_show(struct device *dev, \
3491 struct device_attribute *attr, char *buf) \
3493 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3494 return sprintf(buf, "%d\n", ctrl->field); \
3496 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3498 nvme_show_int_function(cntlid);
3499 nvme_show_int_function(numa_node);
3500 nvme_show_int_function(queue_count);
3501 nvme_show_int_function(sqsize);
3503 static ssize_t nvme_sysfs_delete(struct device *dev,
3504 struct device_attribute *attr, const char *buf,
3507 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3509 if (device_remove_file_self(dev, attr))
3510 nvme_delete_ctrl_sync(ctrl);
3513 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3515 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3516 struct device_attribute *attr,
3519 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3521 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3523 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3525 static ssize_t nvme_sysfs_show_state(struct device *dev,
3526 struct device_attribute *attr,
3529 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3530 static const char *const state_name[] = {
3531 [NVME_CTRL_NEW] = "new",
3532 [NVME_CTRL_LIVE] = "live",
3533 [NVME_CTRL_RESETTING] = "resetting",
3534 [NVME_CTRL_CONNECTING] = "connecting",
3535 [NVME_CTRL_DELETING] = "deleting",
3536 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3537 [NVME_CTRL_DEAD] = "dead",
3540 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3541 state_name[ctrl->state])
3542 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3544 return sprintf(buf, "unknown state\n");
3547 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3549 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3550 struct device_attribute *attr,
3553 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3555 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3557 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3559 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3560 struct device_attribute *attr,
3563 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3565 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3567 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3569 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3570 struct device_attribute *attr,
3573 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3575 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3577 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3579 static ssize_t nvme_sysfs_show_address(struct device *dev,
3580 struct device_attribute *attr,
3583 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3585 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3587 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3589 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3590 struct device_attribute *attr, char *buf)
3592 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3593 struct nvmf_ctrl_options *opts = ctrl->opts;
3595 if (ctrl->opts->max_reconnects == -1)
3596 return sprintf(buf, "off\n");
3597 return sprintf(buf, "%d\n",
3598 opts->max_reconnects * opts->reconnect_delay);
3601 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3602 struct device_attribute *attr, const char *buf, size_t count)
3604 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3605 struct nvmf_ctrl_options *opts = ctrl->opts;
3606 int ctrl_loss_tmo, err;
3608 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3612 else if (ctrl_loss_tmo < 0)
3613 opts->max_reconnects = -1;
3615 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3616 opts->reconnect_delay);
3619 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3620 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3622 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3623 struct device_attribute *attr, char *buf)
3625 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3627 if (ctrl->opts->reconnect_delay == -1)
3628 return sprintf(buf, "off\n");
3629 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3632 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3633 struct device_attribute *attr, const char *buf, size_t count)
3635 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3639 err = kstrtou32(buf, 10, &v);
3643 ctrl->opts->reconnect_delay = v;
3646 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3647 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3649 static struct attribute *nvme_dev_attrs[] = {
3650 &dev_attr_reset_controller.attr,
3651 &dev_attr_rescan_controller.attr,
3652 &dev_attr_model.attr,
3653 &dev_attr_serial.attr,
3654 &dev_attr_firmware_rev.attr,
3655 &dev_attr_cntlid.attr,
3656 &dev_attr_delete_controller.attr,
3657 &dev_attr_transport.attr,
3658 &dev_attr_subsysnqn.attr,
3659 &dev_attr_address.attr,
3660 &dev_attr_state.attr,
3661 &dev_attr_numa_node.attr,
3662 &dev_attr_queue_count.attr,
3663 &dev_attr_sqsize.attr,
3664 &dev_attr_hostnqn.attr,
3665 &dev_attr_hostid.attr,
3666 &dev_attr_ctrl_loss_tmo.attr,
