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 static 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 static 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 blk_mq_complete_request(req);
386 EXPORT_SYMBOL_GPL(nvme_cancel_request);
388 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
391 blk_mq_tagset_busy_iter(ctrl->tagset,
392 nvme_cancel_request, ctrl);
393 blk_mq_tagset_wait_completed_request(ctrl->tagset);
396 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
398 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
400 if (ctrl->admin_tagset) {
401 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
402 nvme_cancel_request, ctrl);
403 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
406 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
408 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
409 enum nvme_ctrl_state new_state)
411 enum nvme_ctrl_state old_state;
413 bool changed = false;
415 spin_lock_irqsave(&ctrl->lock, flags);
417 old_state = ctrl->state;
422 case NVME_CTRL_RESETTING:
423 case NVME_CTRL_CONNECTING:
430 case NVME_CTRL_RESETTING:
440 case NVME_CTRL_CONNECTING:
443 case NVME_CTRL_RESETTING:
450 case NVME_CTRL_DELETING:
453 case NVME_CTRL_RESETTING:
454 case NVME_CTRL_CONNECTING:
461 case NVME_CTRL_DELETING_NOIO:
463 case NVME_CTRL_DELETING:
473 case NVME_CTRL_DELETING:
485 ctrl->state = new_state;
486 wake_up_all(&ctrl->state_wq);
489 spin_unlock_irqrestore(&ctrl->lock, flags);
493 if (ctrl->state == NVME_CTRL_LIVE) {
494 if (old_state == NVME_CTRL_CONNECTING)
495 nvme_stop_failfast_work(ctrl);
496 nvme_kick_requeue_lists(ctrl);
497 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
498 old_state == NVME_CTRL_RESETTING) {
499 nvme_start_failfast_work(ctrl);
503 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
506 * Returns true for sink states that can't ever transition back to live.
508 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
510 switch (ctrl->state) {
513 case NVME_CTRL_RESETTING:
514 case NVME_CTRL_CONNECTING:
516 case NVME_CTRL_DELETING:
517 case NVME_CTRL_DELETING_NOIO:
521 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
527 * Waits for the controller state to be resetting, or returns false if it is
528 * not possible to ever transition to that state.
530 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
532 wait_event(ctrl->state_wq,
533 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
534 nvme_state_terminal(ctrl));
535 return ctrl->state == NVME_CTRL_RESETTING;
537 EXPORT_SYMBOL_GPL(nvme_wait_reset);
539 static void nvme_free_ns_head(struct kref *ref)
541 struct nvme_ns_head *head =
542 container_of(ref, struct nvme_ns_head, ref);
544 nvme_mpath_remove_disk(head);
545 ida_simple_remove(&head->subsys->ns_ida, head->instance);
546 cleanup_srcu_struct(&head->srcu);
547 nvme_put_subsystem(head->subsys);
551 static void nvme_put_ns_head(struct nvme_ns_head *head)
553 kref_put(&head->ref, nvme_free_ns_head);
556 static void nvme_free_ns(struct kref *kref)
558 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
561 nvme_nvm_unregister(ns);
564 nvme_put_ns_head(ns->head);
565 nvme_put_ctrl(ns->ctrl);
569 void nvme_put_ns(struct nvme_ns *ns)
571 kref_put(&ns->kref, nvme_free_ns);
573 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
575 static inline void nvme_clear_nvme_request(struct request *req)
577 if (!(req->rq_flags & RQF_DONTPREP)) {
578 nvme_req(req)->retries = 0;
579 nvme_req(req)->flags = 0;
580 req->rq_flags |= RQF_DONTPREP;
584 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
586 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
589 static inline void nvme_init_request(struct request *req,
590 struct nvme_command *cmd)
592 if (req->q->queuedata)
593 req->timeout = NVME_IO_TIMEOUT;
594 else /* no queuedata implies admin queue */
595 req->timeout = NVME_ADMIN_TIMEOUT;
597 req->cmd_flags |= REQ_FAILFAST_DRIVER;
598 nvme_clear_nvme_request(req);
599 nvme_req(req)->cmd = cmd;
602 struct request *nvme_alloc_request(struct request_queue *q,
603 struct nvme_command *cmd, blk_mq_req_flags_t flags)
607 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
609 nvme_init_request(req, cmd);
612 EXPORT_SYMBOL_GPL(nvme_alloc_request);
614 static struct request *nvme_alloc_request_qid(struct request_queue *q,
615 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
619 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
622 nvme_init_request(req, cmd);
626 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
628 struct nvme_command c;
630 memset(&c, 0, sizeof(c));
632 c.directive.opcode = nvme_admin_directive_send;
633 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
634 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
635 c.directive.dtype = NVME_DIR_IDENTIFY;
636 c.directive.tdtype = NVME_DIR_STREAMS;
637 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
639 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
642 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
644 return nvme_toggle_streams(ctrl, false);
647 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
649 return nvme_toggle_streams(ctrl, true);
652 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
653 struct streams_directive_params *s, u32 nsid)
655 struct nvme_command c;
657 memset(&c, 0, sizeof(c));
658 memset(s, 0, sizeof(*s));
660 c.directive.opcode = nvme_admin_directive_recv;
661 c.directive.nsid = cpu_to_le32(nsid);
662 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
663 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
664 c.directive.dtype = NVME_DIR_STREAMS;
666 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
669 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
671 struct streams_directive_params s;
674 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
679 ret = nvme_enable_streams(ctrl);
683 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
685 goto out_disable_stream;
687 ctrl->nssa = le16_to_cpu(s.nssa);
688 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
689 dev_info(ctrl->device, "too few streams (%u) available\n",
691 goto out_disable_stream;
694 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
695 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
699 nvme_disable_streams(ctrl);
704 * Check if 'req' has a write hint associated with it. If it does, assign
705 * a valid namespace stream to the write.
707 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
708 struct request *req, u16 *control,
711 enum rw_hint streamid = req->write_hint;
713 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
717 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
720 *control |= NVME_RW_DTYPE_STREAMS;
721 *dsmgmt |= streamid << 16;
724 if (streamid < ARRAY_SIZE(req->q->write_hints))
725 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
728 static void nvme_setup_passthrough(struct request *req,
729 struct nvme_command *cmd)
731 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
732 /* passthru commands should let the driver set the SGL flags */
733 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
736 static inline void nvme_setup_flush(struct nvme_ns *ns,
737 struct nvme_command *cmnd)
739 cmnd->common.opcode = nvme_cmd_flush;
740 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
743 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
744 struct nvme_command *cmnd)
746 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
747 struct nvme_dsm_range *range;
751 * Some devices do not consider the DSM 'Number of Ranges' field when
752 * determining how much data to DMA. Always allocate memory for maximum
753 * number of segments to prevent device reading beyond end of buffer.
755 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
757 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
760 * If we fail allocation our range, fallback to the controller
761 * discard page. If that's also busy, it's safe to return
762 * busy, as we know we can make progress once that's freed.
764 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
765 return BLK_STS_RESOURCE;
767 range = page_address(ns->ctrl->discard_page);
770 __rq_for_each_bio(bio, req) {
771 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
772 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
775 range[n].cattr = cpu_to_le32(0);
776 range[n].nlb = cpu_to_le32(nlb);
777 range[n].slba = cpu_to_le64(slba);
782 if (WARN_ON_ONCE(n != segments)) {
783 if (virt_to_page(range) == ns->ctrl->discard_page)
784 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
787 return BLK_STS_IOERR;
790 cmnd->dsm.opcode = nvme_cmd_dsm;
791 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
792 cmnd->dsm.nr = cpu_to_le32(segments - 1);
793 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
795 req->special_vec.bv_page = virt_to_page(range);
796 req->special_vec.bv_offset = offset_in_page(range);
797 req->special_vec.bv_len = alloc_size;
798 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
803 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
804 struct request *req, struct nvme_command *cmnd)
806 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
807 return nvme_setup_discard(ns, req, cmnd);
809 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
810 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
811 cmnd->write_zeroes.slba =
812 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
813 cmnd->write_zeroes.length =
814 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
815 cmnd->write_zeroes.control = 0;
819 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
820 struct request *req, struct nvme_command *cmnd,
823 struct nvme_ctrl *ctrl = ns->ctrl;
827 if (req->cmd_flags & REQ_FUA)
828 control |= NVME_RW_FUA;
829 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
830 control |= NVME_RW_LR;
832 if (req->cmd_flags & REQ_RAHEAD)
833 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
835 cmnd->rw.opcode = op;
836 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
837 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
838 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
840 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
841 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
845 * If formated with metadata, the block layer always provides a
846 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
847 * we enable the PRACT bit for protection information or set the
848 * namespace capacity to zero to prevent any I/O.
850 if (!blk_integrity_rq(req)) {
851 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
852 return BLK_STS_NOTSUPP;
853 control |= NVME_RW_PRINFO_PRACT;
856 switch (ns->pi_type) {
857 case NVME_NS_DPS_PI_TYPE3:
858 control |= NVME_RW_PRINFO_PRCHK_GUARD;
860 case NVME_NS_DPS_PI_TYPE1:
861 case NVME_NS_DPS_PI_TYPE2:
862 control |= NVME_RW_PRINFO_PRCHK_GUARD |
863 NVME_RW_PRINFO_PRCHK_REF;
864 if (op == nvme_cmd_zone_append)
865 control |= NVME_RW_APPEND_PIREMAP;
866 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
871 cmnd->rw.control = cpu_to_le16(control);
872 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
876 void nvme_cleanup_cmd(struct request *req)
878 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
879 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
880 struct page *page = req->special_vec.bv_page;
882 if (page == ctrl->discard_page)
883 clear_bit_unlock(0, &ctrl->discard_page_busy);
885 kfree(page_address(page) + req->special_vec.bv_offset);
888 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
890 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
891 struct nvme_command *cmd)
893 blk_status_t ret = BLK_STS_OK;
895 nvme_clear_nvme_request(req);
897 memset(cmd, 0, sizeof(*cmd));
898 switch (req_op(req)) {
901 nvme_setup_passthrough(req, cmd);
904 nvme_setup_flush(ns, cmd);
906 case REQ_OP_ZONE_RESET_ALL:
907 case REQ_OP_ZONE_RESET:
908 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
910 case REQ_OP_ZONE_OPEN:
911 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
913 case REQ_OP_ZONE_CLOSE:
914 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
916 case REQ_OP_ZONE_FINISH:
917 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
919 case REQ_OP_WRITE_ZEROES:
920 ret = nvme_setup_write_zeroes(ns, req, cmd);
923 ret = nvme_setup_discard(ns, req, cmd);
926 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
929 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
931 case REQ_OP_ZONE_APPEND:
932 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
936 return BLK_STS_IOERR;
939 cmd->common.command_id = req->tag;
940 trace_nvme_setup_cmd(req, cmd);
943 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
945 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
947 struct completion *waiting = rq->end_io_data;
949 rq->end_io_data = NULL;
953 static void nvme_execute_rq_polled(struct request_queue *q,
954 struct gendisk *bd_disk, struct request *rq, int at_head)
956 DECLARE_COMPLETION_ONSTACK(wait);
958 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
960 rq->cmd_flags |= REQ_HIPRI;
961 rq->end_io_data = &wait;
962 blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
964 while (!completion_done(&wait)) {
965 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
971 * Returns 0 on success. If the result is negative, it's a Linux error code;
972 * if the result is positive, it's an NVM Express status code
974 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
975 union nvme_result *result, void *buffer, unsigned bufflen,
976 unsigned timeout, int qid, int at_head,
977 blk_mq_req_flags_t flags, bool poll)
982 if (qid == NVME_QID_ANY)
983 req = nvme_alloc_request(q, cmd, flags);
985 req = nvme_alloc_request_qid(q, cmd, flags, qid);
990 req->timeout = timeout;
992 if (buffer && bufflen) {
993 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
999 nvme_execute_rq_polled(req->q, NULL, req, at_head);
1001 blk_execute_rq(NULL, req, at_head);
1003 *result = nvme_req(req)->result;
1004 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1007 ret = nvme_req(req)->status;
1009 blk_mq_free_request(req);
1012 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1014 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1015 void *buffer, unsigned bufflen)
1017 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1018 NVME_QID_ANY, 0, 0, false);
1020 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1022 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1023 unsigned len, u32 seed, bool write)
1025 struct bio_integrity_payload *bip;
1029 buf = kmalloc(len, GFP_KERNEL);
1034 if (write && copy_from_user(buf, ubuf, len))
1037 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1043 bip->bip_iter.bi_size = len;
1044 bip->bip_iter.bi_sector = seed;
1045 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1046 offset_in_page(buf));
1053 return ERR_PTR(ret);
1056 static u32 nvme_known_admin_effects(u8 opcode)
1059 case nvme_admin_format_nvm:
1060 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1061 NVME_CMD_EFFECTS_CSE_MASK;
1062 case nvme_admin_sanitize_nvm:
1063 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1070 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1075 if (ns->head->effects)
1076 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1077 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1078 dev_warn(ctrl->device,
1079 "IO command:%02x has unhandled effects:%08x\n",
1085 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1086 effects |= nvme_known_admin_effects(opcode);
1090 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1092 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1095 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1098 * For simplicity, IO to all namespaces is quiesced even if the command
1099 * effects say only one namespace is affected.
