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 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
384 blk_mq_complete_request(req);
387 EXPORT_SYMBOL_GPL(nvme_cancel_request);
389 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
392 blk_mq_tagset_busy_iter(ctrl->tagset,
393 nvme_cancel_request, ctrl);
394 blk_mq_tagset_wait_completed_request(ctrl->tagset);
397 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
399 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
401 if (ctrl->admin_tagset) {
402 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
403 nvme_cancel_request, ctrl);
404 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
407 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
409 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
410 enum nvme_ctrl_state new_state)
412 enum nvme_ctrl_state old_state;
414 bool changed = false;
416 spin_lock_irqsave(&ctrl->lock, flags);
418 old_state = ctrl->state;
423 case NVME_CTRL_RESETTING:
424 case NVME_CTRL_CONNECTING:
431 case NVME_CTRL_RESETTING:
441 case NVME_CTRL_CONNECTING:
444 case NVME_CTRL_RESETTING:
451 case NVME_CTRL_DELETING:
454 case NVME_CTRL_RESETTING:
455 case NVME_CTRL_CONNECTING:
462 case NVME_CTRL_DELETING_NOIO:
464 case NVME_CTRL_DELETING:
474 case NVME_CTRL_DELETING:
486 ctrl->state = new_state;
487 wake_up_all(&ctrl->state_wq);
490 spin_unlock_irqrestore(&ctrl->lock, flags);
494 if (ctrl->state == NVME_CTRL_LIVE) {
495 if (old_state == NVME_CTRL_CONNECTING)
496 nvme_stop_failfast_work(ctrl);
497 nvme_kick_requeue_lists(ctrl);
498 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
499 old_state == NVME_CTRL_RESETTING) {
500 nvme_start_failfast_work(ctrl);
504 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
507 * Returns true for sink states that can't ever transition back to live.
509 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
511 switch (ctrl->state) {
514 case NVME_CTRL_RESETTING:
515 case NVME_CTRL_CONNECTING:
517 case NVME_CTRL_DELETING:
518 case NVME_CTRL_DELETING_NOIO:
522 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
528 * Waits for the controller state to be resetting, or returns false if it is
529 * not possible to ever transition to that state.
531 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
533 wait_event(ctrl->state_wq,
534 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
535 nvme_state_terminal(ctrl));
536 return ctrl->state == NVME_CTRL_RESETTING;
538 EXPORT_SYMBOL_GPL(nvme_wait_reset);
540 static void nvme_free_ns_head(struct kref *ref)
542 struct nvme_ns_head *head =
543 container_of(ref, struct nvme_ns_head, ref);
545 nvme_mpath_remove_disk(head);
546 ida_simple_remove(&head->subsys->ns_ida, head->instance);
547 cleanup_srcu_struct(&head->srcu);
548 nvme_put_subsystem(head->subsys);
552 static void nvme_put_ns_head(struct nvme_ns_head *head)
554 kref_put(&head->ref, nvme_free_ns_head);
557 static void nvme_free_ns(struct kref *kref)
559 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
562 nvme_nvm_unregister(ns);
565 nvme_put_ns_head(ns->head);
566 nvme_put_ctrl(ns->ctrl);
570 void nvme_put_ns(struct nvme_ns *ns)
572 kref_put(&ns->kref, nvme_free_ns);
574 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
576 static inline void nvme_clear_nvme_request(struct request *req)
578 struct nvme_command *cmd = nvme_req(req)->cmd;
580 memset(cmd, 0, sizeof(*cmd));
581 nvme_req(req)->retries = 0;
582 nvme_req(req)->flags = 0;
583 req->rq_flags |= RQF_DONTPREP;
586 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
588 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
591 static inline void nvme_init_request(struct request *req,
592 struct nvme_command *cmd)
594 if (req->q->queuedata)
595 req->timeout = NVME_IO_TIMEOUT;
596 else /* no queuedata implies admin queue */
597 req->timeout = NVME_ADMIN_TIMEOUT;
599 /* passthru commands should let the driver set the SGL flags */
600 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
602 req->cmd_flags |= REQ_FAILFAST_DRIVER;
603 nvme_clear_nvme_request(req);
604 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
607 struct request *nvme_alloc_request(struct request_queue *q,
608 struct nvme_command *cmd, blk_mq_req_flags_t flags)
612 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
614 nvme_init_request(req, cmd);
617 EXPORT_SYMBOL_GPL(nvme_alloc_request);
619 static struct request *nvme_alloc_request_qid(struct request_queue *q,
620 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
624 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
627 nvme_init_request(req, cmd);
631 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
633 struct nvme_command c;
635 memset(&c, 0, sizeof(c));
637 c.directive.opcode = nvme_admin_directive_send;
638 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
639 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
640 c.directive.dtype = NVME_DIR_IDENTIFY;
641 c.directive.tdtype = NVME_DIR_STREAMS;
642 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
644 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
647 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
649 return nvme_toggle_streams(ctrl, false);
652 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
654 return nvme_toggle_streams(ctrl, true);
657 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
658 struct streams_directive_params *s, u32 nsid)
660 struct nvme_command c;
662 memset(&c, 0, sizeof(c));
663 memset(s, 0, sizeof(*s));
665 c.directive.opcode = nvme_admin_directive_recv;
666 c.directive.nsid = cpu_to_le32(nsid);
667 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
668 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
669 c.directive.dtype = NVME_DIR_STREAMS;
671 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
674 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
676 struct streams_directive_params s;
679 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
684 ret = nvme_enable_streams(ctrl);
688 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
690 goto out_disable_stream;
692 ctrl->nssa = le16_to_cpu(s.nssa);
693 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
694 dev_info(ctrl->device, "too few streams (%u) available\n",
696 goto out_disable_stream;
699 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
700 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
704 nvme_disable_streams(ctrl);
709 * Check if 'req' has a write hint associated with it. If it does, assign
710 * a valid namespace stream to the write.
712 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
713 struct request *req, u16 *control,
716 enum rw_hint streamid = req->write_hint;
718 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
722 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
725 *control |= NVME_RW_DTYPE_STREAMS;
726 *dsmgmt |= streamid << 16;
729 if (streamid < ARRAY_SIZE(req->q->write_hints))
730 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
733 static inline void nvme_setup_flush(struct nvme_ns *ns,
734 struct nvme_command *cmnd)
736 cmnd->common.opcode = nvme_cmd_flush;
737 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
740 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
741 struct nvme_command *cmnd)
743 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
744 struct nvme_dsm_range *range;
748 * Some devices do not consider the DSM 'Number of Ranges' field when
749 * determining how much data to DMA. Always allocate memory for maximum
750 * number of segments to prevent device reading beyond end of buffer.
752 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
754 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
757 * If we fail allocation our range, fallback to the controller
758 * discard page. If that's also busy, it's safe to return
759 * busy, as we know we can make progress once that's freed.
761 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
762 return BLK_STS_RESOURCE;
764 range = page_address(ns->ctrl->discard_page);
767 __rq_for_each_bio(bio, req) {
768 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
769 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
772 range[n].cattr = cpu_to_le32(0);
773 range[n].nlb = cpu_to_le32(nlb);
774 range[n].slba = cpu_to_le64(slba);
779 if (WARN_ON_ONCE(n != segments)) {
780 if (virt_to_page(range) == ns->ctrl->discard_page)
781 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
784 return BLK_STS_IOERR;
787 cmnd->dsm.opcode = nvme_cmd_dsm;
788 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
789 cmnd->dsm.nr = cpu_to_le32(segments - 1);
790 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
792 req->special_vec.bv_page = virt_to_page(range);
793 req->special_vec.bv_offset = offset_in_page(range);
794 req->special_vec.bv_len = alloc_size;
795 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
800 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
801 struct request *req, struct nvme_command *cmnd)
803 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
804 return nvme_setup_discard(ns, req, cmnd);
806 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
807 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
808 cmnd->write_zeroes.slba =
809 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
810 cmnd->write_zeroes.length =
811 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
812 cmnd->write_zeroes.control = 0;
816 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
817 struct request *req, struct nvme_command *cmnd,
820 struct nvme_ctrl *ctrl = ns->ctrl;
824 if (req->cmd_flags & REQ_FUA)
825 control |= NVME_RW_FUA;
826 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
827 control |= NVME_RW_LR;
829 if (req->cmd_flags & REQ_RAHEAD)
830 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
832 cmnd->rw.opcode = op;
833 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
834 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
835 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
837 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
838 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
842 * If formated with metadata, the block layer always provides a
843 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
844 * we enable the PRACT bit for protection information or set the
845 * namespace capacity to zero to prevent any I/O.
847 if (!blk_integrity_rq(req)) {
848 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
849 return BLK_STS_NOTSUPP;
850 control |= NVME_RW_PRINFO_PRACT;
853 switch (ns->pi_type) {
854 case NVME_NS_DPS_PI_TYPE3:
855 control |= NVME_RW_PRINFO_PRCHK_GUARD;
857 case NVME_NS_DPS_PI_TYPE1:
858 case NVME_NS_DPS_PI_TYPE2:
859 control |= NVME_RW_PRINFO_PRCHK_GUARD |
860 NVME_RW_PRINFO_PRCHK_REF;
861 if (op == nvme_cmd_zone_append)
862 control |= NVME_RW_APPEND_PIREMAP;
863 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
868 cmnd->rw.control = cpu_to_le16(control);
869 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
873 void nvme_cleanup_cmd(struct request *req)
875 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
876 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
877 struct page *page = req->special_vec.bv_page;
879 if (page == ctrl->discard_page)
880 clear_bit_unlock(0, &ctrl->discard_page_busy);
882 kfree(page_address(page) + req->special_vec.bv_offset);
885 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
887 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
889 struct nvme_command *cmd = nvme_req(req)->cmd;
890 blk_status_t ret = BLK_STS_OK;
892 if (!(req->rq_flags & RQF_DONTPREP))
893 nvme_clear_nvme_request(req);
895 switch (req_op(req)) {
898 /* these are setup prior to execution in nvme_init_request() */
901 nvme_setup_flush(ns, cmd);
903 case REQ_OP_ZONE_RESET_ALL:
904 case REQ_OP_ZONE_RESET:
905 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
907 case REQ_OP_ZONE_OPEN:
908 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
910 case REQ_OP_ZONE_CLOSE:
911 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
913 case REQ_OP_ZONE_FINISH:
914 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
916 case REQ_OP_WRITE_ZEROES:
917 ret = nvme_setup_write_zeroes(ns, req, cmd);
920 ret = nvme_setup_discard(ns, req, cmd);
923 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
926 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
928 case REQ_OP_ZONE_APPEND:
929 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
933 return BLK_STS_IOERR;
936 cmd->common.command_id = req->tag;
937 trace_nvme_setup_cmd(req, cmd);
940 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
942 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
944 struct completion *waiting = rq->end_io_data;
946 rq->end_io_data = NULL;
950 static void nvme_execute_rq_polled(struct request_queue *q,
951 struct gendisk *bd_disk, struct request *rq, int at_head)
953 DECLARE_COMPLETION_ONSTACK(wait);
955 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
957 rq->cmd_flags |= REQ_HIPRI;
958 rq->end_io_data = &wait;
959 blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
961 while (!completion_done(&wait)) {
962 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
968 * Returns 0 on success. If the result is negative, it's a Linux error code;
969 * if the result is positive, it's an NVM Express status code
971 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
972 union nvme_result *result, void *buffer, unsigned bufflen,
973 unsigned timeout, int qid, int at_head,
974 blk_mq_req_flags_t flags, bool poll)
979 if (qid == NVME_QID_ANY)
980 req = nvme_alloc_request(q, cmd, flags);
982 req = nvme_alloc_request_qid(q, cmd, flags, qid);
987 req->timeout = timeout;
989 if (buffer && bufflen) {
990 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
996 nvme_execute_rq_polled(req->q, NULL, req, at_head);
998 blk_execute_rq(NULL, req, at_head);
1000 *result = nvme_req(req)->result;
1001 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1004 ret = nvme_req(req)->status;
1006 blk_mq_free_request(req);
1009 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1011 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1012 void *buffer, unsigned bufflen)
1014 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1015 NVME_QID_ANY, 0, 0, false);
1017 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1019 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1020 unsigned len, u32 seed, bool write)
1022 struct bio_integrity_payload *bip;
1026 buf = kmalloc(len, GFP_KERNEL);
1031 if (write && copy_from_user(buf, ubuf, len))
1034 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1040 bip->bip_iter.bi_size = len;
1041 bip->bip_iter.bi_sector = seed;
1042 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1043 offset_in_page(buf));
1050 return ERR_PTR(ret);
1053 static u32 nvme_known_admin_effects(u8 opcode)
1056 case nvme_admin_format_nvm:
1057 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1058 NVME_CMD_EFFECTS_CSE_MASK;
1059 case nvme_admin_sanitize_nvm:
1060 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1067 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1072 if (ns->head->effects)
1073 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1074 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1075 dev_warn_once(ctrl->device,
1076 "IO command:%02x has unhandled effects:%08x\n",
1082 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1083 effects |= nvme_known_admin_effects(opcode);
1087 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1089 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1092 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1095 * For simplicity, IO to all namespaces is quiesced even if the command
1096 * effects say only one namespace is affected.
