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_ctrl_base_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,
97 * Prepare a queue for teardown.
99 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
100 * the capacity to 0 after that to avoid blocking dispatchers that may be
101 * holding bd_butex. This will end buffered writers dirtying pages that can't
104 static void nvme_set_queue_dying(struct nvme_ns *ns)
106 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
109 blk_set_queue_dying(ns->queue);
110 blk_mq_unquiesce_queue(ns->queue);
112 set_capacity_and_notify(ns->disk, 0);
115 void nvme_queue_scan(struct nvme_ctrl *ctrl)
118 * Only new queue scan work when admin and IO queues are both alive
120 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
121 queue_work(nvme_wq, &ctrl->scan_work);
125 * Use this function to proceed with scheduling reset_work for a controller
126 * that had previously been set to the resetting state. This is intended for
127 * code paths that can't be interrupted by other reset attempts. A hot removal
128 * may prevent this from succeeding.
130 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
132 if (ctrl->state != NVME_CTRL_RESETTING)
134 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
138 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
140 static void nvme_failfast_work(struct work_struct *work)
142 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
143 struct nvme_ctrl, failfast_work);
145 if (ctrl->state != NVME_CTRL_CONNECTING)
148 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
149 dev_info(ctrl->device, "failfast expired\n");
150 nvme_kick_requeue_lists(ctrl);
153 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
155 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
158 schedule_delayed_work(&ctrl->failfast_work,
159 ctrl->opts->fast_io_fail_tmo * HZ);
162 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
167 cancel_delayed_work_sync(&ctrl->failfast_work);
168 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
172 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
174 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
176 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
180 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
182 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
186 ret = nvme_reset_ctrl(ctrl);
188 flush_work(&ctrl->reset_work);
189 if (ctrl->state != NVME_CTRL_LIVE)
196 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
198 dev_info(ctrl->device,
199 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
201 flush_work(&ctrl->reset_work);
202 nvme_stop_ctrl(ctrl);
203 nvme_remove_namespaces(ctrl);
204 ctrl->ops->delete_ctrl(ctrl);
205 nvme_uninit_ctrl(ctrl);
208 static void nvme_delete_ctrl_work(struct work_struct *work)
210 struct nvme_ctrl *ctrl =
211 container_of(work, struct nvme_ctrl, delete_work);
213 nvme_do_delete_ctrl(ctrl);
216 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
218 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
220 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
224 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
226 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
229 * Keep a reference until nvme_do_delete_ctrl() complete,
230 * since ->delete_ctrl can free the controller.
233 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
234 nvme_do_delete_ctrl(ctrl);
238 static blk_status_t nvme_error_status(u16 status)
240 switch (status & 0x7ff) {
241 case NVME_SC_SUCCESS:
243 case NVME_SC_CAP_EXCEEDED:
244 return BLK_STS_NOSPC;
245 case NVME_SC_LBA_RANGE:
246 case NVME_SC_CMD_INTERRUPTED:
247 case NVME_SC_NS_NOT_READY:
248 return BLK_STS_TARGET;
249 case NVME_SC_BAD_ATTRIBUTES:
250 case NVME_SC_ONCS_NOT_SUPPORTED:
251 case NVME_SC_INVALID_OPCODE:
252 case NVME_SC_INVALID_FIELD:
253 case NVME_SC_INVALID_NS:
254 return BLK_STS_NOTSUPP;
255 case NVME_SC_WRITE_FAULT:
256 case NVME_SC_READ_ERROR:
257 case NVME_SC_UNWRITTEN_BLOCK:
258 case NVME_SC_ACCESS_DENIED:
259 case NVME_SC_READ_ONLY:
260 case NVME_SC_COMPARE_FAILED:
261 return BLK_STS_MEDIUM;
262 case NVME_SC_GUARD_CHECK:
263 case NVME_SC_APPTAG_CHECK:
264 case NVME_SC_REFTAG_CHECK:
265 case NVME_SC_INVALID_PI:
266 return BLK_STS_PROTECTION;
267 case NVME_SC_RESERVATION_CONFLICT:
268 return BLK_STS_NEXUS;
269 case NVME_SC_HOST_PATH_ERROR:
270 return BLK_STS_TRANSPORT;
271 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
272 return BLK_STS_ZONE_ACTIVE_RESOURCE;
273 case NVME_SC_ZONE_TOO_MANY_OPEN:
274 return BLK_STS_ZONE_OPEN_RESOURCE;
276 return BLK_STS_IOERR;
280 static void nvme_retry_req(struct request *req)
282 unsigned long delay = 0;
285 /* The mask and shift result must be <= 3 */
286 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
288 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
290 nvme_req(req)->retries++;
291 blk_mq_requeue_request(req, false);
292 blk_mq_delay_kick_requeue_list(req->q, delay);
295 enum nvme_disposition {
301 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
303 if (likely(nvme_req(req)->status == 0))
306 if (blk_noretry_request(req) ||
307 (nvme_req(req)->status & NVME_SC_DNR) ||
308 nvme_req(req)->retries >= nvme_max_retries)
311 if (req->cmd_flags & REQ_NVME_MPATH) {
312 if (nvme_is_path_error(nvme_req(req)->status) ||
313 blk_queue_dying(req->q))
316 if (blk_queue_dying(req->q))
323 static inline void nvme_end_req(struct request *req)
325 blk_status_t status = nvme_error_status(nvme_req(req)->status);
327 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
328 req_op(req) == REQ_OP_ZONE_APPEND)
329 req->__sector = nvme_lba_to_sect(req->q->queuedata,
330 le64_to_cpu(nvme_req(req)->result.u64));
332 nvme_trace_bio_complete(req);
333 blk_mq_end_request(req, status);
336 void nvme_complete_rq(struct request *req)
338 trace_nvme_complete_rq(req);
339 nvme_cleanup_cmd(req);
341 if (nvme_req(req)->ctrl->kas)
342 nvme_req(req)->ctrl->comp_seen = true;
344 switch (nvme_decide_disposition(req)) {
352 nvme_failover_req(req);
356 EXPORT_SYMBOL_GPL(nvme_complete_rq);
359 * Called to unwind from ->queue_rq on a failed command submission so that the
360 * multipathing code gets called to potentially failover to another path.
361 * The caller needs to unwind all transport specific resource allocations and
362 * must return propagate the return value.
364 blk_status_t nvme_host_path_error(struct request *req)
366 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
367 blk_mq_set_request_complete(req);
368 nvme_complete_rq(req);
371 EXPORT_SYMBOL_GPL(nvme_host_path_error);
373 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
375 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
376 "Cancelling I/O %d", req->tag);
378 /* don't abort one completed request */
379 if (blk_mq_request_completed(req))
382 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
383 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
384 blk_mq_complete_request(req);
387 EXPORT_SYMBOL_GPL(nvme_cancel_request);
389 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
392 blk_mq_tagset_busy_iter(ctrl->tagset,
393 nvme_cancel_request, ctrl);
394 blk_mq_tagset_wait_completed_request(ctrl->tagset);
397 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
399 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
401 if (ctrl->admin_tagset) {
402 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
403 nvme_cancel_request, ctrl);
404 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
407 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
409 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
410 enum nvme_ctrl_state new_state)
412 enum nvme_ctrl_state old_state;
414 bool changed = false;
416 spin_lock_irqsave(&ctrl->lock, flags);
418 old_state = ctrl->state;
423 case NVME_CTRL_RESETTING:
424 case NVME_CTRL_CONNECTING:
431 case NVME_CTRL_RESETTING:
441 case NVME_CTRL_CONNECTING:
444 case NVME_CTRL_RESETTING:
451 case NVME_CTRL_DELETING:
454 case NVME_CTRL_RESETTING:
455 case NVME_CTRL_CONNECTING:
462 case NVME_CTRL_DELETING_NOIO:
464 case NVME_CTRL_DELETING:
474 case NVME_CTRL_DELETING:
486 ctrl->state = new_state;
487 wake_up_all(&ctrl->state_wq);
490 spin_unlock_irqrestore(&ctrl->lock, flags);
494 if (ctrl->state == NVME_CTRL_LIVE) {
495 if (old_state == NVME_CTRL_CONNECTING)
496 nvme_stop_failfast_work(ctrl);
497 nvme_kick_requeue_lists(ctrl);
498 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
499 old_state == NVME_CTRL_RESETTING) {
500 nvme_start_failfast_work(ctrl);
504 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
507 * Returns true for sink states that can't ever transition back to live.
509 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
511 switch (ctrl->state) {
514 case NVME_CTRL_RESETTING:
515 case NVME_CTRL_CONNECTING:
517 case NVME_CTRL_DELETING:
518 case NVME_CTRL_DELETING_NOIO:
522 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
528 * Waits for the controller state to be resetting, or returns false if it is
529 * not possible to ever transition to that state.
531 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
533 wait_event(ctrl->state_wq,
534 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
535 nvme_state_terminal(ctrl));
536 return ctrl->state == NVME_CTRL_RESETTING;
538 EXPORT_SYMBOL_GPL(nvme_wait_reset);
540 static void nvme_free_ns_head(struct kref *ref)
542 struct nvme_ns_head *head =
543 container_of(ref, struct nvme_ns_head, ref);
545 nvme_mpath_remove_disk(head);
546 ida_simple_remove(&head->subsys->ns_ida, head->instance);
547 cleanup_srcu_struct(&head->srcu);
548 nvme_put_subsystem(head->subsys);
552 static void nvme_put_ns_head(struct nvme_ns_head *head)
554 kref_put(&head->ref, nvme_free_ns_head);
557 static void nvme_free_ns(struct kref *kref)
559 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
562 nvme_nvm_unregister(ns);
565 nvme_put_ns_head(ns->head);
566 nvme_put_ctrl(ns->ctrl);
570 void nvme_put_ns(struct nvme_ns *ns)
572 kref_put(&ns->kref, nvme_free_ns);
574 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
576 static inline void nvme_clear_nvme_request(struct request *req)
578 struct nvme_command *cmd = nvme_req(req)->cmd;
580 memset(cmd, 0, sizeof(*cmd));
581 nvme_req(req)->retries = 0;
582 nvme_req(req)->flags = 0;
583 req->rq_flags |= RQF_DONTPREP;
586 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
588 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
591 static inline void nvme_init_request(struct request *req,
592 struct nvme_command *cmd)
594 if (req->q->queuedata)
595 req->timeout = NVME_IO_TIMEOUT;
596 else /* no queuedata implies admin queue */
597 req->timeout = NVME_ADMIN_TIMEOUT;
599 /* passthru commands should let the driver set the SGL flags */
600 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
602 req->cmd_flags |= REQ_FAILFAST_DRIVER;
603 nvme_clear_nvme_request(req);
604 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
607 struct request *nvme_alloc_request(struct request_queue *q,
608 struct nvme_command *cmd, blk_mq_req_flags_t flags)
612 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
614 nvme_init_request(req, cmd);
617 EXPORT_SYMBOL_GPL(nvme_alloc_request);
619 static struct request *nvme_alloc_request_qid(struct request_queue *q,
620 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
624 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
627 nvme_init_request(req, cmd);
631 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
633 struct nvme_command c;
635 memset(&c, 0, sizeof(c));
637 c.directive.opcode = nvme_admin_directive_send;
638 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
639 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
640 c.directive.dtype = NVME_DIR_IDENTIFY;
641 c.directive.tdtype = NVME_DIR_STREAMS;
642 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
644 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
647 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
649 return nvme_toggle_streams(ctrl, false);
652 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
654 return nvme_toggle_streams(ctrl, true);
657 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
658 struct streams_directive_params *s, u32 nsid)
660 struct nvme_command c;
662 memset(&c, 0, sizeof(c));
663 memset(s, 0, sizeof(*s));
665 c.directive.opcode = nvme_admin_directive_recv;
666 c.directive.nsid = cpu_to_le32(nsid);
667 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
668 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
669 c.directive.dtype = NVME_DIR_STREAMS;
671 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
674 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
676 struct streams_directive_params s;
679 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
684 ret = nvme_enable_streams(ctrl);
688 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
690 goto out_disable_stream;
692 ctrl->nssa = le16_to_cpu(s.nssa);
693 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
694 dev_info(ctrl->device, "too few streams (%u) available\n",
696 goto out_disable_stream;
699 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
700 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
704 nvme_disable_streams(ctrl);
709 * Check if 'req' has a write hint associated with it. If it does, assign
710 * a valid namespace stream to the write.