1101 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1102 mutex_lock(&ctrl->scan_lock);
1103 mutex_lock(&ctrl->subsys->lock);
1104 nvme_mpath_start_freeze(ctrl->subsys);
1105 nvme_mpath_wait_freeze(ctrl->subsys);
1106 nvme_start_freeze(ctrl);
1107 nvme_wait_freeze(ctrl);
1112 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1114 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1115 nvme_unfreeze(ctrl);
1116 nvme_mpath_unfreeze(ctrl->subsys);
1117 mutex_unlock(&ctrl->subsys->lock);
1118 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1119 mutex_unlock(&ctrl->scan_lock);
1121 if (effects & NVME_CMD_EFFECTS_CCC)
1122 nvme_init_identify(ctrl);
1123 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1124 nvme_queue_scan(ctrl);
1125 flush_work(&ctrl->scan_work);
1129 void nvme_execute_passthru_rq(struct request *rq)
1131 struct nvme_command *cmd = nvme_req(rq)->cmd;
1132 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1133 struct nvme_ns *ns = rq->q->queuedata;
1134 struct gendisk *disk = ns ? ns->disk : NULL;
1137 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1138 blk_execute_rq(disk, rq, 0);
1139 nvme_passthru_end(ctrl, effects);
1141 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1143 static int nvme_submit_user_cmd(struct request_queue *q,
1144 struct nvme_command *cmd, void __user *ubuffer,
1145 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1146 u32 meta_seed, u64 *result, unsigned timeout)
1148 bool write = nvme_is_write(cmd);
1149 struct nvme_ns *ns = q->queuedata;
1150 struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1151 struct request *req;
1152 struct bio *bio = NULL;
1156 req = nvme_alloc_request(q, cmd, 0);
1158 return PTR_ERR(req);
1161 req->timeout = timeout;
1162 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1164 if (ubuffer && bufflen) {
1165 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1171 bio_set_dev(bio, bdev);
1172 if (bdev && meta_buffer && meta_len) {
1173 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1176 ret = PTR_ERR(meta);
1179 req->cmd_flags |= REQ_INTEGRITY;
1183 nvme_execute_passthru_rq(req);
1184 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1187 ret = nvme_req(req)->status;
1189 *result = le64_to_cpu(nvme_req(req)->result.u64);
1190 if (meta && !ret && !write) {
1191 if (copy_to_user(meta_buffer, meta, meta_len))
1197 blk_rq_unmap_user(bio);
1199 blk_mq_free_request(req);
1203 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1205 struct nvme_ctrl *ctrl = rq->end_io_data;
1206 unsigned long flags;
1207 bool startka = false;
1209 blk_mq_free_request(rq);
1212 dev_err(ctrl->device,
1213 "failed nvme_keep_alive_end_io error=%d\n",
1218 ctrl->comp_seen = false;
1219 spin_lock_irqsave(&ctrl->lock, flags);
1220 if (ctrl->state == NVME_CTRL_LIVE ||
1221 ctrl->state == NVME_CTRL_CONNECTING)
1223 spin_unlock_irqrestore(&ctrl->lock, flags);
1225 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1228 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1232 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1233 BLK_MQ_REQ_RESERVED);
1237 rq->timeout = ctrl->kato * HZ;
1238 rq->end_io_data = ctrl;
1240 blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1245 static void nvme_keep_alive_work(struct work_struct *work)
1247 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1248 struct nvme_ctrl, ka_work);
1249 bool comp_seen = ctrl->comp_seen;
1251 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1252 dev_dbg(ctrl->device,
1253 "reschedule traffic based keep-alive timer\n");
1254 ctrl->comp_seen = false;
1255 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1259 if (nvme_keep_alive(ctrl)) {
1260 /* allocation failure, reset the controller */
1261 dev_err(ctrl->device, "keep-alive failed\n");
1262 nvme_reset_ctrl(ctrl);
1267 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1269 if (unlikely(ctrl->kato == 0))
1272 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1275 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1277 if (unlikely(ctrl->kato == 0))
1280 cancel_delayed_work_sync(&ctrl->ka_work);
1282 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1285 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1286 * flag, thus sending any new CNS opcodes has a big chance of not working.
1287 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1288 * (but not for any later version).
1290 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1292 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1293 return ctrl->vs < NVME_VS(1, 2, 0);
1294 return ctrl->vs < NVME_VS(1, 1, 0);
1297 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1299 struct nvme_command c = { };
1302 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1303 c.identify.opcode = nvme_admin_identify;
1304 c.identify.cns = NVME_ID_CNS_CTRL;
1306 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1310 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1311 sizeof(struct nvme_id_ctrl));
1317 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1319 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1322 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1323 struct nvme_ns_id_desc *cur, bool *csi_seen)
1325 const char *warn_str = "ctrl returned bogus length:";
1328 switch (cur->nidt) {
1329 case NVME_NIDT_EUI64:
1330 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1331 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1332 warn_str, cur->nidl);
1335 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1336 return NVME_NIDT_EUI64_LEN;
1337 case NVME_NIDT_NGUID:
1338 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1339 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1340 warn_str, cur->nidl);
1343 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1344 return NVME_NIDT_NGUID_LEN;
1345 case NVME_NIDT_UUID:
1346 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1347 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1348 warn_str, cur->nidl);
1351 uuid_copy(&ids->uuid, data + sizeof(*cur));
1352 return NVME_NIDT_UUID_LEN;
1354 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1355 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1356 warn_str, cur->nidl);
1359 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1361 return NVME_NIDT_CSI_LEN;
1363 /* Skip unknown types */
1368 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1369 struct nvme_ns_ids *ids)
1371 struct nvme_command c = { };
1372 bool csi_seen = false;
1373 int status, pos, len;
1376 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1378 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1381 c.identify.opcode = nvme_admin_identify;
1382 c.identify.nsid = cpu_to_le32(nsid);
1383 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1385 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1389 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1390 NVME_IDENTIFY_DATA_SIZE);
1392 dev_warn(ctrl->device,
1393 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1398 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1399 struct nvme_ns_id_desc *cur = data + pos;
1404 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1408 len += sizeof(*cur);
1411 if (nvme_multi_css(ctrl) && !csi_seen) {
1412 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1422 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1423 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1425 struct nvme_command c = { };
1428 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1429 c.identify.opcode = nvme_admin_identify;
1430 c.identify.nsid = cpu_to_le32(nsid);
1431 c.identify.cns = NVME_ID_CNS_NS;
1433 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1437 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1439 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1444 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1447 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1448 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1449 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1450 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1451 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1452 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1461 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1462 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1464 union nvme_result res = { 0 };
1465 struct nvme_command c;
1468 memset(&c, 0, sizeof(c));
1469 c.features.opcode = op;
1470 c.features.fid = cpu_to_le32(fid);
1471 c.features.dword11 = cpu_to_le32(dword11);
1473 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1474 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1475 if (ret >= 0 && result)
1476 *result = le32_to_cpu(res.u32);
1480 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1481 unsigned int dword11, void *buffer, size_t buflen,
1484 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1487 EXPORT_SYMBOL_GPL(nvme_set_features);
1489 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1490 unsigned int dword11, void *buffer, size_t buflen,
1493 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1496 EXPORT_SYMBOL_GPL(nvme_get_features);
1498 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1500 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1502 int status, nr_io_queues;
1504 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1510 * Degraded controllers might return an error when setting the queue
1511 * count. We still want to be able to bring them online and offer
1512 * access to the admin queue, as that might be only way to fix them up.
