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 if (!(req->rq_flags & RQF_DONTPREP)) {
579 nvme_req(req)->retries = 0;
580 nvme_req(req)->flags = 0;
581 req->rq_flags |= RQF_DONTPREP;
585 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
587 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
590 static inline void nvme_init_request(struct request *req,
591 struct nvme_command *cmd)
593 if (req->q->queuedata)
594 req->timeout = NVME_IO_TIMEOUT;
595 else /* no queuedata implies admin queue */
596 req->timeout = NVME_ADMIN_TIMEOUT;
598 req->cmd_flags |= REQ_FAILFAST_DRIVER;
599 nvme_clear_nvme_request(req);
600 nvme_req(req)->cmd = cmd;
603 struct request *nvme_alloc_request(struct request_queue *q,
604 struct nvme_command *cmd, blk_mq_req_flags_t flags)
608 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
610 nvme_init_request(req, cmd);
613 EXPORT_SYMBOL_GPL(nvme_alloc_request);
615 static struct request *nvme_alloc_request_qid(struct request_queue *q,
616 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
620 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
623 nvme_init_request(req, cmd);
627 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
629 struct nvme_command c;
631 memset(&c, 0, sizeof(c));
633 c.directive.opcode = nvme_admin_directive_send;
634 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
635 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
636 c.directive.dtype = NVME_DIR_IDENTIFY;
637 c.directive.tdtype = NVME_DIR_STREAMS;
638 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
640 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
643 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
645 return nvme_toggle_streams(ctrl, false);
648 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
650 return nvme_toggle_streams(ctrl, true);
653 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
654 struct streams_directive_params *s, u32 nsid)
656 struct nvme_command c;
658 memset(&c, 0, sizeof(c));
659 memset(s, 0, sizeof(*s));
661 c.directive.opcode = nvme_admin_directive_recv;
662 c.directive.nsid = cpu_to_le32(nsid);
663 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
664 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
665 c.directive.dtype = NVME_DIR_STREAMS;
667 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
670 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
672 struct streams_directive_params s;
675 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
680 ret = nvme_enable_streams(ctrl);
684 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
686 goto out_disable_stream;
688 ctrl->nssa = le16_to_cpu(s.nssa);
689 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
690 dev_info(ctrl->device, "too few streams (%u) available\n",
692 goto out_disable_stream;
695 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
696 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
700 nvme_disable_streams(ctrl);
705 * Check if 'req' has a write hint associated with it. If it does, assign
706 * a valid namespace stream to the write.
708 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
709 struct request *req, u16 *control,
712 enum rw_hint streamid = req->write_hint;
714 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
718 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
721 *control |= NVME_RW_DTYPE_STREAMS;
722 *dsmgmt |= streamid << 16;
725 if (streamid < ARRAY_SIZE(req->q->write_hints))
726 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
729 static void nvme_setup_passthrough(struct request *req,
730 struct nvme_command *cmd)
732 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
733 /* passthru commands should let the driver set the SGL flags */
734 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
737 static inline void nvme_setup_flush(struct nvme_ns *ns,
738 struct nvme_command *cmnd)
740 cmnd->common.opcode = nvme_cmd_flush;
741 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
744 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
745 struct nvme_command *cmnd)
747 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
748 struct nvme_dsm_range *range;
752 * Some devices do not consider the DSM 'Number of Ranges' field when
753 * determining how much data to DMA. Always allocate memory for maximum
754 * number of segments to prevent device reading beyond end of buffer.
756 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
758 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
761 * If we fail allocation our range, fallback to the controller
762 * discard page. If that's also busy, it's safe to return
763 * busy, as we know we can make progress once that's freed.
765 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
766 return BLK_STS_RESOURCE;
768 range = page_address(ns->ctrl->discard_page);
771 __rq_for_each_bio(bio, req) {
772 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
773 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
776 range[n].cattr = cpu_to_le32(0);
777 range[n].nlb = cpu_to_le32(nlb);
778 range[n].slba = cpu_to_le64(slba);
783 if (WARN_ON_ONCE(n != segments)) {
784 if (virt_to_page(range) == ns->ctrl->discard_page)
785 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
788 return BLK_STS_IOERR;
791 cmnd->dsm.opcode = nvme_cmd_dsm;
792 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
793 cmnd->dsm.nr = cpu_to_le32(segments - 1);
794 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
796 req->special_vec.bv_page = virt_to_page(range);
797 req->special_vec.bv_offset = offset_in_page(range);
798 req->special_vec.bv_len = alloc_size;
799 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
804 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
805 struct request *req, struct nvme_command *cmnd)
807 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
808 return nvme_setup_discard(ns, req, cmnd);
810 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
811 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
812 cmnd->write_zeroes.slba =
813 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
814 cmnd->write_zeroes.length =
815 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
816 cmnd->write_zeroes.control = 0;
820 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
821 struct request *req, struct nvme_command *cmnd,
824 struct nvme_ctrl *ctrl = ns->ctrl;
828 if (req->cmd_flags & REQ_FUA)
829 control |= NVME_RW_FUA;
830 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
831 control |= NVME_RW_LR;
833 if (req->cmd_flags & REQ_RAHEAD)
834 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
836 cmnd->rw.opcode = op;
837 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
838 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
839 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
841 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
842 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
846 * If formated with metadata, the block layer always provides a
847 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
848 * we enable the PRACT bit for protection information or set the
849 * namespace capacity to zero to prevent any I/O.
851 if (!blk_integrity_rq(req)) {
852 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
853 return BLK_STS_NOTSUPP;
854 control |= NVME_RW_PRINFO_PRACT;
857 switch (ns->pi_type) {
858 case NVME_NS_DPS_PI_TYPE3:
859 control |= NVME_RW_PRINFO_PRCHK_GUARD;
861 case NVME_NS_DPS_PI_TYPE1:
862 case NVME_NS_DPS_PI_TYPE2:
863 control |= NVME_RW_PRINFO_PRCHK_GUARD |
864 NVME_RW_PRINFO_PRCHK_REF;
865 if (op == nvme_cmd_zone_append)
866 control |= NVME_RW_APPEND_PIREMAP;
867 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
872 cmnd->rw.control = cpu_to_le16(control);
873 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
877 void nvme_cleanup_cmd(struct request *req)
879 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
880 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
881 struct page *page = req->special_vec.bv_page;
883 if (page == ctrl->discard_page)
884 clear_bit_unlock(0, &ctrl->discard_page_busy);
886 kfree(page_address(page) + req->special_vec.bv_offset);
889 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
891 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
892 struct nvme_command *cmd)
894 blk_status_t ret = BLK_STS_OK;
896 nvme_clear_nvme_request(req);
898 memset(cmd, 0, sizeof(*cmd));
899 switch (req_op(req)) {
902 nvme_setup_passthrough(req, cmd);
905 nvme_setup_flush(ns, cmd);
907 case REQ_OP_ZONE_RESET_ALL:
908 case REQ_OP_ZONE_RESET:
909 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
911 case REQ_OP_ZONE_OPEN:
912 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
914 case REQ_OP_ZONE_CLOSE:
915 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
917 case REQ_OP_ZONE_FINISH:
918 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
920 case REQ_OP_WRITE_ZEROES:
921 ret = nvme_setup_write_zeroes(ns, req, cmd);
924 ret = nvme_setup_discard(ns, req, cmd);
927 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
930 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
932 case REQ_OP_ZONE_APPEND:
933 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
937 return BLK_STS_IOERR;
940 cmd->common.command_id = req->tag;
941 trace_nvme_setup_cmd(req, cmd);
944 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
946 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
948 struct completion *waiting = rq->end_io_data;
950 rq->end_io_data = NULL;
954 static void nvme_execute_rq_polled(struct request_queue *q,
955 struct gendisk *bd_disk, struct request *rq, int at_head)
957 DECLARE_COMPLETION_ONSTACK(wait);
959 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
961 rq->cmd_flags |= REQ_HIPRI;
962 rq->end_io_data = &wait;
963 blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
965 while (!completion_done(&wait)) {
966 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
972 * Returns 0 on success. If the result is negative, it's a Linux error code;
973 * if the result is positive, it's an NVM Express status code
975 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
976 union nvme_result *result, void *buffer, unsigned bufflen,
977 unsigned timeout, int qid, int at_head,
978 blk_mq_req_flags_t flags, bool poll)
983 if (qid == NVME_QID_ANY)
984 req = nvme_alloc_request(q, cmd, flags);
986 req = nvme_alloc_request_qid(q, cmd, flags, qid);
991 req->timeout = timeout;
993 if (buffer && bufflen) {
994 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1000 nvme_execute_rq_polled(req->q, NULL, req, at_head);
1002 blk_execute_rq(NULL, req, at_head);
1004 *result = nvme_req(req)->result;
1005 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1008 ret = nvme_req(req)->status;
1010 blk_mq_free_request(req);
1013 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1015 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1016 void *buffer, unsigned bufflen)
1018 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1019 NVME_QID_ANY, 0, 0, false);
1021 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1023 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
1024 unsigned len, u32 seed, bool write)
1026 struct bio_integrity_payload *bip;
1030 buf = kmalloc(len, GFP_KERNEL);
1035 if (write && copy_from_user(buf, ubuf, len))
1038 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1044 bip->bip_iter.bi_size = len;
1045 bip->bip_iter.bi_sector = seed;
1046 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1047 offset_in_page(buf));
1054 return ERR_PTR(ret);
1057 static u32 nvme_known_admin_effects(u8 opcode)
1060 case nvme_admin_format_nvm:
1061 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1062 NVME_CMD_EFFECTS_CSE_MASK;
1063 case nvme_admin_sanitize_nvm:
1064 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1071 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1076 if (ns->head->effects)
1077 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1078 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1079 dev_warn(ctrl->device,
1080 "IO command:%02x has unhandled effects:%08x\n",
1086 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1087 effects |= nvme_known_admin_effects(opcode);
1091 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1093 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1096 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1099 * For simplicity, IO to all namespaces is quiesced even if the command
1100 * effects say only one namespace is affected.