712 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
713 struct request *req, u16 *control,
716 enum rw_hint streamid = req->write_hint;
718 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
722 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
725 *control |= NVME_RW_DTYPE_STREAMS;
726 *dsmgmt |= streamid << 16;
729 if (streamid < ARRAY_SIZE(req->q->write_hints))
730 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
733 static inline void nvme_setup_flush(struct nvme_ns *ns,
734 struct nvme_command *cmnd)
736 cmnd->common.opcode = nvme_cmd_flush;
737 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
740 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
741 struct nvme_command *cmnd)
743 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
744 struct nvme_dsm_range *range;
748 * Some devices do not consider the DSM 'Number of Ranges' field when
749 * determining how much data to DMA. Always allocate memory for maximum
750 * number of segments to prevent device reading beyond end of buffer.
752 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
754 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
757 * If we fail allocation our range, fallback to the controller
758 * discard page. If that's also busy, it's safe to return
759 * busy, as we know we can make progress once that's freed.
761 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
762 return BLK_STS_RESOURCE;
764 range = page_address(ns->ctrl->discard_page);
767 __rq_for_each_bio(bio, req) {
768 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
769 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
772 range[n].cattr = cpu_to_le32(0);
773 range[n].nlb = cpu_to_le32(nlb);
774 range[n].slba = cpu_to_le64(slba);
779 if (WARN_ON_ONCE(n != segments)) {
780 if (virt_to_page(range) == ns->ctrl->discard_page)
781 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
784 return BLK_STS_IOERR;
787 cmnd->dsm.opcode = nvme_cmd_dsm;
788 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
789 cmnd->dsm.nr = cpu_to_le32(segments - 1);
790 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
792 req->special_vec.bv_page = virt_to_page(range);
793 req->special_vec.bv_offset = offset_in_page(range);
794 req->special_vec.bv_len = alloc_size;
795 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
800 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
801 struct request *req, struct nvme_command *cmnd)
803 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
804 return nvme_setup_discard(ns, req, cmnd);
806 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
807 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
808 cmnd->write_zeroes.slba =
809 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
810 cmnd->write_zeroes.length =
811 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
812 cmnd->write_zeroes.control = 0;
816 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
817 struct request *req, struct nvme_command *cmnd,
820 struct nvme_ctrl *ctrl = ns->ctrl;
824 if (req->cmd_flags & REQ_FUA)
825 control |= NVME_RW_FUA;
826 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
827 control |= NVME_RW_LR;
829 if (req->cmd_flags & REQ_RAHEAD)
830 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
832 cmnd->rw.opcode = op;
833 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
834 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
835 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
837 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
838 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
842 * If formated with metadata, the block layer always provides a
843 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
844 * we enable the PRACT bit for protection information or set the
845 * namespace capacity to zero to prevent any I/O.
847 if (!blk_integrity_rq(req)) {
848 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
849 return BLK_STS_NOTSUPP;
850 control |= NVME_RW_PRINFO_PRACT;
853 switch (ns->pi_type) {
854 case NVME_NS_DPS_PI_TYPE3:
855 control |= NVME_RW_PRINFO_PRCHK_GUARD;
857 case NVME_NS_DPS_PI_TYPE1:
858 case NVME_NS_DPS_PI_TYPE2:
859 control |= NVME_RW_PRINFO_PRCHK_GUARD |
860 NVME_RW_PRINFO_PRCHK_REF;
861 if (op == nvme_cmd_zone_append)
862 control |= NVME_RW_APPEND_PIREMAP;
863 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
868 cmnd->rw.control = cpu_to_le16(control);
869 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
873 void nvme_cleanup_cmd(struct request *req)
875 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
876 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
877 struct page *page = req->special_vec.bv_page;
879 if (page == ctrl->discard_page)
880 clear_bit_unlock(0, &ctrl->discard_page_busy);
882 kfree(page_address(page) + req->special_vec.bv_offset);
885 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
887 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
889 struct nvme_command *cmd = nvme_req(req)->cmd;
890 blk_status_t ret = BLK_STS_OK;
892 if (!(req->rq_flags & RQF_DONTPREP))
893 nvme_clear_nvme_request(req);
895 switch (req_op(req)) {
898 /* these are setup prior to execution in nvme_init_request() */
901 nvme_setup_flush(ns, cmd);
903 case REQ_OP_ZONE_RESET_ALL:
904 case REQ_OP_ZONE_RESET:
905 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
907 case REQ_OP_ZONE_OPEN:
908 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
910 case REQ_OP_ZONE_CLOSE:
911 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
913 case REQ_OP_ZONE_FINISH:
914 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
916 case REQ_OP_WRITE_ZEROES:
917 ret = nvme_setup_write_zeroes(ns, req, cmd);
920 ret = nvme_setup_discard(ns, req, cmd);
923 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
926 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
928 case REQ_OP_ZONE_APPEND:
929 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
933 return BLK_STS_IOERR;
936 cmd->common.command_id = req->tag;
937 trace_nvme_setup_cmd(req, cmd);
940 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
942 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
944 struct completion *waiting = rq->end_io_data;
946 rq->end_io_data = NULL;
950 static void nvme_execute_rq_polled(struct request_queue *q,
951 struct gendisk *bd_disk, struct request *rq, int at_head)
953 DECLARE_COMPLETION_ONSTACK(wait);
955 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
957 rq->cmd_flags |= REQ_HIPRI;
958 rq->end_io_data = &wait;
959 blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
961 while (!completion_done(&wait)) {
962 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
968 * Returns 0 on success. If the result is negative, it's a Linux error code;
969 * if the result is positive, it's an NVM Express status code
971 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
972 union nvme_result *result, void *buffer, unsigned bufflen,
973 unsigned timeout, int qid, int at_head,
974 blk_mq_req_flags_t flags, bool poll)
979 if (qid == NVME_QID_ANY)
980 req = nvme_alloc_request(q, cmd, flags);
982 req = nvme_alloc_request_qid(q, cmd, flags, qid);
987 req->timeout = timeout;
989 if (buffer && bufflen) {
990 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
996 nvme_execute_rq_polled(req->q, NULL, req, at_head);
998 blk_execute_rq(NULL, req, at_head);
1000 *result = nvme_req(req)->result;
1001 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1004 ret = nvme_req(req)->status;
1006 blk_mq_free_request(req);
1009 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1011 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1012 void *buffer, unsigned bufflen)
1014 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1015 NVME_QID_ANY, 0, 0, false);
1017 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1019 static u32 nvme_known_admin_effects(u8 opcode)
1022 case nvme_admin_format_nvm:
1023 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1024 NVME_CMD_EFFECTS_CSE_MASK;
1025 case nvme_admin_sanitize_nvm:
1026 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1033 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1038 if (ns->head->effects)
1039 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1040 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1041 dev_warn_once(ctrl->device,
1042 "IO command:%02x has unhandled effects:%08x\n",
1048 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1049 effects |= nvme_known_admin_effects(opcode);
1053 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1055 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1058 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1061 * For simplicity, IO to all namespaces is quiesced even if the command
1062 * effects say only one namespace is affected.
1064 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1065 mutex_lock(&ctrl->scan_lock);
1066 mutex_lock(&ctrl->subsys->lock);
1067 nvme_mpath_start_freeze(ctrl->subsys);
1068 nvme_mpath_wait_freeze(ctrl->subsys);
1069 nvme_start_freeze(ctrl);
1070 nvme_wait_freeze(ctrl);
1075 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1077 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1078 nvme_unfreeze(ctrl);
1079 nvme_mpath_unfreeze(ctrl->subsys);
1080 mutex_unlock(&ctrl->subsys->lock);
1081 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1082 mutex_unlock(&ctrl->scan_lock);
1084 if (effects & NVME_CMD_EFFECTS_CCC)
1085 nvme_init_ctrl_finish(ctrl);
1086 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1087 nvme_queue_scan(ctrl);
1088 flush_work(&ctrl->scan_work);
1092 void nvme_execute_passthru_rq(struct request *rq)
1094 struct nvme_command *cmd = nvme_req(rq)->cmd;
1095 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1096 struct nvme_ns *ns = rq->q->queuedata;
1097 struct gendisk *disk = ns ? ns->disk : NULL;
1100 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1101 blk_execute_rq(disk, rq, 0);
1102 if (effects) /* nothing to be done for zero cmd effects */
1103 nvme_passthru_end(ctrl, effects);
1105 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1107 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1109 struct nvme_ctrl *ctrl = rq->end_io_data;
1110 unsigned long flags;
1111 bool startka = false;
1113 blk_mq_free_request(rq);
1116 dev_err(ctrl->device,
1117 "failed nvme_keep_alive_end_io error=%d\n",
1122 ctrl->comp_seen = false;
1123 spin_lock_irqsave(&ctrl->lock, flags);
1124 if (ctrl->state == NVME_CTRL_LIVE ||
1125 ctrl->state == NVME_CTRL_CONNECTING)
1127 spin_unlock_irqrestore(&ctrl->lock, flags);
1129 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1132 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1136 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1137 BLK_MQ_REQ_RESERVED);
1141 rq->timeout = ctrl->kato * HZ;
1142 rq->end_io_data = ctrl;
1144 blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1149 static void nvme_keep_alive_work(struct work_struct *work)
1151 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1152 struct nvme_ctrl, ka_work);
1153 bool comp_seen = ctrl->comp_seen;
1155 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1156 dev_dbg(ctrl->device,
1157 "reschedule traffic based keep-alive timer\n");
1158 ctrl->comp_seen = false;
1159 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1163 if (nvme_keep_alive(ctrl)) {
1164 /* allocation failure, reset the controller */
1165 dev_err(ctrl->device, "keep-alive failed\n");
1166 nvme_reset_ctrl(ctrl);
1171 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1173 if (unlikely(ctrl->kato == 0))
1176 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1179 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1181 if (unlikely(ctrl->kato == 0))
1184 cancel_delayed_work_sync(&ctrl->ka_work);
1186 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1189 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1190 * flag, thus sending any new CNS opcodes has a big chance of not working.