1515 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1518 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1519 *count = min(*count, nr_io_queues);
1524 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1526 #define NVME_AEN_SUPPORTED \
1527 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1528 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1530 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1532 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1535 if (!supported_aens)
1538 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1541 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1544 queue_work(nvme_wq, &ctrl->async_event_work);
1548 * Convert integer values from ioctl structures to user pointers, silently
1549 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1552 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1554 if (in_compat_syscall())
1555 ptrval = (compat_uptr_t)ptrval;
1556 return (void __user *)ptrval;
1559 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1561 struct nvme_user_io io;
1562 struct nvme_command c;
1563 unsigned length, meta_len;
1564 void __user *metadata;
1566 if (copy_from_user(&io, uio, sizeof(io)))
1571 switch (io.opcode) {
1572 case nvme_cmd_write:
1574 case nvme_cmd_compare:
1580 length = (io.nblocks + 1) << ns->lba_shift;
1582 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1583 ns->ms == sizeof(struct t10_pi_tuple)) {
1585 * Protection information is stripped/inserted by the
1588 if (nvme_to_user_ptr(io.metadata))
1593 meta_len = (io.nblocks + 1) * ns->ms;
1594 metadata = nvme_to_user_ptr(io.metadata);
1597 if (ns->features & NVME_NS_EXT_LBAS) {
1600 } else if (meta_len) {
1601 if ((io.metadata & 3) || !io.metadata)
1605 memset(&c, 0, sizeof(c));
1606 c.rw.opcode = io.opcode;
1607 c.rw.flags = io.flags;
1608 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1609 c.rw.slba = cpu_to_le64(io.slba);
1610 c.rw.length = cpu_to_le16(io.nblocks);
1611 c.rw.control = cpu_to_le16(io.control);
1612 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1613 c.rw.reftag = cpu_to_le32(io.reftag);
1614 c.rw.apptag = cpu_to_le16(io.apptag);
1615 c.rw.appmask = cpu_to_le16(io.appmask);
1617 return nvme_submit_user_cmd(ns->queue, &c,
1618 nvme_to_user_ptr(io.addr), length,
1619 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1622 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1623 struct nvme_passthru_cmd __user *ucmd)
1625 struct nvme_passthru_cmd cmd;
1626 struct nvme_command c;
1627 unsigned timeout = 0;
1631 if (!capable(CAP_SYS_ADMIN))
1633 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1638 memset(&c, 0, sizeof(c));
1639 c.common.opcode = cmd.opcode;
1640 c.common.flags = cmd.flags;
1641 c.common.nsid = cpu_to_le32(cmd.nsid);
1642 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1643 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1644 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1645 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1646 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1647 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1648 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1649 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1652 timeout = msecs_to_jiffies(cmd.timeout_ms);
1654 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1655 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1656 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1657 0, &result, timeout);
1660 if (put_user(result, &ucmd->result))
1667 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1668 struct nvme_passthru_cmd64 __user *ucmd)
1670 struct nvme_passthru_cmd64 cmd;
1671 struct nvme_command c;
1672 unsigned timeout = 0;
1675 if (!capable(CAP_SYS_ADMIN))
1677 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1682 memset(&c, 0, sizeof(c));
1683 c.common.opcode = cmd.opcode;
1684 c.common.flags = cmd.flags;
1685 c.common.nsid = cpu_to_le32(cmd.nsid);
1686 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1687 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1688 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1689 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1690 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1691 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1692 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1693 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1696 timeout = msecs_to_jiffies(cmd.timeout_ms);
1698 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1699 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1700 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1701 0, &cmd.result, timeout);
1704 if (put_user(cmd.result, &ucmd->result))
1712 * Issue ioctl requests on the first available path. Note that unlike normal
1713 * block layer requests we will not retry failed request on another controller.
1715 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1716 struct nvme_ns_head **head, int *srcu_idx)
1718 #ifdef CONFIG_NVME_MULTIPATH
1719 if (disk->fops == &nvme_ns_head_ops) {
1722 *head = disk->private_data;
1723 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1724 ns = nvme_find_path(*head);
1726 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1732 return disk->private_data;
1735 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1738 srcu_read_unlock(&head->srcu, idx);
1741 static bool is_ctrl_ioctl(unsigned int cmd)
1743 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1745 if (is_sed_ioctl(cmd))
1750 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1752 struct nvme_ns_head *head,
1755 struct nvme_ctrl *ctrl = ns->ctrl;
1758 nvme_get_ctrl(ns->ctrl);
1759 nvme_put_ns_from_disk(head, srcu_idx);
1762 case NVME_IOCTL_ADMIN_CMD:
1763 ret = nvme_user_cmd(ctrl, NULL, argp);
1765 case NVME_IOCTL_ADMIN64_CMD:
1766 ret = nvme_user_cmd64(ctrl, NULL, argp);
1769 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1772 nvme_put_ctrl(ctrl);
1776 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1777 unsigned int cmd, unsigned long arg)
1779 struct nvme_ns_head *head = NULL;
1780 void __user *argp = (void __user *)arg;
1784 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1786 return -EWOULDBLOCK;
1789 * Handle ioctls that apply to the controller instead of the namespace
1790 * seperately and drop the ns SRCU reference early. This avoids a
1791 * deadlock when deleting namespaces using the passthrough interface.
1793 if (is_ctrl_ioctl(cmd))
1794 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1798 force_successful_syscall_return();
1799 ret = ns->head->ns_id;
1801 case NVME_IOCTL_IO_CMD:
1802 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1804 case NVME_IOCTL_SUBMIT_IO:
1805 ret = nvme_submit_io(ns, argp);
1807 case NVME_IOCTL_IO64_CMD:
1808 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1812 ret = nvme_nvm_ioctl(ns, cmd, arg);
1817 nvme_put_ns_from_disk(head, srcu_idx);
1821 #ifdef CONFIG_COMPAT
1822 struct nvme_user_io32 {
1835 } __attribute__((__packed__));
1837 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1839 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1840 unsigned int cmd, unsigned long arg)
1843 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1844 * between 32 bit programs and 64 bit kernel.
1845 * The cause is that the results of sizeof(struct nvme_user_io),
1846 * which is used to define NVME_IOCTL_SUBMIT_IO,
1847 * are not same between 32 bit compiler and 64 bit compiler.
1848 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1849 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1850 * Other IOCTL numbers are same between 32 bit and 64 bit.
1851 * So there is nothing to do regarding to other IOCTL numbers.
1853 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1854 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1856 return nvme_ioctl(bdev, mode, cmd, arg);
1859 #define nvme_compat_ioctl NULL
1860 #endif /* CONFIG_COMPAT */
1862 static int nvme_open(struct block_device *bdev, fmode_t mode)
1864 struct nvme_ns *ns = bdev->bd_disk->private_data;
1866 #ifdef CONFIG_NVME_MULTIPATH
1867 /* should never be called due to GENHD_FL_HIDDEN */
1868 if (WARN_ON_ONCE(ns->head->disk))
1871 if (!kref_get_unless_zero(&ns->kref))
1873 if (!try_module_get(ns->ctrl->ops->module))
1884 static void nvme_release(struct gendisk *disk, fmode_t mode)
1886 struct nvme_ns *ns = disk->private_data;
1888 module_put(ns->ctrl->ops->module);
1892 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1894 /* some standard values */
1895 geo->heads = 1 << 6;
1896 geo->sectors = 1 << 5;
1897 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1901 #ifdef CONFIG_BLK_DEV_INTEGRITY
1902 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1903 u32 max_integrity_segments)
1905 struct blk_integrity integrity;
1907 memset(&integrity, 0, sizeof(integrity));
1909 case NVME_NS_DPS_PI_TYPE3:
1910 integrity.profile = &t10_pi_type3_crc;
1911 integrity.tag_size = sizeof(u16) + sizeof(u32);
1912 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1914 case NVME_NS_DPS_PI_TYPE1:
1915 case NVME_NS_DPS_PI_TYPE2:
1916 integrity.profile = &t10_pi_type1_crc;
1917 integrity.tag_size = sizeof(u16);
1918 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1921 integrity.profile = NULL;
1924 integrity.tuple_size = ms;
1925 blk_integrity_register(disk, &integrity);
1926 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1929 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1930 u32 max_integrity_segments)
1933 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1935 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1937 struct nvme_ctrl *ctrl = ns->ctrl;
1938 struct request_queue *queue = disk->queue;
1939 u32 size = queue_logical_block_size(queue);
1941 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1942 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1946 if (ctrl->nr_streams && ns->sws && ns->sgs)
1947 size *= ns->sws * ns->sgs;
1949 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1950 NVME_DSM_MAX_RANGES);
1952 queue->limits.discard_alignment = 0;
1953 queue->limits.discard_granularity = size;
1955 /* If discard is already enabled, don't reset queue limits */
1956 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1959 blk_queue_max_discard_sectors(queue, UINT_MAX);
1960 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1962 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1963 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1966 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1970 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1971 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1974 * Even though NVMe spec explicitly states that MDTS is not
1975 * applicable to the write-zeroes:- "The restriction does not apply to
1976 * commands that do not transfer data between the host and the
1977 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1978 * In order to be more cautious use controller's max_hw_sectors value
1979 * to configure the maximum sectors for the write-zeroes which is
1980 * configured based on the controller's MDTS field in the
1981 * nvme_init_identify() if available.
1983 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1984 max_blocks = (u64)USHRT_MAX + 1;
1986 max_blocks = ns->ctrl->max_hw_sectors + 1;
1988 blk_queue_max_write_zeroes_sectors(disk->queue,
1989 nvme_lba_to_sect(ns, max_blocks));
1992 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1994 return !uuid_is_null(&ids->uuid) ||
1995 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1996 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1999 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
2001 return uuid_equal(&a->uuid, &b->uuid) &&
2002 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
2003 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
2007 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
2008 u32 *phys_bs, u32 *io_opt)
2010 struct streams_directive_params s;
2013 if (!ctrl->nr_streams)
2016 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2020 ns->sws = le32_to_cpu(s.sws);
2021 ns->sgs = le16_to_cpu(s.sgs);
2024 *phys_bs = ns->sws * (1 << ns->lba_shift);
2026 *io_opt = *phys_bs * ns->sgs;
2032 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2034 struct nvme_ctrl *ctrl = ns->ctrl;
2037 * The PI implementation requires the metadata size to be equal to the
2038 * t10 pi tuple size.
2040 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2041 if (ns->ms == sizeof(struct t10_pi_tuple))
2042 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2046 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2047 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2049 if (ctrl->ops->flags & NVME_F_FABRICS) {
2051 * The NVMe over Fabrics specification only supports metadata as
2052 * part of the extended data LBA. We rely on HCA/HBA support to
2053 * remap the separate metadata buffer from the block layer.
2055 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2057 if (ctrl->max_integrity_segments)
2059 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2062 * For PCIe controllers, we can't easily remap the separate
2063 * metadata buffer from the block layer and thus require a
2064 * separate metadata buffer for block layer metadata/PI support.
2065 * We allow extended LBAs for the passthrough interface, though.
2067 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2068 ns->features |= NVME_NS_EXT_LBAS;
2070 ns->features |= NVME_NS_METADATA_SUPPORTED;
2076 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2077 struct request_queue *q)
2079 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2081 if (ctrl->max_hw_sectors) {
2083 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2085 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2086 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2087 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2089 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2090 blk_queue_dma_alignment(q, 7);
2091 blk_queue_write_cache(q, vwc, vwc);
2094 static void nvme_update_disk_info(struct gendisk *disk,
2095 struct nvme_ns *ns, struct nvme_id_ns *id)
2097 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2098 unsigned short bs = 1 << ns->lba_shift;
2099 u32 atomic_bs, phys_bs, io_opt = 0;
2102 * The block layer can't support LBA sizes larger than the page size
2103 * yet, so catch this early and don't allow block I/O.
2105 if (ns->lba_shift > PAGE_SHIFT) {
2110 blk_integrity_unregister(disk);
2112 atomic_bs = phys_bs = bs;
2113 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2114 if (id->nabo == 0) {
2116 * Bit 1 indicates whether NAWUPF is defined for this namespace
2117 * and whether it should be used instead of AWUPF. If NAWUPF ==
2118 * 0 then AWUPF must be used instead.
2120 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2121 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2123 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2126 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2127 /* NPWG = Namespace Preferred Write Granularity */
2128 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2129 /* NOWS = Namespace Optimal Write Size */
2130 io_opt = bs * (1 + le16_to_cpu(id->nows));
2133 blk_queue_logical_block_size(disk->queue, bs);
2135 * Linux filesystems assume writing a single physical block is
2136 * an atomic operation. Hence limit the physical block size to the
2137 * value of the Atomic Write Unit Power Fail parameter.