1102 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1103 mutex_lock(&ctrl->scan_lock);
1104 mutex_lock(&ctrl->subsys->lock);
1105 nvme_mpath_start_freeze(ctrl->subsys);
1106 nvme_mpath_wait_freeze(ctrl->subsys);
1107 nvme_start_freeze(ctrl);
1108 nvme_wait_freeze(ctrl);
1113 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1115 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1116 nvme_unfreeze(ctrl);
1117 nvme_mpath_unfreeze(ctrl->subsys);
1118 mutex_unlock(&ctrl->subsys->lock);
1119 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1120 mutex_unlock(&ctrl->scan_lock);
1122 if (effects & NVME_CMD_EFFECTS_CCC)
1123 nvme_init_identify(ctrl);
1124 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1125 nvme_queue_scan(ctrl);
1126 flush_work(&ctrl->scan_work);
1130 void nvme_execute_passthru_rq(struct request *rq)
1132 struct nvme_command *cmd = nvme_req(rq)->cmd;
1133 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1134 struct nvme_ns *ns = rq->q->queuedata;
1135 struct gendisk *disk = ns ? ns->disk : NULL;
1138 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1139 blk_execute_rq(disk, rq, 0);
1140 nvme_passthru_end(ctrl, effects);
1142 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1144 static int nvme_submit_user_cmd(struct request_queue *q,
1145 struct nvme_command *cmd, void __user *ubuffer,
1146 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1147 u32 meta_seed, u64 *result, unsigned timeout)
1149 bool write = nvme_is_write(cmd);
1150 struct nvme_ns *ns = q->queuedata;
1151 struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1152 struct request *req;
1153 struct bio *bio = NULL;
1157 req = nvme_alloc_request(q, cmd, 0);
1159 return PTR_ERR(req);
1162 req->timeout = timeout;
1163 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1165 if (ubuffer && bufflen) {
1166 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1172 bio_set_dev(bio, bdev);
1173 if (bdev && meta_buffer && meta_len) {
1174 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1177 ret = PTR_ERR(meta);
1180 req->cmd_flags |= REQ_INTEGRITY;
1184 nvme_execute_passthru_rq(req);
1185 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1188 ret = nvme_req(req)->status;
1190 *result = le64_to_cpu(nvme_req(req)->result.u64);
1191 if (meta && !ret && !write) {
1192 if (copy_to_user(meta_buffer, meta, meta_len))
1198 blk_rq_unmap_user(bio);
1200 blk_mq_free_request(req);
1204 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1206 struct nvme_ctrl *ctrl = rq->end_io_data;
1207 unsigned long flags;
1208 bool startka = false;
1210 blk_mq_free_request(rq);
1213 dev_err(ctrl->device,
1214 "failed nvme_keep_alive_end_io error=%d\n",
1219 ctrl->comp_seen = false;
1220 spin_lock_irqsave(&ctrl->lock, flags);
1221 if (ctrl->state == NVME_CTRL_LIVE ||
1222 ctrl->state == NVME_CTRL_CONNECTING)
1224 spin_unlock_irqrestore(&ctrl->lock, flags);
1226 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1229 static void nvme_keep_alive_work(struct work_struct *work)
1231 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1232 struct nvme_ctrl, ka_work);
1233 bool comp_seen = ctrl->comp_seen;
1236 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1237 dev_dbg(ctrl->device,
1238 "reschedule traffic based keep-alive timer\n");
1239 ctrl->comp_seen = false;
1240 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1244 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1245 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1247 /* allocation failure, reset the controller */
1248 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1249 nvme_reset_ctrl(ctrl);
1253 rq->timeout = ctrl->kato * HZ;
1254 rq->end_io_data = ctrl;
1255 blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1258 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1260 if (unlikely(ctrl->kato == 0))
1263 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1266 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1268 if (unlikely(ctrl->kato == 0))
1271 cancel_delayed_work_sync(&ctrl->ka_work);
1273 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1276 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1277 * flag, thus sending any new CNS opcodes has a big chance of not working.
1278 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1279 * (but not for any later version).
1281 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1283 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1284 return ctrl->vs < NVME_VS(1, 2, 0);
1285 return ctrl->vs < NVME_VS(1, 1, 0);
1288 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1290 struct nvme_command c = { };
1293 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1294 c.identify.opcode = nvme_admin_identify;
1295 c.identify.cns = NVME_ID_CNS_CTRL;
1297 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1301 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1302 sizeof(struct nvme_id_ctrl));
1308 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1310 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1313 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1314 struct nvme_ns_id_desc *cur, bool *csi_seen)
1316 const char *warn_str = "ctrl returned bogus length:";
1319 switch (cur->nidt) {
1320 case NVME_NIDT_EUI64:
1321 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1322 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1323 warn_str, cur->nidl);
1326 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1327 return NVME_NIDT_EUI64_LEN;
1328 case NVME_NIDT_NGUID:
1329 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1330 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1331 warn_str, cur->nidl);
1334 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1335 return NVME_NIDT_NGUID_LEN;
1336 case NVME_NIDT_UUID:
1337 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1338 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1339 warn_str, cur->nidl);
1342 uuid_copy(&ids->uuid, data + sizeof(*cur));
1343 return NVME_NIDT_UUID_LEN;
1345 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1346 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1347 warn_str, cur->nidl);
1350 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1352 return NVME_NIDT_CSI_LEN;
1354 /* Skip unknown types */
1359 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1360 struct nvme_ns_ids *ids)
1362 struct nvme_command c = { };
1363 bool csi_seen = false;
1364 int status, pos, len;
1367 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1369 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1372 c.identify.opcode = nvme_admin_identify;
1373 c.identify.nsid = cpu_to_le32(nsid);
1374 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1376 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1380 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1381 NVME_IDENTIFY_DATA_SIZE);
1383 dev_warn(ctrl->device,
1384 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1389 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1390 struct nvme_ns_id_desc *cur = data + pos;
1395 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1399 len += sizeof(*cur);
1402 if (nvme_multi_css(ctrl) && !csi_seen) {
1403 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1413 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1414 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1416 struct nvme_command c = { };
1419 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1420 c.identify.opcode = nvme_admin_identify;
1421 c.identify.nsid = cpu_to_le32(nsid);
1422 c.identify.cns = NVME_ID_CNS_NS;
1424 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1428 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1430 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1434 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1435 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1438 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1439 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1440 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1441 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1442 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1443 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1452 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1453 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1455 union nvme_result res = { 0 };
1456 struct nvme_command c;
1459 memset(&c, 0, sizeof(c));
1460 c.features.opcode = op;
1461 c.features.fid = cpu_to_le32(fid);
1462 c.features.dword11 = cpu_to_le32(dword11);
1464 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1465 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1466 if (ret >= 0 && result)
1467 *result = le32_to_cpu(res.u32);
1471 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1472 unsigned int dword11, void *buffer, size_t buflen,
1475 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1478 EXPORT_SYMBOL_GPL(nvme_set_features);
1480 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1481 unsigned int dword11, void *buffer, size_t buflen,
1484 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1487 EXPORT_SYMBOL_GPL(nvme_get_features);
1489 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1491 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1493 int status, nr_io_queues;
1495 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1501 * Degraded controllers might return an error when setting the queue
1502 * count. We still want to be able to bring them online and offer
1503 * access to the admin queue, as that might be only way to fix them up.
1506 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1509 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1510 *count = min(*count, nr_io_queues);
1515 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1517 #define NVME_AEN_SUPPORTED \
1518 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1519 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1521 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1523 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1526 if (!supported_aens)
1529 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1532 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1535 queue_work(nvme_wq, &ctrl->async_event_work);
1539 * Convert integer values from ioctl structures to user pointers, silently
1540 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1543 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1545 if (in_compat_syscall())
1546 ptrval = (compat_uptr_t)ptrval;
1547 return (void __user *)ptrval;
1550 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1552 struct nvme_user_io io;
1553 struct nvme_command c;
1554 unsigned length, meta_len;
1555 void __user *metadata;
1557 if (copy_from_user(&io, uio, sizeof(io)))
1562 switch (io.opcode) {
1563 case nvme_cmd_write:
1565 case nvme_cmd_compare:
1571 length = (io.nblocks + 1) << ns->lba_shift;
1573 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1574 ns->ms == sizeof(struct t10_pi_tuple)) {
1576 * Protection information is stripped/inserted by the
1579 if (nvme_to_user_ptr(io.metadata))
1584 meta_len = (io.nblocks + 1) * ns->ms;
1585 metadata = nvme_to_user_ptr(io.metadata);
1588 if (ns->features & NVME_NS_EXT_LBAS) {
1591 } else if (meta_len) {
1592 if ((io.metadata & 3) || !io.metadata)
1596 memset(&c, 0, sizeof(c));
1597 c.rw.opcode = io.opcode;
1598 c.rw.flags = io.flags;
1599 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1600 c.rw.slba = cpu_to_le64(io.slba);
1601 c.rw.length = cpu_to_le16(io.nblocks);
1602 c.rw.control = cpu_to_le16(io.control);
1603 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1604 c.rw.reftag = cpu_to_le32(io.reftag);
1605 c.rw.apptag = cpu_to_le16(io.apptag);
1606 c.rw.appmask = cpu_to_le16(io.appmask);
1608 return nvme_submit_user_cmd(ns->queue, &c,
1609 nvme_to_user_ptr(io.addr), length,
1610 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1613 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1614 struct nvme_passthru_cmd __user *ucmd)
1616 struct nvme_passthru_cmd cmd;
1617 struct nvme_command c;
1618 unsigned timeout = 0;
1622 if (!capable(CAP_SYS_ADMIN))
1624 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1629 memset(&c, 0, sizeof(c));
1630 c.common.opcode = cmd.opcode;
1631 c.common.flags = cmd.flags;
1632 c.common.nsid = cpu_to_le32(cmd.nsid);
1633 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1634 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1635 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1636 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1637 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1638 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1639 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1640 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1643 timeout = msecs_to_jiffies(cmd.timeout_ms);
1645 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1646 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1647 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1648 0, &result, timeout);
1651 if (put_user(result, &ucmd->result))
1658 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1659 struct nvme_passthru_cmd64 __user *ucmd)
1661 struct nvme_passthru_cmd64 cmd;
1662 struct nvme_command c;
1663 unsigned timeout = 0;
1666 if (!capable(CAP_SYS_ADMIN))
1668 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1673 memset(&c, 0, sizeof(c));
1674 c.common.opcode = cmd.opcode;
1675 c.common.flags = cmd.flags;
1676 c.common.nsid = cpu_to_le32(cmd.nsid);
1677 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1678 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1679 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1680 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1681 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1682 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1683 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1684 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1687 timeout = msecs_to_jiffies(cmd.timeout_ms);
1689 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1690 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1691 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1692 0, &cmd.result, timeout);
1695 if (put_user(cmd.result, &ucmd->result))
1703 * Issue ioctl requests on the first available path. Note that unlike normal
1704 * block layer requests we will not retry failed request on another controller.