1191 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1192 * (but not for any later version).
1194 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1196 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1197 return ctrl->vs < NVME_VS(1, 2, 0);
1198 return ctrl->vs < NVME_VS(1, 1, 0);
1201 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1203 struct nvme_command c = { };
1206 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1207 c.identify.opcode = nvme_admin_identify;
1208 c.identify.cns = NVME_ID_CNS_CTRL;
1210 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1214 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1215 sizeof(struct nvme_id_ctrl));
1221 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1223 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1226 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1227 struct nvme_ns_id_desc *cur, bool *csi_seen)
1229 const char *warn_str = "ctrl returned bogus length:";
1232 switch (cur->nidt) {
1233 case NVME_NIDT_EUI64:
1234 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1235 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1236 warn_str, cur->nidl);
1239 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1240 return NVME_NIDT_EUI64_LEN;
1241 case NVME_NIDT_NGUID:
1242 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1243 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1244 warn_str, cur->nidl);
1247 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1248 return NVME_NIDT_NGUID_LEN;
1249 case NVME_NIDT_UUID:
1250 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1251 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1252 warn_str, cur->nidl);
1255 uuid_copy(&ids->uuid, data + sizeof(*cur));
1256 return NVME_NIDT_UUID_LEN;
1258 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1259 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1260 warn_str, cur->nidl);
1263 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1265 return NVME_NIDT_CSI_LEN;
1267 /* Skip unknown types */
1272 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1273 struct nvme_ns_ids *ids)
1275 struct nvme_command c = { };
1276 bool csi_seen = false;
1277 int status, pos, len;
1280 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1282 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1285 c.identify.opcode = nvme_admin_identify;
1286 c.identify.nsid = cpu_to_le32(nsid);
1287 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1289 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1293 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1294 NVME_IDENTIFY_DATA_SIZE);
1296 dev_warn(ctrl->device,
1297 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1302 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1303 struct nvme_ns_id_desc *cur = data + pos;
1308 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1312 len += sizeof(*cur);
1315 if (nvme_multi_css(ctrl) && !csi_seen) {
1316 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1326 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1327 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1329 struct nvme_command c = { };
1332 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1333 c.identify.opcode = nvme_admin_identify;
1334 c.identify.nsid = cpu_to_le32(nsid);
1335 c.identify.cns = NVME_ID_CNS_NS;
1337 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1341 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1343 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1347 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1348 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1351 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1352 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1353 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1354 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1355 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1356 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1365 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1366 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1368 union nvme_result res = { 0 };
1369 struct nvme_command c;
1372 memset(&c, 0, sizeof(c));
1373 c.features.opcode = op;
1374 c.features.fid = cpu_to_le32(fid);
1375 c.features.dword11 = cpu_to_le32(dword11);
1377 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1378 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1379 if (ret >= 0 && result)
1380 *result = le32_to_cpu(res.u32);
1384 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1385 unsigned int dword11, void *buffer, size_t buflen,
1388 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1391 EXPORT_SYMBOL_GPL(nvme_set_features);
1393 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1394 unsigned int dword11, void *buffer, size_t buflen,
1397 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1400 EXPORT_SYMBOL_GPL(nvme_get_features);
1402 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1404 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1406 int status, nr_io_queues;
1408 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1414 * Degraded controllers might return an error when setting the queue
1415 * count. We still want to be able to bring them online and offer
1416 * access to the admin queue, as that might be only way to fix them up.
1419 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1422 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1423 *count = min(*count, nr_io_queues);
1428 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1430 #define NVME_AEN_SUPPORTED \
1431 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1432 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1434 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1436 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1439 if (!supported_aens)
1442 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1445 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1448 queue_work(nvme_wq, &ctrl->async_event_work);
1452 * Issue ioctl requests on the first available path. Note that unlike normal
1453 * block layer requests we will not retry failed request on another controller.
1455 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1456 struct nvme_ns_head **head, int *srcu_idx)
1458 #ifdef CONFIG_NVME_MULTIPATH
1459 if (disk->fops == &nvme_ns_head_ops) {
1462 *head = disk->private_data;
1463 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1464 ns = nvme_find_path(*head);
1466 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1472 return disk->private_data;
1475 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1478 srcu_read_unlock(&head->srcu, idx);
1481 static int nvme_open(struct block_device *bdev, fmode_t mode)
1483 struct nvme_ns *ns = bdev->bd_disk->private_data;
1485 /* should never be called due to GENHD_FL_HIDDEN */
1486 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1488 if (!kref_get_unless_zero(&ns->kref))
1490 if (!try_module_get(ns->ctrl->ops->module))
1501 static void nvme_release(struct gendisk *disk, fmode_t mode)
1503 struct nvme_ns *ns = disk->private_data;
1505 module_put(ns->ctrl->ops->module);
1509 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1511 /* some standard values */
1512 geo->heads = 1 << 6;
1513 geo->sectors = 1 << 5;
1514 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1518 #ifdef CONFIG_BLK_DEV_INTEGRITY
1519 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1520 u32 max_integrity_segments)
1522 struct blk_integrity integrity;
1524 memset(&integrity, 0, sizeof(integrity));
1526 case NVME_NS_DPS_PI_TYPE3:
1527 integrity.profile = &t10_pi_type3_crc;
1528 integrity.tag_size = sizeof(u16) + sizeof(u32);
1529 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1531 case NVME_NS_DPS_PI_TYPE1:
1532 case NVME_NS_DPS_PI_TYPE2:
1533 integrity.profile = &t10_pi_type1_crc;
1534 integrity.tag_size = sizeof(u16);
1535 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1538 integrity.profile = NULL;
1541 integrity.tuple_size = ms;
1542 blk_integrity_register(disk, &integrity);
1543 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1546 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1547 u32 max_integrity_segments)
1550 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1552 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1554 struct nvme_ctrl *ctrl = ns->ctrl;
1555 struct request_queue *queue = disk->queue;
1556 u32 size = queue_logical_block_size(queue);
1558 if (ctrl->max_discard_sectors == 0) {
1559 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1563 if (ctrl->nr_streams && ns->sws && ns->sgs)
1564 size *= ns->sws * ns->sgs;
1566 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1567 NVME_DSM_MAX_RANGES);
1569 queue->limits.discard_alignment = 0;
1570 queue->limits.discard_granularity = size;
1572 /* If discard is already enabled, don't reset queue limits */
1573 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1576 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1577 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1579 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1580 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1583 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1585 return !uuid_is_null(&ids->uuid) ||
1586 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1587 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1590 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1592 return uuid_equal(&a->uuid, &b->uuid) &&
1593 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1594 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1598 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1599 u32 *phys_bs, u32 *io_opt)
1601 struct streams_directive_params s;
1604 if (!ctrl->nr_streams)
1607 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1611 ns->sws = le32_to_cpu(s.sws);
1612 ns->sgs = le16_to_cpu(s.sgs);
1615 *phys_bs = ns->sws * (1 << ns->lba_shift);
1617 *io_opt = *phys_bs * ns->sgs;
1623 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1625 struct nvme_ctrl *ctrl = ns->ctrl;
1628 * The PI implementation requires the metadata size to be equal to the
1629 * t10 pi tuple size.
1631 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1632 if (ns->ms == sizeof(struct t10_pi_tuple))
1633 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1637 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1638 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1640 if (ctrl->ops->flags & NVME_F_FABRICS) {
1642 * The NVMe over Fabrics specification only supports metadata as
1643 * part of the extended data LBA. We rely on HCA/HBA support to
1644 * remap the separate metadata buffer from the block layer.
1646 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1648 if (ctrl->max_integrity_segments)
1650 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1653 * For PCIe controllers, we can't easily remap the separate
1654 * metadata buffer from the block layer and thus require a
1655 * separate metadata buffer for block layer metadata/PI support.
1656 * We allow extended LBAs for the passthrough interface, though.
1658 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1659 ns->features |= NVME_NS_EXT_LBAS;
1661 ns->features |= NVME_NS_METADATA_SUPPORTED;
1667 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1668 struct request_queue *q)
1670 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1672 if (ctrl->max_hw_sectors) {
1674 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1676 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1677 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1678 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1680 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1681 blk_queue_dma_alignment(q, 7);
1682 blk_queue_write_cache(q, vwc, vwc);
1685 static void nvme_update_disk_info(struct gendisk *disk,
1686 struct nvme_ns *ns, struct nvme_id_ns *id)
1688 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1689 unsigned short bs = 1 << ns->lba_shift;
1690 u32 atomic_bs, phys_bs, io_opt = 0;
1693 * The block layer can't support LBA sizes larger than the page size
1694 * yet, so catch this early and don't allow block I/O.
1696 if (ns->lba_shift > PAGE_SHIFT) {
1701 blk_integrity_unregister(disk);
1703 atomic_bs = phys_bs = bs;
1704 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1705 if (id->nabo == 0) {
1707 * Bit 1 indicates whether NAWUPF is defined for this namespace
1708 * and whether it should be used instead of AWUPF. If NAWUPF ==
1709 * 0 then AWUPF must be used instead.
1711 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1712 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1714 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1717 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1718 /* NPWG = Namespace Preferred Write Granularity */
1719 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1720 /* NOWS = Namespace Optimal Write Size */
1721 io_opt = bs * (1 + le16_to_cpu(id->nows));
1724 blk_queue_logical_block_size(disk->queue, bs);
1726 * Linux filesystems assume writing a single physical block is
1727 * an atomic operation. Hence limit the physical block size to the
1728 * value of the Atomic Write Unit Power Fail parameter.
1730 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1731 blk_queue_io_min(disk->queue, phys_bs);
1732 blk_queue_io_opt(disk->queue, io_opt);
1735 * Register a metadata profile for PI, or the plain non-integrity NVMe
1736 * metadata masquerading as Type 0 if supported, otherwise reject block
1737 * I/O to namespaces with metadata except when the namespace supports
1738 * PI, as it can strip/insert in that case.