2139 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2140 blk_queue_io_min(disk->queue, phys_bs);
2141 blk_queue_io_opt(disk->queue, io_opt);
2144 * Register a metadata profile for PI, or the plain non-integrity NVMe
2145 * metadata masquerading as Type 0 if supported, otherwise reject block
2146 * I/O to namespaces with metadata except when the namespace supports
2147 * PI, as it can strip/insert in that case.
2150 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2151 (ns->features & NVME_NS_METADATA_SUPPORTED))
2152 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2153 ns->ctrl->max_integrity_segments);
2154 else if (!nvme_ns_has_pi(ns))
2158 set_capacity_and_notify(disk, capacity);
2160 nvme_config_discard(disk, ns);
2161 nvme_config_write_zeroes(disk, ns);
2163 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2164 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2167 static inline bool nvme_first_scan(struct gendisk *disk)
2169 /* nvme_alloc_ns() scans the disk prior to adding it */
2170 return !(disk->flags & GENHD_FL_UP);
2173 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2175 struct nvme_ctrl *ctrl = ns->ctrl;
2178 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2179 is_power_of_2(ctrl->max_hw_sectors))
2180 iob = ctrl->max_hw_sectors;
2182 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2187 if (!is_power_of_2(iob)) {
2188 if (nvme_first_scan(ns->disk))
2189 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2190 ns->disk->disk_name, iob);
2194 if (blk_queue_is_zoned(ns->disk->queue)) {
2195 if (nvme_first_scan(ns->disk))
2196 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2197 ns->disk->disk_name);
2201 blk_queue_chunk_sectors(ns->queue, iob);
2204 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2206 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2209 blk_mq_freeze_queue(ns->disk->queue);
2210 ns->lba_shift = id->lbaf[lbaf].ds;
2211 nvme_set_queue_limits(ns->ctrl, ns->queue);
2213 ret = nvme_configure_metadata(ns, id);
2216 nvme_set_chunk_sectors(ns, id);
2217 nvme_update_disk_info(ns->disk, ns, id);
2219 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2220 ret = nvme_update_zone_info(ns, lbaf);
2225 blk_mq_unfreeze_queue(ns->disk->queue);
2227 if (blk_queue_is_zoned(ns->queue)) {
2228 ret = nvme_revalidate_zones(ns);
2229 if (ret && !nvme_first_scan(ns->disk))
2233 #ifdef CONFIG_NVME_MULTIPATH
2234 if (ns->head->disk) {
2235 blk_mq_freeze_queue(ns->head->disk->queue);
2236 nvme_update_disk_info(ns->head->disk, ns, id);
2237 blk_stack_limits(&ns->head->disk->queue->limits,
2238 &ns->queue->limits, 0);
2239 blk_queue_update_readahead(ns->head->disk->queue);
2240 blk_mq_unfreeze_queue(ns->head->disk->queue);
2246 blk_mq_unfreeze_queue(ns->disk->queue);
2250 static char nvme_pr_type(enum pr_type type)
2253 case PR_WRITE_EXCLUSIVE:
2255 case PR_EXCLUSIVE_ACCESS:
2257 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2259 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2261 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2263 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2270 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2271 u64 key, u64 sa_key, u8 op)
2273 struct nvme_ns_head *head = NULL;
2275 struct nvme_command c;
2277 u8 data[16] = { 0, };
2279 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2281 return -EWOULDBLOCK;
2283 put_unaligned_le64(key, &data[0]);
2284 put_unaligned_le64(sa_key, &data[8]);
2286 memset(&c, 0, sizeof(c));
2287 c.common.opcode = op;
2288 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2289 c.common.cdw10 = cpu_to_le32(cdw10);
2291 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2292 nvme_put_ns_from_disk(head, srcu_idx);
2296 static int nvme_pr_register(struct block_device *bdev, u64 old,
2297 u64 new, unsigned flags)
2301 if (flags & ~PR_FL_IGNORE_KEY)
2304 cdw10 = old ? 2 : 0;
2305 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2306 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2307 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2310 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2311 enum pr_type type, unsigned flags)
2315 if (flags & ~PR_FL_IGNORE_KEY)
2318 cdw10 = nvme_pr_type(type) << 8;
2319 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2320 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2323 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2324 enum pr_type type, bool abort)
2326 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2327 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2330 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2332 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2333 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2336 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2338 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2339 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2342 static const struct pr_ops nvme_pr_ops = {
2343 .pr_register = nvme_pr_register,
2344 .pr_reserve = nvme_pr_reserve,
2345 .pr_release = nvme_pr_release,
2346 .pr_preempt = nvme_pr_preempt,
2347 .pr_clear = nvme_pr_clear,
2350 #ifdef CONFIG_BLK_SED_OPAL
2351 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2354 struct nvme_ctrl *ctrl = data;
2355 struct nvme_command cmd;
2357 memset(&cmd, 0, sizeof(cmd));
2359 cmd.common.opcode = nvme_admin_security_send;
2361 cmd.common.opcode = nvme_admin_security_recv;
2362 cmd.common.nsid = 0;
2363 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2364 cmd.common.cdw11 = cpu_to_le32(len);
2366 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2367 NVME_QID_ANY, 1, 0, false);
2369 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2370 #endif /* CONFIG_BLK_SED_OPAL */
2372 static const struct block_device_operations nvme_bdev_ops = {
2373 .owner = THIS_MODULE,
2374 .ioctl = nvme_ioctl,
2375 .compat_ioctl = nvme_compat_ioctl,
2377 .release = nvme_release,
2378 .getgeo = nvme_getgeo,
2379 .report_zones = nvme_report_zones,
2380 .pr_ops = &nvme_pr_ops,
2383 #ifdef CONFIG_NVME_MULTIPATH
2384 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2386 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2388 if (!kref_get_unless_zero(&head->ref))
2393 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2395 nvme_put_ns_head(disk->private_data);
2398 const struct block_device_operations nvme_ns_head_ops = {
2399 .owner = THIS_MODULE,
2400 .submit_bio = nvme_ns_head_submit_bio,
2401 .open = nvme_ns_head_open,
2402 .release = nvme_ns_head_release,
2403 .ioctl = nvme_ioctl,
2404 .compat_ioctl = nvme_compat_ioctl,
2405 .getgeo = nvme_getgeo,
2406 .report_zones = nvme_report_zones,
2407 .pr_ops = &nvme_pr_ops,
2409 #endif /* CONFIG_NVME_MULTIPATH */
2411 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2413 unsigned long timeout =
2414 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2415 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2418 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2421 if ((csts & NVME_CSTS_RDY) == bit)
2424 usleep_range(1000, 2000);
2425 if (fatal_signal_pending(current))
2427 if (time_after(jiffies, timeout)) {
2428 dev_err(ctrl->device,
2429 "Device not ready; aborting %s, CSTS=0x%x\n",
2430 enabled ? "initialisation" : "reset", csts);
2439 * If the device has been passed off to us in an enabled state, just clear
2440 * the enabled bit. The spec says we should set the 'shutdown notification
2441 * bits', but doing so may cause the device to complete commands to the
2442 * admin queue ... and we don't know what memory that might be pointing at!
2444 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2448 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2449 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2451 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2455 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2456 msleep(NVME_QUIRK_DELAY_AMOUNT);
2458 return nvme_wait_ready(ctrl, ctrl->cap, false);
2460 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2462 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2464 unsigned dev_page_min;
2467 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2469 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2472 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2474 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2475 dev_err(ctrl->device,
2476 "Minimum device page size %u too large for host (%u)\n",
2477 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2481 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2482 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2484 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2485 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2486 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2487 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2488 ctrl->ctrl_config |= NVME_CC_ENABLE;
2490 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2493 return nvme_wait_ready(ctrl, ctrl->cap, true);
2495 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2497 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2499 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2503 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2504 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2506 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2510 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2511 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2515 if (fatal_signal_pending(current))
2517 if (time_after(jiffies, timeout)) {
2518 dev_err(ctrl->device,
2519 "Device shutdown incomplete; abort shutdown\n");
2526 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2528 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2533 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2536 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2537 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2540 dev_warn_once(ctrl->device,
2541 "could not set timestamp (%d)\n", ret);
2545 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2547 struct nvme_feat_host_behavior *host;
2550 /* Don't bother enabling the feature if retry delay is not reported */
2554 host = kzalloc(sizeof(*host), GFP_KERNEL);
2558 host->acre = NVME_ENABLE_ACRE;
2559 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2560 host, sizeof(*host), NULL);
2565 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2568 * APST (Autonomous Power State Transition) lets us program a
2569 * table of power state transitions that the controller will
2570 * perform automatically. We configure it with a simple
2571 * heuristic: we are willing to spend at most 2% of the time
2572 * transitioning between power states. Therefore, when running
2573 * in any given state, we will enter the next lower-power
2574 * non-operational state after waiting 50 * (enlat + exlat)
2575 * microseconds, as long as that state's exit latency is under
2576 * the requested maximum latency.
2578 * We will not autonomously enter any non-operational state for
2579 * which the total latency exceeds ps_max_latency_us. Users
2580 * can set ps_max_latency_us to zero to turn off APST.
2584 struct nvme_feat_auto_pst *table;
2590 * If APST isn't supported or if we haven't been initialized yet,
2591 * then don't do anything.
2596 if (ctrl->npss > 31) {
2597 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2601 table = kzalloc(sizeof(*table), GFP_KERNEL);
2605 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2606 /* Turn off APST. */
2608 dev_dbg(ctrl->device, "APST disabled\n");
2610 __le64 target = cpu_to_le64(0);
2614 * Walk through all states from lowest- to highest-power.
2615 * According to the spec, lower-numbered states use more
2616 * power. NPSS, despite the name, is the index of the
2617 * lowest-power state, not the number of states.
2619 for (state = (int)ctrl->npss; state >= 0; state--) {
2620 u64 total_latency_us, exit_latency_us, transition_ms;
2623 table->entries[state] = target;
2626 * Don't allow transitions to the deepest state
2627 * if it's quirked off.
2629 if (state == ctrl->npss &&
2630 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2634 * Is this state a useful non-operational state for
2635 * higher-power states to autonomously transition to?
2637 if (!(ctrl->psd[state].flags &
2638 NVME_PS_FLAGS_NON_OP_STATE))
2642 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2643 if (exit_latency_us > ctrl->ps_max_latency_us)
2648 le32_to_cpu(ctrl->psd[state].entry_lat);
2651 * This state is good. Use it as the APST idle
2652 * target for higher power states.