1706 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1707 struct nvme_ns_head **head, int *srcu_idx)
1709 #ifdef CONFIG_NVME_MULTIPATH
1710 if (disk->fops == &nvme_ns_head_ops) {
1713 *head = disk->private_data;
1714 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1715 ns = nvme_find_path(*head);
1717 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1723 return disk->private_data;
1726 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1729 srcu_read_unlock(&head->srcu, idx);
1732 static bool is_ctrl_ioctl(unsigned int cmd)
1734 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1736 if (is_sed_ioctl(cmd))
1741 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1743 struct nvme_ns_head *head,
1746 struct nvme_ctrl *ctrl = ns->ctrl;
1749 nvme_get_ctrl(ns->ctrl);
1750 nvme_put_ns_from_disk(head, srcu_idx);
1753 case NVME_IOCTL_ADMIN_CMD:
1754 ret = nvme_user_cmd(ctrl, NULL, argp);
1756 case NVME_IOCTL_ADMIN64_CMD:
1757 ret = nvme_user_cmd64(ctrl, NULL, argp);
1760 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1763 nvme_put_ctrl(ctrl);
1767 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1768 unsigned int cmd, unsigned long arg)
1770 struct nvme_ns_head *head = NULL;
1771 void __user *argp = (void __user *)arg;
1775 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1777 return -EWOULDBLOCK;
1780 * Handle ioctls that apply to the controller instead of the namespace
1781 * seperately and drop the ns SRCU reference early. This avoids a
1782 * deadlock when deleting namespaces using the passthrough interface.
1784 if (is_ctrl_ioctl(cmd))
1785 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1789 force_successful_syscall_return();
1790 ret = ns->head->ns_id;
1792 case NVME_IOCTL_IO_CMD:
1793 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1795 case NVME_IOCTL_SUBMIT_IO:
1796 ret = nvme_submit_io(ns, argp);
1798 case NVME_IOCTL_IO64_CMD:
1799 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1803 ret = nvme_nvm_ioctl(ns, cmd, arg);
1808 nvme_put_ns_from_disk(head, srcu_idx);
1812 #ifdef CONFIG_COMPAT
1813 struct nvme_user_io32 {
1826 } __attribute__((__packed__));
1828 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1830 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1831 unsigned int cmd, unsigned long arg)
1834 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1835 * between 32 bit programs and 64 bit kernel.
1836 * The cause is that the results of sizeof(struct nvme_user_io),
1837 * which is used to define NVME_IOCTL_SUBMIT_IO,
1838 * are not same between 32 bit compiler and 64 bit compiler.
1839 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1840 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1841 * Other IOCTL numbers are same between 32 bit and 64 bit.
1842 * So there is nothing to do regarding to other IOCTL numbers.
1844 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1845 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1847 return nvme_ioctl(bdev, mode, cmd, arg);
1850 #define nvme_compat_ioctl NULL
1851 #endif /* CONFIG_COMPAT */
1853 static int nvme_open(struct block_device *bdev, fmode_t mode)
1855 struct nvme_ns *ns = bdev->bd_disk->private_data;
1857 #ifdef CONFIG_NVME_MULTIPATH
1858 /* should never be called due to GENHD_FL_HIDDEN */
1859 if (WARN_ON_ONCE(ns->head->disk))
1862 if (!kref_get_unless_zero(&ns->kref))
1864 if (!try_module_get(ns->ctrl->ops->module))
1875 static void nvme_release(struct gendisk *disk, fmode_t mode)
1877 struct nvme_ns *ns = disk->private_data;
1879 module_put(ns->ctrl->ops->module);
1883 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1885 /* some standard values */
1886 geo->heads = 1 << 6;
1887 geo->sectors = 1 << 5;
1888 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1892 #ifdef CONFIG_BLK_DEV_INTEGRITY
1893 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1894 u32 max_integrity_segments)
1896 struct blk_integrity integrity;
1898 memset(&integrity, 0, sizeof(integrity));
1900 case NVME_NS_DPS_PI_TYPE3:
1901 integrity.profile = &t10_pi_type3_crc;
1902 integrity.tag_size = sizeof(u16) + sizeof(u32);
1903 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1905 case NVME_NS_DPS_PI_TYPE1:
1906 case NVME_NS_DPS_PI_TYPE2:
1907 integrity.profile = &t10_pi_type1_crc;
1908 integrity.tag_size = sizeof(u16);
1909 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1912 integrity.profile = NULL;
1915 integrity.tuple_size = ms;
1916 blk_integrity_register(disk, &integrity);
1917 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1920 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1921 u32 max_integrity_segments)
1924 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1926 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1928 struct nvme_ctrl *ctrl = ns->ctrl;
1929 struct request_queue *queue = disk->queue;
1930 u32 size = queue_logical_block_size(queue);
1932 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1933 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1937 if (ctrl->nr_streams && ns->sws && ns->sgs)
1938 size *= ns->sws * ns->sgs;
1940 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1941 NVME_DSM_MAX_RANGES);
1943 queue->limits.discard_alignment = 0;
1944 queue->limits.discard_granularity = size;
1946 /* If discard is already enabled, don't reset queue limits */
1947 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1950 blk_queue_max_discard_sectors(queue, UINT_MAX);
1951 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1953 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1954 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1958 * Even though NVMe spec explicitly states that MDTS is not applicable to the
1959 * write-zeroes, we are cautious and limit the size to the controllers
1960 * max_hw_sectors value, which is based on the MDTS field and possibly other
1963 static void nvme_config_write_zeroes(struct request_queue *q,
1964 struct nvme_ctrl *ctrl)
1966 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
1967 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1968 blk_queue_max_write_zeroes_sectors(q, ctrl->max_hw_sectors);
1971 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1973 return !uuid_is_null(&ids->uuid) ||
1974 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1975 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1978 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1980 return uuid_equal(&a->uuid, &b->uuid) &&
1981 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1982 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1986 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1987 u32 *phys_bs, u32 *io_opt)
1989 struct streams_directive_params s;
1992 if (!ctrl->nr_streams)
1995 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1999 ns->sws = le32_to_cpu(s.sws);
2000 ns->sgs = le16_to_cpu(s.sgs);
2003 *phys_bs = ns->sws * (1 << ns->lba_shift);
2005 *io_opt = *phys_bs * ns->sgs;
2011 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2013 struct nvme_ctrl *ctrl = ns->ctrl;
2016 * The PI implementation requires the metadata size to be equal to the
2017 * t10 pi tuple size.
2019 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2020 if (ns->ms == sizeof(struct t10_pi_tuple))
2021 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2025 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2026 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2028 if (ctrl->ops->flags & NVME_F_FABRICS) {
2030 * The NVMe over Fabrics specification only supports metadata as
2031 * part of the extended data LBA. We rely on HCA/HBA support to
2032 * remap the separate metadata buffer from the block layer.
2034 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2036 if (ctrl->max_integrity_segments)
2038 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2041 * For PCIe controllers, we can't easily remap the separate
2042 * metadata buffer from the block layer and thus require a
2043 * separate metadata buffer for block layer metadata/PI support.
2044 * We allow extended LBAs for the passthrough interface, though.
2046 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2047 ns->features |= NVME_NS_EXT_LBAS;
2049 ns->features |= NVME_NS_METADATA_SUPPORTED;
2055 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2056 struct request_queue *q)
2058 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2060 if (ctrl->max_hw_sectors) {
2062 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2064 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2065 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2066 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2068 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2069 blk_queue_dma_alignment(q, 7);
2070 blk_queue_write_cache(q, vwc, vwc);
2073 static void nvme_update_disk_info(struct gendisk *disk,
2074 struct nvme_ns *ns, struct nvme_id_ns *id)
2076 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2077 unsigned short bs = 1 << ns->lba_shift;
2078 u32 atomic_bs, phys_bs, io_opt = 0;
2081 * The block layer can't support LBA sizes larger than the page size
2082 * yet, so catch this early and don't allow block I/O.
2084 if (ns->lba_shift > PAGE_SHIFT) {
2089 blk_integrity_unregister(disk);
2091 atomic_bs = phys_bs = bs;
2092 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2093 if (id->nabo == 0) {
2095 * Bit 1 indicates whether NAWUPF is defined for this namespace
2096 * and whether it should be used instead of AWUPF. If NAWUPF ==
2097 * 0 then AWUPF must be used instead.
2099 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2100 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2102 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2105 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2106 /* NPWG = Namespace Preferred Write Granularity */
2107 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2108 /* NOWS = Namespace Optimal Write Size */
2109 io_opt = bs * (1 + le16_to_cpu(id->nows));
2112 blk_queue_logical_block_size(disk->queue, bs);
2114 * Linux filesystems assume writing a single physical block is
2115 * an atomic operation. Hence limit the physical block size to the
2116 * value of the Atomic Write Unit Power Fail parameter.
2118 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2119 blk_queue_io_min(disk->queue, phys_bs);
2120 blk_queue_io_opt(disk->queue, io_opt);
2123 * Register a metadata profile for PI, or the plain non-integrity NVMe
2124 * metadata masquerading as Type 0 if supported, otherwise reject block
2125 * I/O to namespaces with metadata except when the namespace supports
2126 * PI, as it can strip/insert in that case.