1741 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1742 (ns->features & NVME_NS_METADATA_SUPPORTED))
1743 nvme_init_integrity(disk, ns->ms, ns->pi_type,
1744 ns->ctrl->max_integrity_segments);
1745 else if (!nvme_ns_has_pi(ns))
1749 set_capacity_and_notify(disk, capacity);
1751 nvme_config_discard(disk, ns);
1752 blk_queue_max_write_zeroes_sectors(disk->queue,
1753 ns->ctrl->max_zeroes_sectors);
1755 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1756 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1759 static inline bool nvme_first_scan(struct gendisk *disk)
1761 /* nvme_alloc_ns() scans the disk prior to adding it */
1762 return !(disk->flags & GENHD_FL_UP);
1765 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1767 struct nvme_ctrl *ctrl = ns->ctrl;
1770 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1771 is_power_of_2(ctrl->max_hw_sectors))
1772 iob = ctrl->max_hw_sectors;
1774 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1779 if (!is_power_of_2(iob)) {
1780 if (nvme_first_scan(ns->disk))
1781 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1782 ns->disk->disk_name, iob);
1786 if (blk_queue_is_zoned(ns->disk->queue)) {
1787 if (nvme_first_scan(ns->disk))
1788 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1789 ns->disk->disk_name);
1793 blk_queue_chunk_sectors(ns->queue, iob);
1796 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1798 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1801 blk_mq_freeze_queue(ns->disk->queue);
1802 ns->lba_shift = id->lbaf[lbaf].ds;
1803 nvme_set_queue_limits(ns->ctrl, ns->queue);
1805 ret = nvme_configure_metadata(ns, id);
1808 nvme_set_chunk_sectors(ns, id);
1809 nvme_update_disk_info(ns->disk, ns, id);
1811 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1812 ret = nvme_update_zone_info(ns, lbaf);
1817 blk_mq_unfreeze_queue(ns->disk->queue);
1819 if (blk_queue_is_zoned(ns->queue)) {
1820 ret = nvme_revalidate_zones(ns);
1821 if (ret && !nvme_first_scan(ns->disk))
1825 if (nvme_ns_head_multipath(ns->head)) {
1826 blk_mq_freeze_queue(ns->head->disk->queue);
1827 nvme_update_disk_info(ns->head->disk, ns, id);
1828 blk_stack_limits(&ns->head->disk->queue->limits,
1829 &ns->queue->limits, 0);
1830 blk_queue_update_readahead(ns->head->disk->queue);
1831 blk_mq_unfreeze_queue(ns->head->disk->queue);
1836 blk_mq_unfreeze_queue(ns->disk->queue);
1840 static char nvme_pr_type(enum pr_type type)
1843 case PR_WRITE_EXCLUSIVE:
1845 case PR_EXCLUSIVE_ACCESS:
1847 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1849 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1851 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1853 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1860 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1861 u64 key, u64 sa_key, u8 op)
1863 struct nvme_ns_head *head = NULL;
1865 struct nvme_command c;
1867 u8 data[16] = { 0, };
1869 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1871 return -EWOULDBLOCK;
1873 put_unaligned_le64(key, &data[0]);
1874 put_unaligned_le64(sa_key, &data[8]);
1876 memset(&c, 0, sizeof(c));
1877 c.common.opcode = op;
1878 c.common.nsid = cpu_to_le32(ns->head->ns_id);
1879 c.common.cdw10 = cpu_to_le32(cdw10);
1881 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
1882 nvme_put_ns_from_disk(head, srcu_idx);
1886 static int nvme_pr_register(struct block_device *bdev, u64 old,
1887 u64 new, unsigned flags)
1891 if (flags & ~PR_FL_IGNORE_KEY)
1894 cdw10 = old ? 2 : 0;
1895 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
1896 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
1897 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
1900 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
1901 enum pr_type type, unsigned flags)
1905 if (flags & ~PR_FL_IGNORE_KEY)
1908 cdw10 = nvme_pr_type(type) << 8;
1909 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
1910 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
1913 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
1914 enum pr_type type, bool abort)
1916 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
1918 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
1921 static int nvme_pr_clear(struct block_device *bdev, u64 key)
1923 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
1925 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
1928 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1930 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
1932 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
1935 static const struct pr_ops nvme_pr_ops = {
1936 .pr_register = nvme_pr_register,
1937 .pr_reserve = nvme_pr_reserve,
1938 .pr_release = nvme_pr_release,
1939 .pr_preempt = nvme_pr_preempt,
1940 .pr_clear = nvme_pr_clear,
1943 #ifdef CONFIG_BLK_SED_OPAL
1944 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
1947 struct nvme_ctrl *ctrl = data;
1948 struct nvme_command cmd;
1950 memset(&cmd, 0, sizeof(cmd));
1952 cmd.common.opcode = nvme_admin_security_send;
1954 cmd.common.opcode = nvme_admin_security_recv;
1955 cmd.common.nsid = 0;
1956 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
1957 cmd.common.cdw11 = cpu_to_le32(len);
1959 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
1960 NVME_QID_ANY, 1, 0, false);
1962 EXPORT_SYMBOL_GPL(nvme_sec_submit);
1963 #endif /* CONFIG_BLK_SED_OPAL */
1965 static const struct block_device_operations nvme_bdev_ops = {
1966 .owner = THIS_MODULE,
1967 .ioctl = nvme_ioctl,
1969 .release = nvme_release,
1970 .getgeo = nvme_getgeo,
1971 .report_zones = nvme_report_zones,
1972 .pr_ops = &nvme_pr_ops,
1975 #ifdef CONFIG_NVME_MULTIPATH
1976 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
1978 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1980 if (!kref_get_unless_zero(&head->ref))
1985 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
1987 nvme_put_ns_head(disk->private_data);
1990 struct nvme_ctrl *nvme_find_get_live_ctrl(struct nvme_subsystem *subsys)
1992 struct nvme_ctrl *ctrl;
1995 ret = mutex_lock_killable(&nvme_subsystems_lock);
1997 return ERR_PTR(ret);
1998 list_for_each_entry(ctrl, &subsys->ctrls, subsys_entry) {
1999 if (ctrl->state == NVME_CTRL_LIVE)
2002 mutex_unlock(&nvme_subsystems_lock);
2003 return ERR_PTR(-EWOULDBLOCK);
2005 nvme_get_ctrl(ctrl);
2006 mutex_unlock(&nvme_subsystems_lock);
2010 const struct block_device_operations nvme_ns_head_ops = {
2011 .owner = THIS_MODULE,
2012 .submit_bio = nvme_ns_head_submit_bio,
2013 .open = nvme_ns_head_open,
2014 .release = nvme_ns_head_release,
2015 .ioctl = nvme_ns_head_ioctl,
2016 .getgeo = nvme_getgeo,
2017 .report_zones = nvme_report_zones,
2018 .pr_ops = &nvme_pr_ops,
2020 #endif /* CONFIG_NVME_MULTIPATH */
2022 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2024 unsigned long timeout =
2025 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2026 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2029 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2032 if ((csts & NVME_CSTS_RDY) == bit)
2035 usleep_range(1000, 2000);
2036 if (fatal_signal_pending(current))
2038 if (time_after(jiffies, timeout)) {
2039 dev_err(ctrl->device,
2040 "Device not ready; aborting %s, CSTS=0x%x\n",
2041 enabled ? "initialisation" : "reset", csts);
2050 * If the device has been passed off to us in an enabled state, just clear
2051 * the enabled bit. The spec says we should set the 'shutdown notification
2052 * bits', but doing so may cause the device to complete commands to the
2053 * admin queue ... and we don't know what memory that might be pointing at!
2055 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2059 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2060 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2062 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2066 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2067 msleep(NVME_QUIRK_DELAY_AMOUNT);
2069 return nvme_wait_ready(ctrl, ctrl->cap, false);
2071 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2073 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2075 unsigned dev_page_min;
2078 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2080 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2083 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2085 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2086 dev_err(ctrl->device,
2087 "Minimum device page size %u too large for host (%u)\n",
2088 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2092 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2093 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2095 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2096 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2097 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2098 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2099 ctrl->ctrl_config |= NVME_CC_ENABLE;
2101 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2104 return nvme_wait_ready(ctrl, ctrl->cap, true);
2106 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2108 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2110 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2114 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2115 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2117 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2121 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2122 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2126 if (fatal_signal_pending(current))
2128 if (time_after(jiffies, timeout)) {
2129 dev_err(ctrl->device,
2130 "Device shutdown incomplete; abort shutdown\n");
2137 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2139 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2144 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2147 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2148 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2151 dev_warn_once(ctrl->device,
2152 "could not set timestamp (%d)\n", ret);
2156 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2158 struct nvme_feat_host_behavior *host;
2161 /* Don't bother enabling the feature if retry delay is not reported */
2165 host = kzalloc(sizeof(*host), GFP_KERNEL);
2169 host->acre = NVME_ENABLE_ACRE;
2170 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2171 host, sizeof(*host), NULL);
2176 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2179 * APST (Autonomous Power State Transition) lets us program a
2180 * table of power state transitions that the controller will
2181 * perform automatically. We configure it with a simple
2182 * heuristic: we are willing to spend at most 2% of the time
2183 * transitioning between power states. Therefore, when running
2184 * in any given state, we will enter the next lower-power
2185 * non-operational state after waiting 50 * (enlat + exlat)
2186 * microseconds, as long as that state's exit latency is under
2187 * the requested maximum latency.
2189 * We will not autonomously enter any non-operational state for
2190 * which the total latency exceeds ps_max_latency_us. Users
2191 * can set ps_max_latency_us to zero to turn off APST.
2195 struct nvme_feat_auto_pst *table;
2201 * If APST isn't supported or if we haven't been initialized yet,
2202 * then don't do anything.
2207 if (ctrl->npss > 31) {
2208 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2212 table = kzalloc(sizeof(*table), GFP_KERNEL);
2216 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2217 /* Turn off APST. */
2219 dev_dbg(ctrl->device, "APST disabled\n");
2221 __le64 target = cpu_to_le64(0);
2225 * Walk through all states from lowest- to highest-power.
2226 * According to the spec, lower-numbered states use more
2227 * power. NPSS, despite the name, is the index of the
2228 * lowest-power state, not the number of states.
2230 for (state = (int)ctrl->npss; state >= 0; state--) {
2231 u64 total_latency_us, exit_latency_us, transition_ms;
2234 table->entries[state] = target;
2237 * Don't allow transitions to the deepest state
2238 * if it's quirked off.
2240 if (state == ctrl->npss &&
2241 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2245 * Is this state a useful non-operational state for
2246 * higher-power states to autonomously transition to?
2248 if (!(ctrl->psd[state].flags &
2249 NVME_PS_FLAGS_NON_OP_STATE))
2253 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2254 if (exit_latency_us > ctrl->ps_max_latency_us)
2259 le32_to_cpu(ctrl->psd[state].entry_lat);
2262 * This state is good. Use it as the APST idle
2263 * target for higher power states.