2654 transition_ms = total_latency_us + 19;
2655 do_div(transition_ms, 20);
2656 if (transition_ms > (1 << 24) - 1)
2657 transition_ms = (1 << 24) - 1;
2659 target = cpu_to_le64((state << 3) |
2660 (transition_ms << 8));
2665 if (total_latency_us > max_lat_us)
2666 max_lat_us = total_latency_us;
2672 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2674 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2675 max_ps, max_lat_us, (int)sizeof(*table), table);
2679 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2680 table, sizeof(*table), NULL);
2682 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2688 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2690 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2694 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2695 case PM_QOS_LATENCY_ANY:
2703 if (ctrl->ps_max_latency_us != latency) {
2704 ctrl->ps_max_latency_us = latency;
2705 nvme_configure_apst(ctrl);
2709 struct nvme_core_quirk_entry {
2711 * NVMe model and firmware strings are padded with spaces. For
2712 * simplicity, strings in the quirk table are padded with NULLs
2718 unsigned long quirks;
2721 static const struct nvme_core_quirk_entry core_quirks[] = {
2724 * This Toshiba device seems to die using any APST states. See:
2725 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2728 .mn = "THNSF5256GPUK TOSHIBA",
2729 .quirks = NVME_QUIRK_NO_APST,
2733 * This LiteON CL1-3D*-Q11 firmware version has a race
2734 * condition associated with actions related to suspend to idle
2735 * LiteON has resolved the problem in future firmware
2739 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2743 /* match is null-terminated but idstr is space-padded. */
2744 static bool string_matches(const char *idstr, const char *match, size_t len)
2751 matchlen = strlen(match);
2752 WARN_ON_ONCE(matchlen > len);
2754 if (memcmp(idstr, match, matchlen))
2757 for (; matchlen < len; matchlen++)
2758 if (idstr[matchlen] != ' ')
2764 static bool quirk_matches(const struct nvme_id_ctrl *id,
2765 const struct nvme_core_quirk_entry *q)
2767 return q->vid == le16_to_cpu(id->vid) &&
2768 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2769 string_matches(id->fr, q->fr, sizeof(id->fr));
2772 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2773 struct nvme_id_ctrl *id)
2778 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2779 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2780 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2781 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2785 if (ctrl->vs >= NVME_VS(1, 2, 1))
2786 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2789 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2790 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2791 "nqn.2014.08.org.nvmexpress:%04x%04x",
2792 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2793 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2794 off += sizeof(id->sn);
2795 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2796 off += sizeof(id->mn);
2797 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2800 static void nvme_release_subsystem(struct device *dev)
2802 struct nvme_subsystem *subsys =
2803 container_of(dev, struct nvme_subsystem, dev);
2805 if (subsys->instance >= 0)
2806 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2810 static void nvme_destroy_subsystem(struct kref *ref)
2812 struct nvme_subsystem *subsys =
2813 container_of(ref, struct nvme_subsystem, ref);
2815 mutex_lock(&nvme_subsystems_lock);
2816 list_del(&subsys->entry);
2817 mutex_unlock(&nvme_subsystems_lock);
2819 ida_destroy(&subsys->ns_ida);
2820 device_del(&subsys->dev);
2821 put_device(&subsys->dev);
2824 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2826 kref_put(&subsys->ref, nvme_destroy_subsystem);
2829 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2831 struct nvme_subsystem *subsys;
2833 lockdep_assert_held(&nvme_subsystems_lock);
2836 * Fail matches for discovery subsystems. This results
2837 * in each discovery controller bound to a unique subsystem.
2838 * This avoids issues with validating controller values
2839 * that can only be true when there is a single unique subsystem.
2840 * There may be multiple and completely independent entities
2841 * that provide discovery controllers.
2843 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2846 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2847 if (strcmp(subsys->subnqn, subsysnqn))
2849 if (!kref_get_unless_zero(&subsys->ref))
2857 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2858 struct device_attribute subsys_attr_##_name = \
2859 __ATTR(_name, _mode, _show, NULL)
2861 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2862 struct device_attribute *attr,
2865 struct nvme_subsystem *subsys =
2866 container_of(dev, struct nvme_subsystem, dev);
2868 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2870 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2872 #define nvme_subsys_show_str_function(field) \
2873 static ssize_t subsys_##field##_show(struct device *dev, \
2874 struct device_attribute *attr, char *buf) \
2876 struct nvme_subsystem *subsys = \
2877 container_of(dev, struct nvme_subsystem, dev); \
2878 return sprintf(buf, "%.*s\n", \
2879 (int)sizeof(subsys->field), subsys->field); \
2881 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2883 nvme_subsys_show_str_function(model);
2884 nvme_subsys_show_str_function(serial);
2885 nvme_subsys_show_str_function(firmware_rev);
2887 static struct attribute *nvme_subsys_attrs[] = {
2888 &subsys_attr_model.attr,
2889 &subsys_attr_serial.attr,
2890 &subsys_attr_firmware_rev.attr,
2891 &subsys_attr_subsysnqn.attr,
2892 #ifdef CONFIG_NVME_MULTIPATH
2893 &subsys_attr_iopolicy.attr,
2898 static const struct attribute_group nvme_subsys_attrs_group = {
2899 .attrs = nvme_subsys_attrs,
2902 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2903 &nvme_subsys_attrs_group,
2907 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2909 return ctrl->opts && ctrl->opts->discovery_nqn;
2912 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2913 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2915 struct nvme_ctrl *tmp;
2917 lockdep_assert_held(&nvme_subsystems_lock);
2919 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2920 if (nvme_state_terminal(tmp))
2923 if (tmp->cntlid == ctrl->cntlid) {
2924 dev_err(ctrl->device,
2925 "Duplicate cntlid %u with %s, rejecting\n",
2926 ctrl->cntlid, dev_name(tmp->device));
2930 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2931 nvme_discovery_ctrl(ctrl))
2934 dev_err(ctrl->device,
2935 "Subsystem does not support multiple controllers\n");
2942 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2944 struct nvme_subsystem *subsys, *found;
2947 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2951 subsys->instance = -1;
2952 mutex_init(&subsys->lock);
2953 kref_init(&subsys->ref);
2954 INIT_LIST_HEAD(&subsys->ctrls);
2955 INIT_LIST_HEAD(&subsys->nsheads);
2956 nvme_init_subnqn(subsys, ctrl, id);
2957 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2958 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2959 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2960 subsys->vendor_id = le16_to_cpu(id->vid);
2961 subsys->cmic = id->cmic;
2962 subsys->awupf = le16_to_cpu(id->awupf);
2963 #ifdef CONFIG_NVME_MULTIPATH
2964 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2967 subsys->dev.class = nvme_subsys_class;
2968 subsys->dev.release = nvme_release_subsystem;
2969 subsys->dev.groups = nvme_subsys_attrs_groups;
2970 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2971 device_initialize(&subsys->dev);
2973 mutex_lock(&nvme_subsystems_lock);
2974 found = __nvme_find_get_subsystem(subsys->subnqn);
2976 put_device(&subsys->dev);
2979 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2981 goto out_put_subsystem;
2984 ret = device_add(&subsys->dev);
2986 dev_err(ctrl->device,
2987 "failed to register subsystem device.\n");
2988 put_device(&subsys->dev);
2991 ida_init(&subsys->ns_ida);
2992 list_add_tail(&subsys->entry, &nvme_subsystems);
2995 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2996 dev_name(ctrl->device));
2998 dev_err(ctrl->device,
2999 "failed to create sysfs link from subsystem.\n");
3000 goto out_put_subsystem;
3004 subsys->instance = ctrl->instance;
3005 ctrl->subsys = subsys;
3006 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
3007 mutex_unlock(&nvme_subsystems_lock);
3011 nvme_put_subsystem(subsys);
3013 mutex_unlock(&nvme_subsystems_lock);
3017 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3018 void *log, size_t size, u64 offset)
3020 struct nvme_command c = { };
3021 u32 dwlen = nvme_bytes_to_numd(size);
3023 c.get_log_page.opcode = nvme_admin_get_log_page;
3024 c.get_log_page.nsid = cpu_to_le32(nsid);
3025 c.get_log_page.lid = log_page;
3026 c.get_log_page.lsp = lsp;
3027 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3028 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3029 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3030 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3031 c.get_log_page.csi = csi;
3033 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3036 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3037 struct nvme_effects_log **log)
3039 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3045 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3049 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3050 cel, sizeof(*cel), 0);
3056 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3063 * Initialize the cached copies of the Identify data and various controller
3064 * register in our nvme_ctrl structure. This should be called as soon as
3065 * the admin queue is fully up and running.
3067 int nvme_init_identify(struct nvme_ctrl *ctrl)
3069 struct nvme_id_ctrl *id;
3070 int ret, page_shift;
3072 bool prev_apst_enabled;
3074 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3076 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3079 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3080 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3082 if (ctrl->vs >= NVME_VS(1, 1, 0))
3083 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3085 ret = nvme_identify_ctrl(ctrl, &id);
3087 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3091 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3092 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3097 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3098 ctrl->cntlid = le16_to_cpu(id->cntlid);
3100 if (!ctrl->identified) {
3103 ret = nvme_init_subsystem(ctrl, id);
3108 * Check for quirks. Quirk can depend on firmware version,
3109 * so, in principle, the set of quirks present can change
3110 * across a reset. As a possible future enhancement, we
3111 * could re-scan for quirks every time we reinitialize
3112 * the device, but we'd have to make sure that the driver
3113 * behaves intelligently if the quirks change.