2129 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2130 (ns->features & NVME_NS_METADATA_SUPPORTED))
2131 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2132 ns->ctrl->max_integrity_segments);
2133 else if (!nvme_ns_has_pi(ns))
2137 set_capacity_and_notify(disk, capacity);
2139 nvme_config_discard(disk, ns);
2140 nvme_config_write_zeroes(disk->queue, ns->ctrl);
2142 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2143 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2146 static inline bool nvme_first_scan(struct gendisk *disk)
2148 /* nvme_alloc_ns() scans the disk prior to adding it */
2149 return !(disk->flags & GENHD_FL_UP);
2152 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2154 struct nvme_ctrl *ctrl = ns->ctrl;
2157 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2158 is_power_of_2(ctrl->max_hw_sectors))
2159 iob = ctrl->max_hw_sectors;
2161 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2166 if (!is_power_of_2(iob)) {
2167 if (nvme_first_scan(ns->disk))
2168 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2169 ns->disk->disk_name, iob);
2173 if (blk_queue_is_zoned(ns->disk->queue)) {
2174 if (nvme_first_scan(ns->disk))
2175 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2176 ns->disk->disk_name);
2180 blk_queue_chunk_sectors(ns->queue, iob);
2183 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2185 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2188 blk_mq_freeze_queue(ns->disk->queue);
2189 ns->lba_shift = id->lbaf[lbaf].ds;
2190 nvme_set_queue_limits(ns->ctrl, ns->queue);
2192 ret = nvme_configure_metadata(ns, id);
2195 nvme_set_chunk_sectors(ns, id);
2196 nvme_update_disk_info(ns->disk, ns, id);
2198 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2199 ret = nvme_update_zone_info(ns, lbaf);
2204 blk_mq_unfreeze_queue(ns->disk->queue);
2206 if (blk_queue_is_zoned(ns->queue)) {
2207 ret = nvme_revalidate_zones(ns);
2208 if (ret && !nvme_first_scan(ns->disk))
2212 #ifdef CONFIG_NVME_MULTIPATH
2213 if (ns->head->disk) {
2214 blk_mq_freeze_queue(ns->head->disk->queue);
2215 nvme_update_disk_info(ns->head->disk, ns, id);
2216 blk_stack_limits(&ns->head->disk->queue->limits,
2217 &ns->queue->limits, 0);
2218 blk_queue_update_readahead(ns->head->disk->queue);
2219 blk_mq_unfreeze_queue(ns->head->disk->queue);
2225 blk_mq_unfreeze_queue(ns->disk->queue);
2229 static char nvme_pr_type(enum pr_type type)
2232 case PR_WRITE_EXCLUSIVE:
2234 case PR_EXCLUSIVE_ACCESS:
2236 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2238 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2240 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2242 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2249 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2250 u64 key, u64 sa_key, u8 op)
2252 struct nvme_ns_head *head = NULL;
2254 struct nvme_command c;
2256 u8 data[16] = { 0, };
2258 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2260 return -EWOULDBLOCK;
2262 put_unaligned_le64(key, &data[0]);
2263 put_unaligned_le64(sa_key, &data[8]);
2265 memset(&c, 0, sizeof(c));
2266 c.common.opcode = op;
2267 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2268 c.common.cdw10 = cpu_to_le32(cdw10);
2270 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2271 nvme_put_ns_from_disk(head, srcu_idx);
2275 static int nvme_pr_register(struct block_device *bdev, u64 old,
2276 u64 new, unsigned flags)
2280 if (flags & ~PR_FL_IGNORE_KEY)
2283 cdw10 = old ? 2 : 0;
2284 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2285 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2286 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2289 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2290 enum pr_type type, unsigned flags)
2294 if (flags & ~PR_FL_IGNORE_KEY)
2297 cdw10 = nvme_pr_type(type) << 8;
2298 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2299 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2302 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2303 enum pr_type type, bool abort)
2305 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2306 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2309 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2311 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2312 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2315 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2317 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2318 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2321 static const struct pr_ops nvme_pr_ops = {
2322 .pr_register = nvme_pr_register,
2323 .pr_reserve = nvme_pr_reserve,
2324 .pr_release = nvme_pr_release,
2325 .pr_preempt = nvme_pr_preempt,
2326 .pr_clear = nvme_pr_clear,
2329 #ifdef CONFIG_BLK_SED_OPAL
2330 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2333 struct nvme_ctrl *ctrl = data;
2334 struct nvme_command cmd;
2336 memset(&cmd, 0, sizeof(cmd));
2338 cmd.common.opcode = nvme_admin_security_send;
2340 cmd.common.opcode = nvme_admin_security_recv;
2341 cmd.common.nsid = 0;
2342 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2343 cmd.common.cdw11 = cpu_to_le32(len);
2345 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2346 NVME_QID_ANY, 1, 0, false);
2348 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2349 #endif /* CONFIG_BLK_SED_OPAL */
2351 static const struct block_device_operations nvme_bdev_ops = {
2352 .owner = THIS_MODULE,
2353 .ioctl = nvme_ioctl,
2354 .compat_ioctl = nvme_compat_ioctl,
2356 .release = nvme_release,
2357 .getgeo = nvme_getgeo,
2358 .report_zones = nvme_report_zones,
2359 .pr_ops = &nvme_pr_ops,
2362 #ifdef CONFIG_NVME_MULTIPATH
2363 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2365 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2367 if (!kref_get_unless_zero(&head->ref))
2372 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2374 nvme_put_ns_head(disk->private_data);
2377 const struct block_device_operations nvme_ns_head_ops = {
2378 .owner = THIS_MODULE,
2379 .submit_bio = nvme_ns_head_submit_bio,
2380 .open = nvme_ns_head_open,
2381 .release = nvme_ns_head_release,
2382 .ioctl = nvme_ioctl,
2383 .compat_ioctl = nvme_compat_ioctl,
2384 .getgeo = nvme_getgeo,
2385 .report_zones = nvme_report_zones,
2386 .pr_ops = &nvme_pr_ops,
2388 #endif /* CONFIG_NVME_MULTIPATH */
2390 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2392 unsigned long timeout =
2393 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2394 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2397 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2400 if ((csts & NVME_CSTS_RDY) == bit)
2403 usleep_range(1000, 2000);
2404 if (fatal_signal_pending(current))
2406 if (time_after(jiffies, timeout)) {
2407 dev_err(ctrl->device,
2408 "Device not ready; aborting %s, CSTS=0x%x\n",
2409 enabled ? "initialisation" : "reset", csts);
2418 * If the device has been passed off to us in an enabled state, just clear
2419 * the enabled bit. The spec says we should set the 'shutdown notification
2420 * bits', but doing so may cause the device to complete commands to the
2421 * admin queue ... and we don't know what memory that might be pointing at!
2423 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2427 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2428 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2430 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2434 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2435 msleep(NVME_QUIRK_DELAY_AMOUNT);
2437 return nvme_wait_ready(ctrl, ctrl->cap, false);
2439 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2441 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2443 unsigned dev_page_min;
2446 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2448 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2451 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2453 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2454 dev_err(ctrl->device,
2455 "Minimum device page size %u too large for host (%u)\n",
2456 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2460 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2461 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2463 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2464 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2465 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2466 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2467 ctrl->ctrl_config |= NVME_CC_ENABLE;
2469 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2472 return nvme_wait_ready(ctrl, ctrl->cap, true);
2474 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2476 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2478 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2482 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2483 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2485 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2489 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2490 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2494 if (fatal_signal_pending(current))
2496 if (time_after(jiffies, timeout)) {
2497 dev_err(ctrl->device,
2498 "Device shutdown incomplete; abort shutdown\n");
2505 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2507 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2512 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2515 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2516 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2519 dev_warn_once(ctrl->device,
2520 "could not set timestamp (%d)\n", ret);
2524 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2526 struct nvme_feat_host_behavior *host;
2529 /* Don't bother enabling the feature if retry delay is not reported */
2533 host = kzalloc(sizeof(*host), GFP_KERNEL);
2537 host->acre = NVME_ENABLE_ACRE;
2538 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2539 host, sizeof(*host), NULL);
2544 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2547 * APST (Autonomous Power State Transition) lets us program a
2548 * table of power state transitions that the controller will
2549 * perform automatically. We configure it with a simple
2550 * heuristic: we are willing to spend at most 2% of the time
2551 * transitioning between power states. Therefore, when running
2552 * in any given state, we will enter the next lower-power
2553 * non-operational state after waiting 50 * (enlat + exlat)
2554 * microseconds, as long as that state's exit latency is under
2555 * the requested maximum latency.
2557 * We will not autonomously enter any non-operational state for
2558 * which the total latency exceeds ps_max_latency_us. Users
2559 * can set ps_max_latency_us to zero to turn off APST.
2563 struct nvme_feat_auto_pst *table;
2569 * If APST isn't supported or if we haven't been initialized yet,
2570 * then don't do anything.
2575 if (ctrl->npss > 31) {
2576 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2580 table = kzalloc(sizeof(*table), GFP_KERNEL);
2584 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2585 /* Turn off APST. */
2587 dev_dbg(ctrl->device, "APST disabled\n");
2589 __le64 target = cpu_to_le64(0);
2593 * Walk through all states from lowest- to highest-power.
2594 * According to the spec, lower-numbered states use more
2595 * power. NPSS, despite the name, is the index of the
2596 * lowest-power state, not the number of states.
2598 for (state = (int)ctrl->npss; state >= 0; state--) {
2599 u64 total_latency_us, exit_latency_us, transition_ms;
2602 table->entries[state] = target;
2605 * Don't allow transitions to the deepest state
2606 * if it's quirked off.
2608 if (state == ctrl->npss &&
2609 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2613 * Is this state a useful non-operational state for
2614 * higher-power states to autonomously transition to?
2616 if (!(ctrl->psd[state].flags &
2617 NVME_PS_FLAGS_NON_OP_STATE))
2621 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2622 if (exit_latency_us > ctrl->ps_max_latency_us)
2627 le32_to_cpu(ctrl->psd[state].entry_lat);
2630 * This state is good. Use it as the APST idle
2631 * target for higher power states.