1098 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1099 mutex_lock(&ctrl->scan_lock);
1100 mutex_lock(&ctrl->subsys->lock);
1101 nvme_mpath_start_freeze(ctrl->subsys);
1102 nvme_mpath_wait_freeze(ctrl->subsys);
1103 nvme_start_freeze(ctrl);
1104 nvme_wait_freeze(ctrl);
1109 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1111 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1112 nvme_unfreeze(ctrl);
1113 nvme_mpath_unfreeze(ctrl->subsys);
1114 mutex_unlock(&ctrl->subsys->lock);
1115 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1116 mutex_unlock(&ctrl->scan_lock);
1118 if (effects & NVME_CMD_EFFECTS_CCC)
1119 nvme_init_ctrl_finish(ctrl);
1120 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1121 nvme_queue_scan(ctrl);
1122 flush_work(&ctrl->scan_work);
1126 void nvme_execute_passthru_rq(struct request *rq)
1128 struct nvme_command *cmd = nvme_req(rq)->cmd;
1129 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1130 struct nvme_ns *ns = rq->q->queuedata;
1131 struct gendisk *disk = ns ? ns->disk : NULL;
1134 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1135 blk_execute_rq(disk, rq, 0);
1136 if (effects) /* nothing to be done for zero cmd effects */
1137 nvme_passthru_end(ctrl, effects);
1139 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1141 static int nvme_submit_user_cmd(struct request_queue *q,
1142 struct nvme_command *cmd, void __user *ubuffer,
1143 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1144 u32 meta_seed, u64 *result, unsigned timeout)
1146 bool write = nvme_is_write(cmd);
1147 struct nvme_ns *ns = q->queuedata;
1148 struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1149 struct request *req;
1150 struct bio *bio = NULL;
1154 req = nvme_alloc_request(q, cmd, 0);
1156 return PTR_ERR(req);
1159 req->timeout = timeout;
1160 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1162 if (ubuffer && bufflen) {
1163 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1169 bio_set_dev(bio, bdev);
1170 if (bdev && meta_buffer && meta_len) {
1171 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1174 ret = PTR_ERR(meta);
1177 req->cmd_flags |= REQ_INTEGRITY;
1181 nvme_execute_passthru_rq(req);
1182 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1185 ret = nvme_req(req)->status;
1187 *result = le64_to_cpu(nvme_req(req)->result.u64);
1188 if (meta && !ret && !write) {
1189 if (copy_to_user(meta_buffer, meta, meta_len))
1195 blk_rq_unmap_user(bio);
1197 blk_mq_free_request(req);
1201 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1203 struct nvme_ctrl *ctrl = rq->end_io_data;
1204 unsigned long flags;
1205 bool startka = false;
1207 blk_mq_free_request(rq);
1210 dev_err(ctrl->device,
1211 "failed nvme_keep_alive_end_io error=%d\n",
1216 ctrl->comp_seen = false;
1217 spin_lock_irqsave(&ctrl->lock, flags);
1218 if (ctrl->state == NVME_CTRL_LIVE ||
1219 ctrl->state == NVME_CTRL_CONNECTING)
1221 spin_unlock_irqrestore(&ctrl->lock, flags);
1223 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1226 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1230 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1231 BLK_MQ_REQ_RESERVED);
1235 rq->timeout = ctrl->kato * HZ;
1236 rq->end_io_data = ctrl;
1238 blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1243 static void nvme_keep_alive_work(struct work_struct *work)
1245 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1246 struct nvme_ctrl, ka_work);
1247 bool comp_seen = ctrl->comp_seen;
1249 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1250 dev_dbg(ctrl->device,
1251 "reschedule traffic based keep-alive timer\n");
1252 ctrl->comp_seen = false;
1253 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1257 if (nvme_keep_alive(ctrl)) {
1258 /* allocation failure, reset the controller */
1259 dev_err(ctrl->device, "keep-alive failed\n");
1260 nvme_reset_ctrl(ctrl);
1265 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1267 if (unlikely(ctrl->kato == 0))
1270 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1273 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1275 if (unlikely(ctrl->kato == 0))
1278 cancel_delayed_work_sync(&ctrl->ka_work);
1280 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1283 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1284 * flag, thus sending any new CNS opcodes has a big chance of not working.
1285 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1286 * (but not for any later version).
1288 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1290 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1291 return ctrl->vs < NVME_VS(1, 2, 0);
1292 return ctrl->vs < NVME_VS(1, 1, 0);
1295 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1297 struct nvme_command c = { };
1300 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1301 c.identify.opcode = nvme_admin_identify;
1302 c.identify.cns = NVME_ID_CNS_CTRL;
1304 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1308 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1309 sizeof(struct nvme_id_ctrl));
1315 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1317 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1320 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1321 struct nvme_ns_id_desc *cur, bool *csi_seen)
1323 const char *warn_str = "ctrl returned bogus length:";
1326 switch (cur->nidt) {
1327 case NVME_NIDT_EUI64:
1328 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1329 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1330 warn_str, cur->nidl);
1333 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1334 return NVME_NIDT_EUI64_LEN;
1335 case NVME_NIDT_NGUID:
1336 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1337 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1338 warn_str, cur->nidl);
1341 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1342 return NVME_NIDT_NGUID_LEN;
1343 case NVME_NIDT_UUID:
1344 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1345 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1346 warn_str, cur->nidl);
1349 uuid_copy(&ids->uuid, data + sizeof(*cur));
1350 return NVME_NIDT_UUID_LEN;
1352 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1353 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1354 warn_str, cur->nidl);
1357 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1359 return NVME_NIDT_CSI_LEN;
1361 /* Skip unknown types */
1366 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1367 struct nvme_ns_ids *ids)
1369 struct nvme_command c = { };
1370 bool csi_seen = false;
1371 int status, pos, len;
1374 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1376 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1379 c.identify.opcode = nvme_admin_identify;
1380 c.identify.nsid = cpu_to_le32(nsid);
1381 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1383 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1387 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1388 NVME_IDENTIFY_DATA_SIZE);
1390 dev_warn(ctrl->device,
1391 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1396 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1397 struct nvme_ns_id_desc *cur = data + pos;
1402 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1406 len += sizeof(*cur);
1409 if (nvme_multi_css(ctrl) && !csi_seen) {
1410 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1420 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1421 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1423 struct nvme_command c = { };
1426 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1427 c.identify.opcode = nvme_admin_identify;
1428 c.identify.nsid = cpu_to_le32(nsid);
1429 c.identify.cns = NVME_ID_CNS_NS;
1431 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1435 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1437 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1441 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1442 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1445 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1446 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1447 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1448 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1449 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1450 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1459 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1460 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1462 union nvme_result res = { 0 };
1463 struct nvme_command c;
1466 memset(&c, 0, sizeof(c));
1467 c.features.opcode = op;
1468 c.features.fid = cpu_to_le32(fid);
1469 c.features.dword11 = cpu_to_le32(dword11);
1471 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1472 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1473 if (ret >= 0 && result)
1474 *result = le32_to_cpu(res.u32);
1478 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1479 unsigned int dword11, void *buffer, size_t buflen,
1482 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1485 EXPORT_SYMBOL_GPL(nvme_set_features);
1487 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1488 unsigned int dword11, void *buffer, size_t buflen,
1491 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1494 EXPORT_SYMBOL_GPL(nvme_get_features);
1496 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1498 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1500 int status, nr_io_queues;
1502 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1508 * Degraded controllers might return an error when setting the queue
1509 * count. We still want to be able to bring them online and offer
1510 * access to the admin queue, as that might be only way to fix them up.