3667 &dev_attr_reconnect_delay.attr,
3671 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3672 struct attribute *a, int n)
3674 struct device *dev = container_of(kobj, struct device, kobj);
3675 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3677 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3679 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3681 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3683 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3685 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3687 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3693 static struct attribute_group nvme_dev_attrs_group = {
3694 .attrs = nvme_dev_attrs,
3695 .is_visible = nvme_dev_attrs_are_visible,
3698 static const struct attribute_group *nvme_dev_attr_groups[] = {
3699 &nvme_dev_attrs_group,
3703 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3706 struct nvme_ns_head *h;
3708 lockdep_assert_held(&subsys->lock);
3710 list_for_each_entry(h, &subsys->nsheads, entry) {
3711 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3718 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3719 struct nvme_ns_head *new)
3721 struct nvme_ns_head *h;
3723 lockdep_assert_held(&subsys->lock);
3725 list_for_each_entry(h, &subsys->nsheads, entry) {
3726 if (nvme_ns_ids_valid(&new->ids) &&
3727 nvme_ns_ids_equal(&new->ids, &h->ids))
3734 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3735 unsigned nsid, struct nvme_ns_ids *ids)
3737 struct nvme_ns_head *head;
3738 size_t size = sizeof(*head);
3741 #ifdef CONFIG_NVME_MULTIPATH
3742 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3745 head = kzalloc(size, GFP_KERNEL);
3748 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3751 head->instance = ret;
3752 INIT_LIST_HEAD(&head->list);
3753 ret = init_srcu_struct(&head->srcu);
3755 goto out_ida_remove;
3756 head->subsys = ctrl->subsys;
3759 kref_init(&head->ref);
3761 ret = __nvme_check_ids(ctrl->subsys, head);
3763 dev_err(ctrl->device,
3764 "duplicate IDs for nsid %d\n", nsid);
3765 goto out_cleanup_srcu;
3768 if (head->ids.csi) {
3769 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3771 goto out_cleanup_srcu;
3773 head->effects = ctrl->effects;
3775 ret = nvme_mpath_alloc_disk(ctrl, head);
3777 goto out_cleanup_srcu;
3779 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3781 kref_get(&ctrl->subsys->ref);
3785 cleanup_srcu_struct(&head->srcu);
3787 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3792 ret = blk_status_to_errno(nvme_error_status(ret));
3793 return ERR_PTR(ret);
3796 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3797 struct nvme_id_ns *id)
3799 struct nvme_ctrl *ctrl = ns->ctrl;
3800 bool is_shared = id->nmic & NVME_NS_NMIC_SHARED;
3801 struct nvme_ns_head *head = NULL;
3802 struct nvme_ns_ids ids;
3805 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3809 return blk_status_to_errno(nvme_error_status(ret));
3812 mutex_lock(&ctrl->subsys->lock);
3813 head = nvme_find_ns_head(ctrl->subsys, nsid);
3815 head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3817 ret = PTR_ERR(head);
3820 head->shared = is_shared;
3823 if (!is_shared || !head->shared) {
3824 dev_err(ctrl->device,
3825 "Duplicate unshared namespace %d\n", nsid);
3826 goto out_put_ns_head;
3828 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3829 dev_err(ctrl->device,
3830 "IDs don't match for shared namespace %d\n",
3832 goto out_put_ns_head;
3836 list_add_tail(&ns->siblings, &head->list);
3838 mutex_unlock(&ctrl->subsys->lock);
3842 nvme_put_ns_head(head);
3844 mutex_unlock(&ctrl->subsys->lock);
3848 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3850 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3851 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3853 return nsa->head->ns_id - nsb->head->ns_id;
3856 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3858 struct nvme_ns *ns, *ret = NULL;
3860 down_read(&ctrl->namespaces_rwsem);
3861 list_for_each_entry(ns, &ctrl->namespaces, list) {
3862 if (ns->head->ns_id == nsid) {
3863 if (!kref_get_unless_zero(&ns->kref))
3868 if (ns->head->ns_id > nsid)
3871 up_read(&ctrl->namespaces_rwsem);
3874 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3876 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3879 struct gendisk *disk;
3880 struct nvme_id_ns *id;
3881 char disk_name[DISK_NAME_LEN];
3882 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3884 if (nvme_identify_ns(ctrl, nsid, &id))
3887 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3891 ns->queue = blk_mq_init_queue(ctrl->tagset);
3892 if (IS_ERR(ns->queue))
3895 if (ctrl->opts && ctrl->opts->data_digest)
3896 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3898 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3899 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3900 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3902 ns->queue->queuedata = ns;
3904 kref_init(&ns->kref);
3906 ret = nvme_init_ns_head(ns, nsid, id);
3908 goto out_free_queue;