2633 transition_ms = total_latency_us + 19;
2634 do_div(transition_ms, 20);
2635 if (transition_ms > (1 << 24) - 1)
2636 transition_ms = (1 << 24) - 1;
2638 target = cpu_to_le64((state << 3) |
2639 (transition_ms << 8));
2644 if (total_latency_us > max_lat_us)
2645 max_lat_us = total_latency_us;
2651 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2653 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2654 max_ps, max_lat_us, (int)sizeof(*table), table);
2658 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2659 table, sizeof(*table), NULL);
2661 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2667 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2669 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2673 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2674 case PM_QOS_LATENCY_ANY:
2682 if (ctrl->ps_max_latency_us != latency) {
2683 ctrl->ps_max_latency_us = latency;
2684 nvme_configure_apst(ctrl);
2688 struct nvme_core_quirk_entry {
2690 * NVMe model and firmware strings are padded with spaces. For
2691 * simplicity, strings in the quirk table are padded with NULLs
2697 unsigned long quirks;
2700 static const struct nvme_core_quirk_entry core_quirks[] = {
2703 * This Toshiba device seems to die using any APST states. See:
2704 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2707 .mn = "THNSF5256GPUK TOSHIBA",
2708 .quirks = NVME_QUIRK_NO_APST,
2712 * This LiteON CL1-3D*-Q11 firmware version has a race
2713 * condition associated with actions related to suspend to idle
2714 * LiteON has resolved the problem in future firmware
2718 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2722 /* match is null-terminated but idstr is space-padded. */
2723 static bool string_matches(const char *idstr, const char *match, size_t len)
2730 matchlen = strlen(match);
2731 WARN_ON_ONCE(matchlen > len);
2733 if (memcmp(idstr, match, matchlen))
2736 for (; matchlen < len; matchlen++)
2737 if (idstr[matchlen] != ' ')
2743 static bool quirk_matches(const struct nvme_id_ctrl *id,
2744 const struct nvme_core_quirk_entry *q)
2746 return q->vid == le16_to_cpu(id->vid) &&
2747 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2748 string_matches(id->fr, q->fr, sizeof(id->fr));
2751 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2752 struct nvme_id_ctrl *id)
2757 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2758 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2759 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2760 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2764 if (ctrl->vs >= NVME_VS(1, 2, 1))
2765 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2768 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2769 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2770 "nqn.2014.08.org.nvmexpress:%04x%04x",
2771 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2772 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2773 off += sizeof(id->sn);
2774 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2775 off += sizeof(id->mn);
2776 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2779 static void nvme_release_subsystem(struct device *dev)
2781 struct nvme_subsystem *subsys =
2782 container_of(dev, struct nvme_subsystem, dev);
2784 if (subsys->instance >= 0)
2785 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2789 static void nvme_destroy_subsystem(struct kref *ref)
2791 struct nvme_subsystem *subsys =
2792 container_of(ref, struct nvme_subsystem, ref);
2794 mutex_lock(&nvme_subsystems_lock);
2795 list_del(&subsys->entry);
2796 mutex_unlock(&nvme_subsystems_lock);
2798 ida_destroy(&subsys->ns_ida);
2799 device_del(&subsys->dev);
2800 put_device(&subsys->dev);
2803 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2805 kref_put(&subsys->ref, nvme_destroy_subsystem);
2808 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2810 struct nvme_subsystem *subsys;
2812 lockdep_assert_held(&nvme_subsystems_lock);
2815 * Fail matches for discovery subsystems. This results
2816 * in each discovery controller bound to a unique subsystem.
2817 * This avoids issues with validating controller values
2818 * that can only be true when there is a single unique subsystem.
2819 * There may be multiple and completely independent entities
2820 * that provide discovery controllers.
2822 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2825 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2826 if (strcmp(subsys->subnqn, subsysnqn))
2828 if (!kref_get_unless_zero(&subsys->ref))
2836 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2837 struct device_attribute subsys_attr_##_name = \
2838 __ATTR(_name, _mode, _show, NULL)
2840 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2841 struct device_attribute *attr,
2844 struct nvme_subsystem *subsys =
2845 container_of(dev, struct nvme_subsystem, dev);
2847 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2849 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2851 #define nvme_subsys_show_str_function(field) \
2852 static ssize_t subsys_##field##_show(struct device *dev, \
2853 struct device_attribute *attr, char *buf) \
2855 struct nvme_subsystem *subsys = \
2856 container_of(dev, struct nvme_subsystem, dev); \
2857 return sprintf(buf, "%.*s\n", \
2858 (int)sizeof(subsys->field), subsys->field); \
2860 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2862 nvme_subsys_show_str_function(model);
2863 nvme_subsys_show_str_function(serial);
2864 nvme_subsys_show_str_function(firmware_rev);
2866 static struct attribute *nvme_subsys_attrs[] = {
2867 &subsys_attr_model.attr,
2868 &subsys_attr_serial.attr,
2869 &subsys_attr_firmware_rev.attr,
2870 &subsys_attr_subsysnqn.attr,
2871 #ifdef CONFIG_NVME_MULTIPATH
2872 &subsys_attr_iopolicy.attr,
2877 static const struct attribute_group nvme_subsys_attrs_group = {
2878 .attrs = nvme_subsys_attrs,
2881 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2882 &nvme_subsys_attrs_group,
2886 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2888 return ctrl->opts && ctrl->opts->discovery_nqn;
2891 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2892 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2894 struct nvme_ctrl *tmp;
2896 lockdep_assert_held(&nvme_subsystems_lock);
2898 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2899 if (nvme_state_terminal(tmp))
2902 if (tmp->cntlid == ctrl->cntlid) {
2903 dev_err(ctrl->device,
2904 "Duplicate cntlid %u with %s, rejecting\n",
2905 ctrl->cntlid, dev_name(tmp->device));
2909 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2910 nvme_discovery_ctrl(ctrl))
2913 dev_err(ctrl->device,
2914 "Subsystem does not support multiple controllers\n");
2921 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2923 struct nvme_subsystem *subsys, *found;
2926 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2930 subsys->instance = -1;
2931 mutex_init(&subsys->lock);
2932 kref_init(&subsys->ref);
2933 INIT_LIST_HEAD(&subsys->ctrls);
2934 INIT_LIST_HEAD(&subsys->nsheads);
2935 nvme_init_subnqn(subsys, ctrl, id);
2936 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2937 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2938 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2939 subsys->vendor_id = le16_to_cpu(id->vid);
2940 subsys->cmic = id->cmic;
2941 subsys->awupf = le16_to_cpu(id->awupf);
2942 #ifdef CONFIG_NVME_MULTIPATH
2943 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2946 subsys->dev.class = nvme_subsys_class;
2947 subsys->dev.release = nvme_release_subsystem;
2948 subsys->dev.groups = nvme_subsys_attrs_groups;
2949 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2950 device_initialize(&subsys->dev);
2952 mutex_lock(&nvme_subsystems_lock);
2953 found = __nvme_find_get_subsystem(subsys->subnqn);
2955 put_device(&subsys->dev);
2958 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2960 goto out_put_subsystem;
2963 ret = device_add(&subsys->dev);
2965 dev_err(ctrl->device,
2966 "failed to register subsystem device.\n");
2967 put_device(&subsys->dev);
2970 ida_init(&subsys->ns_ida);
2971 list_add_tail(&subsys->entry, &nvme_subsystems);
2974 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2975 dev_name(ctrl->device));
2977 dev_err(ctrl->device,
2978 "failed to create sysfs link from subsystem.\n");
2979 goto out_put_subsystem;
2983 subsys->instance = ctrl->instance;
2984 ctrl->subsys = subsys;
2985 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2986 mutex_unlock(&nvme_subsystems_lock);
2990 nvme_put_subsystem(subsys);
2992 mutex_unlock(&nvme_subsystems_lock);
2996 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2997 void *log, size_t size, u64 offset)
2999 struct nvme_command c = { };
3000 u32 dwlen = nvme_bytes_to_numd(size);
3002 c.get_log_page.opcode = nvme_admin_get_log_page;
3003 c.get_log_page.nsid = cpu_to_le32(nsid);
3004 c.get_log_page.lid = log_page;
3005 c.get_log_page.lsp = lsp;
3006 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3007 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3008 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3009 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3010 c.get_log_page.csi = csi;
3012 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3015 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3016 struct nvme_effects_log **log)
3018 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3024 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3028 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3029 cel, sizeof(*cel), 0);
3035 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3042 * Initialize the cached copies of the Identify data and various controller
3043 * register in our nvme_ctrl structure. This should be called as soon as
3044 * the admin queue is fully up and running.
3046 int nvme_init_identify(struct nvme_ctrl *ctrl)
3048 struct nvme_id_ctrl *id;
3049 int ret, page_shift;
3051 bool prev_apst_enabled;
3053 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3055 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3058 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3059 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3061 if (ctrl->vs >= NVME_VS(1, 1, 0))
3062 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3064 ret = nvme_identify_ctrl(ctrl, &id);
3066 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3070 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3071 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3076 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3077 ctrl->cntlid = le16_to_cpu(id->cntlid);
3079 if (!ctrl->identified) {
3082 ret = nvme_init_subsystem(ctrl, id);
3087 * Check for quirks. Quirk can depend on firmware version,
3088 * so, in principle, the set of quirks present can change
3089 * across a reset. As a possible future enhancement, we
3090 * could re-scan for quirks every time we reinitialize
3091 * the device, but we'd have to make sure that the driver
3092 * behaves intelligently if the quirks change.