1513 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1516 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1517 *count = min(*count, nr_io_queues);
1522 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1524 #define NVME_AEN_SUPPORTED \
1525 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1526 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1528 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1530 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1533 if (!supported_aens)
1536 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1539 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1542 queue_work(nvme_wq, &ctrl->async_event_work);
1546 * Convert integer values from ioctl structures to user pointers, silently
1547 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1550 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1552 if (in_compat_syscall())
1553 ptrval = (compat_uptr_t)ptrval;
1554 return (void __user *)ptrval;
1557 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1559 struct nvme_user_io io;
1560 struct nvme_command c;
1561 unsigned length, meta_len;
1562 void __user *metadata;
1564 if (copy_from_user(&io, uio, sizeof(io)))
1569 switch (io.opcode) {
1570 case nvme_cmd_write:
1572 case nvme_cmd_compare:
1578 length = (io.nblocks + 1) << ns->lba_shift;
1580 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1581 ns->ms == sizeof(struct t10_pi_tuple)) {
1583 * Protection information is stripped/inserted by the
1586 if (nvme_to_user_ptr(io.metadata))
1591 meta_len = (io.nblocks + 1) * ns->ms;
1592 metadata = nvme_to_user_ptr(io.metadata);
1595 if (ns->features & NVME_NS_EXT_LBAS) {
1598 } else if (meta_len) {
1599 if ((io.metadata & 3) || !io.metadata)
1603 memset(&c, 0, sizeof(c));
1604 c.rw.opcode = io.opcode;
1605 c.rw.flags = io.flags;
1606 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1607 c.rw.slba = cpu_to_le64(io.slba);
1608 c.rw.length = cpu_to_le16(io.nblocks);
1609 c.rw.control = cpu_to_le16(io.control);
1610 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1611 c.rw.reftag = cpu_to_le32(io.reftag);
1612 c.rw.apptag = cpu_to_le16(io.apptag);
1613 c.rw.appmask = cpu_to_le16(io.appmask);
1615 return nvme_submit_user_cmd(ns->queue, &c,
1616 nvme_to_user_ptr(io.addr), length,
1617 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1620 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1621 struct nvme_passthru_cmd __user *ucmd)
1623 struct nvme_passthru_cmd cmd;
1624 struct nvme_command c;
1625 unsigned timeout = 0;
1629 if (!capable(CAP_SYS_ADMIN))
1631 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1635 if (ns && cmd.nsid != ns->head->ns_id) {
1636 dev_err(ctrl->device,
1637 "%s: nsid (%u) in cmd does not match nsid (%u) of namespace\n",
1638 current->comm, cmd.nsid, ns->head->ns_id);
1642 memset(&c, 0, sizeof(c));
1643 c.common.opcode = cmd.opcode;
1644 c.common.flags = cmd.flags;
1645 c.common.nsid = cpu_to_le32(cmd.nsid);
1646 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1647 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1648 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1649 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1650 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1651 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1652 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1653 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1656 timeout = msecs_to_jiffies(cmd.timeout_ms);
1658 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1659 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1660 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1661 0, &result, timeout);
1664 if (put_user(result, &ucmd->result))
1671 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1672 struct nvme_passthru_cmd64 __user *ucmd)
1674 struct nvme_passthru_cmd64 cmd;
1675 struct nvme_command c;
1676 unsigned timeout = 0;
1679 if (!capable(CAP_SYS_ADMIN))
1681 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1685 if (ns && cmd.nsid != ns->head->ns_id) {
1686 dev_err(ctrl->device,
1687 "%s: nsid (%u) in cmd does not match nsid (%u) of namespace\n",
1688 current->comm, cmd.nsid, ns->head->ns_id);
1692 memset(&c, 0, sizeof(c));
1693 c.common.opcode = cmd.opcode;
1694 c.common.flags = cmd.flags;
1695 c.common.nsid = cpu_to_le32(cmd.nsid);
1696 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1697 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1698 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1699 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1700 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1701 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1702 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1703 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1706 timeout = msecs_to_jiffies(cmd.timeout_ms);
1708 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1709 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1710 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1711 0, &cmd.result, timeout);
1714 if (put_user(cmd.result, &ucmd->result))
1722 * Issue ioctl requests on the first available path. Note that unlike normal
1723 * block layer requests we will not retry failed request on another controller.
1725 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1726 struct nvme_ns_head **head, int *srcu_idx)
1728 #ifdef CONFIG_NVME_MULTIPATH
1729 if (disk->fops == &nvme_ns_head_ops) {
1732 *head = disk->private_data;
1733 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1734 ns = nvme_find_path(*head);
1736 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1742 return disk->private_data;
1745 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1748 srcu_read_unlock(&head->srcu, idx);
1751 static bool is_ctrl_ioctl(unsigned int cmd)
1753 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1755 if (is_sed_ioctl(cmd))
1760 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1762 struct nvme_ns_head *head,
1765 struct nvme_ctrl *ctrl = ns->ctrl;
1768 nvme_get_ctrl(ns->ctrl);
1769 nvme_put_ns_from_disk(head, srcu_idx);
1772 case NVME_IOCTL_ADMIN_CMD:
1773 ret = nvme_user_cmd(ctrl, NULL, argp);
1775 case NVME_IOCTL_ADMIN64_CMD:
1776 ret = nvme_user_cmd64(ctrl, NULL, argp);
1779 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1782 nvme_put_ctrl(ctrl);
1786 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1787 unsigned int cmd, unsigned long arg)
1789 struct nvme_ns_head *head = NULL;
1790 void __user *argp = (void __user *)arg;
1794 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1796 return -EWOULDBLOCK;
1799 * Handle ioctls that apply to the controller instead of the namespace
1800 * seperately and drop the ns SRCU reference early. This avoids a
1801 * deadlock when deleting namespaces using the passthrough interface.
1803 if (is_ctrl_ioctl(cmd))
1804 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1808 force_successful_syscall_return();
1809 ret = ns->head->ns_id;
1811 case NVME_IOCTL_IO_CMD:
1812 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1814 case NVME_IOCTL_SUBMIT_IO:
1815 ret = nvme_submit_io(ns, argp);
1817 case NVME_IOCTL_IO64_CMD:
1818 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1822 ret = nvme_nvm_ioctl(ns, cmd, arg);
1827 nvme_put_ns_from_disk(head, srcu_idx);
1831 #ifdef CONFIG_COMPAT
1832 struct nvme_user_io32 {
1845 } __attribute__((__packed__));
1847 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1849 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1850 unsigned int cmd, unsigned long arg)
1853 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1854 * between 32 bit programs and 64 bit kernel.
1855 * The cause is that the results of sizeof(struct nvme_user_io),
1856 * which is used to define NVME_IOCTL_SUBMIT_IO,
1857 * are not same between 32 bit compiler and 64 bit compiler.
1858 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1859 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1860 * Other IOCTL numbers are same between 32 bit and 64 bit.
1861 * So there is nothing to do regarding to other IOCTL numbers.
1863 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1864 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1866 return nvme_ioctl(bdev, mode, cmd, arg);
1869 #define nvme_compat_ioctl NULL
1870 #endif /* CONFIG_COMPAT */
1872 static int nvme_open(struct block_device *bdev, fmode_t mode)
1874 struct nvme_ns *ns = bdev->bd_disk->private_data;
1876 /* should never be called due to GENHD_FL_HIDDEN */
1877 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1879 if (!kref_get_unless_zero(&ns->kref))
1881 if (!try_module_get(ns->ctrl->ops->module))
1892 static void nvme_release(struct gendisk *disk, fmode_t mode)
1894 struct nvme_ns *ns = disk->private_data;
1896 module_put(ns->ctrl->ops->module);
1900 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1902 /* some standard values */
1903 geo->heads = 1 << 6;
1904 geo->sectors = 1 << 5;
1905 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1909 #ifdef CONFIG_BLK_DEV_INTEGRITY
1910 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1911 u32 max_integrity_segments)
1913 struct blk_integrity integrity;
1915 memset(&integrity, 0, sizeof(integrity));
1917 case NVME_NS_DPS_PI_TYPE3:
1918 integrity.profile = &t10_pi_type3_crc;
1919 integrity.tag_size = sizeof(u16) + sizeof(u32);
1920 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1922 case NVME_NS_DPS_PI_TYPE1:
1923 case NVME_NS_DPS_PI_TYPE2:
1924 integrity.profile = &t10_pi_type1_crc;
1925 integrity.tag_size = sizeof(u16);
1926 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1929 integrity.profile = NULL;
1932 integrity.tuple_size = ms;
1933 blk_integrity_register(disk, &integrity);
1934 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1937 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1938 u32 max_integrity_segments)
1941 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1943 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1945 struct nvme_ctrl *ctrl = ns->ctrl;
1946 struct request_queue *queue = disk->queue;
1947 u32 size = queue_logical_block_size(queue);
1949 if (ctrl->max_discard_sectors == 0) {
1950 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1954 if (ctrl->nr_streams && ns->sws && ns->sgs)
1955 size *= ns->sws * ns->sgs;
1957 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1958 NVME_DSM_MAX_RANGES);
1960 queue->limits.discard_alignment = 0;
1961 queue->limits.discard_granularity = size;
1963 /* If discard is already enabled, don't reset queue limits */
1964 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1967 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1968 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1970 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1971 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1974 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1976 return !uuid_is_null(&ids->uuid) ||
1977 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1978 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1981 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1983 return uuid_equal(&a->uuid, &b->uuid) &&
1984 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1985 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1989 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1990 u32 *phys_bs, u32 *io_opt)
1992 struct streams_directive_params s;
1995 if (!ctrl->nr_streams)
1998 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
2002 ns->sws = le32_to_cpu(s.sws);
2003 ns->sgs = le16_to_cpu(s.sgs);
2006 *phys_bs = ns->sws * (1 << ns->lba_shift);
2008 *io_opt = *phys_bs * ns->sgs;
2014 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2016 struct nvme_ctrl *ctrl = ns->ctrl;
2019 * The PI implementation requires the metadata size to be equal to the
2020 * t10 pi tuple size.
2022 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2023 if (ns->ms == sizeof(struct t10_pi_tuple))
2024 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2028 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2029 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2031 if (ctrl->ops->flags & NVME_F_FABRICS) {
2033 * The NVMe over Fabrics specification only supports metadata as
2034 * part of the extended data LBA. We rely on HCA/HBA support to
2035 * remap the separate metadata buffer from the block layer.
2037 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2039 if (ctrl->max_integrity_segments)
2041 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2044 * For PCIe controllers, we can't easily remap the separate
2045 * metadata buffer from the block layer and thus require a
2046 * separate metadata buffer for block layer metadata/PI support.
2047 * We allow extended LBAs for the passthrough interface, though.
2049 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2050 ns->features |= NVME_NS_EXT_LBAS;
2052 ns->features |= NVME_NS_METADATA_SUPPORTED;
2058 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2059 struct request_queue *q)
2061 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2063 if (ctrl->max_hw_sectors) {
2065 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2067 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2068 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2069 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2071 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2072 blk_queue_dma_alignment(q, 7);
2073 blk_queue_write_cache(q, vwc, vwc);
2076 static void nvme_update_disk_info(struct gendisk *disk,
2077 struct nvme_ns *ns, struct nvme_id_ns *id)
2079 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2080 unsigned short bs = 1 << ns->lba_shift;
2081 u32 atomic_bs, phys_bs, io_opt = 0;
2084 * The block layer can't support LBA sizes larger than the page size
2085 * yet, so catch this early and don't allow block I/O.
2087 if (ns->lba_shift > PAGE_SHIFT) {
2092 blk_integrity_unregister(disk);
2094 atomic_bs = phys_bs = bs;
2095 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2096 if (id->nabo == 0) {
2098 * Bit 1 indicates whether NAWUPF is defined for this namespace
2099 * and whether it should be used instead of AWUPF. If NAWUPF ==
2100 * 0 then AWUPF must be used instead.
2102 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2103 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2105 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2108 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2109 /* NPWG = Namespace Preferred Write Granularity */
2110 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2111 /* NOWS = Namespace Optimal Write Size */
2112 io_opt = bs * (1 + le16_to_cpu(id->nows));
2115 blk_queue_logical_block_size(disk->queue, bs);
2117 * Linux filesystems assume writing a single physical block is
2118 * an atomic operation. Hence limit the physical block size to the
2119 * value of the Atomic Write Unit Power Fail parameter.
2121 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2122 blk_queue_io_min(disk->queue, phys_bs);
2123 blk_queue_io_opt(disk->queue, io_opt);
2126 * Register a metadata profile for PI, or the plain non-integrity NVMe
2127 * metadata masquerading as Type 0 if supported, otherwise reject block
2128 * I/O to namespaces with metadata except when the namespace supports
2129 * PI, as it can strip/insert in that case.