3115 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3116 if (quirk_matches(id, &core_quirks[i]))
3117 ctrl->quirks |= core_quirks[i].quirks;
3121 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3122 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3123 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3126 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3127 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3128 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3130 ctrl->oacs = le16_to_cpu(id->oacs);
3131 ctrl->oncs = le16_to_cpu(id->oncs);
3132 ctrl->mtfa = le16_to_cpu(id->mtfa);
3133 ctrl->oaes = le32_to_cpu(id->oaes);
3134 ctrl->wctemp = le16_to_cpu(id->wctemp);
3135 ctrl->cctemp = le16_to_cpu(id->cctemp);
3137 atomic_set(&ctrl->abort_limit, id->acl + 1);
3138 ctrl->vwc = id->vwc;
3140 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3142 max_hw_sectors = UINT_MAX;
3143 ctrl->max_hw_sectors =
3144 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3146 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3147 ctrl->sgls = le32_to_cpu(id->sgls);
3148 ctrl->kas = le16_to_cpu(id->kas);
3149 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3150 ctrl->ctratt = le32_to_cpu(id->ctratt);
3154 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3156 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3157 shutdown_timeout, 60);
3159 if (ctrl->shutdown_timeout != shutdown_timeout)
3160 dev_info(ctrl->device,
3161 "Shutdown timeout set to %u seconds\n",
3162 ctrl->shutdown_timeout);
3164 ctrl->shutdown_timeout = shutdown_timeout;
3166 ctrl->npss = id->npss;
3167 ctrl->apsta = id->apsta;
3168 prev_apst_enabled = ctrl->apst_enabled;
3169 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3170 if (force_apst && id->apsta) {
3171 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3172 ctrl->apst_enabled = true;
3174 ctrl->apst_enabled = false;
3177 ctrl->apst_enabled = id->apsta;
3179 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3181 if (ctrl->ops->flags & NVME_F_FABRICS) {
3182 ctrl->icdoff = le16_to_cpu(id->icdoff);
3183 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3184 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3185 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3188 * In fabrics we need to verify the cntlid matches the
3191 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3192 dev_err(ctrl->device,
3193 "Mismatching cntlid: Connect %u vs Identify "
3195 ctrl->cntlid, le16_to_cpu(id->cntlid));
3200 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3201 dev_err(ctrl->device,
3202 "keep-alive support is mandatory for fabrics\n");
3207 ctrl->hmpre = le32_to_cpu(id->hmpre);
3208 ctrl->hmmin = le32_to_cpu(id->hmmin);
3209 ctrl->hmminds = le32_to_cpu(id->hmminds);
3210 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3213 ret = nvme_mpath_init(ctrl, id);
3219 if (ctrl->apst_enabled && !prev_apst_enabled)
3220 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3221 else if (!ctrl->apst_enabled && prev_apst_enabled)
3222 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3224 ret = nvme_configure_apst(ctrl);
3228 ret = nvme_configure_timestamp(ctrl);
3232 ret = nvme_configure_directives(ctrl);
3236 ret = nvme_configure_acre(ctrl);
3240 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3241 ret = nvme_hwmon_init(ctrl);
3246 ctrl->identified = true;
3254 EXPORT_SYMBOL_GPL(nvme_init_identify);
3256 static int nvme_dev_open(struct inode *inode, struct file *file)
3258 struct nvme_ctrl *ctrl =
3259 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3261 switch (ctrl->state) {
3262 case NVME_CTRL_LIVE:
3265 return -EWOULDBLOCK;
3268 nvme_get_ctrl(ctrl);
3269 if (!try_module_get(ctrl->ops->module)) {
3270 nvme_put_ctrl(ctrl);
3274 file->private_data = ctrl;
3278 static int nvme_dev_release(struct inode *inode, struct file *file)
3280 struct nvme_ctrl *ctrl =
3281 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3283 module_put(ctrl->ops->module);
3284 nvme_put_ctrl(ctrl);
3288 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3293 down_read(&ctrl->namespaces_rwsem);
3294 if (list_empty(&ctrl->namespaces)) {
3299 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3300 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3301 dev_warn(ctrl->device,
3302 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3307 dev_warn(ctrl->device,
3308 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3309 kref_get(&ns->kref);
3310 up_read(&ctrl->namespaces_rwsem);
3312 ret = nvme_user_cmd(ctrl, ns, argp);
3317 up_read(&ctrl->namespaces_rwsem);
3321 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3324 struct nvme_ctrl *ctrl = file->private_data;
3325 void __user *argp = (void __user *)arg;
3328 case NVME_IOCTL_ADMIN_CMD:
3329 return nvme_user_cmd(ctrl, NULL, argp);
3330 case NVME_IOCTL_ADMIN64_CMD:
3331 return nvme_user_cmd64(ctrl, NULL, argp);
3332 case NVME_IOCTL_IO_CMD:
3333 return nvme_dev_user_cmd(ctrl, argp);
3334 case NVME_IOCTL_RESET:
3335 dev_warn(ctrl->device, "resetting controller\n");
3336 return nvme_reset_ctrl_sync(ctrl);
3337 case NVME_IOCTL_SUBSYS_RESET:
3338 return nvme_reset_subsystem(ctrl);
3339 case NVME_IOCTL_RESCAN:
3340 nvme_queue_scan(ctrl);
3347 static const struct file_operations nvme_dev_fops = {
3348 .owner = THIS_MODULE,
3349 .open = nvme_dev_open,
3350 .release = nvme_dev_release,
3351 .unlocked_ioctl = nvme_dev_ioctl,
3352 .compat_ioctl = compat_ptr_ioctl,
3355 static ssize_t nvme_sysfs_reset(struct device *dev,
3356 struct device_attribute *attr, const char *buf,
3359 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3362 ret = nvme_reset_ctrl_sync(ctrl);
3367 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3369 static ssize_t nvme_sysfs_rescan(struct device *dev,
3370 struct device_attribute *attr, const char *buf,
3373 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3375 nvme_queue_scan(ctrl);
3378 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3380 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3382 struct gendisk *disk = dev_to_disk(dev);
3384 if (disk->fops == &nvme_bdev_ops)
3385 return nvme_get_ns_from_dev(dev)->head;
3387 return disk->private_data;
3390 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3393 struct nvme_ns_head *head = dev_to_ns_head(dev);
3394 struct nvme_ns_ids *ids = &head->ids;
3395 struct nvme_subsystem *subsys = head->subsys;
3396 int serial_len = sizeof(subsys->serial);
3397 int model_len = sizeof(subsys->model);
3399 if (!uuid_is_null(&ids->uuid))
3400 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3402 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3403 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3405 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3406 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3408 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3409 subsys->serial[serial_len - 1] == '\0'))
3411 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3412 subsys->model[model_len - 1] == '\0'))
3415 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3416 serial_len, subsys->serial, model_len, subsys->model,
3419 static DEVICE_ATTR_RO(wwid);
3421 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3424 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3426 static DEVICE_ATTR_RO(nguid);
3428 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3431 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3433 /* For backward compatibility expose the NGUID to userspace if
3434 * we have no UUID set
3436 if (uuid_is_null(&ids->uuid)) {
3437 printk_ratelimited(KERN_WARNING
3438 "No UUID available providing old NGUID\n");
3439 return sprintf(buf, "%pU\n", ids->nguid);
3441 return sprintf(buf, "%pU\n", &ids->uuid);
3443 static DEVICE_ATTR_RO(uuid);
3445 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3448 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3450 static DEVICE_ATTR_RO(eui);
3452 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3455 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3457 static DEVICE_ATTR_RO(nsid);
3459 static struct attribute *nvme_ns_id_attrs[] = {
3460 &dev_attr_wwid.attr,
3461 &dev_attr_uuid.attr,
3462 &dev_attr_nguid.attr,
3464 &dev_attr_nsid.attr,
3465 #ifdef CONFIG_NVME_MULTIPATH
3466 &dev_attr_ana_grpid.attr,
3467 &dev_attr_ana_state.attr,
3472 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3473 struct attribute *a, int n)
3475 struct device *dev = container_of(kobj, struct device, kobj);
3476 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3478 if (a == &dev_attr_uuid.attr) {
3479 if (uuid_is_null(&ids->uuid) &&
3480 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3483 if (a == &dev_attr_nguid.attr) {
3484 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3487 if (a == &dev_attr_eui.attr) {
3488 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3491 #ifdef CONFIG_NVME_MULTIPATH
3492 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3493 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3495 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3502 static const struct attribute_group nvme_ns_id_attr_group = {
3503 .attrs = nvme_ns_id_attrs,
3504 .is_visible = nvme_ns_id_attrs_are_visible,
3507 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3508 &nvme_ns_id_attr_group,
3510 &nvme_nvm_attr_group,
3515 #define nvme_show_str_function(field) \
3516 static ssize_t field##_show(struct device *dev, \
3517 struct device_attribute *attr, char *buf) \
3519 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3520 return sprintf(buf, "%.*s\n", \
3521 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3523 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3525 nvme_show_str_function(model);
3526 nvme_show_str_function(serial);
3527 nvme_show_str_function(firmware_rev);
3529 #define nvme_show_int_function(field) \
3530 static ssize_t field##_show(struct device *dev, \
3531 struct device_attribute *attr, char *buf) \
3533 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3534 return sprintf(buf, "%d\n", ctrl->field); \
3536 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3538 nvme_show_int_function(cntlid);
3539 nvme_show_int_function(numa_node);
3540 nvme_show_int_function(queue_count);
3541 nvme_show_int_function(sqsize);
3543 static ssize_t nvme_sysfs_delete(struct device *dev,
3544 struct device_attribute *attr, const char *buf,
3547 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3549 if (device_remove_file_self(dev, attr))
3550 nvme_delete_ctrl_sync(ctrl);
3553 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3555 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3556 struct device_attribute *attr,
3559 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3561 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3563 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3565 static ssize_t nvme_sysfs_show_state(struct device *dev,
3566 struct device_attribute *attr,
3569 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3570 static const char *const state_name[] = {
3571 [NVME_CTRL_NEW] = "new",
3572 [NVME_CTRL_LIVE] = "live",
3573 [NVME_CTRL_RESETTING] = "resetting",
3574 [NVME_CTRL_CONNECTING] = "connecting",
3575 [NVME_CTRL_DELETING] = "deleting",
3576 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3577 [NVME_CTRL_DEAD] = "dead",
3580 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3581 state_name[ctrl->state])
3582 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3584 return sprintf(buf, "unknown state\n");
3587 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3589 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3590 struct device_attribute *attr,
3593 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3595 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3597 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3599 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3600 struct device_attribute *attr,
3603 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3605 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3607 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3609 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3610 struct device_attribute *attr,
3613 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3615 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3617 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3619 static ssize_t nvme_sysfs_show_address(struct device *dev,
3620 struct device_attribute *attr,
3623 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3625 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3627 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3629 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3630 struct device_attribute *attr, char *buf)
3632 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3633 struct nvmf_ctrl_options *opts = ctrl->opts;
3635 if (ctrl->opts->max_reconnects == -1)
3636 return sprintf(buf, "off\n");
3637 return sprintf(buf, "%d\n",
3638 opts->max_reconnects * opts->reconnect_delay);
3641 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3642 struct device_attribute *attr, const char *buf, size_t count)
3644 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3645 struct nvmf_ctrl_options *opts = ctrl->opts;
3646 int ctrl_loss_tmo, err;