3909 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3911 disk = alloc_disk_node(0, node);
3915 disk->fops = &nvme_fops;
3916 disk->private_data = ns;
3917 disk->queue = ns->queue;
3918 disk->flags = flags;
3919 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3922 if (nvme_update_ns_info(ns, id))
3925 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3926 ret = nvme_nvm_register(ns, disk_name, node);
3928 dev_warn(ctrl->device, "LightNVM init failure\n");
3933 down_write(&ctrl->namespaces_rwsem);
3934 list_add_tail(&ns->list, &ctrl->namespaces);
3935 up_write(&ctrl->namespaces_rwsem);
3937 nvme_get_ctrl(ctrl);
3939 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3941 nvme_mpath_add_disk(ns, id);
3942 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3947 /* prevent double queue cleanup */
3948 ns->disk->queue = NULL;
3951 mutex_lock(&ctrl->subsys->lock);
3952 list_del_rcu(&ns->siblings);
3953 if (list_empty(&ns->head->list))
3954 list_del_init(&ns->head->entry);
3955 mutex_unlock(&ctrl->subsys->lock);
3956 nvme_put_ns_head(ns->head);
3958 blk_cleanup_queue(ns->queue);
3965 static void nvme_ns_remove(struct nvme_ns *ns)
3967 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3970 nvme_fault_inject_fini(&ns->fault_inject);
3972 mutex_lock(&ns->ctrl->subsys->lock);
3973 list_del_rcu(&ns->siblings);
3974 if (list_empty(&ns->head->list))
3975 list_del_init(&ns->head->entry);
3976 mutex_unlock(&ns->ctrl->subsys->lock);
3978 synchronize_rcu(); /* guarantee not available in head->list */
3979 nvme_mpath_clear_current_path(ns);
3980 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3982 if (ns->disk->flags & GENHD_FL_UP) {
3983 del_gendisk(ns->disk);
3984 blk_cleanup_queue(ns->queue);
3985 if (blk_get_integrity(ns->disk))
3986 blk_integrity_unregister(ns->disk);
3989 down_write(&ns->ctrl->namespaces_rwsem);
3990 list_del_init(&ns->list);
3991 up_write(&ns->ctrl->namespaces_rwsem);
3993 nvme_mpath_check_last_path(ns);
3997 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3999 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4007 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4012 ns = nvme_find_get_ns(ctrl, nsid);
4014 nvme_alloc_ns(ctrl, nsid);
4018 ret = nvme_validate_ns(ns);
4019 revalidate_disk_size(ns->disk, ret == 0);
4025 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4028 struct nvme_ns *ns, *next;
4031 down_write(&ctrl->namespaces_rwsem);
4032 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4033 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4034 list_move_tail(&ns->list, &rm_list);
4036 up_write(&ctrl->namespaces_rwsem);
4038 list_for_each_entry_safe(ns, next, &rm_list, list)
4043 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4045 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4050 if (nvme_ctrl_limited_cns(ctrl))
4053 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4058 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
4062 for (i = 0; i < nr_entries; i++) {
4063 u32 nsid = le32_to_cpu(ns_list[i]);
4065 if (!nsid) /* end of the list? */
4067 nvme_validate_or_alloc_ns(ctrl, nsid);
4068 while (++prev < nsid)
4069 nvme_ns_remove_by_nsid(ctrl, prev);
4073 nvme_remove_invalid_namespaces(ctrl, prev);
4079 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4081 struct nvme_id_ctrl *id;
4084 if (nvme_identify_ctrl(ctrl, &id))
4086 nn = le32_to_cpu(id->nn);
4089 for (i = 1; i <= nn; i++)
4090 nvme_validate_or_alloc_ns(ctrl, i);
4092 nvme_remove_invalid_namespaces(ctrl, nn);
4095 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4097 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4101 log = kzalloc(log_size, GFP_KERNEL);
4106 * We need to read the log to clear the AEN, but we don't want to rely
4107 * on it for the changed namespace information as userspace could have
4108 * raced with us in reading the log page, which could cause us to miss
4111 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4112 NVME_CSI_NVM, log, log_size, 0);
4114 dev_warn(ctrl->device,
4115 "reading changed ns log failed: %d\n", error);
4120 static void nvme_scan_work(struct work_struct *work)
4122 struct nvme_ctrl *ctrl =
4123 container_of(work, struct nvme_ctrl, scan_work);
4125 /* No tagset on a live ctrl means IO queues could not created */
4126 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4129 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4130 dev_info(ctrl->device, "rescanning namespaces.\n");
4131 nvme_clear_changed_ns_log(ctrl);
4134 mutex_lock(&ctrl->scan_lock);
4135 if (nvme_scan_ns_list(ctrl) != 0)
4136 nvme_scan_ns_sequential(ctrl);
4137 mutex_unlock(&ctrl->scan_lock);
4139 down_write(&ctrl->namespaces_rwsem);
4140 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4141 up_write(&ctrl->namespaces_rwsem);
4145 * This function iterates the namespace list unlocked to allow recovery from
4146 * controller failure. It is up to the caller to ensure the namespace list is
4147 * not modified by scan work while this function is executing.