2132 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2133 (ns->features & NVME_NS_METADATA_SUPPORTED))
2134 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2135 ns->ctrl->max_integrity_segments);
2136 else if (!nvme_ns_has_pi(ns))
2140 set_capacity_and_notify(disk, capacity);
2142 nvme_config_discard(disk, ns);
2143 blk_queue_max_write_zeroes_sectors(disk->queue,
2144 ns->ctrl->max_zeroes_sectors);
2146 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2147 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2150 static inline bool nvme_first_scan(struct gendisk *disk)
2152 /* nvme_alloc_ns() scans the disk prior to adding it */
2153 return !(disk->flags & GENHD_FL_UP);
2156 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2158 struct nvme_ctrl *ctrl = ns->ctrl;
2161 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2162 is_power_of_2(ctrl->max_hw_sectors))
2163 iob = ctrl->max_hw_sectors;
2165 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2170 if (!is_power_of_2(iob)) {
2171 if (nvme_first_scan(ns->disk))
2172 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2173 ns->disk->disk_name, iob);
2177 if (blk_queue_is_zoned(ns->disk->queue)) {
2178 if (nvme_first_scan(ns->disk))
2179 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2180 ns->disk->disk_name);
2184 blk_queue_chunk_sectors(ns->queue, iob);
2187 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2189 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2192 blk_mq_freeze_queue(ns->disk->queue);
2193 ns->lba_shift = id->lbaf[lbaf].ds;
2194 nvme_set_queue_limits(ns->ctrl, ns->queue);
2196 ret = nvme_configure_metadata(ns, id);
2199 nvme_set_chunk_sectors(ns, id);
2200 nvme_update_disk_info(ns->disk, ns, id);
2202 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2203 ret = nvme_update_zone_info(ns, lbaf);
2208 blk_mq_unfreeze_queue(ns->disk->queue);
2210 if (blk_queue_is_zoned(ns->queue)) {
2211 ret = nvme_revalidate_zones(ns);
2212 if (ret && !nvme_first_scan(ns->disk))
2216 if (nvme_ns_head_multipath(ns->head)) {
2217 blk_mq_freeze_queue(ns->head->disk->queue);
2218 nvme_update_disk_info(ns->head->disk, ns, id);
2219 blk_stack_limits(&ns->head->disk->queue->limits,
2220 &ns->queue->limits, 0);
2221 blk_queue_update_readahead(ns->head->disk->queue);
2222 blk_mq_unfreeze_queue(ns->head->disk->queue);
2227 blk_mq_unfreeze_queue(ns->disk->queue);
2231 static char nvme_pr_type(enum pr_type type)
2234 case PR_WRITE_EXCLUSIVE:
2236 case PR_EXCLUSIVE_ACCESS:
2238 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2240 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2242 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2244 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2251 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2252 u64 key, u64 sa_key, u8 op)
2254 struct nvme_ns_head *head = NULL;
2256 struct nvme_command c;
2258 u8 data[16] = { 0, };
2260 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2262 return -EWOULDBLOCK;
2264 put_unaligned_le64(key, &data[0]);
2265 put_unaligned_le64(sa_key, &data[8]);
2267 memset(&c, 0, sizeof(c));
2268 c.common.opcode = op;
2269 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2270 c.common.cdw10 = cpu_to_le32(cdw10);
2272 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2273 nvme_put_ns_from_disk(head, srcu_idx);
2277 static int nvme_pr_register(struct block_device *bdev, u64 old,
2278 u64 new, unsigned flags)
2282 if (flags & ~PR_FL_IGNORE_KEY)
2285 cdw10 = old ? 2 : 0;
2286 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2287 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2288 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2291 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2292 enum pr_type type, unsigned flags)
2296 if (flags & ~PR_FL_IGNORE_KEY)
2299 cdw10 = nvme_pr_type(type) << 8;
2300 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2301 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2304 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2305 enum pr_type type, bool abort)
2307 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2309 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2312 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2314 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2316 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2319 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2321 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2323 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2326 static const struct pr_ops nvme_pr_ops = {
2327 .pr_register = nvme_pr_register,
2328 .pr_reserve = nvme_pr_reserve,
2329 .pr_release = nvme_pr_release,
2330 .pr_preempt = nvme_pr_preempt,
2331 .pr_clear = nvme_pr_clear,
2334 #ifdef CONFIG_BLK_SED_OPAL
2335 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2338 struct nvme_ctrl *ctrl = data;
2339 struct nvme_command cmd;
2341 memset(&cmd, 0, sizeof(cmd));
2343 cmd.common.opcode = nvme_admin_security_send;
2345 cmd.common.opcode = nvme_admin_security_recv;
2346 cmd.common.nsid = 0;
2347 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2348 cmd.common.cdw11 = cpu_to_le32(len);
2350 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2351 NVME_QID_ANY, 1, 0, false);
2353 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2354 #endif /* CONFIG_BLK_SED_OPAL */
2356 static const struct block_device_operations nvme_bdev_ops = {
2357 .owner = THIS_MODULE,
2358 .ioctl = nvme_ioctl,
2359 .compat_ioctl = nvme_compat_ioctl,
2361 .release = nvme_release,
2362 .getgeo = nvme_getgeo,
2363 .report_zones = nvme_report_zones,
2364 .pr_ops = &nvme_pr_ops,
2367 #ifdef CONFIG_NVME_MULTIPATH
2368 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2370 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2372 if (!kref_get_unless_zero(&head->ref))
2377 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2379 nvme_put_ns_head(disk->private_data);
2382 const struct block_device_operations nvme_ns_head_ops = {
2383 .owner = THIS_MODULE,
2384 .submit_bio = nvme_ns_head_submit_bio,
2385 .open = nvme_ns_head_open,
2386 .release = nvme_ns_head_release,
2387 .ioctl = nvme_ioctl,
2388 .compat_ioctl = nvme_compat_ioctl,
2389 .getgeo = nvme_getgeo,
2390 .report_zones = nvme_report_zones,
2391 .pr_ops = &nvme_pr_ops,
2393 #endif /* CONFIG_NVME_MULTIPATH */
2395 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2397 unsigned long timeout =
2398 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2399 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2402 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2405 if ((csts & NVME_CSTS_RDY) == bit)
2408 usleep_range(1000, 2000);
2409 if (fatal_signal_pending(current))
2411 if (time_after(jiffies, timeout)) {
2412 dev_err(ctrl->device,
2413 "Device not ready; aborting %s, CSTS=0x%x\n",
2414 enabled ? "initialisation" : "reset", csts);
2423 * If the device has been passed off to us in an enabled state, just clear
2424 * the enabled bit. The spec says we should set the 'shutdown notification
2425 * bits', but doing so may cause the device to complete commands to the
2426 * admin queue ... and we don't know what memory that might be pointing at!
2428 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2432 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2433 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2435 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2439 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2440 msleep(NVME_QUIRK_DELAY_AMOUNT);
2442 return nvme_wait_ready(ctrl, ctrl->cap, false);
2444 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2446 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2448 unsigned dev_page_min;
2451 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2453 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2456 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2458 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2459 dev_err(ctrl->device,
2460 "Minimum device page size %u too large for host (%u)\n",
2461 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2465 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2466 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2468 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2469 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2470 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2471 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2472 ctrl->ctrl_config |= NVME_CC_ENABLE;
2474 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2477 return nvme_wait_ready(ctrl, ctrl->cap, true);
2479 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2481 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2483 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2487 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2488 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2490 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2494 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2495 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2499 if (fatal_signal_pending(current))
2501 if (time_after(jiffies, timeout)) {
2502 dev_err(ctrl->device,
2503 "Device shutdown incomplete; abort shutdown\n");
2510 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2512 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2517 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2520 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2521 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2524 dev_warn_once(ctrl->device,
2525 "could not set timestamp (%d)\n", ret);
2529 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2531 struct nvme_feat_host_behavior *host;
2534 /* Don't bother enabling the feature if retry delay is not reported */
2538 host = kzalloc(sizeof(*host), GFP_KERNEL);
2542 host->acre = NVME_ENABLE_ACRE;
2543 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2544 host, sizeof(*host), NULL);
2549 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2552 * APST (Autonomous Power State Transition) lets us program a
2553 * table of power state transitions that the controller will
2554 * perform automatically. We configure it with a simple
2555 * heuristic: we are willing to spend at most 2% of the time
2556 * transitioning between power states. Therefore, when running
2557 * in any given state, we will enter the next lower-power
2558 * non-operational state after waiting 50 * (enlat + exlat)
2559 * microseconds, as long as that state's exit latency is under
2560 * the requested maximum latency.
2562 * We will not autonomously enter any non-operational state for
2563 * which the total latency exceeds ps_max_latency_us. Users
2564 * can set ps_max_latency_us to zero to turn off APST.
2568 struct nvme_feat_auto_pst *table;
2574 * If APST isn't supported or if we haven't been initialized yet,
2575 * then don't do anything.
2580 if (ctrl->npss > 31) {
2581 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2585 table = kzalloc(sizeof(*table), GFP_KERNEL);
2589 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2590 /* Turn off APST. */
2592 dev_dbg(ctrl->device, "APST disabled\n");
2594 __le64 target = cpu_to_le64(0);
2598 * Walk through all states from lowest- to highest-power.
2599 * According to the spec, lower-numbered states use more
2600 * power. NPSS, despite the name, is the index of the
2601 * lowest-power state, not the number of states.
2603 for (state = (int)ctrl->npss; state >= 0; state--) {
2604 u64 total_latency_us, exit_latency_us, transition_ms;
2607 table->entries[state] = target;
2610 * Don't allow transitions to the deepest state
2611 * if it's quirked off.
2613 if (state == ctrl->npss &&
2614 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2618 * Is this state a useful non-operational state for
2619 * higher-power states to autonomously transition to?
2621 if (!(ctrl->psd[state].flags &
2622 NVME_PS_FLAGS_NON_OP_STATE))
2626 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2627 if (exit_latency_us > ctrl->ps_max_latency_us)
2632 le32_to_cpu(ctrl->psd[state].entry_lat);
2635 * This state is good. Use it as the APST idle
2636 * target for higher power states.
2638 transition_ms = total_latency_us + 19;
2639 do_div(transition_ms, 20);
2640 if (transition_ms > (1 << 24) - 1)
2641 transition_ms = (1 << 24) - 1;
2643 target = cpu_to_le64((state << 3) |
2644 (transition_ms << 8));
2649 if (total_latency_us > max_lat_us)
2650 max_lat_us = total_latency_us;
2656 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2658 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2659 max_ps, max_lat_us, (int)sizeof(*table), table);
2663 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2664 table, sizeof(*table), NULL);
2666 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2672 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2674 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2678 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2679 case PM_QOS_LATENCY_ANY:
2687 if (ctrl->ps_max_latency_us != latency) {
2688 ctrl->ps_max_latency_us = latency;
2689 nvme_configure_apst(ctrl);
2693 struct nvme_core_quirk_entry {
2695 * NVMe model and firmware strings are padded with spaces. For
2696 * simplicity, strings in the quirk table are padded with NULLs
2702 unsigned long quirks;
2705 static const struct nvme_core_quirk_entry core_quirks[] = {
2708 * This Toshiba device seems to die using any APST states. See:
2709 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2712 .mn = "THNSF5256GPUK TOSHIBA",
2713 .quirks = NVME_QUIRK_NO_APST,
2717 * This LiteON CL1-3D*-Q11 firmware version has a race
2718 * condition associated with actions related to suspend to idle
2719 * LiteON has resolved the problem in future firmware
2723 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2727 /* match is null-terminated but idstr is space-padded. */
2728 static bool string_matches(const char *idstr, const char *match, size_t len)
2735 matchlen = strlen(match);
2736 WARN_ON_ONCE(matchlen > len);
2738 if (memcmp(idstr, match, matchlen))
2741 for (; matchlen < len; matchlen++)
2742 if (idstr[matchlen] != ' ')
2748 static bool quirk_matches(const struct nvme_id_ctrl *id,
2749 const struct nvme_core_quirk_entry *q)
2751 return q->vid == le16_to_cpu(id->vid) &&
2752 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2753 string_matches(id->fr, q->fr, sizeof(id->fr));
2756 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2757 struct nvme_id_ctrl *id)
2762 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2763 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2764 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2765 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2769 if (ctrl->vs >= NVME_VS(1, 2, 1))
2770 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2773 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2774 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2775 "nqn.2014.08.org.nvmexpress:%04x%04x",
2776 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2777 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2778 off += sizeof(id->sn);
2779 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2780 off += sizeof(id->mn);
2781 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2784 static void nvme_release_subsystem(struct device *dev)
2786 struct nvme_subsystem *subsys =
2787 container_of(dev, struct nvme_subsystem, dev);
2789 if (subsys->instance >= 0)
2790 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2794 static void nvme_destroy_subsystem(struct kref *ref)
2796 struct nvme_subsystem *subsys =
2797 container_of(ref, struct nvme_subsystem, ref);
2799 mutex_lock(&nvme_subsystems_lock);
2800 list_del(&subsys->entry);
2801 mutex_unlock(&nvme_subsystems_lock);
2803 ida_destroy(&subsys->ns_ida);
2804 device_del(&subsys->dev);
2805 put_device(&subsys->dev);
2808 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2810 kref_put(&subsys->ref, nvme_destroy_subsystem);
2813 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2815 struct nvme_subsystem *subsys;
2817 lockdep_assert_held(&nvme_subsystems_lock);
2820 * Fail matches for discovery subsystems. This results
2821 * in each discovery controller bound to a unique subsystem.