2265 transition_ms = total_latency_us + 19;
2266 do_div(transition_ms, 20);
2267 if (transition_ms > (1 << 24) - 1)
2268 transition_ms = (1 << 24) - 1;
2270 target = cpu_to_le64((state << 3) |
2271 (transition_ms << 8));
2276 if (total_latency_us > max_lat_us)
2277 max_lat_us = total_latency_us;
2283 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2285 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2286 max_ps, max_lat_us, (int)sizeof(*table), table);
2290 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2291 table, sizeof(*table), NULL);
2293 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2299 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2301 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2305 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2306 case PM_QOS_LATENCY_ANY:
2314 if (ctrl->ps_max_latency_us != latency) {
2315 ctrl->ps_max_latency_us = latency;
2316 nvme_configure_apst(ctrl);
2320 struct nvme_core_quirk_entry {
2322 * NVMe model and firmware strings are padded with spaces. For
2323 * simplicity, strings in the quirk table are padded with NULLs
2329 unsigned long quirks;
2332 static const struct nvme_core_quirk_entry core_quirks[] = {
2335 * This Toshiba device seems to die using any APST states. See:
2336 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2339 .mn = "THNSF5256GPUK TOSHIBA",
2340 .quirks = NVME_QUIRK_NO_APST,
2344 * This LiteON CL1-3D*-Q11 firmware version has a race
2345 * condition associated with actions related to suspend to idle
2346 * LiteON has resolved the problem in future firmware
2350 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2354 /* match is null-terminated but idstr is space-padded. */
2355 static bool string_matches(const char *idstr, const char *match, size_t len)
2362 matchlen = strlen(match);
2363 WARN_ON_ONCE(matchlen > len);
2365 if (memcmp(idstr, match, matchlen))
2368 for (; matchlen < len; matchlen++)
2369 if (idstr[matchlen] != ' ')
2375 static bool quirk_matches(const struct nvme_id_ctrl *id,
2376 const struct nvme_core_quirk_entry *q)
2378 return q->vid == le16_to_cpu(id->vid) &&
2379 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2380 string_matches(id->fr, q->fr, sizeof(id->fr));
2383 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2384 struct nvme_id_ctrl *id)
2389 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2390 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2391 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2392 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2396 if (ctrl->vs >= NVME_VS(1, 2, 1))
2397 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2400 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2401 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2402 "nqn.2014.08.org.nvmexpress:%04x%04x",
2403 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2404 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2405 off += sizeof(id->sn);
2406 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2407 off += sizeof(id->mn);
2408 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2411 static void nvme_release_subsystem(struct device *dev)
2413 struct nvme_subsystem *subsys =
2414 container_of(dev, struct nvme_subsystem, dev);
2416 if (subsys->instance >= 0)
2417 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2421 static void nvme_destroy_subsystem(struct kref *ref)
2423 struct nvme_subsystem *subsys =
2424 container_of(ref, struct nvme_subsystem, ref);
2426 mutex_lock(&nvme_subsystems_lock);
2427 list_del(&subsys->entry);
2428 mutex_unlock(&nvme_subsystems_lock);
2430 ida_destroy(&subsys->ns_ida);
2431 device_del(&subsys->dev);
2432 put_device(&subsys->dev);
2435 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2437 kref_put(&subsys->ref, nvme_destroy_subsystem);
2440 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2442 struct nvme_subsystem *subsys;
2444 lockdep_assert_held(&nvme_subsystems_lock);
2447 * Fail matches for discovery subsystems. This results
2448 * in each discovery controller bound to a unique subsystem.
2449 * This avoids issues with validating controller values
2450 * that can only be true when there is a single unique subsystem.
2451 * There may be multiple and completely independent entities
2452 * that provide discovery controllers.
2454 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2457 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2458 if (strcmp(subsys->subnqn, subsysnqn))
2460 if (!kref_get_unless_zero(&subsys->ref))
2468 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2469 struct device_attribute subsys_attr_##_name = \
2470 __ATTR(_name, _mode, _show, NULL)
2472 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2473 struct device_attribute *attr,
2476 struct nvme_subsystem *subsys =
2477 container_of(dev, struct nvme_subsystem, dev);
2479 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2481 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2483 #define nvme_subsys_show_str_function(field) \
2484 static ssize_t subsys_##field##_show(struct device *dev, \
2485 struct device_attribute *attr, char *buf) \
2487 struct nvme_subsystem *subsys = \
2488 container_of(dev, struct nvme_subsystem, dev); \
2489 return sysfs_emit(buf, "%.*s\n", \
2490 (int)sizeof(subsys->field), subsys->field); \
2492 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2494 nvme_subsys_show_str_function(model);
2495 nvme_subsys_show_str_function(serial);
2496 nvme_subsys_show_str_function(firmware_rev);
2498 static struct attribute *nvme_subsys_attrs[] = {
2499 &subsys_attr_model.attr,
2500 &subsys_attr_serial.attr,
2501 &subsys_attr_firmware_rev.attr,
2502 &subsys_attr_subsysnqn.attr,
2503 #ifdef CONFIG_NVME_MULTIPATH
2504 &subsys_attr_iopolicy.attr,
2509 static const struct attribute_group nvme_subsys_attrs_group = {
2510 .attrs = nvme_subsys_attrs,
2513 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2514 &nvme_subsys_attrs_group,
2518 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2520 return ctrl->opts && ctrl->opts->discovery_nqn;
2523 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2524 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2526 struct nvme_ctrl *tmp;
2528 lockdep_assert_held(&nvme_subsystems_lock);
2530 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2531 if (nvme_state_terminal(tmp))
2534 if (tmp->cntlid == ctrl->cntlid) {
2535 dev_err(ctrl->device,
2536 "Duplicate cntlid %u with %s, rejecting\n",
2537 ctrl->cntlid, dev_name(tmp->device));
2541 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2542 nvme_discovery_ctrl(ctrl))
2545 dev_err(ctrl->device,
2546 "Subsystem does not support multiple controllers\n");
2553 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2555 struct nvme_subsystem *subsys, *found;
2558 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2562 subsys->instance = -1;
2563 mutex_init(&subsys->lock);
2564 kref_init(&subsys->ref);
2565 INIT_LIST_HEAD(&subsys->ctrls);
2566 INIT_LIST_HEAD(&subsys->nsheads);
2567 nvme_init_subnqn(subsys, ctrl, id);
2568 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2569 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2570 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2571 subsys->vendor_id = le16_to_cpu(id->vid);
2572 subsys->cmic = id->cmic;
2573 subsys->awupf = le16_to_cpu(id->awupf);
2574 #ifdef CONFIG_NVME_MULTIPATH
2575 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2578 subsys->dev.class = nvme_subsys_class;
2579 subsys->dev.release = nvme_release_subsystem;
2580 subsys->dev.groups = nvme_subsys_attrs_groups;
2581 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2582 device_initialize(&subsys->dev);
2584 mutex_lock(&nvme_subsystems_lock);
2585 found = __nvme_find_get_subsystem(subsys->subnqn);
2587 put_device(&subsys->dev);
2590 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2592 goto out_put_subsystem;
2595 ret = device_add(&subsys->dev);
2597 dev_err(ctrl->device,
2598 "failed to register subsystem device.\n");
2599 put_device(&subsys->dev);
2602 ida_init(&subsys->ns_ida);
2603 list_add_tail(&subsys->entry, &nvme_subsystems);
2606 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2607 dev_name(ctrl->device));
2609 dev_err(ctrl->device,
2610 "failed to create sysfs link from subsystem.\n");
2611 goto out_put_subsystem;
2615 subsys->instance = ctrl->instance;
2616 ctrl->subsys = subsys;
2617 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2618 mutex_unlock(&nvme_subsystems_lock);
2622 nvme_put_subsystem(subsys);
2624 mutex_unlock(&nvme_subsystems_lock);
2628 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2629 void *log, size_t size, u64 offset)
2631 struct nvme_command c = { };
2632 u32 dwlen = nvme_bytes_to_numd(size);
2634 c.get_log_page.opcode = nvme_admin_get_log_page;
2635 c.get_log_page.nsid = cpu_to_le32(nsid);
2636 c.get_log_page.lid = log_page;
2637 c.get_log_page.lsp = lsp;
2638 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2639 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2640 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2641 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2642 c.get_log_page.csi = csi;
2644 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2647 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2648 struct nvme_effects_log **log)
2650 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2656 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2660 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2661 cel, sizeof(*cel), 0);
2667 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2673 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2675 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2677 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2682 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2684 struct nvme_command c = { };
2685 struct nvme_id_ctrl_nvm *id;
2688 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2689 ctrl->max_discard_sectors = UINT_MAX;
2690 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2692 ctrl->max_discard_sectors = 0;
2693 ctrl->max_discard_segments = 0;
2697 * Even though NVMe spec explicitly states that MDTS is not applicable
2698 * to the write-zeroes, we are cautious and limit the size to the
2699 * controllers max_hw_sectors value, which is based on the MDTS field
2700 * and possibly other limiting factors.
2702 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2703 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2704 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2706 ctrl->max_zeroes_sectors = 0;
2708 if (nvme_ctrl_limited_cns(ctrl))
2711 id = kzalloc(sizeof(*id), GFP_KERNEL);
2715 c.identify.opcode = nvme_admin_identify;
2716 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2717 c.identify.csi = NVME_CSI_NVM;
2719 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2724 ctrl->max_discard_segments = id->dmrl;
2726 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2728 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2735 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2737 struct nvme_id_ctrl *id;
2739 bool prev_apst_enabled;
2742 ret = nvme_identify_ctrl(ctrl, &id);
2744 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2748 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2749 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2754 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2755 ctrl->cntlid = le16_to_cpu(id->cntlid);
2757 if (!ctrl->identified) {
2760 ret = nvme_init_subsystem(ctrl, id);
2765 * Check for quirks. Quirk can depend on firmware version,
2766 * so, in principle, the set of quirks present can change
2767 * across a reset. As a possible future enhancement, we
2768 * could re-scan for quirks every time we reinitialize
2769 * the device, but we'd have to make sure that the driver
2770 * behaves intelligently if the quirks change.
2772 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2773 if (quirk_matches(id, &core_quirks[i]))
2774 ctrl->quirks |= core_quirks[i].quirks;
2778 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2779 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2780 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2783 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2784 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2785 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2787 ctrl->oacs = le16_to_cpu(id->oacs);
2788 ctrl->oncs = le16_to_cpu(id->oncs);
2789 ctrl->mtfa = le16_to_cpu(id->mtfa);
2790 ctrl->oaes = le32_to_cpu(id->oaes);
2791 ctrl->wctemp = le16_to_cpu(id->wctemp);
2792 ctrl->cctemp = le16_to_cpu(id->cctemp);
2794 atomic_set(&ctrl->abort_limit, id->acl + 1);
2795 ctrl->vwc = id->vwc;
2797 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2799 max_hw_sectors = UINT_MAX;
2800 ctrl->max_hw_sectors =
2801 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2803 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2804 ctrl->sgls = le32_to_cpu(id->sgls);
2805 ctrl->kas = le16_to_cpu(id->kas);
2806 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2807 ctrl->ctratt = le32_to_cpu(id->ctratt);
2811 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2813 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2814 shutdown_timeout, 60);
2816 if (ctrl->shutdown_timeout != shutdown_timeout)
2817 dev_info(ctrl->device,
2818 "Shutdown timeout set to %u seconds\n",
2819 ctrl->shutdown_timeout);
2821 ctrl->shutdown_timeout = shutdown_timeout;
2823 ctrl->npss = id->npss;
2824 ctrl->apsta = id->apsta;
2825 prev_apst_enabled = ctrl->apst_enabled;
2826 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
2827 if (force_apst && id->apsta) {
2828 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
2829 ctrl->apst_enabled = true;
2831 ctrl->apst_enabled = false;
2834 ctrl->apst_enabled = id->apsta;
2836 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
2838 if (ctrl->ops->flags & NVME_F_FABRICS) {
2839 ctrl->icdoff = le16_to_cpu(id->icdoff);
2840 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
2841 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
2842 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
2845 * In fabrics we need to verify the cntlid matches the
2848 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
2849 dev_err(ctrl->device,
2850 "Mismatching cntlid: Connect %u vs Identify "
2852 ctrl->cntlid, le16_to_cpu(id->cntlid));
2857 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
2858 dev_err(ctrl->device,
2859 "keep-alive support is mandatory for fabrics\n");
2864 ctrl->hmpre = le32_to_cpu(id->hmpre);
2865 ctrl->hmmin = le32_to_cpu(id->hmmin);
2866 ctrl->hmminds = le32_to_cpu(id->hmminds);
2867 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
2870 ret = nvme_mpath_init(ctrl, id);
2874 if (ctrl->apst_enabled && !prev_apst_enabled)
2875 dev_pm_qos_expose_latency_tolerance(ctrl->device);
2876 else if (!ctrl->apst_enabled && prev_apst_enabled)
2877 dev_pm_qos_hide_latency_tolerance(ctrl->device);
2885 * Initialize the cached copies of the Identify data and various controller
2886 * register in our nvme_ctrl structure. This should be called as soon as
2887 * the admin queue is fully up and running.