4149 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4151 struct nvme_ns *ns, *next;
4155 * make sure to requeue I/O to all namespaces as these
4156 * might result from the scan itself and must complete
4157 * for the scan_work to make progress
4159 nvme_mpath_clear_ctrl_paths(ctrl);
4161 /* prevent racing with ns scanning */
4162 flush_work(&ctrl->scan_work);
4165 * The dead states indicates the controller was not gracefully
4166 * disconnected. In that case, we won't be able to flush any data while
4167 * removing the namespaces' disks; fail all the queues now to avoid
4168 * potentially having to clean up the failed sync later.
4170 if (ctrl->state == NVME_CTRL_DEAD)
4171 nvme_kill_queues(ctrl);
4173 /* this is a no-op when called from the controller reset handler */
4174 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4176 down_write(&ctrl->namespaces_rwsem);
4177 list_splice_init(&ctrl->namespaces, &ns_list);
4178 up_write(&ctrl->namespaces_rwsem);
4180 list_for_each_entry_safe(ns, next, &ns_list, list)
4183 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4185 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4187 struct nvme_ctrl *ctrl =
4188 container_of(dev, struct nvme_ctrl, ctrl_device);
4189 struct nvmf_ctrl_options *opts = ctrl->opts;
4192 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4197 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4201 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4202 opts->trsvcid ?: "none");
4206 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4207 opts->host_traddr ?: "none");
4212 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4214 char *envp[2] = { NULL, NULL };
4215 u32 aen_result = ctrl->aen_result;
4217 ctrl->aen_result = 0;
4221 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4224 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4228 static void nvme_async_event_work(struct work_struct *work)
4230 struct nvme_ctrl *ctrl =
4231 container_of(work, struct nvme_ctrl, async_event_work);
4233 nvme_aen_uevent(ctrl);
4234 ctrl->ops->submit_async_event(ctrl);
4237 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4242 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4248 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4251 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4253 struct nvme_fw_slot_info_log *log;
4255 log = kmalloc(sizeof(*log), GFP_KERNEL);
4259 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4260 log, sizeof(*log), 0))
4261 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4265 static void nvme_fw_act_work(struct work_struct *work)
4267 struct nvme_ctrl *ctrl = container_of(work,
4268 struct nvme_ctrl, fw_act_work);
4269 unsigned long fw_act_timeout;
4272 fw_act_timeout = jiffies +
4273 msecs_to_jiffies(ctrl->mtfa * 100);
4275 fw_act_timeout = jiffies +
4276 msecs_to_jiffies(admin_timeout * 1000);
4278 nvme_stop_queues(ctrl);
4279 while (nvme_ctrl_pp_status(ctrl)) {
4280 if (time_after(jiffies, fw_act_timeout)) {
4281 dev_warn(ctrl->device,
4282 "Fw activation timeout, reset controller\n");
4283 nvme_try_sched_reset(ctrl);
4289 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4292 nvme_start_queues(ctrl);
4293 /* read FW slot information to clear the AER */
4294 nvme_get_fw_slot_info(ctrl);
4297 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4299 u32 aer_notice_type = (result & 0xff00) >> 8;
4301 trace_nvme_async_event(ctrl, aer_notice_type);
4303 switch (aer_notice_type) {
4304 case NVME_AER_NOTICE_NS_CHANGED:
4305 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4306 nvme_queue_scan(ctrl);
4308 case NVME_AER_NOTICE_FW_ACT_STARTING:
4310 * We are (ab)using the RESETTING state to prevent subsequent
4311 * recovery actions from interfering with the controller's
4312 * firmware activation.