3648 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3652 else if (ctrl_loss_tmo < 0)
3653 opts->max_reconnects = -1;
3655 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3656 opts->reconnect_delay);
3659 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3660 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3662 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3663 struct device_attribute *attr, char *buf)
3665 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3667 if (ctrl->opts->reconnect_delay == -1)
3668 return sprintf(buf, "off\n");
3669 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3672 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3673 struct device_attribute *attr, const char *buf, size_t count)
3675 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3679 err = kstrtou32(buf, 10, &v);
3683 ctrl->opts->reconnect_delay = v;
3686 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3687 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3689 static struct attribute *nvme_dev_attrs[] = {
3690 &dev_attr_reset_controller.attr,
3691 &dev_attr_rescan_controller.attr,
3692 &dev_attr_model.attr,
3693 &dev_attr_serial.attr,
3694 &dev_attr_firmware_rev.attr,
3695 &dev_attr_cntlid.attr,
3696 &dev_attr_delete_controller.attr,
3697 &dev_attr_transport.attr,
3698 &dev_attr_subsysnqn.attr,
3699 &dev_attr_address.attr,
3700 &dev_attr_state.attr,
3701 &dev_attr_numa_node.attr,
3702 &dev_attr_queue_count.attr,
3703 &dev_attr_sqsize.attr,
3704 &dev_attr_hostnqn.attr,
3705 &dev_attr_hostid.attr,
3706 &dev_attr_ctrl_loss_tmo.attr,
3707 &dev_attr_reconnect_delay.attr,
3711 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3712 struct attribute *a, int n)
3714 struct device *dev = container_of(kobj, struct device, kobj);
3715 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3717 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3719 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3721 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3723 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3725 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3727 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3733 static const struct attribute_group nvme_dev_attrs_group = {
3734 .attrs = nvme_dev_attrs,
3735 .is_visible = nvme_dev_attrs_are_visible,
3738 static const struct attribute_group *nvme_dev_attr_groups[] = {
3739 &nvme_dev_attrs_group,
3743 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3746 struct nvme_ns_head *h;
3748 lockdep_assert_held(&subsys->lock);
3750 list_for_each_entry(h, &subsys->nsheads, entry) {
3751 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3758 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3759 struct nvme_ns_head *new)
3761 struct nvme_ns_head *h;
3763 lockdep_assert_held(&subsys->lock);
3765 list_for_each_entry(h, &subsys->nsheads, entry) {
3766 if (nvme_ns_ids_valid(&new->ids) &&
3767 nvme_ns_ids_equal(&new->ids, &h->ids))
3774 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3775 unsigned nsid, struct nvme_ns_ids *ids)
3777 struct nvme_ns_head *head;
3778 size_t size = sizeof(*head);
3781 #ifdef CONFIG_NVME_MULTIPATH
3782 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3785 head = kzalloc(size, GFP_KERNEL);
3788 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3791 head->instance = ret;
3792 INIT_LIST_HEAD(&head->list);
3793 ret = init_srcu_struct(&head->srcu);
3795 goto out_ida_remove;
3796 head->subsys = ctrl->subsys;
3799 kref_init(&head->ref);
3801 ret = __nvme_check_ids(ctrl->subsys, head);
3803 dev_err(ctrl->device,
3804 "duplicate IDs for nsid %d\n", nsid);
3805 goto out_cleanup_srcu;
3808 if (head->ids.csi) {
3809 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3811 goto out_cleanup_srcu;
3813 head->effects = ctrl->effects;
3815 ret = nvme_mpath_alloc_disk(ctrl, head);
3817 goto out_cleanup_srcu;
3819 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3821 kref_get(&ctrl->subsys->ref);
3825 cleanup_srcu_struct(&head->srcu);
3827 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3832 ret = blk_status_to_errno(nvme_error_status(ret));
3833 return ERR_PTR(ret);
3836 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3837 struct nvme_ns_ids *ids, bool is_shared)
3839 struct nvme_ctrl *ctrl = ns->ctrl;
3840 struct nvme_ns_head *head = NULL;
3843 mutex_lock(&ctrl->subsys->lock);
3844 head = nvme_find_ns_head(ctrl->subsys, nsid);
3846 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3848 ret = PTR_ERR(head);
3851 head->shared = is_shared;
3854 if (!is_shared || !head->shared) {
3855 dev_err(ctrl->device,
3856 "Duplicate unshared namespace %d\n", nsid);
3857 goto out_put_ns_head;
3859 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3860 dev_err(ctrl->device,
3861 "IDs don't match for shared namespace %d\n",
3863 goto out_put_ns_head;
3867 list_add_tail_rcu(&ns->siblings, &head->list);
3869 mutex_unlock(&ctrl->subsys->lock);
3873 nvme_put_ns_head(head);
3875 mutex_unlock(&ctrl->subsys->lock);
3879 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3881 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3882 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3884 return nsa->head->ns_id - nsb->head->ns_id;
3887 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3889 struct nvme_ns *ns, *ret = NULL;
3891 down_read(&ctrl->namespaces_rwsem);
3892 list_for_each_entry(ns, &ctrl->namespaces, list) {
3893 if (ns->head->ns_id == nsid) {
3894 if (!kref_get_unless_zero(&ns->kref))
3899 if (ns->head->ns_id > nsid)
3902 up_read(&ctrl->namespaces_rwsem);
3905 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3907 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3908 struct nvme_ns_ids *ids)
3911 struct gendisk *disk;
3912 struct nvme_id_ns *id;
3913 char disk_name[DISK_NAME_LEN];
3914 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3916 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3919 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3923 ns->queue = blk_mq_init_queue(ctrl->tagset);
3924 if (IS_ERR(ns->queue))
3927 if (ctrl->opts && ctrl->opts->data_digest)
3928 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3930 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3931 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3932 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3934 ns->queue->queuedata = ns;
3936 kref_init(&ns->kref);
3938 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3939 goto out_free_queue;
3940 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3942 disk = alloc_disk_node(0, node);
3946 disk->fops = &nvme_bdev_ops;
3947 disk->private_data = ns;
3948 disk->queue = ns->queue;
3949 disk->flags = flags;
3950 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3953 if (nvme_update_ns_info(ns, id))
3956 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3957 if (nvme_nvm_register(ns, disk_name, node)) {
3958 dev_warn(ctrl->device, "LightNVM init failure\n");
3963 down_write(&ctrl->namespaces_rwsem);
3964 list_add_tail(&ns->list, &ctrl->namespaces);
3965 up_write(&ctrl->namespaces_rwsem);
3967 nvme_get_ctrl(ctrl);
3969 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3971 nvme_mpath_add_disk(ns, id);
3972 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3977 /* prevent double queue cleanup */
3978 ns->disk->queue = NULL;
3981 mutex_lock(&ctrl->subsys->lock);
3982 list_del_rcu(&ns->siblings);
3983 if (list_empty(&ns->head->list))
3984 list_del_init(&ns->head->entry);
3985 mutex_unlock(&ctrl->subsys->lock);
3986 nvme_put_ns_head(ns->head);
3988 blk_cleanup_queue(ns->queue);
3995 static void nvme_ns_remove(struct nvme_ns *ns)
3997 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4000 set_capacity(ns->disk, 0);
4001 nvme_fault_inject_fini(&ns->fault_inject);
4003 mutex_lock(&ns->ctrl->subsys->lock);
4004 list_del_rcu(&ns->siblings);
4005 if (list_empty(&ns->head->list))
4006 list_del_init(&ns->head->entry);
4007 mutex_unlock(&ns->ctrl->subsys->lock);
4009 synchronize_rcu(); /* guarantee not available in head->list */
4010 nvme_mpath_clear_current_path(ns);
4011 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4013 if (ns->disk->flags & GENHD_FL_UP) {
4014 del_gendisk(ns->disk);
4015 blk_cleanup_queue(ns->queue);
4016 if (blk_get_integrity(ns->disk))
4017 blk_integrity_unregister(ns->disk);
4020 down_write(&ns->ctrl->namespaces_rwsem);
4021 list_del_init(&ns->list);
4022 up_write(&ns->ctrl->namespaces_rwsem);
4024 nvme_mpath_check_last_path(ns);
4028 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4030 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4038 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4040 struct nvme_id_ns *id;
4043 if (test_bit(NVME_NS_DEAD, &ns->flags))
4046 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4051 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4052 dev_err(ns->ctrl->device,
4053 "identifiers changed for nsid %d\n", ns->head->ns_id);
4057 ret = nvme_update_ns_info(ns, id);
4063 * Only remove the namespace if we got a fatal error back from the
4064 * device, otherwise ignore the error and just move on.
4066 * TODO: we should probably schedule a delayed retry here.
4068 if (ret && ret != -ENOMEM && !(ret > 0 && !(ret & NVME_SC_DNR)))
4072 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4074 struct nvme_ns_ids ids = { };
4077 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4080 ns = nvme_find_get_ns(ctrl, nsid);
4082 nvme_validate_ns(ns, &ids);
4089 nvme_alloc_ns(ctrl, nsid, &ids);
4092 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4093 dev_warn(ctrl->device,
4094 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4098 nvme_alloc_ns(ctrl, nsid, &ids);
4101 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4107 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4110 struct nvme_ns *ns, *next;
4113 down_write(&ctrl->namespaces_rwsem);
4114 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4115 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4116 list_move_tail(&ns->list, &rm_list);
4118 up_write(&ctrl->namespaces_rwsem);
4120 list_for_each_entry_safe(ns, next, &rm_list, list)
4125 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4127 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4132 if (nvme_ctrl_limited_cns(ctrl))
4135 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4140 struct nvme_command cmd = {
4141 .identify.opcode = nvme_admin_identify,
4142 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4143 .identify.nsid = cpu_to_le32(prev),
4146 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4147 NVME_IDENTIFY_DATA_SIZE);
4149 dev_warn(ctrl->device,
4150 "Identify NS List failed (status=0x%x)\n", ret);
4154 for (i = 0; i < nr_entries; i++) {
4155 u32 nsid = le32_to_cpu(ns_list[i]);
4157 if (!nsid) /* end of the list? */
4159 nvme_validate_or_alloc_ns(ctrl, nsid);
4160 while (++prev < nsid)
4161 nvme_ns_remove_by_nsid(ctrl, prev);
4165 nvme_remove_invalid_namespaces(ctrl, prev);
4171 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4173 struct nvme_id_ctrl *id;
4176 if (nvme_identify_ctrl(ctrl, &id))
4178 nn = le32_to_cpu(id->nn);
4181 for (i = 1; i <= nn; i++)
4182 nvme_validate_or_alloc_ns(ctrl, i);
4184 nvme_remove_invalid_namespaces(ctrl, nn);
4187 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4189 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4193 log = kzalloc(log_size, GFP_KERNEL);
4198 * We need to read the log to clear the AEN, but we don't want to rely
4199 * on it for the changed namespace information as userspace could have
4200 * raced with us in reading the log page, which could cause us to miss
4203 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4204 NVME_CSI_NVM, log, log_size, 0);
4206 dev_warn(ctrl->device,
4207 "reading changed ns log failed: %d\n", error);
4212 static void nvme_scan_work(struct work_struct *work)
4214 struct nvme_ctrl *ctrl =
4215 container_of(work, struct nvme_ctrl, scan_work);
4217 /* No tagset on a live ctrl means IO queues could not created */
4218 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4221 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4222 dev_info(ctrl->device, "rescanning namespaces.\n");
4223 nvme_clear_changed_ns_log(ctrl);
4226 mutex_lock(&ctrl->scan_lock);
4227 if (nvme_scan_ns_list(ctrl) != 0)
4228 nvme_scan_ns_sequential(ctrl);
4229 mutex_unlock(&ctrl->scan_lock);
4231 down_write(&ctrl->namespaces_rwsem);
4232 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4233 up_write(&ctrl->namespaces_rwsem);
4237 * This function iterates the namespace list unlocked to allow recovery from
4238 * controller failure. It is up to the caller to ensure the namespace list is
4239 * not modified by scan work while this function is executing.
4241 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4243 struct nvme_ns *ns, *next;
4247 * make sure to requeue I/O to all namespaces as these
4248 * might result from the scan itself and must complete
4249 * for the scan_work to make progress
4251 nvme_mpath_clear_ctrl_paths(ctrl);
4253 /* prevent racing with ns scanning */
4254 flush_work(&ctrl->scan_work);
4257 * The dead states indicates the controller was not gracefully
4258 * disconnected. In that case, we won't be able to flush any data while
4259 * removing the namespaces' disks; fail all the queues now to avoid
4260 * potentially having to clean up the failed sync later.