2822 * This avoids issues with validating controller values
2823 * that can only be true when there is a single unique subsystem.
2824 * There may be multiple and completely independent entities
2825 * that provide discovery controllers.
2827 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2830 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2831 if (strcmp(subsys->subnqn, subsysnqn))
2833 if (!kref_get_unless_zero(&subsys->ref))
2841 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2842 struct device_attribute subsys_attr_##_name = \
2843 __ATTR(_name, _mode, _show, NULL)
2845 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2846 struct device_attribute *attr,
2849 struct nvme_subsystem *subsys =
2850 container_of(dev, struct nvme_subsystem, dev);
2852 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2854 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2856 #define nvme_subsys_show_str_function(field) \
2857 static ssize_t subsys_##field##_show(struct device *dev, \
2858 struct device_attribute *attr, char *buf) \
2860 struct nvme_subsystem *subsys = \
2861 container_of(dev, struct nvme_subsystem, dev); \
2862 return sysfs_emit(buf, "%.*s\n", \
2863 (int)sizeof(subsys->field), subsys->field); \
2865 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2867 nvme_subsys_show_str_function(model);
2868 nvme_subsys_show_str_function(serial);
2869 nvme_subsys_show_str_function(firmware_rev);
2871 static struct attribute *nvme_subsys_attrs[] = {
2872 &subsys_attr_model.attr,
2873 &subsys_attr_serial.attr,
2874 &subsys_attr_firmware_rev.attr,
2875 &subsys_attr_subsysnqn.attr,
2876 #ifdef CONFIG_NVME_MULTIPATH
2877 &subsys_attr_iopolicy.attr,
2882 static const struct attribute_group nvme_subsys_attrs_group = {
2883 .attrs = nvme_subsys_attrs,
2886 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2887 &nvme_subsys_attrs_group,
2891 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2893 return ctrl->opts && ctrl->opts->discovery_nqn;
2896 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2897 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2899 struct nvme_ctrl *tmp;
2901 lockdep_assert_held(&nvme_subsystems_lock);
2903 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2904 if (nvme_state_terminal(tmp))
2907 if (tmp->cntlid == ctrl->cntlid) {
2908 dev_err(ctrl->device,
2909 "Duplicate cntlid %u with %s, rejecting\n",
2910 ctrl->cntlid, dev_name(tmp->device));
2914 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2915 nvme_discovery_ctrl(ctrl))
2918 dev_err(ctrl->device,
2919 "Subsystem does not support multiple controllers\n");
2926 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2928 struct nvme_subsystem *subsys, *found;
2931 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2935 subsys->instance = -1;
2936 mutex_init(&subsys->lock);
2937 kref_init(&subsys->ref);
2938 INIT_LIST_HEAD(&subsys->ctrls);
2939 INIT_LIST_HEAD(&subsys->nsheads);
2940 nvme_init_subnqn(subsys, ctrl, id);
2941 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2942 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2943 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2944 subsys->vendor_id = le16_to_cpu(id->vid);
2945 subsys->cmic = id->cmic;
2946 subsys->awupf = le16_to_cpu(id->awupf);
2947 #ifdef CONFIG_NVME_MULTIPATH
2948 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2951 subsys->dev.class = nvme_subsys_class;
2952 subsys->dev.release = nvme_release_subsystem;
2953 subsys->dev.groups = nvme_subsys_attrs_groups;
2954 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2955 device_initialize(&subsys->dev);
2957 mutex_lock(&nvme_subsystems_lock);
2958 found = __nvme_find_get_subsystem(subsys->subnqn);
2960 put_device(&subsys->dev);
2963 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2965 goto out_put_subsystem;
2968 ret = device_add(&subsys->dev);
2970 dev_err(ctrl->device,
2971 "failed to register subsystem device.\n");
2972 put_device(&subsys->dev);
2975 ida_init(&subsys->ns_ida);
2976 list_add_tail(&subsys->entry, &nvme_subsystems);
2979 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2980 dev_name(ctrl->device));
2982 dev_err(ctrl->device,
2983 "failed to create sysfs link from subsystem.\n");
2984 goto out_put_subsystem;
2988 subsys->instance = ctrl->instance;
2989 ctrl->subsys = subsys;
2990 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2991 mutex_unlock(&nvme_subsystems_lock);
2995 nvme_put_subsystem(subsys);
2997 mutex_unlock(&nvme_subsystems_lock);
3001 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3002 void *log, size_t size, u64 offset)
3004 struct nvme_command c = { };
3005 u32 dwlen = nvme_bytes_to_numd(size);
3007 c.get_log_page.opcode = nvme_admin_get_log_page;
3008 c.get_log_page.nsid = cpu_to_le32(nsid);
3009 c.get_log_page.lid = log_page;
3010 c.get_log_page.lsp = lsp;
3011 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3012 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3013 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3014 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3015 c.get_log_page.csi = csi;
3017 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3020 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3021 struct nvme_effects_log **log)
3023 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3029 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3033 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3034 cel, sizeof(*cel), 0);
3040 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3046 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3048 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3050 if (check_shl_overflow(1U, units + page_shift - 9, &val))
3055 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3057 struct nvme_command c = { };
3058 struct nvme_id_ctrl_nvm *id;
3061 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3062 ctrl->max_discard_sectors = UINT_MAX;
3063 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3065 ctrl->max_discard_sectors = 0;
3066 ctrl->max_discard_segments = 0;
3070 * Even though NVMe spec explicitly states that MDTS is not applicable
3071 * to the write-zeroes, we are cautious and limit the size to the
3072 * controllers max_hw_sectors value, which is based on the MDTS field
3073 * and possibly other limiting factors.
3075 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3076 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3077 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3079 ctrl->max_zeroes_sectors = 0;
3081 if (nvme_ctrl_limited_cns(ctrl))
3084 id = kzalloc(sizeof(*id), GFP_KERNEL);
3088 c.identify.opcode = nvme_admin_identify;
3089 c.identify.cns = NVME_ID_CNS_CS_CTRL;
3090 c.identify.csi = NVME_CSI_NVM;
3092 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3097 ctrl->max_discard_segments = id->dmrl;
3099 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
3101 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3108 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3110 struct nvme_id_ctrl *id;
3112 bool prev_apst_enabled;
3115 ret = nvme_identify_ctrl(ctrl, &id);
3117 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3121 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3122 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3127 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3128 ctrl->cntlid = le16_to_cpu(id->cntlid);
3130 if (!ctrl->identified) {
3133 ret = nvme_init_subsystem(ctrl, id);
3138 * Check for quirks. Quirk can depend on firmware version,
3139 * so, in principle, the set of quirks present can change
3140 * across a reset. As a possible future enhancement, we
3141 * could re-scan for quirks every time we reinitialize
3142 * the device, but we'd have to make sure that the driver
3143 * behaves intelligently if the quirks change.
3145 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3146 if (quirk_matches(id, &core_quirks[i]))
3147 ctrl->quirks |= core_quirks[i].quirks;
3151 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3152 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3153 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3156 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3157 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3158 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3160 ctrl->oacs = le16_to_cpu(id->oacs);
3161 ctrl->oncs = le16_to_cpu(id->oncs);
3162 ctrl->mtfa = le16_to_cpu(id->mtfa);
3163 ctrl->oaes = le32_to_cpu(id->oaes);
3164 ctrl->wctemp = le16_to_cpu(id->wctemp);
3165 ctrl->cctemp = le16_to_cpu(id->cctemp);
3167 atomic_set(&ctrl->abort_limit, id->acl + 1);
3168 ctrl->vwc = id->vwc;
3170 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3172 max_hw_sectors = UINT_MAX;
3173 ctrl->max_hw_sectors =
3174 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3176 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3177 ctrl->sgls = le32_to_cpu(id->sgls);
3178 ctrl->kas = le16_to_cpu(id->kas);
3179 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3180 ctrl->ctratt = le32_to_cpu(id->ctratt);
3184 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3186 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3187 shutdown_timeout, 60);
3189 if (ctrl->shutdown_timeout != shutdown_timeout)
3190 dev_info(ctrl->device,
3191 "Shutdown timeout set to %u seconds\n",
3192 ctrl->shutdown_timeout);
3194 ctrl->shutdown_timeout = shutdown_timeout;
3196 ctrl->npss = id->npss;
3197 ctrl->apsta = id->apsta;
3198 prev_apst_enabled = ctrl->apst_enabled;
3199 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3200 if (force_apst && id->apsta) {
3201 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3202 ctrl->apst_enabled = true;
3204 ctrl->apst_enabled = false;
3207 ctrl->apst_enabled = id->apsta;
3209 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3211 if (ctrl->ops->flags & NVME_F_FABRICS) {
3212 ctrl->icdoff = le16_to_cpu(id->icdoff);
3213 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3214 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3215 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3218 * In fabrics we need to verify the cntlid matches the
3221 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3222 dev_err(ctrl->device,
3223 "Mismatching cntlid: Connect %u vs Identify "
3225 ctrl->cntlid, le16_to_cpu(id->cntlid));
3230 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3231 dev_err(ctrl->device,
3232 "keep-alive support is mandatory for fabrics\n");
3237 ctrl->hmpre = le32_to_cpu(id->hmpre);
3238 ctrl->hmmin = le32_to_cpu(id->hmmin);
3239 ctrl->hmminds = le32_to_cpu(id->hmminds);
3240 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3243 ret = nvme_mpath_init(ctrl, id);
3247 if (ctrl->apst_enabled && !prev_apst_enabled)
3248 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3249 else if (!ctrl->apst_enabled && prev_apst_enabled)
3250 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3258 * Initialize the cached copies of the Identify data and various controller
3259 * register in our nvme_ctrl structure. This should be called as soon as
3260 * the admin queue is fully up and running.