2889 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
2893 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
2895 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
2899 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
2901 if (ctrl->vs >= NVME_VS(1, 1, 0))
2902 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
2904 ret = nvme_init_identify(ctrl);
2908 ret = nvme_init_non_mdts_limits(ctrl);
2912 ret = nvme_configure_apst(ctrl);
2916 ret = nvme_configure_timestamp(ctrl);
2920 ret = nvme_configure_directives(ctrl);
2924 ret = nvme_configure_acre(ctrl);
2928 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
2929 ret = nvme_hwmon_init(ctrl);
2934 ctrl->identified = true;
2938 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
2940 static int nvme_dev_open(struct inode *inode, struct file *file)
2942 struct nvme_ctrl *ctrl =
2943 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2945 switch (ctrl->state) {
2946 case NVME_CTRL_LIVE:
2949 return -EWOULDBLOCK;
2952 nvme_get_ctrl(ctrl);
2953 if (!try_module_get(ctrl->ops->module)) {
2954 nvme_put_ctrl(ctrl);
2958 file->private_data = ctrl;
2962 static int nvme_dev_release(struct inode *inode, struct file *file)
2964 struct nvme_ctrl *ctrl =
2965 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
2967 module_put(ctrl->ops->module);
2968 nvme_put_ctrl(ctrl);
2972 static const struct file_operations nvme_dev_fops = {
2973 .owner = THIS_MODULE,
2974 .open = nvme_dev_open,
2975 .release = nvme_dev_release,
2976 .unlocked_ioctl = nvme_dev_ioctl,
2977 .compat_ioctl = compat_ptr_ioctl,
2980 static ssize_t nvme_sysfs_reset(struct device *dev,
2981 struct device_attribute *attr, const char *buf,
2984 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2987 ret = nvme_reset_ctrl_sync(ctrl);
2992 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
2994 static ssize_t nvme_sysfs_rescan(struct device *dev,
2995 struct device_attribute *attr, const char *buf,
2998 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3000 nvme_queue_scan(ctrl);
3003 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3005 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3007 struct gendisk *disk = dev_to_disk(dev);
3009 if (disk->fops == &nvme_bdev_ops)
3010 return nvme_get_ns_from_dev(dev)->head;
3012 return disk->private_data;
3015 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3018 struct nvme_ns_head *head = dev_to_ns_head(dev);
3019 struct nvme_ns_ids *ids = &head->ids;
3020 struct nvme_subsystem *subsys = head->subsys;
3021 int serial_len = sizeof(subsys->serial);
3022 int model_len = sizeof(subsys->model);
3024 if (!uuid_is_null(&ids->uuid))
3025 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3027 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3028 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3030 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3031 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3033 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3034 subsys->serial[serial_len - 1] == '\0'))
3036 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3037 subsys->model[model_len - 1] == '\0'))
3040 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3041 serial_len, subsys->serial, model_len, subsys->model,
3044 static DEVICE_ATTR_RO(wwid);
3046 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3049 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3051 static DEVICE_ATTR_RO(nguid);
3053 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3056 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3058 /* For backward compatibility expose the NGUID to userspace if
3059 * we have no UUID set
3061 if (uuid_is_null(&ids->uuid)) {
3062 printk_ratelimited(KERN_WARNING
3063 "No UUID available providing old NGUID\n");
3064 return sysfs_emit(buf, "%pU\n", ids->nguid);
3066 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3068 static DEVICE_ATTR_RO(uuid);
3070 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3073 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3075 static DEVICE_ATTR_RO(eui);
3077 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3080 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3082 static DEVICE_ATTR_RO(nsid);
3084 static struct attribute *nvme_ns_id_attrs[] = {
3085 &dev_attr_wwid.attr,
3086 &dev_attr_uuid.attr,
3087 &dev_attr_nguid.attr,
3089 &dev_attr_nsid.attr,
3090 #ifdef CONFIG_NVME_MULTIPATH
3091 &dev_attr_ana_grpid.attr,
3092 &dev_attr_ana_state.attr,
3097 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3098 struct attribute *a, int n)
3100 struct device *dev = container_of(kobj, struct device, kobj);
3101 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3103 if (a == &dev_attr_uuid.attr) {
3104 if (uuid_is_null(&ids->uuid) &&
3105 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3108 if (a == &dev_attr_nguid.attr) {
3109 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3112 if (a == &dev_attr_eui.attr) {
3113 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3116 #ifdef CONFIG_NVME_MULTIPATH
3117 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3118 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3120 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3127 static const struct attribute_group nvme_ns_id_attr_group = {
3128 .attrs = nvme_ns_id_attrs,
3129 .is_visible = nvme_ns_id_attrs_are_visible,
3132 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3133 &nvme_ns_id_attr_group,
3135 &nvme_nvm_attr_group,
3140 #define nvme_show_str_function(field) \
3141 static ssize_t field##_show(struct device *dev, \
3142 struct device_attribute *attr, char *buf) \
3144 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3145 return sysfs_emit(buf, "%.*s\n", \
3146 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3148 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3150 nvme_show_str_function(model);
3151 nvme_show_str_function(serial);
3152 nvme_show_str_function(firmware_rev);
3154 #define nvme_show_int_function(field) \
3155 static ssize_t field##_show(struct device *dev, \
3156 struct device_attribute *attr, char *buf) \
3158 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3159 return sysfs_emit(buf, "%d\n", ctrl->field); \
3161 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3163 nvme_show_int_function(cntlid);
3164 nvme_show_int_function(numa_node);
3165 nvme_show_int_function(queue_count);
3166 nvme_show_int_function(sqsize);
3168 static ssize_t nvme_sysfs_delete(struct device *dev,
3169 struct device_attribute *attr, const char *buf,
3172 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3174 if (device_remove_file_self(dev, attr))
3175 nvme_delete_ctrl_sync(ctrl);
3178 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3180 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3181 struct device_attribute *attr,
3184 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3186 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3188 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3190 static ssize_t nvme_sysfs_show_state(struct device *dev,
3191 struct device_attribute *attr,
3194 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3195 static const char *const state_name[] = {
3196 [NVME_CTRL_NEW] = "new",
3197 [NVME_CTRL_LIVE] = "live",
3198 [NVME_CTRL_RESETTING] = "resetting",
3199 [NVME_CTRL_CONNECTING] = "connecting",
3200 [NVME_CTRL_DELETING] = "deleting",
3201 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3202 [NVME_CTRL_DEAD] = "dead",
3205 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3206 state_name[ctrl->state])
3207 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3209 return sysfs_emit(buf, "unknown state\n");
3212 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3214 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3215 struct device_attribute *attr,
3218 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3220 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3222 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3224 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3225 struct device_attribute *attr,
3228 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3230 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3232 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3234 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3235 struct device_attribute *attr,
3238 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3240 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3242 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3244 static ssize_t nvme_sysfs_show_address(struct device *dev,
3245 struct device_attribute *attr,
3248 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3250 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3252 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3254 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3255 struct device_attribute *attr, char *buf)
3257 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3258 struct nvmf_ctrl_options *opts = ctrl->opts;
3260 if (ctrl->opts->max_reconnects == -1)
3261 return sysfs_emit(buf, "off\n");
3262 return sysfs_emit(buf, "%d\n",
3263 opts->max_reconnects * opts->reconnect_delay);
3266 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3267 struct device_attribute *attr, const char *buf, size_t count)
3269 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3270 struct nvmf_ctrl_options *opts = ctrl->opts;
3271 int ctrl_loss_tmo, err;
3273 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3277 if (ctrl_loss_tmo < 0)
3278 opts->max_reconnects = -1;
3280 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3281 opts->reconnect_delay);
3284 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3285 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3287 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3288 struct device_attribute *attr, char *buf)
3290 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3292 if (ctrl->opts->reconnect_delay == -1)
3293 return sysfs_emit(buf, "off\n");
3294 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3297 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3298 struct device_attribute *attr, const char *buf, size_t count)
3300 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3304 err = kstrtou32(buf, 10, &v);
3308 ctrl->opts->reconnect_delay = v;
3311 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3312 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3314 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3315 struct device_attribute *attr, char *buf)
3317 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3319 if (ctrl->opts->fast_io_fail_tmo == -1)
3320 return sysfs_emit(buf, "off\n");
3321 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3324 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3325 struct device_attribute *attr, const char *buf, size_t count)
3327 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3328 struct nvmf_ctrl_options *opts = ctrl->opts;
3329 int fast_io_fail_tmo, err;
3331 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3335 if (fast_io_fail_tmo < 0)
3336 opts->fast_io_fail_tmo = -1;
3338 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3341 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3342 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3344 static struct attribute *nvme_dev_attrs[] = {
3345 &dev_attr_reset_controller.attr,
3346 &dev_attr_rescan_controller.attr,
3347 &dev_attr_model.attr,
3348 &dev_attr_serial.attr,
3349 &dev_attr_firmware_rev.attr,
3350 &dev_attr_cntlid.attr,
3351 &dev_attr_delete_controller.attr,
3352 &dev_attr_transport.attr,
3353 &dev_attr_subsysnqn.attr,
3354 &dev_attr_address.attr,
3355 &dev_attr_state.attr,
3356 &dev_attr_numa_node.attr,
3357 &dev_attr_queue_count.attr,
3358 &dev_attr_sqsize.attr,
3359 &dev_attr_hostnqn.attr,
3360 &dev_attr_hostid.attr,
3361 &dev_attr_ctrl_loss_tmo.attr,
3362 &dev_attr_reconnect_delay.attr,
3363 &dev_attr_fast_io_fail_tmo.attr,
3367 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3368 struct attribute *a, int n)
3370 struct device *dev = container_of(kobj, struct device, kobj);
3371 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3373 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3375 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3377 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3379 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3381 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3383 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3389 static const struct attribute_group nvme_dev_attrs_group = {
3390 .attrs = nvme_dev_attrs,
3391 .is_visible = nvme_dev_attrs_are_visible,
3394 static const struct attribute_group *nvme_dev_attr_groups[] = {
3395 &nvme_dev_attrs_group,
3399 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3402 struct nvme_ns_head *h;
3404 lockdep_assert_held(&subsys->lock);
3406 list_for_each_entry(h, &subsys->nsheads, entry) {
3407 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3414 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3415 struct nvme_ns_head *new)
3417 struct nvme_ns_head *h;
3419 lockdep_assert_held(&subsys->lock);
3421 list_for_each_entry(h, &subsys->nsheads, entry) {
3422 if (nvme_ns_ids_valid(&new->ids) &&
3423 nvme_ns_ids_equal(&new->ids, &h->ids))
3430 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3431 unsigned nsid, struct nvme_ns_ids *ids)
3433 struct nvme_ns_head *head;
3434 size_t size = sizeof(*head);
3437 #ifdef CONFIG_NVME_MULTIPATH
3438 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3441 head = kzalloc(size, GFP_KERNEL);
3444 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3447 head->instance = ret;
3448 INIT_LIST_HEAD(&head->list);
3449 ret = init_srcu_struct(&head->srcu);
3451 goto out_ida_remove;
3452 head->subsys = ctrl->subsys;
3455 kref_init(&head->ref);
3457 ret = __nvme_check_ids(ctrl->subsys, head);
3459 dev_err(ctrl->device,
3460 "duplicate IDs for nsid %d\n", nsid);
3461 goto out_cleanup_srcu;
3464 if (head->ids.csi) {
3465 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3467 goto out_cleanup_srcu;
3469 head->effects = ctrl->effects;
3471 ret = nvme_mpath_alloc_disk(ctrl, head);
3473 goto out_cleanup_srcu;
3475 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3477 kref_get(&ctrl->subsys->ref);
3481 cleanup_srcu_struct(&head->srcu);
3483 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3488 ret = blk_status_to_errno(nvme_error_status(ret));
3489 return ERR_PTR(ret);
3492 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3493 struct nvme_ns_ids *ids, bool is_shared)
3495 struct nvme_ctrl *ctrl = ns->ctrl;
3496 struct nvme_ns_head *head = NULL;
3499 mutex_lock(&ctrl->subsys->lock);
3500 head = nvme_find_ns_head(ctrl->subsys, nsid);
3502 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3504 ret = PTR_ERR(head);
3507 head->shared = is_shared;
3510 if (!is_shared || !head->shared) {
3511 dev_err(ctrl->device,
3512 "Duplicate unshared namespace %d\n", nsid);
3513 goto out_put_ns_head;
3515 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3516 dev_err(ctrl->device,
3517 "IDs don't match for shared namespace %d\n",
3519 goto out_put_ns_head;
3523 list_add_tail_rcu(&ns->siblings, &head->list);
3525 mutex_unlock(&ctrl->subsys->lock);
3529 nvme_put_ns_head(head);
3531 mutex_unlock(&ctrl->subsys->lock);
3535 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3537 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3538 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3540 return nsa->head->ns_id - nsb->head->ns_id;
3543 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3545 struct nvme_ns *ns, *ret = NULL;
3547 down_read(&ctrl->namespaces_rwsem);
3548 list_for_each_entry(ns, &ctrl->namespaces, list) {
3549 if (ns->head->ns_id == nsid) {
3550 if (!kref_get_unless_zero(&ns->kref))
3555 if (ns->head->ns_id > nsid)
3558 up_read(&ctrl->namespaces_rwsem);
3561 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3563 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3564 struct nvme_ns_ids *ids)
3567 struct gendisk *disk;
3568 struct nvme_id_ns *id;
3569 int node = ctrl->numa_node;
3571 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3574 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3578 ns->queue = blk_mq_init_queue(ctrl->tagset);
3579 if (IS_ERR(ns->queue))
3582 if (ctrl->opts && ctrl->opts->data_digest)
3583 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3585 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3586 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3587 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3589 ns->queue->queuedata = ns;
3591 kref_init(&ns->kref);
3593 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3594 goto out_free_queue;
3596 disk = alloc_disk_node(0, node);
3600 disk->fops = &nvme_bdev_ops;
3601 disk->private_data = ns;
3602 disk->queue = ns->queue;
3603 disk->flags = GENHD_FL_EXT_DEVT;
3605 * Without the multipath code enabled, multiple controller per
3606 * subsystems are visible as devices and thus we cannot use the
3607 * subsystem instance.