4262 if (ctrl->state == NVME_CTRL_DEAD)
4263 nvme_kill_queues(ctrl);
4265 /* this is a no-op when called from the controller reset handler */
4266 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4268 down_write(&ctrl->namespaces_rwsem);
4269 list_splice_init(&ctrl->namespaces, &ns_list);
4270 up_write(&ctrl->namespaces_rwsem);
4272 list_for_each_entry_safe(ns, next, &ns_list, list)
4275 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4277 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4279 struct nvme_ctrl *ctrl =
4280 container_of(dev, struct nvme_ctrl, ctrl_device);
4281 struct nvmf_ctrl_options *opts = ctrl->opts;
4284 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4289 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4293 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4294 opts->trsvcid ?: "none");
4298 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4299 opts->host_traddr ?: "none");
4304 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4306 char *envp[2] = { NULL, NULL };
4307 u32 aen_result = ctrl->aen_result;
4309 ctrl->aen_result = 0;
4313 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4316 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4320 static void nvme_async_event_work(struct work_struct *work)
4322 struct nvme_ctrl *ctrl =
4323 container_of(work, struct nvme_ctrl, async_event_work);
4325 nvme_aen_uevent(ctrl);
4326 ctrl->ops->submit_async_event(ctrl);
4329 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4334 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4340 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4343 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4345 struct nvme_fw_slot_info_log *log;
4347 log = kmalloc(sizeof(*log), GFP_KERNEL);
4351 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4352 log, sizeof(*log), 0))
4353 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4357 static void nvme_fw_act_work(struct work_struct *work)
4359 struct nvme_ctrl *ctrl = container_of(work,
4360 struct nvme_ctrl, fw_act_work);
4361 unsigned long fw_act_timeout;
4364 fw_act_timeout = jiffies +
4365 msecs_to_jiffies(ctrl->mtfa * 100);
4367 fw_act_timeout = jiffies +
4368 msecs_to_jiffies(admin_timeout * 1000);
4370 nvme_stop_queues(ctrl);
4371 while (nvme_ctrl_pp_status(ctrl)) {
4372 if (time_after(jiffies, fw_act_timeout)) {
4373 dev_warn(ctrl->device,
4374 "Fw activation timeout, reset controller\n");
4375 nvme_try_sched_reset(ctrl);
4381 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4384 nvme_start_queues(ctrl);
4385 /* read FW slot information to clear the AER */
4386 nvme_get_fw_slot_info(ctrl);
4389 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4391 u32 aer_notice_type = (result & 0xff00) >> 8;
4393 trace_nvme_async_event(ctrl, aer_notice_type);
4395 switch (aer_notice_type) {
4396 case NVME_AER_NOTICE_NS_CHANGED:
4397 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4398 nvme_queue_scan(ctrl);
4400 case NVME_AER_NOTICE_FW_ACT_STARTING:
4402 * We are (ab)using the RESETTING state to prevent subsequent
4403 * recovery actions from interfering with the controller's
4404 * firmware activation.
4406 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4407 queue_work(nvme_wq, &ctrl->fw_act_work);
4409 #ifdef CONFIG_NVME_MULTIPATH
4410 case NVME_AER_NOTICE_ANA:
4411 if (!ctrl->ana_log_buf)
4413 queue_work(nvme_wq, &ctrl->ana_work);
4416 case NVME_AER_NOTICE_DISC_CHANGED:
4417 ctrl->aen_result = result;
4420 dev_warn(ctrl->device, "async event result %08x\n", result);
4424 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4425 volatile union nvme_result *res)
4427 u32 result = le32_to_cpu(res->u32);
4428 u32 aer_type = result & 0x07;
4430 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4434 case NVME_AER_NOTICE:
4435 nvme_handle_aen_notice(ctrl, result);
4437 case NVME_AER_ERROR:
4438 case NVME_AER_SMART:
4441 trace_nvme_async_event(ctrl, aer_type);
4442 ctrl->aen_result = result;
4447 queue_work(nvme_wq, &ctrl->async_event_work);
4449 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4451 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4453 nvme_mpath_stop(ctrl);
4454 nvme_stop_keep_alive(ctrl);
4455 nvme_stop_failfast_work(ctrl);
4456 flush_work(&ctrl->async_event_work);
4457 cancel_work_sync(&ctrl->fw_act_work);
4459 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4461 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4463 nvme_start_keep_alive(ctrl);
4465 nvme_enable_aen(ctrl);
4467 if (ctrl->queue_count > 1) {
4468 nvme_queue_scan(ctrl);
4469 nvme_start_queues(ctrl);
4472 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4474 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4476 nvme_hwmon_exit(ctrl);
4477 nvme_fault_inject_fini(&ctrl->fault_inject);
4478 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4479 cdev_device_del(&ctrl->cdev, ctrl->device);
4480 nvme_put_ctrl(ctrl);
4482 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4484 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4486 struct nvme_effects_log *cel;
4489 xa_for_each(&ctrl->cels, i, cel) {
4490 xa_erase(&ctrl->cels, i);
4494 xa_destroy(&ctrl->cels);
4497 static void nvme_free_ctrl(struct device *dev)
4499 struct nvme_ctrl *ctrl =
4500 container_of(dev, struct nvme_ctrl, ctrl_device);
4501 struct nvme_subsystem *subsys = ctrl->subsys;
4503 if (!subsys || ctrl->instance != subsys->instance)
4504 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4506 nvme_free_cels(ctrl);
4507 nvme_mpath_uninit(ctrl);
4508 __free_page(ctrl->discard_page);
4511 mutex_lock(&nvme_subsystems_lock);
4512 list_del(&ctrl->subsys_entry);
4513 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4514 mutex_unlock(&nvme_subsystems_lock);
4517 ctrl->ops->free_ctrl(ctrl);
4520 nvme_put_subsystem(subsys);
4524 * Initialize a NVMe controller structures. This needs to be called during
4525 * earliest initialization so that we have the initialized structured around
4528 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4529 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4533 ctrl->state = NVME_CTRL_NEW;
4534 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4535 spin_lock_init(&ctrl->lock);
4536 mutex_init(&ctrl->scan_lock);
4537 INIT_LIST_HEAD(&ctrl->namespaces);
4538 xa_init(&ctrl->cels);
4539 init_rwsem(&ctrl->namespaces_rwsem);
4542 ctrl->quirks = quirks;
4543 ctrl->numa_node = NUMA_NO_NODE;
4544 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4545 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4546 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4547 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4548 init_waitqueue_head(&ctrl->state_wq);
4550 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4551 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4552 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4553 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4555 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4557 ctrl->discard_page = alloc_page(GFP_KERNEL);
4558 if (!ctrl->discard_page) {
4563 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4566 ctrl->instance = ret;
4568 device_initialize(&ctrl->ctrl_device);
4569 ctrl->device = &ctrl->ctrl_device;
4570 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4572 ctrl->device->class = nvme_class;
4573 ctrl->device->parent = ctrl->dev;
4574 ctrl->device->groups = nvme_dev_attr_groups;
4575 ctrl->device->release = nvme_free_ctrl;
4576 dev_set_drvdata(ctrl->device, ctrl);
4577 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4579 goto out_release_instance;
4581 nvme_get_ctrl(ctrl);
4582 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4583 ctrl->cdev.owner = ops->module;
4584 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4589 * Initialize latency tolerance controls. The sysfs files won't
4590 * be visible to userspace unless the device actually supports APST.
4592 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4593 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4594 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4596 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4600 nvme_put_ctrl(ctrl);
4601 kfree_const(ctrl->device->kobj.name);
4602 out_release_instance:
4603 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4605 if (ctrl->discard_page)
4606 __free_page(ctrl->discard_page);
4609 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4612 * nvme_kill_queues(): Ends all namespace queues
4613 * @ctrl: the dead controller that needs to end
4615 * Call this function when the driver determines it is unable to get the
4616 * controller in a state capable of servicing IO.
4618 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4622 down_read(&ctrl->namespaces_rwsem);
4624 /* Forcibly unquiesce queues to avoid blocking dispatch */
4625 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4626 blk_mq_unquiesce_queue(ctrl->admin_q);
4628 list_for_each_entry(ns, &ctrl->namespaces, list)
4629 nvme_set_queue_dying(ns);
4631 up_read(&ctrl->namespaces_rwsem);
4633 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4635 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4639 down_read(&ctrl->namespaces_rwsem);
4640 list_for_each_entry(ns, &ctrl->namespaces, list)
4641 blk_mq_unfreeze_queue(ns->queue);
4642 up_read(&ctrl->namespaces_rwsem);
4644 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4646 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4650 down_read(&ctrl->namespaces_rwsem);
4651 list_for_each_entry(ns, &ctrl->namespaces, list) {
4652 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4656 up_read(&ctrl->namespaces_rwsem);
4659 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4661 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4665 down_read(&ctrl->namespaces_rwsem);
4666 list_for_each_entry(ns, &ctrl->namespaces, list)
4667 blk_mq_freeze_queue_wait(ns->queue);
4668 up_read(&ctrl->namespaces_rwsem);
4670 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4672 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4676 down_read(&ctrl->namespaces_rwsem);
4677 list_for_each_entry(ns, &ctrl->namespaces, list)
4678 blk_freeze_queue_start(ns->queue);
4679 up_read(&ctrl->namespaces_rwsem);
4681 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4683 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4687 down_read(&ctrl->namespaces_rwsem);
4688 list_for_each_entry(ns, &ctrl->namespaces, list)
4689 blk_mq_quiesce_queue(ns->queue);
4690 up_read(&ctrl->namespaces_rwsem);
4692 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4694 void nvme_start_queues(struct nvme_ctrl *ctrl)
4698 down_read(&ctrl->namespaces_rwsem);
4699 list_for_each_entry(ns, &ctrl->namespaces, list)
4700 blk_mq_unquiesce_queue(ns->queue);
4701 up_read(&ctrl->namespaces_rwsem);
4703 EXPORT_SYMBOL_GPL(nvme_start_queues);
4705 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4709 down_read(&ctrl->namespaces_rwsem);
4710 list_for_each_entry(ns, &ctrl->namespaces, list)
4711 blk_sync_queue(ns->queue);
4712 up_read(&ctrl->namespaces_rwsem);
4714 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4716 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4718 nvme_sync_io_queues(ctrl);
4720 blk_sync_queue(ctrl->admin_q);
4722 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4724 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4726 if (file->f_op != &nvme_dev_fops)
4728 return file->private_data;
4730 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4733 * Check we didn't inadvertently grow the command structure sizes:
4735 static inline void _nvme_check_size(void)
4737 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4738 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4739 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4740 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4741 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4742 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4743 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4744 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4745 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4746 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4747 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4748 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4749 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4750 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4751 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4752 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4753 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4754 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4755 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4759 static int __init nvme_core_init(void)
4761 int result = -ENOMEM;
4765 nvme_wq = alloc_workqueue("nvme-wq",
4766 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4770 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4771 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4775 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4776 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4777 if (!nvme_delete_wq)
4778 goto destroy_reset_wq;
4780 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4781 NVME_MINORS, "nvme");
4783 goto destroy_delete_wq;
4785 nvme_class = class_create(THIS_MODULE, "nvme");
4786 if (IS_ERR(nvme_class)) {
4787 result = PTR_ERR(nvme_class);
4788 goto unregister_chrdev;
4790 nvme_class->dev_uevent = nvme_class_uevent;
4792 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4793 if (IS_ERR(nvme_subsys_class)) {
4794 result = PTR_ERR(nvme_subsys_class);
4800 class_destroy(nvme_class);
4802 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4804 destroy_workqueue(nvme_delete_wq);
4806 destroy_workqueue(nvme_reset_wq);
4808 destroy_workqueue(nvme_wq);
4813 static void __exit nvme_core_exit(void)
4815 class_destroy(nvme_subsys_class);
4816 class_destroy(nvme_class);
4817 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4818 destroy_workqueue(nvme_delete_wq);
4819 destroy_workqueue(nvme_reset_wq);
4820 destroy_workqueue(nvme_wq);
4821 ida_destroy(&nvme_instance_ida);
4824 MODULE_LICENSE("GPL");
4825 MODULE_VERSION("1.0");
4826 module_init(nvme_core_init);
4827 module_exit(nvme_core_exit);