3262 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3266 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3268 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3272 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3274 if (ctrl->vs >= NVME_VS(1, 1, 0))
3275 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3277 ret = nvme_init_identify(ctrl);
3281 ret = nvme_init_non_mdts_limits(ctrl);
3285 ret = nvme_configure_apst(ctrl);
3289 ret = nvme_configure_timestamp(ctrl);
3293 ret = nvme_configure_directives(ctrl);
3297 ret = nvme_configure_acre(ctrl);
3301 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3302 ret = nvme_hwmon_init(ctrl);
3307 ctrl->identified = true;
3311 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3313 static int nvme_dev_open(struct inode *inode, struct file *file)
3315 struct nvme_ctrl *ctrl =
3316 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3318 switch (ctrl->state) {
3319 case NVME_CTRL_LIVE:
3322 return -EWOULDBLOCK;
3325 nvme_get_ctrl(ctrl);
3326 if (!try_module_get(ctrl->ops->module)) {
3327 nvme_put_ctrl(ctrl);
3331 file->private_data = ctrl;
3335 static int nvme_dev_release(struct inode *inode, struct file *file)
3337 struct nvme_ctrl *ctrl =
3338 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3340 module_put(ctrl->ops->module);
3341 nvme_put_ctrl(ctrl);
3345 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3350 down_read(&ctrl->namespaces_rwsem);
3351 if (list_empty(&ctrl->namespaces)) {
3356 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3357 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3358 dev_warn(ctrl->device,
3359 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3364 dev_warn(ctrl->device,
3365 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3366 kref_get(&ns->kref);
3367 up_read(&ctrl->namespaces_rwsem);
3369 ret = nvme_user_cmd(ctrl, ns, argp);
3374 up_read(&ctrl->namespaces_rwsem);
3378 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3381 struct nvme_ctrl *ctrl = file->private_data;
3382 void __user *argp = (void __user *)arg;
3385 case NVME_IOCTL_ADMIN_CMD:
3386 return nvme_user_cmd(ctrl, NULL, argp);
3387 case NVME_IOCTL_ADMIN64_CMD:
3388 return nvme_user_cmd64(ctrl, NULL, argp);
3389 case NVME_IOCTL_IO_CMD:
3390 return nvme_dev_user_cmd(ctrl, argp);
3391 case NVME_IOCTL_RESET:
3392 dev_warn(ctrl->device, "resetting controller\n");
3393 return nvme_reset_ctrl_sync(ctrl);
3394 case NVME_IOCTL_SUBSYS_RESET:
3395 return nvme_reset_subsystem(ctrl);
3396 case NVME_IOCTL_RESCAN:
3397 nvme_queue_scan(ctrl);
3404 static const struct file_operations nvme_dev_fops = {
3405 .owner = THIS_MODULE,
3406 .open = nvme_dev_open,
3407 .release = nvme_dev_release,
3408 .unlocked_ioctl = nvme_dev_ioctl,
3409 .compat_ioctl = compat_ptr_ioctl,
3412 static ssize_t nvme_sysfs_reset(struct device *dev,
3413 struct device_attribute *attr, const char *buf,
3416 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3419 ret = nvme_reset_ctrl_sync(ctrl);
3424 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3426 static ssize_t nvme_sysfs_rescan(struct device *dev,
3427 struct device_attribute *attr, const char *buf,
3430 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3432 nvme_queue_scan(ctrl);
3435 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3437 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3439 struct gendisk *disk = dev_to_disk(dev);
3441 if (disk->fops == &nvme_bdev_ops)
3442 return nvme_get_ns_from_dev(dev)->head;
3444 return disk->private_data;
3447 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3450 struct nvme_ns_head *head = dev_to_ns_head(dev);
3451 struct nvme_ns_ids *ids = &head->ids;
3452 struct nvme_subsystem *subsys = head->subsys;
3453 int serial_len = sizeof(subsys->serial);
3454 int model_len = sizeof(subsys->model);
3456 if (!uuid_is_null(&ids->uuid))
3457 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3459 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3460 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3462 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3463 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3465 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3466 subsys->serial[serial_len - 1] == '\0'))
3468 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3469 subsys->model[model_len - 1] == '\0'))
3472 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3473 serial_len, subsys->serial, model_len, subsys->model,
3476 static DEVICE_ATTR_RO(wwid);
3478 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3481 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3483 static DEVICE_ATTR_RO(nguid);
3485 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3488 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3490 /* For backward compatibility expose the NGUID to userspace if
3491 * we have no UUID set
3493 if (uuid_is_null(&ids->uuid)) {
3494 printk_ratelimited(KERN_WARNING
3495 "No UUID available providing old NGUID\n");
3496 return sysfs_emit(buf, "%pU\n", ids->nguid);
3498 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3500 static DEVICE_ATTR_RO(uuid);
3502 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3505 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3507 static DEVICE_ATTR_RO(eui);
3509 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3512 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3514 static DEVICE_ATTR_RO(nsid);
3516 static struct attribute *nvme_ns_id_attrs[] = {
3517 &dev_attr_wwid.attr,
3518 &dev_attr_uuid.attr,
3519 &dev_attr_nguid.attr,
3521 &dev_attr_nsid.attr,
3522 #ifdef CONFIG_NVME_MULTIPATH
3523 &dev_attr_ana_grpid.attr,
3524 &dev_attr_ana_state.attr,
3529 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3530 struct attribute *a, int n)
3532 struct device *dev = container_of(kobj, struct device, kobj);
3533 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3535 if (a == &dev_attr_uuid.attr) {
3536 if (uuid_is_null(&ids->uuid) &&
3537 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3540 if (a == &dev_attr_nguid.attr) {
3541 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3544 if (a == &dev_attr_eui.attr) {
3545 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3548 #ifdef CONFIG_NVME_MULTIPATH
3549 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3550 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3552 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3559 static const struct attribute_group nvme_ns_id_attr_group = {
3560 .attrs = nvme_ns_id_attrs,
3561 .is_visible = nvme_ns_id_attrs_are_visible,
3564 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3565 &nvme_ns_id_attr_group,
3567 &nvme_nvm_attr_group,
3572 #define nvme_show_str_function(field) \
3573 static ssize_t field##_show(struct device *dev, \
3574 struct device_attribute *attr, char *buf) \
3576 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3577 return sysfs_emit(buf, "%.*s\n", \
3578 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3580 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3582 nvme_show_str_function(model);
3583 nvme_show_str_function(serial);
3584 nvme_show_str_function(firmware_rev);
3586 #define nvme_show_int_function(field) \
3587 static ssize_t field##_show(struct device *dev, \
3588 struct device_attribute *attr, char *buf) \
3590 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3591 return sysfs_emit(buf, "%d\n", ctrl->field); \
3593 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3595 nvme_show_int_function(cntlid);
3596 nvme_show_int_function(numa_node);
3597 nvme_show_int_function(queue_count);
3598 nvme_show_int_function(sqsize);
3600 static ssize_t nvme_sysfs_delete(struct device *dev,
3601 struct device_attribute *attr, const char *buf,
3604 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3606 if (device_remove_file_self(dev, attr))
3607 nvme_delete_ctrl_sync(ctrl);
3610 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3612 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3613 struct device_attribute *attr,
3616 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3618 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3620 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3622 static ssize_t nvme_sysfs_show_state(struct device *dev,
3623 struct device_attribute *attr,
3626 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3627 static const char *const state_name[] = {
3628 [NVME_CTRL_NEW] = "new",
3629 [NVME_CTRL_LIVE] = "live",
3630 [NVME_CTRL_RESETTING] = "resetting",
3631 [NVME_CTRL_CONNECTING] = "connecting",
3632 [NVME_CTRL_DELETING] = "deleting",
3633 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3634 [NVME_CTRL_DEAD] = "dead",
3637 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3638 state_name[ctrl->state])
3639 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3641 return sysfs_emit(buf, "unknown state\n");
3644 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3646 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3647 struct device_attribute *attr,
3650 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3652 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3654 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3656 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3657 struct device_attribute *attr,
3660 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3662 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3664 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3666 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3667 struct device_attribute *attr,
3670 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3672 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3674 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3676 static ssize_t nvme_sysfs_show_address(struct device *dev,
3677 struct device_attribute *attr,
3680 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3682 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3684 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3686 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3687 struct device_attribute *attr, char *buf)
3689 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3690 struct nvmf_ctrl_options *opts = ctrl->opts;
3692 if (ctrl->opts->max_reconnects == -1)
3693 return sysfs_emit(buf, "off\n");
3694 return sysfs_emit(buf, "%d\n",
3695 opts->max_reconnects * opts->reconnect_delay);
3698 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3699 struct device_attribute *attr, const char *buf, size_t count)
3701 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3702 struct nvmf_ctrl_options *opts = ctrl->opts;
3703 int ctrl_loss_tmo, err;
3705 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3709 if (ctrl_loss_tmo < 0)
3710 opts->max_reconnects = -1;
3712 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3713 opts->reconnect_delay);
3716 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3717 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3719 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3720 struct device_attribute *attr, char *buf)
3722 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3724 if (ctrl->opts->reconnect_delay == -1)
3725 return sysfs_emit(buf, "off\n");
3726 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3729 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3730 struct device_attribute *attr, const char *buf, size_t count)
3732 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3736 err = kstrtou32(buf, 10, &v);
3740 ctrl->opts->reconnect_delay = v;
3743 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3744 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3746 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3747 struct device_attribute *attr, char *buf)
3749 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3751 if (ctrl->opts->fast_io_fail_tmo == -1)
3752 return sysfs_emit(buf, "off\n");
3753 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3756 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3757 struct device_attribute *attr, const char *buf, size_t count)
3759 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3760 struct nvmf_ctrl_options *opts = ctrl->opts;
3761 int fast_io_fail_tmo, err;
3763 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3767 if (fast_io_fail_tmo < 0)
3768 opts->fast_io_fail_tmo = -1;
3770 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3773 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3774 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3776 static struct attribute *nvme_dev_attrs[] = {
3777 &dev_attr_reset_controller.attr,
3778 &dev_attr_rescan_controller.attr,