3094 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3095 if (quirk_matches(id, &core_quirks[i]))
3096 ctrl->quirks |= core_quirks[i].quirks;
3100 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3101 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3102 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3105 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3106 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3107 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3109 ctrl->oacs = le16_to_cpu(id->oacs);
3110 ctrl->oncs = le16_to_cpu(id->oncs);
3111 ctrl->mtfa = le16_to_cpu(id->mtfa);
3112 ctrl->oaes = le32_to_cpu(id->oaes);
3113 ctrl->wctemp = le16_to_cpu(id->wctemp);
3114 ctrl->cctemp = le16_to_cpu(id->cctemp);
3116 atomic_set(&ctrl->abort_limit, id->acl + 1);
3117 ctrl->vwc = id->vwc;
3119 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3121 max_hw_sectors = UINT_MAX;
3122 ctrl->max_hw_sectors =
3123 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3125 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3126 ctrl->sgls = le32_to_cpu(id->sgls);
3127 ctrl->kas = le16_to_cpu(id->kas);
3128 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3129 ctrl->ctratt = le32_to_cpu(id->ctratt);
3133 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3135 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3136 shutdown_timeout, 60);
3138 if (ctrl->shutdown_timeout != shutdown_timeout)
3139 dev_info(ctrl->device,
3140 "Shutdown timeout set to %u seconds\n",
3141 ctrl->shutdown_timeout);
3143 ctrl->shutdown_timeout = shutdown_timeout;
3145 ctrl->npss = id->npss;
3146 ctrl->apsta = id->apsta;
3147 prev_apst_enabled = ctrl->apst_enabled;
3148 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3149 if (force_apst && id->apsta) {
3150 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3151 ctrl->apst_enabled = true;
3153 ctrl->apst_enabled = false;
3156 ctrl->apst_enabled = id->apsta;
3158 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3160 if (ctrl->ops->flags & NVME_F_FABRICS) {
3161 ctrl->icdoff = le16_to_cpu(id->icdoff);
3162 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3163 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3164 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3167 * In fabrics we need to verify the cntlid matches the
3170 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3171 dev_err(ctrl->device,
3172 "Mismatching cntlid: Connect %u vs Identify "
3174 ctrl->cntlid, le16_to_cpu(id->cntlid));
3179 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3180 dev_err(ctrl->device,
3181 "keep-alive support is mandatory for fabrics\n");
3186 ctrl->hmpre = le32_to_cpu(id->hmpre);
3187 ctrl->hmmin = le32_to_cpu(id->hmmin);
3188 ctrl->hmminds = le32_to_cpu(id->hmminds);
3189 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3192 ret = nvme_mpath_init(ctrl, id);
3198 if (ctrl->apst_enabled && !prev_apst_enabled)
3199 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3200 else if (!ctrl->apst_enabled && prev_apst_enabled)
3201 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3203 ret = nvme_configure_apst(ctrl);
3207 ret = nvme_configure_timestamp(ctrl);
3211 ret = nvme_configure_directives(ctrl);
3215 ret = nvme_configure_acre(ctrl);
3219 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3220 ret = nvme_hwmon_init(ctrl);
3225 ctrl->identified = true;
3233 EXPORT_SYMBOL_GPL(nvme_init_identify);
3235 static int nvme_dev_open(struct inode *inode, struct file *file)
3237 struct nvme_ctrl *ctrl =
3238 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3240 switch (ctrl->state) {
3241 case NVME_CTRL_LIVE:
3244 return -EWOULDBLOCK;
3247 nvme_get_ctrl(ctrl);
3248 if (!try_module_get(ctrl->ops->module)) {
3249 nvme_put_ctrl(ctrl);
3253 file->private_data = ctrl;
3257 static int nvme_dev_release(struct inode *inode, struct file *file)
3259 struct nvme_ctrl *ctrl =
3260 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3262 module_put(ctrl->ops->module);
3263 nvme_put_ctrl(ctrl);
3267 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3272 down_read(&ctrl->namespaces_rwsem);
3273 if (list_empty(&ctrl->namespaces)) {
3278 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3279 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3280 dev_warn(ctrl->device,
3281 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3286 dev_warn(ctrl->device,
3287 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3288 kref_get(&ns->kref);
3289 up_read(&ctrl->namespaces_rwsem);
3291 ret = nvme_user_cmd(ctrl, ns, argp);
3296 up_read(&ctrl->namespaces_rwsem);
3300 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3303 struct nvme_ctrl *ctrl = file->private_data;
3304 void __user *argp = (void __user *)arg;
3307 case NVME_IOCTL_ADMIN_CMD:
3308 return nvme_user_cmd(ctrl, NULL, argp);
3309 case NVME_IOCTL_ADMIN64_CMD:
3310 return nvme_user_cmd64(ctrl, NULL, argp);
3311 case NVME_IOCTL_IO_CMD:
3312 return nvme_dev_user_cmd(ctrl, argp);
3313 case NVME_IOCTL_RESET:
3314 dev_warn(ctrl->device, "resetting controller\n");
3315 return nvme_reset_ctrl_sync(ctrl);
3316 case NVME_IOCTL_SUBSYS_RESET:
3317 return nvme_reset_subsystem(ctrl);
3318 case NVME_IOCTL_RESCAN:
3319 nvme_queue_scan(ctrl);
3326 static const struct file_operations nvme_dev_fops = {
3327 .owner = THIS_MODULE,
3328 .open = nvme_dev_open,
3329 .release = nvme_dev_release,
3330 .unlocked_ioctl = nvme_dev_ioctl,
3331 .compat_ioctl = compat_ptr_ioctl,
3334 static ssize_t nvme_sysfs_reset(struct device *dev,
3335 struct device_attribute *attr, const char *buf,
3338 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3341 ret = nvme_reset_ctrl_sync(ctrl);
3346 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3348 static ssize_t nvme_sysfs_rescan(struct device *dev,
3349 struct device_attribute *attr, const char *buf,
3352 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3354 nvme_queue_scan(ctrl);
3357 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3359 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3361 struct gendisk *disk = dev_to_disk(dev);
3363 if (disk->fops == &nvme_bdev_ops)
3364 return nvme_get_ns_from_dev(dev)->head;
3366 return disk->private_data;
3369 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3372 struct nvme_ns_head *head = dev_to_ns_head(dev);
3373 struct nvme_ns_ids *ids = &head->ids;
3374 struct nvme_subsystem *subsys = head->subsys;
3375 int serial_len = sizeof(subsys->serial);
3376 int model_len = sizeof(subsys->model);
3378 if (!uuid_is_null(&ids->uuid))
3379 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3381 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3382 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3384 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3385 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3387 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3388 subsys->serial[serial_len - 1] == '\0'))
3390 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3391 subsys->model[model_len - 1] == '\0'))
3394 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3395 serial_len, subsys->serial, model_len, subsys->model,
3398 static DEVICE_ATTR_RO(wwid);
3400 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3403 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3405 static DEVICE_ATTR_RO(nguid);
3407 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3410 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3412 /* For backward compatibility expose the NGUID to userspace if
3413 * we have no UUID set
3415 if (uuid_is_null(&ids->uuid)) {
3416 printk_ratelimited(KERN_WARNING
3417 "No UUID available providing old NGUID\n");
3418 return sprintf(buf, "%pU\n", ids->nguid);
3420 return sprintf(buf, "%pU\n", &ids->uuid);
3422 static DEVICE_ATTR_RO(uuid);
3424 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3427 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3429 static DEVICE_ATTR_RO(eui);
3431 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3434 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3436 static DEVICE_ATTR_RO(nsid);
3438 static struct attribute *nvme_ns_id_attrs[] = {
3439 &dev_attr_wwid.attr,
3440 &dev_attr_uuid.attr,
3441 &dev_attr_nguid.attr,
3443 &dev_attr_nsid.attr,
3444 #ifdef CONFIG_NVME_MULTIPATH
3445 &dev_attr_ana_grpid.attr,
3446 &dev_attr_ana_state.attr,
3451 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3452 struct attribute *a, int n)
3454 struct device *dev = container_of(kobj, struct device, kobj);
3455 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3457 if (a == &dev_attr_uuid.attr) {
3458 if (uuid_is_null(&ids->uuid) &&
3459 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3462 if (a == &dev_attr_nguid.attr) {
3463 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3466 if (a == &dev_attr_eui.attr) {
3467 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3470 #ifdef CONFIG_NVME_MULTIPATH
3471 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3472 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3474 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3481 static const struct attribute_group nvme_ns_id_attr_group = {
3482 .attrs = nvme_ns_id_attrs,
3483 .is_visible = nvme_ns_id_attrs_are_visible,
3486 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3487 &nvme_ns_id_attr_group,
3489 &nvme_nvm_attr_group,
3494 #define nvme_show_str_function(field) \
3495 static ssize_t field##_show(struct device *dev, \
3496 struct device_attribute *attr, char *buf) \
3498 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3499 return sprintf(buf, "%.*s\n", \
3500 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3502 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3504 nvme_show_str_function(model);
3505 nvme_show_str_function(serial);
3506 nvme_show_str_function(firmware_rev);
3508 #define nvme_show_int_function(field) \
3509 static ssize_t field##_show(struct device *dev, \
3510 struct device_attribute *attr, char *buf) \
3512 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3513 return sprintf(buf, "%d\n", ctrl->field); \
3515 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3517 nvme_show_int_function(cntlid);
3518 nvme_show_int_function(numa_node);
3519 nvme_show_int_function(queue_count);
3520 nvme_show_int_function(sqsize);
3522 static ssize_t nvme_sysfs_delete(struct device *dev,
3523 struct device_attribute *attr, const char *buf,
3526 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3528 if (device_remove_file_self(dev, attr))
3529 nvme_delete_ctrl_sync(ctrl);
3532 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3534 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3535 struct device_attribute *attr,
3538 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3540 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3542 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3544 static ssize_t nvme_sysfs_show_state(struct device *dev,
3545 struct device_attribute *attr,
3548 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3549 static const char *const state_name[] = {
3550 [NVME_CTRL_NEW] = "new",
3551 [NVME_CTRL_LIVE] = "live",
3552 [NVME_CTRL_RESETTING] = "resetting",
3553 [NVME_CTRL_CONNECTING] = "connecting",
3554 [NVME_CTRL_DELETING] = "deleting",