4314 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4315 queue_work(nvme_wq, &ctrl->fw_act_work);
4317 #ifdef CONFIG_NVME_MULTIPATH
4318 case NVME_AER_NOTICE_ANA:
4319 if (!ctrl->ana_log_buf)
4321 queue_work(nvme_wq, &ctrl->ana_work);
4324 case NVME_AER_NOTICE_DISC_CHANGED:
4325 ctrl->aen_result = result;
4328 dev_warn(ctrl->device, "async event result %08x\n", result);
4332 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4333 volatile union nvme_result *res)
4335 u32 result = le32_to_cpu(res->u32);
4336 u32 aer_type = result & 0x07;
4338 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4342 case NVME_AER_NOTICE:
4343 nvme_handle_aen_notice(ctrl, result);
4345 case NVME_AER_ERROR:
4346 case NVME_AER_SMART:
4349 trace_nvme_async_event(ctrl, aer_type);
4350 ctrl->aen_result = result;
4355 queue_work(nvme_wq, &ctrl->async_event_work);
4357 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4359 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4361 nvme_mpath_stop(ctrl);
4362 nvme_stop_keep_alive(ctrl);
4363 flush_work(&ctrl->async_event_work);
4364 cancel_work_sync(&ctrl->fw_act_work);
4366 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4368 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4370 nvme_start_keep_alive(ctrl);
4372 nvme_enable_aen(ctrl);
4374 if (ctrl->queue_count > 1) {
4375 nvme_queue_scan(ctrl);
4376 nvme_start_queues(ctrl);
4379 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4381 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4383 nvme_fault_inject_fini(&ctrl->fault_inject);
4384 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4385 cdev_device_del(&ctrl->cdev, ctrl->device);
4386 nvme_put_ctrl(ctrl);
4388 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4390 static void nvme_free_ctrl(struct device *dev)
4392 struct nvme_ctrl *ctrl =
4393 container_of(dev, struct nvme_ctrl, ctrl_device);
4394 struct nvme_subsystem *subsys = ctrl->subsys;
4396 if (!subsys || ctrl->instance != subsys->instance)
4397 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4399 xa_destroy(&ctrl->cels);
4401 nvme_mpath_uninit(ctrl);
4402 __free_page(ctrl->discard_page);
4405 mutex_lock(&nvme_subsystems_lock);
4406 list_del(&ctrl->subsys_entry);
4407 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4408 mutex_unlock(&nvme_subsystems_lock);
4411 ctrl->ops->free_ctrl(ctrl);
4414 nvme_put_subsystem(subsys);
4418 * Initialize a NVMe controller structures. This needs to be called during
4419 * earliest initialization so that we have the initialized structured around
4422 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4423 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4427 ctrl->state = NVME_CTRL_NEW;
4428 spin_lock_init(&ctrl->lock);
4429 mutex_init(&ctrl->scan_lock);
4430 INIT_LIST_HEAD(&ctrl->namespaces);
4431 xa_init(&ctrl->cels);
4432 init_rwsem(&ctrl->namespaces_rwsem);
4435 ctrl->quirks = quirks;
4436 ctrl->numa_node = NUMA_NO_NODE;
4437 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4438 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4439 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4440 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4441 init_waitqueue_head(&ctrl->state_wq);
4443 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4444 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4445 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4447 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4449 ctrl->discard_page = alloc_page(GFP_KERNEL);
4450 if (!ctrl->discard_page) {
4455 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4458 ctrl->instance = ret;
4460 device_initialize(&ctrl->ctrl_device);
4461 ctrl->device = &ctrl->ctrl_device;
4462 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4463 ctrl->device->class = nvme_class;
4464 ctrl->device->parent = ctrl->dev;
4465 ctrl->device->groups = nvme_dev_attr_groups;
4466 ctrl->device->release = nvme_free_ctrl;
4467 dev_set_drvdata(ctrl->device, ctrl);
4468 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4470 goto out_release_instance;
4472 nvme_get_ctrl(ctrl);
4473 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4474 ctrl->cdev.owner = ops->module;
4475 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4480 * Initialize latency tolerance controls. The sysfs files won't
4481 * be visible to userspace unless the device actually supports APST.
4483 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4484 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4485 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4487 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4491 nvme_put_ctrl(ctrl);
4492 kfree_const(ctrl->device->kobj.name);
4493 out_release_instance:
4494 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4496 if (ctrl->discard_page)
4497 __free_page(ctrl->discard_page);
4500 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4503 * nvme_kill_queues(): Ends all namespace queues
4504 * @ctrl: the dead controller that needs to end
4506 * Call this function when the driver determines it is unable to get the
4507 * controller in a state capable of servicing IO.