3779 &dev_attr_model.attr,
3780 &dev_attr_serial.attr,
3781 &dev_attr_firmware_rev.attr,
3782 &dev_attr_cntlid.attr,
3783 &dev_attr_delete_controller.attr,
3784 &dev_attr_transport.attr,
3785 &dev_attr_subsysnqn.attr,
3786 &dev_attr_address.attr,
3787 &dev_attr_state.attr,
3788 &dev_attr_numa_node.attr,
3789 &dev_attr_queue_count.attr,
3790 &dev_attr_sqsize.attr,
3791 &dev_attr_hostnqn.attr,
3792 &dev_attr_hostid.attr,
3793 &dev_attr_ctrl_loss_tmo.attr,
3794 &dev_attr_reconnect_delay.attr,
3795 &dev_attr_fast_io_fail_tmo.attr,
3799 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3800 struct attribute *a, int n)
3802 struct device *dev = container_of(kobj, struct device, kobj);
3803 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3805 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3807 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3809 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3811 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3813 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3815 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3821 static const struct attribute_group nvme_dev_attrs_group = {
3822 .attrs = nvme_dev_attrs,
3823 .is_visible = nvme_dev_attrs_are_visible,
3826 static const struct attribute_group *nvme_dev_attr_groups[] = {
3827 &nvme_dev_attrs_group,
3831 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3834 struct nvme_ns_head *h;
3836 lockdep_assert_held(&subsys->lock);
3838 list_for_each_entry(h, &subsys->nsheads, entry) {
3839 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3846 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3847 struct nvme_ns_head *new)
3849 struct nvme_ns_head *h;
3851 lockdep_assert_held(&subsys->lock);
3853 list_for_each_entry(h, &subsys->nsheads, entry) {
3854 if (nvme_ns_ids_valid(&new->ids) &&
3855 nvme_ns_ids_equal(&new->ids, &h->ids))
3862 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3863 unsigned nsid, struct nvme_ns_ids *ids)
3865 struct nvme_ns_head *head;
3866 size_t size = sizeof(*head);
3869 #ifdef CONFIG_NVME_MULTIPATH
3870 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3873 head = kzalloc(size, GFP_KERNEL);
3876 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3879 head->instance = ret;
3880 INIT_LIST_HEAD(&head->list);
3881 ret = init_srcu_struct(&head->srcu);
3883 goto out_ida_remove;
3884 head->subsys = ctrl->subsys;
3887 kref_init(&head->ref);
3889 ret = __nvme_check_ids(ctrl->subsys, head);
3891 dev_err(ctrl->device,
3892 "duplicate IDs for nsid %d\n", nsid);
3893 goto out_cleanup_srcu;
3896 if (head->ids.csi) {
3897 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3899 goto out_cleanup_srcu;
3901 head->effects = ctrl->effects;
3903 ret = nvme_mpath_alloc_disk(ctrl, head);
3905 goto out_cleanup_srcu;
3907 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3909 kref_get(&ctrl->subsys->ref);
3913 cleanup_srcu_struct(&head->srcu);
3915 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3920 ret = blk_status_to_errno(nvme_error_status(ret));
3921 return ERR_PTR(ret);
3924 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3925 struct nvme_ns_ids *ids, bool is_shared)
3927 struct nvme_ctrl *ctrl = ns->ctrl;
3928 struct nvme_ns_head *head = NULL;
3931 mutex_lock(&ctrl->subsys->lock);
3932 head = nvme_find_ns_head(ctrl->subsys, nsid);
3934 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3936 ret = PTR_ERR(head);
3939 head->shared = is_shared;
3942 if (!is_shared || !head->shared) {
3943 dev_err(ctrl->device,
3944 "Duplicate unshared namespace %d\n", nsid);
3945 goto out_put_ns_head;
3947 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3948 dev_err(ctrl->device,
3949 "IDs don't match for shared namespace %d\n",
3951 goto out_put_ns_head;
3955 list_add_tail_rcu(&ns->siblings, &head->list);
3957 mutex_unlock(&ctrl->subsys->lock);
3961 nvme_put_ns_head(head);
3963 mutex_unlock(&ctrl->subsys->lock);
3967 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3969 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3970 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3972 return nsa->head->ns_id - nsb->head->ns_id;
3975 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3977 struct nvme_ns *ns, *ret = NULL;
3979 down_read(&ctrl->namespaces_rwsem);
3980 list_for_each_entry(ns, &ctrl->namespaces, list) {
3981 if (ns->head->ns_id == nsid) {
3982 if (!kref_get_unless_zero(&ns->kref))
3987 if (ns->head->ns_id > nsid)
3990 up_read(&ctrl->namespaces_rwsem);
3993 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3995 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3996 struct nvme_ns_ids *ids)
3999 struct gendisk *disk;
4000 struct nvme_id_ns *id;
4001 int node = ctrl->numa_node;
4003 if (nvme_identify_ns(ctrl, nsid, ids, &id))
4006 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4010 ns->queue = blk_mq_init_queue(ctrl->tagset);
4011 if (IS_ERR(ns->queue))
4014 if (ctrl->opts && ctrl->opts->data_digest)
4015 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4017 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4018 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
4019 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4021 ns->queue->queuedata = ns;
4023 kref_init(&ns->kref);
4025 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
4026 goto out_free_queue;
4028 disk = alloc_disk_node(0, node);
4032 disk->fops = &nvme_bdev_ops;
4033 disk->private_data = ns;
4034 disk->queue = ns->queue;
4035 disk->flags = GENHD_FL_EXT_DEVT;
4037 * Without the multipath code enabled, multiple controller per
4038 * subsystems are visible as devices and thus we cannot use the
4039 * subsystem instance.
4041 if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
4042 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4043 ns->head->instance);
4046 if (nvme_update_ns_info(ns, id))
4049 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
4050 if (nvme_nvm_register(ns, disk->disk_name, node)) {
4051 dev_warn(ctrl->device, "LightNVM init failure\n");
4056 down_write(&ctrl->namespaces_rwsem);
4057 list_add_tail(&ns->list, &ctrl->namespaces);
4058 up_write(&ctrl->namespaces_rwsem);
4060 nvme_get_ctrl(ctrl);
4062 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
4064 nvme_mpath_add_disk(ns, id);
4065 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4070 /* prevent double queue cleanup */
4071 ns->disk->queue = NULL;
4074 mutex_lock(&ctrl->subsys->lock);
4075 list_del_rcu(&ns->siblings);
4076 if (list_empty(&ns->head->list))
4077 list_del_init(&ns->head->entry);
4078 mutex_unlock(&ctrl->subsys->lock);
4079 nvme_put_ns_head(ns->head);
4081 blk_cleanup_queue(ns->queue);
4088 static void nvme_ns_remove(struct nvme_ns *ns)
4090 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4093 set_capacity(ns->disk, 0);
4094 nvme_fault_inject_fini(&ns->fault_inject);
4096 mutex_lock(&ns->ctrl->subsys->lock);
4097 list_del_rcu(&ns->siblings);
4098 if (list_empty(&ns->head->list))
4099 list_del_init(&ns->head->entry);
4100 mutex_unlock(&ns->ctrl->subsys->lock);
4102 synchronize_rcu(); /* guarantee not available in head->list */
4103 nvme_mpath_clear_current_path(ns);
4104 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4106 if (ns->disk->flags & GENHD_FL_UP) {
4107 del_gendisk(ns->disk);
4108 blk_cleanup_queue(ns->queue);
4109 if (blk_get_integrity(ns->disk))
4110 blk_integrity_unregister(ns->disk);
4113 down_write(&ns->ctrl->namespaces_rwsem);
4114 list_del_init(&ns->list);
4115 up_write(&ns->ctrl->namespaces_rwsem);
4117 nvme_mpath_check_last_path(ns);
4121 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4123 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4131 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4133 struct nvme_id_ns *id;
4134 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4136 if (test_bit(NVME_NS_DEAD, &ns->flags))
4139 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4143 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4144 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4145 dev_err(ns->ctrl->device,
4146 "identifiers changed for nsid %d\n", ns->head->ns_id);
4150 ret = nvme_update_ns_info(ns, id);
4156 * Only remove the namespace if we got a fatal error back from the
4157 * device, otherwise ignore the error and just move on.
4159 * TODO: we should probably schedule a delayed retry here.
4161 if (ret > 0 && (ret & NVME_SC_DNR))
4165 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4167 struct nvme_ns_ids ids = { };
4170 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4173 ns = nvme_find_get_ns(ctrl, nsid);
4175 nvme_validate_ns(ns, &ids);
4182 nvme_alloc_ns(ctrl, nsid, &ids);
4185 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4186 dev_warn(ctrl->device,
4187 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4191 if (!nvme_multi_css(ctrl)) {
4192 dev_warn(ctrl->device,
4193 "command set not reported for nsid: %d\n",
4197 nvme_alloc_ns(ctrl, nsid, &ids);
4200 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4206 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4209 struct nvme_ns *ns, *next;
4212 down_write(&ctrl->namespaces_rwsem);
4213 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4214 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4215 list_move_tail(&ns->list, &rm_list);
4217 up_write(&ctrl->namespaces_rwsem);
4219 list_for_each_entry_safe(ns, next, &rm_list, list)
4224 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4226 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4231 if (nvme_ctrl_limited_cns(ctrl))
4234 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4239 struct nvme_command cmd = {
4240 .identify.opcode = nvme_admin_identify,
4241 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4242 .identify.nsid = cpu_to_le32(prev),
4245 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4246 NVME_IDENTIFY_DATA_SIZE);
4248 dev_warn(ctrl->device,
4249 "Identify NS List failed (status=0x%x)\n", ret);
4253 for (i = 0; i < nr_entries; i++) {
4254 u32 nsid = le32_to_cpu(ns_list[i]);
4256 if (!nsid) /* end of the list? */
4258 nvme_validate_or_alloc_ns(ctrl, nsid);
4259 while (++prev < nsid)
4260 nvme_ns_remove_by_nsid(ctrl, prev);
4264 nvme_remove_invalid_namespaces(ctrl, prev);
4270 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4272 struct nvme_id_ctrl *id;
4275 if (nvme_identify_ctrl(ctrl, &id))
4277 nn = le32_to_cpu(id->nn);
4280 for (i = 1; i <= nn; i++)
4281 nvme_validate_or_alloc_ns(ctrl, i);
4283 nvme_remove_invalid_namespaces(ctrl, nn);
4286 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4288 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4292 log = kzalloc(log_size, GFP_KERNEL);
4297 * We need to read the log to clear the AEN, but we don't want to rely
4298 * on it for the changed namespace information as userspace could have
4299 * raced with us in reading the log page, which could cause us to miss
4302 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4303 NVME_CSI_NVM, log, log_size, 0);
4305 dev_warn(ctrl->device,
4306 "reading changed ns log failed: %d\n", error);
4311 static void nvme_scan_work(struct work_struct *work)
4313 struct nvme_ctrl *ctrl =
4314 container_of(work, struct nvme_ctrl, scan_work);
4316 /* No tagset on a live ctrl means IO queues could not created */
4317 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4320 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4321 dev_info(ctrl->device, "rescanning namespaces.\n");
4322 nvme_clear_changed_ns_log(ctrl);
4325 mutex_lock(&ctrl->scan_lock);
4326 if (nvme_scan_ns_list(ctrl) != 0)
4327 nvme_scan_ns_sequential(ctrl);
4328 mutex_unlock(&ctrl->scan_lock);
4330 down_write(&ctrl->namespaces_rwsem);
4331 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4332 up_write(&ctrl->namespaces_rwsem);
4336 * This function iterates the namespace list unlocked to allow recovery from
4337 * controller failure. It is up to the caller to ensure the namespace list is
4338 * not modified by scan work while this function is executing.
4340 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4342 struct nvme_ns *ns, *next;
4346 * make sure to requeue I/O to all namespaces as these
4347 * might result from the scan itself and must complete
4348 * for the scan_work to make progress
4350 nvme_mpath_clear_ctrl_paths(ctrl);
4352 /* prevent racing with ns scanning */
4353 flush_work(&ctrl->scan_work);
4356 * The dead states indicates the controller was not gracefully
4357 * disconnected. In that case, we won't be able to flush any data while
4358 * removing the namespaces' disks; fail all the queues now to avoid
4359 * potentially having to clean up the failed sync later.