3555 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3556 [NVME_CTRL_DEAD] = "dead",
3559 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3560 state_name[ctrl->state])
3561 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3563 return sprintf(buf, "unknown state\n");
3566 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3568 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3569 struct device_attribute *attr,
3572 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3574 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3576 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3578 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3579 struct device_attribute *attr,
3582 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3584 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3586 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3588 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3589 struct device_attribute *attr,
3592 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3594 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3596 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3598 static ssize_t nvme_sysfs_show_address(struct device *dev,
3599 struct device_attribute *attr,
3602 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3604 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3606 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3608 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3609 struct device_attribute *attr, char *buf)
3611 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3612 struct nvmf_ctrl_options *opts = ctrl->opts;
3614 if (ctrl->opts->max_reconnects == -1)
3615 return sprintf(buf, "off\n");
3616 return sprintf(buf, "%d\n",
3617 opts->max_reconnects * opts->reconnect_delay);
3620 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3621 struct device_attribute *attr, const char *buf, size_t count)
3623 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3624 struct nvmf_ctrl_options *opts = ctrl->opts;
3625 int ctrl_loss_tmo, err;
3627 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3631 else if (ctrl_loss_tmo < 0)
3632 opts->max_reconnects = -1;
3634 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3635 opts->reconnect_delay);
3638 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3639 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3641 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3642 struct device_attribute *attr, char *buf)
3644 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3646 if (ctrl->opts->reconnect_delay == -1)
3647 return sprintf(buf, "off\n");
3648 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3651 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3652 struct device_attribute *attr, const char *buf, size_t count)
3654 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3658 err = kstrtou32(buf, 10, &v);
3662 ctrl->opts->reconnect_delay = v;
3665 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3666 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3668 static struct attribute *nvme_dev_attrs[] = {
3669 &dev_attr_reset_controller.attr,
3670 &dev_attr_rescan_controller.attr,
3671 &dev_attr_model.attr,
3672 &dev_attr_serial.attr,
3673 &dev_attr_firmware_rev.attr,
3674 &dev_attr_cntlid.attr,
3675 &dev_attr_delete_controller.attr,
3676 &dev_attr_transport.attr,
3677 &dev_attr_subsysnqn.attr,
3678 &dev_attr_address.attr,
3679 &dev_attr_state.attr,
3680 &dev_attr_numa_node.attr,
3681 &dev_attr_queue_count.attr,
3682 &dev_attr_sqsize.attr,
3683 &dev_attr_hostnqn.attr,
3684 &dev_attr_hostid.attr,
3685 &dev_attr_ctrl_loss_tmo.attr,
3686 &dev_attr_reconnect_delay.attr,
3690 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3691 struct attribute *a, int n)
3693 struct device *dev = container_of(kobj, struct device, kobj);
3694 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3696 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3698 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3700 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3702 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3704 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3706 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3712 static const struct attribute_group nvme_dev_attrs_group = {
3713 .attrs = nvme_dev_attrs,
3714 .is_visible = nvme_dev_attrs_are_visible,
3717 static const struct attribute_group *nvme_dev_attr_groups[] = {
3718 &nvme_dev_attrs_group,
3722 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3725 struct nvme_ns_head *h;
3727 lockdep_assert_held(&subsys->lock);
3729 list_for_each_entry(h, &subsys->nsheads, entry) {
3730 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3737 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3738 struct nvme_ns_head *new)
3740 struct nvme_ns_head *h;
3742 lockdep_assert_held(&subsys->lock);
3744 list_for_each_entry(h, &subsys->nsheads, entry) {
3745 if (nvme_ns_ids_valid(&new->ids) &&
3746 nvme_ns_ids_equal(&new->ids, &h->ids))
3753 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3754 unsigned nsid, struct nvme_ns_ids *ids)
3756 struct nvme_ns_head *head;
3757 size_t size = sizeof(*head);
3760 #ifdef CONFIG_NVME_MULTIPATH
3761 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3764 head = kzalloc(size, GFP_KERNEL);
3767 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3770 head->instance = ret;
3771 INIT_LIST_HEAD(&head->list);
3772 ret = init_srcu_struct(&head->srcu);
3774 goto out_ida_remove;
3775 head->subsys = ctrl->subsys;
3778 kref_init(&head->ref);
3780 ret = __nvme_check_ids(ctrl->subsys, head);
3782 dev_err(ctrl->device,
3783 "duplicate IDs for nsid %d\n", nsid);
3784 goto out_cleanup_srcu;
3787 if (head->ids.csi) {
3788 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3790 goto out_cleanup_srcu;
3792 head->effects = ctrl->effects;
3794 ret = nvme_mpath_alloc_disk(ctrl, head);
3796 goto out_cleanup_srcu;
3798 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3800 kref_get(&ctrl->subsys->ref);
3804 cleanup_srcu_struct(&head->srcu);
3806 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3811 ret = blk_status_to_errno(nvme_error_status(ret));
3812 return ERR_PTR(ret);
3815 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3816 struct nvme_ns_ids *ids, bool is_shared)
3818 struct nvme_ctrl *ctrl = ns->ctrl;
3819 struct nvme_ns_head *head = NULL;
3822 mutex_lock(&ctrl->subsys->lock);
3823 head = nvme_find_ns_head(ctrl->subsys, nsid);
3825 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3827 ret = PTR_ERR(head);
3830 head->shared = is_shared;
3833 if (!is_shared || !head->shared) {
3834 dev_err(ctrl->device,
3835 "Duplicate unshared namespace %d\n", nsid);
3836 goto out_put_ns_head;
3838 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3839 dev_err(ctrl->device,
3840 "IDs don't match for shared namespace %d\n",
3842 goto out_put_ns_head;
3846 list_add_tail_rcu(&ns->siblings, &head->list);
3848 mutex_unlock(&ctrl->subsys->lock);
3852 nvme_put_ns_head(head);
3854 mutex_unlock(&ctrl->subsys->lock);
3858 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3860 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3861 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3863 return nsa->head->ns_id - nsb->head->ns_id;
3866 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3868 struct nvme_ns *ns, *ret = NULL;
3870 down_read(&ctrl->namespaces_rwsem);
3871 list_for_each_entry(ns, &ctrl->namespaces, list) {
3872 if (ns->head->ns_id == nsid) {
3873 if (!kref_get_unless_zero(&ns->kref))
3878 if (ns->head->ns_id > nsid)
3881 up_read(&ctrl->namespaces_rwsem);
3884 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3886 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3887 struct nvme_ns_ids *ids)
3890 struct gendisk *disk;
3891 struct nvme_id_ns *id;
3892 char disk_name[DISK_NAME_LEN];
3893 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3895 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3898 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3902 ns->queue = blk_mq_init_queue(ctrl->tagset);
3903 if (IS_ERR(ns->queue))
3906 if (ctrl->opts && ctrl->opts->data_digest)
3907 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3909 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3910 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3911 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3913 ns->queue->queuedata = ns;
3915 kref_init(&ns->kref);
3917 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3918 goto out_free_queue;
3919 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3921 disk = alloc_disk_node(0, node);
3925 disk->fops = &nvme_bdev_ops;
3926 disk->private_data = ns;
3927 disk->queue = ns->queue;
3928 disk->flags = flags;
3929 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3932 if (nvme_update_ns_info(ns, id))
3935 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3936 if (nvme_nvm_register(ns, disk_name, node)) {
3937 dev_warn(ctrl->device, "LightNVM init failure\n");
3942 down_write(&ctrl->namespaces_rwsem);
3943 list_add_tail(&ns->list, &ctrl->namespaces);
3944 up_write(&ctrl->namespaces_rwsem);
3946 nvme_get_ctrl(ctrl);
3948 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3950 nvme_mpath_add_disk(ns, id);
3951 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3956 /* prevent double queue cleanup */
3957 ns->disk->queue = NULL;
3960 mutex_lock(&ctrl->subsys->lock);
3961 list_del_rcu(&ns->siblings);
3962 if (list_empty(&ns->head->list))
3963 list_del_init(&ns->head->entry);
3964 mutex_unlock(&ctrl->subsys->lock);
3965 nvme_put_ns_head(ns->head);
3967 blk_cleanup_queue(ns->queue);
3974 static void nvme_ns_remove(struct nvme_ns *ns)
3976 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3979 set_capacity(ns->disk, 0);
3980 nvme_fault_inject_fini(&ns->fault_inject);
3982 mutex_lock(&ns->ctrl->subsys->lock);
3983 list_del_rcu(&ns->siblings);
3984 if (list_empty(&ns->head->list))
3985 list_del_init(&ns->head->entry);
3986 mutex_unlock(&ns->ctrl->subsys->lock);
3988 synchronize_rcu(); /* guarantee not available in head->list */
3989 nvme_mpath_clear_current_path(ns);
3990 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3992 if (ns->disk->flags & GENHD_FL_UP) {
3993 del_gendisk(ns->disk);
3994 blk_cleanup_queue(ns->queue);
3995 if (blk_get_integrity(ns->disk))
3996 blk_integrity_unregister(ns->disk);
3999 down_write(&ns->ctrl->namespaces_rwsem);
4000 list_del_init(&ns->list);
4001 up_write(&ns->ctrl->namespaces_rwsem);
4003 nvme_mpath_check_last_path(ns);
4007 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4009 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4017 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4019 struct nvme_id_ns *id;
4020 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4022 if (test_bit(NVME_NS_DEAD, &ns->flags))
4025 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4029 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4030 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4031 dev_err(ns->ctrl->device,
4032 "identifiers changed for nsid %d\n", ns->head->ns_id);
4036 ret = nvme_update_ns_info(ns, id);
4042 * Only remove the namespace if we got a fatal error back from the
4043 * device, otherwise ignore the error and just move on.
4045 * TODO: we should probably schedule a delayed retry here.