4509 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4513 down_read(&ctrl->namespaces_rwsem);
4515 /* Forcibly unquiesce queues to avoid blocking dispatch */
4516 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4517 blk_mq_unquiesce_queue(ctrl->admin_q);
4519 list_for_each_entry(ns, &ctrl->namespaces, list)
4520 nvme_set_queue_dying(ns);
4522 up_read(&ctrl->namespaces_rwsem);
4524 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4526 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4530 down_read(&ctrl->namespaces_rwsem);
4531 list_for_each_entry(ns, &ctrl->namespaces, list)
4532 blk_mq_unfreeze_queue(ns->queue);
4533 up_read(&ctrl->namespaces_rwsem);
4535 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4537 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4541 down_read(&ctrl->namespaces_rwsem);
4542 list_for_each_entry(ns, &ctrl->namespaces, list) {
4543 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4547 up_read(&ctrl->namespaces_rwsem);
4550 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4552 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4556 down_read(&ctrl->namespaces_rwsem);
4557 list_for_each_entry(ns, &ctrl->namespaces, list)
4558 blk_mq_freeze_queue_wait(ns->queue);
4559 up_read(&ctrl->namespaces_rwsem);
4561 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4563 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4567 down_read(&ctrl->namespaces_rwsem);
4568 list_for_each_entry(ns, &ctrl->namespaces, list)
4569 blk_freeze_queue_start(ns->queue);
4570 up_read(&ctrl->namespaces_rwsem);
4572 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4574 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4578 down_read(&ctrl->namespaces_rwsem);
4579 list_for_each_entry(ns, &ctrl->namespaces, list)
4580 blk_mq_quiesce_queue(ns->queue);
4581 up_read(&ctrl->namespaces_rwsem);
4583 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4585 void nvme_start_queues(struct nvme_ctrl *ctrl)
4589 down_read(&ctrl->namespaces_rwsem);
4590 list_for_each_entry(ns, &ctrl->namespaces, list)
4591 blk_mq_unquiesce_queue(ns->queue);
4592 up_read(&ctrl->namespaces_rwsem);
4594 EXPORT_SYMBOL_GPL(nvme_start_queues);
4597 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4601 down_read(&ctrl->namespaces_rwsem);
4602 list_for_each_entry(ns, &ctrl->namespaces, list)
4603 blk_sync_queue(ns->queue);
4604 up_read(&ctrl->namespaces_rwsem);
4607 blk_sync_queue(ctrl->admin_q);
4609 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4611 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4613 if (file->f_op != &nvme_dev_fops)
4615 return file->private_data;
4617 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4620 * Check we didn't inadvertently grow the command structure sizes:
4622 static inline void _nvme_check_size(void)
4624 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4625 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4626 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4627 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4628 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4629 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4630 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4631 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4632 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4633 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4634 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4635 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4636 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4637 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4638 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4639 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4640 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4641 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4642 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4646 static int __init nvme_core_init(void)
4648 int result = -ENOMEM;
4652 nvme_wq = alloc_workqueue("nvme-wq",
4653 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4657 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4658 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4662 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4663 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4664 if (!nvme_delete_wq)
4665 goto destroy_reset_wq;
4667 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4669 goto destroy_delete_wq;
4671 nvme_class = class_create(THIS_MODULE, "nvme");
4672 if (IS_ERR(nvme_class)) {
4673 result = PTR_ERR(nvme_class);
4674 goto unregister_chrdev;
4676 nvme_class->dev_uevent = nvme_class_uevent;
4678 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4679 if (IS_ERR(nvme_subsys_class)) {
4680 result = PTR_ERR(nvme_subsys_class);
4686 class_destroy(nvme_class);
4688 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4690 destroy_workqueue(nvme_delete_wq);
4692 destroy_workqueue(nvme_reset_wq);
4694 destroy_workqueue(nvme_wq);
4699 static void __exit nvme_core_exit(void)
4701 class_destroy(nvme_subsys_class);
4702 class_destroy(nvme_class);
4703 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4704 destroy_workqueue(nvme_delete_wq);
4705 destroy_workqueue(nvme_reset_wq);
4706 destroy_workqueue(nvme_wq);
4707 ida_destroy(&nvme_instance_ida);
4710 MODULE_LICENSE("GPL");
4711 MODULE_VERSION("1.0");
4712 module_init(nvme_core_init);
4713 module_exit(nvme_core_exit);