4361 if (ctrl->state == NVME_CTRL_DEAD)
4362 nvme_kill_queues(ctrl);
4364 /* this is a no-op when called from the controller reset handler */
4365 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4367 down_write(&ctrl->namespaces_rwsem);
4368 list_splice_init(&ctrl->namespaces, &ns_list);
4369 up_write(&ctrl->namespaces_rwsem);
4371 list_for_each_entry_safe(ns, next, &ns_list, list)
4374 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4376 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4378 struct nvme_ctrl *ctrl =
4379 container_of(dev, struct nvme_ctrl, ctrl_device);
4380 struct nvmf_ctrl_options *opts = ctrl->opts;
4383 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4388 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4392 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4393 opts->trsvcid ?: "none");
4397 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4398 opts->host_traddr ?: "none");
4403 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4405 char *envp[2] = { NULL, NULL };
4406 u32 aen_result = ctrl->aen_result;
4408 ctrl->aen_result = 0;
4412 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4415 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4419 static void nvme_async_event_work(struct work_struct *work)
4421 struct nvme_ctrl *ctrl =
4422 container_of(work, struct nvme_ctrl, async_event_work);
4424 nvme_aen_uevent(ctrl);
4425 ctrl->ops->submit_async_event(ctrl);
4428 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4433 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4439 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4442 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4444 struct nvme_fw_slot_info_log *log;
4446 log = kmalloc(sizeof(*log), GFP_KERNEL);
4450 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4451 log, sizeof(*log), 0))
4452 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4456 static void nvme_fw_act_work(struct work_struct *work)
4458 struct nvme_ctrl *ctrl = container_of(work,
4459 struct nvme_ctrl, fw_act_work);
4460 unsigned long fw_act_timeout;
4463 fw_act_timeout = jiffies +
4464 msecs_to_jiffies(ctrl->mtfa * 100);
4466 fw_act_timeout = jiffies +
4467 msecs_to_jiffies(admin_timeout * 1000);
4469 nvme_stop_queues(ctrl);
4470 while (nvme_ctrl_pp_status(ctrl)) {
4471 if (time_after(jiffies, fw_act_timeout)) {
4472 dev_warn(ctrl->device,
4473 "Fw activation timeout, reset controller\n");
4474 nvme_try_sched_reset(ctrl);
4480 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4483 nvme_start_queues(ctrl);
4484 /* read FW slot information to clear the AER */
4485 nvme_get_fw_slot_info(ctrl);
4488 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4490 u32 aer_notice_type = (result & 0xff00) >> 8;
4492 trace_nvme_async_event(ctrl, aer_notice_type);
4494 switch (aer_notice_type) {
4495 case NVME_AER_NOTICE_NS_CHANGED:
4496 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4497 nvme_queue_scan(ctrl);
4499 case NVME_AER_NOTICE_FW_ACT_STARTING:
4501 * We are (ab)using the RESETTING state to prevent subsequent
4502 * recovery actions from interfering with the controller's
4503 * firmware activation.
4505 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4506 queue_work(nvme_wq, &ctrl->fw_act_work);
4508 #ifdef CONFIG_NVME_MULTIPATH
4509 case NVME_AER_NOTICE_ANA:
4510 if (!ctrl->ana_log_buf)
4512 queue_work(nvme_wq, &ctrl->ana_work);
4515 case NVME_AER_NOTICE_DISC_CHANGED:
4516 ctrl->aen_result = result;
4519 dev_warn(ctrl->device, "async event result %08x\n", result);
4523 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4524 volatile union nvme_result *res)
4526 u32 result = le32_to_cpu(res->u32);
4527 u32 aer_type = result & 0x07;
4529 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4533 case NVME_AER_NOTICE:
4534 nvme_handle_aen_notice(ctrl, result);
4536 case NVME_AER_ERROR:
4537 case NVME_AER_SMART:
4540 trace_nvme_async_event(ctrl, aer_type);
4541 ctrl->aen_result = result;
4546 queue_work(nvme_wq, &ctrl->async_event_work);
4548 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4550 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4552 nvme_mpath_stop(ctrl);
4553 nvme_stop_keep_alive(ctrl);
4554 nvme_stop_failfast_work(ctrl);
4555 flush_work(&ctrl->async_event_work);
4556 cancel_work_sync(&ctrl->fw_act_work);
4558 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4560 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4562 nvme_start_keep_alive(ctrl);
4564 nvme_enable_aen(ctrl);
4566 if (ctrl->queue_count > 1) {
4567 nvme_queue_scan(ctrl);
4568 nvme_start_queues(ctrl);
4571 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4573 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4575 nvme_hwmon_exit(ctrl);
4576 nvme_fault_inject_fini(&ctrl->fault_inject);
4577 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4578 cdev_device_del(&ctrl->cdev, ctrl->device);
4579 nvme_put_ctrl(ctrl);
4581 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4583 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4585 struct nvme_effects_log *cel;
4588 xa_for_each(&ctrl->cels, i, cel) {
4589 xa_erase(&ctrl->cels, i);
4593 xa_destroy(&ctrl->cels);
4596 static void nvme_free_ctrl(struct device *dev)
4598 struct nvme_ctrl *ctrl =
4599 container_of(dev, struct nvme_ctrl, ctrl_device);
4600 struct nvme_subsystem *subsys = ctrl->subsys;
4602 if (!subsys || ctrl->instance != subsys->instance)
4603 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4605 nvme_free_cels(ctrl);
4606 nvme_mpath_uninit(ctrl);
4607 __free_page(ctrl->discard_page);
4610 mutex_lock(&nvme_subsystems_lock);
4611 list_del(&ctrl->subsys_entry);
4612 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4613 mutex_unlock(&nvme_subsystems_lock);
4616 ctrl->ops->free_ctrl(ctrl);
4619 nvme_put_subsystem(subsys);
4623 * Initialize a NVMe controller structures. This needs to be called during
4624 * earliest initialization so that we have the initialized structured around
4627 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4628 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4632 ctrl->state = NVME_CTRL_NEW;
4633 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4634 spin_lock_init(&ctrl->lock);
4635 mutex_init(&ctrl->scan_lock);
4636 INIT_LIST_HEAD(&ctrl->namespaces);
4637 xa_init(&ctrl->cels);
4638 init_rwsem(&ctrl->namespaces_rwsem);
4641 ctrl->quirks = quirks;
4642 ctrl->numa_node = NUMA_NO_NODE;
4643 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4644 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4645 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4646 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4647 init_waitqueue_head(&ctrl->state_wq);
4649 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4650 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4651 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4652 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4654 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4656 ctrl->discard_page = alloc_page(GFP_KERNEL);
4657 if (!ctrl->discard_page) {
4662 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4665 ctrl->instance = ret;
4667 device_initialize(&ctrl->ctrl_device);
4668 ctrl->device = &ctrl->ctrl_device;
4669 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4671 ctrl->device->class = nvme_class;
4672 ctrl->device->parent = ctrl->dev;
4673 ctrl->device->groups = nvme_dev_attr_groups;
4674 ctrl->device->release = nvme_free_ctrl;
4675 dev_set_drvdata(ctrl->device, ctrl);
4676 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4678 goto out_release_instance;
4680 nvme_get_ctrl(ctrl);
4681 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4682 ctrl->cdev.owner = ops->module;
4683 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4688 * Initialize latency tolerance controls. The sysfs files won't
4689 * be visible to userspace unless the device actually supports APST.
4691 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4692 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4693 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4695 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4699 nvme_put_ctrl(ctrl);
4700 kfree_const(ctrl->device->kobj.name);
4701 out_release_instance:
4702 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4704 if (ctrl->discard_page)
4705 __free_page(ctrl->discard_page);
4708 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4711 * nvme_kill_queues(): Ends all namespace queues
4712 * @ctrl: the dead controller that needs to end
4714 * Call this function when the driver determines it is unable to get the
4715 * controller in a state capable of servicing IO.
4717 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4721 down_read(&ctrl->namespaces_rwsem);
4723 /* Forcibly unquiesce queues to avoid blocking dispatch */
4724 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4725 blk_mq_unquiesce_queue(ctrl->admin_q);
4727 list_for_each_entry(ns, &ctrl->namespaces, list)
4728 nvme_set_queue_dying(ns);
4730 up_read(&ctrl->namespaces_rwsem);
4732 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4734 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4738 down_read(&ctrl->namespaces_rwsem);
4739 list_for_each_entry(ns, &ctrl->namespaces, list)
4740 blk_mq_unfreeze_queue(ns->queue);
4741 up_read(&ctrl->namespaces_rwsem);
4743 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4745 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4749 down_read(&ctrl->namespaces_rwsem);
4750 list_for_each_entry(ns, &ctrl->namespaces, list) {
4751 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4755 up_read(&ctrl->namespaces_rwsem);
4758 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4760 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4764 down_read(&ctrl->namespaces_rwsem);
4765 list_for_each_entry(ns, &ctrl->namespaces, list)
4766 blk_mq_freeze_queue_wait(ns->queue);
4767 up_read(&ctrl->namespaces_rwsem);
4769 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4771 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4775 down_read(&ctrl->namespaces_rwsem);
4776 list_for_each_entry(ns, &ctrl->namespaces, list)
4777 blk_freeze_queue_start(ns->queue);
4778 up_read(&ctrl->namespaces_rwsem);
4780 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4782 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4786 down_read(&ctrl->namespaces_rwsem);
4787 list_for_each_entry(ns, &ctrl->namespaces, list)
4788 blk_mq_quiesce_queue(ns->queue);
4789 up_read(&ctrl->namespaces_rwsem);
4791 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4793 void nvme_start_queues(struct nvme_ctrl *ctrl)
4797 down_read(&ctrl->namespaces_rwsem);
4798 list_for_each_entry(ns, &ctrl->namespaces, list)
4799 blk_mq_unquiesce_queue(ns->queue);
4800 up_read(&ctrl->namespaces_rwsem);
4802 EXPORT_SYMBOL_GPL(nvme_start_queues);
4804 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4808 down_read(&ctrl->namespaces_rwsem);
4809 list_for_each_entry(ns, &ctrl->namespaces, list)
4810 blk_sync_queue(ns->queue);
4811 up_read(&ctrl->namespaces_rwsem);
4813 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4815 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4817 nvme_sync_io_queues(ctrl);
4819 blk_sync_queue(ctrl->admin_q);
4821 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4823 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4825 if (file->f_op != &nvme_dev_fops)
4827 return file->private_data;
4829 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4832 * Check we didn't inadvertently grow the command structure sizes:
4834 static inline void _nvme_check_size(void)
4836 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4837 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4838 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4839 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4840 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4841 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4842 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4843 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4844 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4845 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4846 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4847 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4848 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4849 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4850 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4851 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4852 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4853 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4854 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4855 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4859 static int __init nvme_core_init(void)
4861 int result = -ENOMEM;
4865 nvme_wq = alloc_workqueue("nvme-wq",
4866 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4870 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4871 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4875 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4876 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4877 if (!nvme_delete_wq)
4878 goto destroy_reset_wq;
4880 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4881 NVME_MINORS, "nvme");
4883 goto destroy_delete_wq;
4885 nvme_class = class_create(THIS_MODULE, "nvme");
4886 if (IS_ERR(nvme_class)) {
4887 result = PTR_ERR(nvme_class);
4888 goto unregister_chrdev;
4890 nvme_class->dev_uevent = nvme_class_uevent;
4892 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4893 if (IS_ERR(nvme_subsys_class)) {
4894 result = PTR_ERR(nvme_subsys_class);
4900 class_destroy(nvme_class);
4902 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4904 destroy_workqueue(nvme_delete_wq);
4906 destroy_workqueue(nvme_reset_wq);
4908 destroy_workqueue(nvme_wq);
4913 static void __exit nvme_core_exit(void)
4915 class_destroy(nvme_subsys_class);
4916 class_destroy(nvme_class);
4917 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4918 destroy_workqueue(nvme_delete_wq);
4919 destroy_workqueue(nvme_reset_wq);
4920 destroy_workqueue(nvme_wq);
4921 ida_destroy(&nvme_instance_ida);
4924 MODULE_LICENSE("GPL");
4925 MODULE_VERSION("1.0");
4926 module_init(nvme_core_init);
4927 module_exit(nvme_core_exit);