3609 if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3610 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3611 ns->head->instance);
3614 if (nvme_update_ns_info(ns, id))
3617 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3618 if (nvme_nvm_register(ns, disk->disk_name, node)) {
3619 dev_warn(ctrl->device, "LightNVM init failure\n");
3624 down_write(&ctrl->namespaces_rwsem);
3625 list_add_tail(&ns->list, &ctrl->namespaces);
3626 up_write(&ctrl->namespaces_rwsem);
3628 nvme_get_ctrl(ctrl);
3630 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3632 nvme_mpath_add_disk(ns, id);
3633 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3638 /* prevent double queue cleanup */
3639 ns->disk->queue = NULL;
3642 mutex_lock(&ctrl->subsys->lock);
3643 list_del_rcu(&ns->siblings);
3644 if (list_empty(&ns->head->list))
3645 list_del_init(&ns->head->entry);
3646 mutex_unlock(&ctrl->subsys->lock);
3647 nvme_put_ns_head(ns->head);
3649 blk_cleanup_queue(ns->queue);
3656 static void nvme_ns_remove(struct nvme_ns *ns)
3658 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3661 set_capacity(ns->disk, 0);
3662 nvme_fault_inject_fini(&ns->fault_inject);
3664 mutex_lock(&ns->ctrl->subsys->lock);
3665 list_del_rcu(&ns->siblings);
3666 if (list_empty(&ns->head->list))
3667 list_del_init(&ns->head->entry);
3668 mutex_unlock(&ns->ctrl->subsys->lock);
3670 synchronize_rcu(); /* guarantee not available in head->list */
3671 nvme_mpath_clear_current_path(ns);
3672 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3674 if (ns->disk->flags & GENHD_FL_UP) {
3675 del_gendisk(ns->disk);
3676 blk_cleanup_queue(ns->queue);
3677 if (blk_get_integrity(ns->disk))
3678 blk_integrity_unregister(ns->disk);
3681 down_write(&ns->ctrl->namespaces_rwsem);
3682 list_del_init(&ns->list);
3683 up_write(&ns->ctrl->namespaces_rwsem);
3685 nvme_mpath_check_last_path(ns);
3689 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3691 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3699 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3701 struct nvme_id_ns *id;
3702 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3704 if (test_bit(NVME_NS_DEAD, &ns->flags))
3707 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3711 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3712 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3713 dev_err(ns->ctrl->device,
3714 "identifiers changed for nsid %d\n", ns->head->ns_id);
3718 ret = nvme_update_ns_info(ns, id);
3724 * Only remove the namespace if we got a fatal error back from the
3725 * device, otherwise ignore the error and just move on.
3727 * TODO: we should probably schedule a delayed retry here.
3729 if (ret > 0 && (ret & NVME_SC_DNR))
3733 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3735 struct nvme_ns_ids ids = { };
3738 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3741 ns = nvme_find_get_ns(ctrl, nsid);
3743 nvme_validate_ns(ns, &ids);
3750 nvme_alloc_ns(ctrl, nsid, &ids);
3753 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
3754 dev_warn(ctrl->device,
3755 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
3759 if (!nvme_multi_css(ctrl)) {
3760 dev_warn(ctrl->device,
3761 "command set not reported for nsid: %d\n",
3765 nvme_alloc_ns(ctrl, nsid, &ids);
3768 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
3774 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3777 struct nvme_ns *ns, *next;
3780 down_write(&ctrl->namespaces_rwsem);
3781 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3782 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
3783 list_move_tail(&ns->list, &rm_list);
3785 up_write(&ctrl->namespaces_rwsem);
3787 list_for_each_entry_safe(ns, next, &rm_list, list)
3792 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3794 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3799 if (nvme_ctrl_limited_cns(ctrl))
3802 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3807 struct nvme_command cmd = {
3808 .identify.opcode = nvme_admin_identify,
3809 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
3810 .identify.nsid = cpu_to_le32(prev),
3813 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3814 NVME_IDENTIFY_DATA_SIZE);
3816 dev_warn(ctrl->device,
3817 "Identify NS List failed (status=0x%x)\n", ret);
3821 for (i = 0; i < nr_entries; i++) {
3822 u32 nsid = le32_to_cpu(ns_list[i]);
3824 if (!nsid) /* end of the list? */
3826 nvme_validate_or_alloc_ns(ctrl, nsid);
3827 while (++prev < nsid)
3828 nvme_ns_remove_by_nsid(ctrl, prev);
3832 nvme_remove_invalid_namespaces(ctrl, prev);
3838 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3840 struct nvme_id_ctrl *id;
3843 if (nvme_identify_ctrl(ctrl, &id))
3845 nn = le32_to_cpu(id->nn);
3848 for (i = 1; i <= nn; i++)
3849 nvme_validate_or_alloc_ns(ctrl, i);
3851 nvme_remove_invalid_namespaces(ctrl, nn);
3854 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3856 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3860 log = kzalloc(log_size, GFP_KERNEL);
3865 * We need to read the log to clear the AEN, but we don't want to rely
3866 * on it for the changed namespace information as userspace could have
3867 * raced with us in reading the log page, which could cause us to miss
3870 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3871 NVME_CSI_NVM, log, log_size, 0);
3873 dev_warn(ctrl->device,
3874 "reading changed ns log failed: %d\n", error);
3879 static void nvme_scan_work(struct work_struct *work)
3881 struct nvme_ctrl *ctrl =
3882 container_of(work, struct nvme_ctrl, scan_work);
3884 /* No tagset on a live ctrl means IO queues could not created */
3885 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3888 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3889 dev_info(ctrl->device, "rescanning namespaces.\n");
3890 nvme_clear_changed_ns_log(ctrl);
3893 mutex_lock(&ctrl->scan_lock);
3894 if (nvme_scan_ns_list(ctrl) != 0)
3895 nvme_scan_ns_sequential(ctrl);
3896 mutex_unlock(&ctrl->scan_lock);
3898 down_write(&ctrl->namespaces_rwsem);
3899 list_sort(NULL, &ctrl->namespaces, ns_cmp);
3900 up_write(&ctrl->namespaces_rwsem);
3904 * This function iterates the namespace list unlocked to allow recovery from
3905 * controller failure. It is up to the caller to ensure the namespace list is
3906 * not modified by scan work while this function is executing.
3908 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3910 struct nvme_ns *ns, *next;
3914 * make sure to requeue I/O to all namespaces as these
3915 * might result from the scan itself and must complete
3916 * for the scan_work to make progress
3918 nvme_mpath_clear_ctrl_paths(ctrl);
3920 /* prevent racing with ns scanning */
3921 flush_work(&ctrl->scan_work);
3924 * The dead states indicates the controller was not gracefully
3925 * disconnected. In that case, we won't be able to flush any data while
3926 * removing the namespaces' disks; fail all the queues now to avoid
3927 * potentially having to clean up the failed sync later.
3929 if (ctrl->state == NVME_CTRL_DEAD)
3930 nvme_kill_queues(ctrl);
3932 /* this is a no-op when called from the controller reset handler */
3933 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
3935 down_write(&ctrl->namespaces_rwsem);
3936 list_splice_init(&ctrl->namespaces, &ns_list);
3937 up_write(&ctrl->namespaces_rwsem);
3939 list_for_each_entry_safe(ns, next, &ns_list, list)
3942 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3944 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
3946 struct nvme_ctrl *ctrl =
3947 container_of(dev, struct nvme_ctrl, ctrl_device);
3948 struct nvmf_ctrl_options *opts = ctrl->opts;
3951 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
3956 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
3960 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
3961 opts->trsvcid ?: "none");
3965 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
3966 opts->host_traddr ?: "none");
3971 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
3973 char *envp[2] = { NULL, NULL };
3974 u32 aen_result = ctrl->aen_result;
3976 ctrl->aen_result = 0;
3980 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
3983 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
3987 static void nvme_async_event_work(struct work_struct *work)
3989 struct nvme_ctrl *ctrl =
3990 container_of(work, struct nvme_ctrl, async_event_work);
3992 nvme_aen_uevent(ctrl);
3993 ctrl->ops->submit_async_event(ctrl);
3996 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4001 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4007 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4010 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4012 struct nvme_fw_slot_info_log *log;
4014 log = kmalloc(sizeof(*log), GFP_KERNEL);
4018 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4019 log, sizeof(*log), 0))
4020 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4024 static void nvme_fw_act_work(struct work_struct *work)
4026 struct nvme_ctrl *ctrl = container_of(work,
4027 struct nvme_ctrl, fw_act_work);
4028 unsigned long fw_act_timeout;
4031 fw_act_timeout = jiffies +
4032 msecs_to_jiffies(ctrl->mtfa * 100);
4034 fw_act_timeout = jiffies +
4035 msecs_to_jiffies(admin_timeout * 1000);
4037 nvme_stop_queues(ctrl);
4038 while (nvme_ctrl_pp_status(ctrl)) {
4039 if (time_after(jiffies, fw_act_timeout)) {
4040 dev_warn(ctrl->device,
4041 "Fw activation timeout, reset controller\n");
4042 nvme_try_sched_reset(ctrl);
4048 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4051 nvme_start_queues(ctrl);
4052 /* read FW slot information to clear the AER */
4053 nvme_get_fw_slot_info(ctrl);
4056 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4058 u32 aer_notice_type = (result & 0xff00) >> 8;
4060 trace_nvme_async_event(ctrl, aer_notice_type);
4062 switch (aer_notice_type) {
4063 case NVME_AER_NOTICE_NS_CHANGED:
4064 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4065 nvme_queue_scan(ctrl);
4067 case NVME_AER_NOTICE_FW_ACT_STARTING:
4069 * We are (ab)using the RESETTING state to prevent subsequent
4070 * recovery actions from interfering with the controller's
4071 * firmware activation.
4073 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4074 queue_work(nvme_wq, &ctrl->fw_act_work);
4076 #ifdef CONFIG_NVME_MULTIPATH
4077 case NVME_AER_NOTICE_ANA:
4078 if (!ctrl->ana_log_buf)
4080 queue_work(nvme_wq, &ctrl->ana_work);
4083 case NVME_AER_NOTICE_DISC_CHANGED:
4084 ctrl->aen_result = result;
4087 dev_warn(ctrl->device, "async event result %08x\n", result);
4091 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4092 volatile union nvme_result *res)
4094 u32 result = le32_to_cpu(res->u32);
4095 u32 aer_type = result & 0x07;
4097 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4101 case NVME_AER_NOTICE:
4102 nvme_handle_aen_notice(ctrl, result);
4104 case NVME_AER_ERROR:
4105 case NVME_AER_SMART:
4108 trace_nvme_async_event(ctrl, aer_type);
4109 ctrl->aen_result = result;
4114 queue_work(nvme_wq, &ctrl->async_event_work);
4116 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4118 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4120 nvme_mpath_stop(ctrl);
4121 nvme_stop_keep_alive(ctrl);
4122 nvme_stop_failfast_work(ctrl);
4123 flush_work(&ctrl->async_event_work);
4124 cancel_work_sync(&ctrl->fw_act_work);
4126 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4128 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4130 nvme_start_keep_alive(ctrl);
4132 nvme_enable_aen(ctrl);
4134 if (ctrl->queue_count > 1) {
4135 nvme_queue_scan(ctrl);
4136 nvme_start_queues(ctrl);
4139 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4141 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4143 nvme_hwmon_exit(ctrl);
4144 nvme_fault_inject_fini(&ctrl->fault_inject);
4145 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4146 cdev_device_del(&ctrl->cdev, ctrl->device);
4147 nvme_put_ctrl(ctrl);
4149 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4151 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4153 struct nvme_effects_log *cel;
4156 xa_for_each(&ctrl->cels, i, cel) {
4157 xa_erase(&ctrl->cels, i);
4161 xa_destroy(&ctrl->cels);
4164 static void nvme_free_ctrl(struct device *dev)
4166 struct nvme_ctrl *ctrl =
4167 container_of(dev, struct nvme_ctrl, ctrl_device);
4168 struct nvme_subsystem *subsys = ctrl->subsys;
4170 if (!subsys || ctrl->instance != subsys->instance)
4171 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4173 nvme_free_cels(ctrl);
4174 nvme_mpath_uninit(ctrl);
4175 __free_page(ctrl->discard_page);
4178 mutex_lock(&nvme_subsystems_lock);
4179 list_del(&ctrl->subsys_entry);
4180 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4181 mutex_unlock(&nvme_subsystems_lock);
4184 ctrl->ops->free_ctrl(ctrl);
4187 nvme_put_subsystem(subsys);
4191 * Initialize a NVMe controller structures. This needs to be called during
4192 * earliest initialization so that we have the initialized structured around
4195 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4196 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4200 ctrl->state = NVME_CTRL_NEW;
4201 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4202 spin_lock_init(&ctrl->lock);
4203 mutex_init(&ctrl->scan_lock);
4204 INIT_LIST_HEAD(&ctrl->namespaces);
4205 xa_init(&ctrl->cels);
4206 init_rwsem(&ctrl->namespaces_rwsem);
4209 ctrl->quirks = quirks;
4210 ctrl->numa_node = NUMA_NO_NODE;
4211 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4212 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4213 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4214 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4215 init_waitqueue_head(&ctrl->state_wq);
4217 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4218 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4219 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4220 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4222 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4224 ctrl->discard_page = alloc_page(GFP_KERNEL);
4225 if (!ctrl->discard_page) {
4230 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4233 ctrl->instance = ret;
4235 device_initialize(&ctrl->ctrl_device);
4236 ctrl->device = &ctrl->ctrl_device;
4237 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4239 ctrl->device->class = nvme_class;
4240 ctrl->device->parent = ctrl->dev;
4241 ctrl->device->groups = nvme_dev_attr_groups;
4242 ctrl->device->release = nvme_free_ctrl;
4243 dev_set_drvdata(ctrl->device, ctrl);
4244 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4246 goto out_release_instance;
4248 nvme_get_ctrl(ctrl);
4249 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4250 ctrl->cdev.owner = ops->module;
4251 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4256 * Initialize latency tolerance controls. The sysfs files won't
4257 * be visible to userspace unless the device actually supports APST.
4259 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4260 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4261 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4263 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4267 nvme_put_ctrl(ctrl);
4268 kfree_const(ctrl->device->kobj.name);
4269 out_release_instance:
4270 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4272 if (ctrl->discard_page)
4273 __free_page(ctrl->discard_page);
4276 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4279 * nvme_kill_queues(): Ends all namespace queues
4280 * @ctrl: the dead controller that needs to end
4282 * Call this function when the driver determines it is unable to get the
4283 * controller in a state capable of servicing IO.
4285 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4289 down_read(&ctrl->namespaces_rwsem);
4291 /* Forcibly unquiesce queues to avoid blocking dispatch */
4292 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4293 blk_mq_unquiesce_queue(ctrl->admin_q);
4295 list_for_each_entry(ns, &ctrl->namespaces, list)
4296 nvme_set_queue_dying(ns);
4298 up_read(&ctrl->namespaces_rwsem);
4300 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4302 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4306 down_read(&ctrl->namespaces_rwsem);
4307 list_for_each_entry(ns, &ctrl->namespaces, list)
4308 blk_mq_unfreeze_queue(ns->queue);
4309 up_read(&ctrl->namespaces_rwsem);
4311 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4313 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4317 down_read(&ctrl->namespaces_rwsem);
4318 list_for_each_entry(ns, &ctrl->namespaces, list) {
4319 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4323 up_read(&ctrl->namespaces_rwsem);
4326 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4328 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4332 down_read(&ctrl->namespaces_rwsem);
4333 list_for_each_entry(ns, &ctrl->namespaces, list)
4334 blk_mq_freeze_queue_wait(ns->queue);
4335 up_read(&ctrl->namespaces_rwsem);
4337 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4339 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4343 down_read(&ctrl->namespaces_rwsem);
4344 list_for_each_entry(ns, &ctrl->namespaces, list)
4345 blk_freeze_queue_start(ns->queue);
4346 up_read(&ctrl->namespaces_rwsem);
4348 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4350 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4354 down_read(&ctrl->namespaces_rwsem);
4355 list_for_each_entry(ns, &ctrl->namespaces, list)
4356 blk_mq_quiesce_queue(ns->queue);
4357 up_read(&ctrl->namespaces_rwsem);
4359 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4361 void nvme_start_queues(struct nvme_ctrl *ctrl)
4365 down_read(&ctrl->namespaces_rwsem);
4366 list_for_each_entry(ns, &ctrl->namespaces, list)
4367 blk_mq_unquiesce_queue(ns->queue);
4368 up_read(&ctrl->namespaces_rwsem);
4370 EXPORT_SYMBOL_GPL(nvme_start_queues);
4372 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4376 down_read(&ctrl->namespaces_rwsem);
4377 list_for_each_entry(ns, &ctrl->namespaces, list)
4378 blk_sync_queue(ns->queue);
4379 up_read(&ctrl->namespaces_rwsem);
4381 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4383 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4385 nvme_sync_io_queues(ctrl);
4387 blk_sync_queue(ctrl->admin_q);
4389 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4391 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4393 if (file->f_op != &nvme_dev_fops)
4395 return file->private_data;
4397 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4400 * Check we didn't inadvertently grow the command structure sizes:
4402 static inline void _nvme_check_size(void)
4404 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4405 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4406 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4407 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4408 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4409 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4410 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4411 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4412 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4413 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4414 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4415 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4416 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4417 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4418 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4419 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4420 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4421 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4422 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4423 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4427 static int __init nvme_core_init(void)
4429 int result = -ENOMEM;
4433 nvme_wq = alloc_workqueue("nvme-wq",
4434 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4438 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4439 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4443 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4444 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4445 if (!nvme_delete_wq)
4446 goto destroy_reset_wq;
4448 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4449 NVME_MINORS, "nvme");
4451 goto destroy_delete_wq;
4453 nvme_class = class_create(THIS_MODULE, "nvme");
4454 if (IS_ERR(nvme_class)) {
4455 result = PTR_ERR(nvme_class);
4456 goto unregister_chrdev;
4458 nvme_class->dev_uevent = nvme_class_uevent;
4460 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4461 if (IS_ERR(nvme_subsys_class)) {
4462 result = PTR_ERR(nvme_subsys_class);
4468 class_destroy(nvme_class);
4470 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4472 destroy_workqueue(nvme_delete_wq);
4474 destroy_workqueue(nvme_reset_wq);
4476 destroy_workqueue(nvme_wq);
4481 static void __exit nvme_core_exit(void)
4483 class_destroy(nvme_subsys_class);
4484 class_destroy(nvme_class);
4485 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4486 destroy_workqueue(nvme_delete_wq);
4487 destroy_workqueue(nvme_reset_wq);
4488 destroy_workqueue(nvme_wq);
4489 ida_destroy(&nvme_instance_ida);
4492 MODULE_LICENSE("GPL");
4493 MODULE_VERSION("1.0");
4494 module_init(nvme_core_init);
4495 module_exit(nvme_core_exit);