4047 if (ret > 0 && (ret & NVME_SC_DNR))
4051 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4053 struct nvme_ns_ids ids = { };
4056 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4059 ns = nvme_find_get_ns(ctrl, nsid);
4061 nvme_validate_ns(ns, &ids);
4068 nvme_alloc_ns(ctrl, nsid, &ids);
4071 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4072 dev_warn(ctrl->device,
4073 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4077 if (!nvme_multi_css(ctrl)) {
4078 dev_warn(ctrl->device,
4079 "command set not reported for nsid: %d\n",
4083 nvme_alloc_ns(ctrl, nsid, &ids);
4086 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4092 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4095 struct nvme_ns *ns, *next;
4098 down_write(&ctrl->namespaces_rwsem);
4099 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4100 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4101 list_move_tail(&ns->list, &rm_list);
4103 up_write(&ctrl->namespaces_rwsem);
4105 list_for_each_entry_safe(ns, next, &rm_list, list)
4110 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4112 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4117 if (nvme_ctrl_limited_cns(ctrl))
4120 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4125 struct nvme_command cmd = {
4126 .identify.opcode = nvme_admin_identify,
4127 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4128 .identify.nsid = cpu_to_le32(prev),
4131 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4132 NVME_IDENTIFY_DATA_SIZE);
4134 dev_warn(ctrl->device,
4135 "Identify NS List failed (status=0x%x)\n", ret);
4139 for (i = 0; i < nr_entries; i++) {
4140 u32 nsid = le32_to_cpu(ns_list[i]);
4142 if (!nsid) /* end of the list? */
4144 nvme_validate_or_alloc_ns(ctrl, nsid);
4145 while (++prev < nsid)
4146 nvme_ns_remove_by_nsid(ctrl, prev);
4150 nvme_remove_invalid_namespaces(ctrl, prev);
4156 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4158 struct nvme_id_ctrl *id;
4161 if (nvme_identify_ctrl(ctrl, &id))
4163 nn = le32_to_cpu(id->nn);
4166 for (i = 1; i <= nn; i++)
4167 nvme_validate_or_alloc_ns(ctrl, i);
4169 nvme_remove_invalid_namespaces(ctrl, nn);
4172 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4174 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4178 log = kzalloc(log_size, GFP_KERNEL);
4183 * We need to read the log to clear the AEN, but we don't want to rely
4184 * on it for the changed namespace information as userspace could have
4185 * raced with us in reading the log page, which could cause us to miss
4188 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4189 NVME_CSI_NVM, log, log_size, 0);
4191 dev_warn(ctrl->device,
4192 "reading changed ns log failed: %d\n", error);
4197 static void nvme_scan_work(struct work_struct *work)
4199 struct nvme_ctrl *ctrl =
4200 container_of(work, struct nvme_ctrl, scan_work);
4202 /* No tagset on a live ctrl means IO queues could not created */
4203 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4206 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4207 dev_info(ctrl->device, "rescanning namespaces.\n");
4208 nvme_clear_changed_ns_log(ctrl);
4211 mutex_lock(&ctrl->scan_lock);
4212 if (nvme_scan_ns_list(ctrl) != 0)
4213 nvme_scan_ns_sequential(ctrl);
4214 mutex_unlock(&ctrl->scan_lock);
4216 down_write(&ctrl->namespaces_rwsem);
4217 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4218 up_write(&ctrl->namespaces_rwsem);
4222 * This function iterates the namespace list unlocked to allow recovery from
4223 * controller failure. It is up to the caller to ensure the namespace list is
4224 * not modified by scan work while this function is executing.
4226 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4228 struct nvme_ns *ns, *next;
4232 * make sure to requeue I/O to all namespaces as these
4233 * might result from the scan itself and must complete
4234 * for the scan_work to make progress
4236 nvme_mpath_clear_ctrl_paths(ctrl);
4238 /* prevent racing with ns scanning */
4239 flush_work(&ctrl->scan_work);
4242 * The dead states indicates the controller was not gracefully
4243 * disconnected. In that case, we won't be able to flush any data while
4244 * removing the namespaces' disks; fail all the queues now to avoid
4245 * potentially having to clean up the failed sync later.
4247 if (ctrl->state == NVME_CTRL_DEAD)
4248 nvme_kill_queues(ctrl);
4250 /* this is a no-op when called from the controller reset handler */
4251 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4253 down_write(&ctrl->namespaces_rwsem);
4254 list_splice_init(&ctrl->namespaces, &ns_list);
4255 up_write(&ctrl->namespaces_rwsem);
4257 list_for_each_entry_safe(ns, next, &ns_list, list)
4260 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4262 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4264 struct nvme_ctrl *ctrl =
4265 container_of(dev, struct nvme_ctrl, ctrl_device);
4266 struct nvmf_ctrl_options *opts = ctrl->opts;
4269 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4274 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4278 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4279 opts->trsvcid ?: "none");
4283 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4284 opts->host_traddr ?: "none");
4289 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4291 char *envp[2] = { NULL, NULL };
4292 u32 aen_result = ctrl->aen_result;
4294 ctrl->aen_result = 0;
4298 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4301 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4305 static void nvme_async_event_work(struct work_struct *work)
4307 struct nvme_ctrl *ctrl =
4308 container_of(work, struct nvme_ctrl, async_event_work);
4310 nvme_aen_uevent(ctrl);
4311 ctrl->ops->submit_async_event(ctrl);
4314 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4319 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4325 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4328 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4330 struct nvme_fw_slot_info_log *log;
4332 log = kmalloc(sizeof(*log), GFP_KERNEL);
4336 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4337 log, sizeof(*log), 0))
4338 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4342 static void nvme_fw_act_work(struct work_struct *work)
4344 struct nvme_ctrl *ctrl = container_of(work,
4345 struct nvme_ctrl, fw_act_work);
4346 unsigned long fw_act_timeout;
4349 fw_act_timeout = jiffies +
4350 msecs_to_jiffies(ctrl->mtfa * 100);
4352 fw_act_timeout = jiffies +
4353 msecs_to_jiffies(admin_timeout * 1000);
4355 nvme_stop_queues(ctrl);
4356 while (nvme_ctrl_pp_status(ctrl)) {
4357 if (time_after(jiffies, fw_act_timeout)) {
4358 dev_warn(ctrl->device,
4359 "Fw activation timeout, reset controller\n");
4360 nvme_try_sched_reset(ctrl);
4366 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4369 nvme_start_queues(ctrl);
4370 /* read FW slot information to clear the AER */
4371 nvme_get_fw_slot_info(ctrl);
4374 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4376 u32 aer_notice_type = (result & 0xff00) >> 8;
4378 trace_nvme_async_event(ctrl, aer_notice_type);
4380 switch (aer_notice_type) {
4381 case NVME_AER_NOTICE_NS_CHANGED:
4382 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4383 nvme_queue_scan(ctrl);
4385 case NVME_AER_NOTICE_FW_ACT_STARTING:
4387 * We are (ab)using the RESETTING state to prevent subsequent
4388 * recovery actions from interfering with the controller's
4389 * firmware activation.
4391 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4392 queue_work(nvme_wq, &ctrl->fw_act_work);
4394 #ifdef CONFIG_NVME_MULTIPATH
4395 case NVME_AER_NOTICE_ANA:
4396 if (!ctrl->ana_log_buf)
4398 queue_work(nvme_wq, &ctrl->ana_work);
4401 case NVME_AER_NOTICE_DISC_CHANGED:
4402 ctrl->aen_result = result;
4405 dev_warn(ctrl->device, "async event result %08x\n", result);
4409 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4410 volatile union nvme_result *res)
4412 u32 result = le32_to_cpu(res->u32);
4413 u32 aer_type = result & 0x07;
4415 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4419 case NVME_AER_NOTICE:
4420 nvme_handle_aen_notice(ctrl, result);
4422 case NVME_AER_ERROR:
4423 case NVME_AER_SMART:
4426 trace_nvme_async_event(ctrl, aer_type);
4427 ctrl->aen_result = result;
4432 queue_work(nvme_wq, &ctrl->async_event_work);
4434 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4436 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4438 nvme_mpath_stop(ctrl);
4439 nvme_stop_keep_alive(ctrl);
4440 nvme_stop_failfast_work(ctrl);
4441 flush_work(&ctrl->async_event_work);
4442 cancel_work_sync(&ctrl->fw_act_work);
4444 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4446 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4448 nvme_start_keep_alive(ctrl);
4450 nvme_enable_aen(ctrl);
4452 if (ctrl->queue_count > 1) {
4453 nvme_queue_scan(ctrl);
4454 nvme_start_queues(ctrl);
4457 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4459 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4461 nvme_hwmon_exit(ctrl);
4462 nvme_fault_inject_fini(&ctrl->fault_inject);
4463 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4464 cdev_device_del(&ctrl->cdev, ctrl->device);
4465 nvme_put_ctrl(ctrl);
4467 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4469 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4471 struct nvme_effects_log *cel;
4474 xa_for_each(&ctrl->cels, i, cel) {
4475 xa_erase(&ctrl->cels, i);
4479 xa_destroy(&ctrl->cels);
4482 static void nvme_free_ctrl(struct device *dev)
4484 struct nvme_ctrl *ctrl =
4485 container_of(dev, struct nvme_ctrl, ctrl_device);
4486 struct nvme_subsystem *subsys = ctrl->subsys;
4488 if (!subsys || ctrl->instance != subsys->instance)
4489 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4491 nvme_free_cels(ctrl);
4492 nvme_mpath_uninit(ctrl);
4493 __free_page(ctrl->discard_page);
4496 mutex_lock(&nvme_subsystems_lock);
4497 list_del(&ctrl->subsys_entry);
4498 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4499 mutex_unlock(&nvme_subsystems_lock);
4502 ctrl->ops->free_ctrl(ctrl);
4505 nvme_put_subsystem(subsys);
4509 * Initialize a NVMe controller structures. This needs to be called during
4510 * earliest initialization so that we have the initialized structured around
4513 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4514 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4518 ctrl->state = NVME_CTRL_NEW;
4519 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4520 spin_lock_init(&ctrl->lock);
4521 mutex_init(&ctrl->scan_lock);
4522 INIT_LIST_HEAD(&ctrl->namespaces);
4523 xa_init(&ctrl->cels);
4524 init_rwsem(&ctrl->namespaces_rwsem);
4527 ctrl->quirks = quirks;
4528 ctrl->numa_node = NUMA_NO_NODE;
4529 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4530 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4531 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4532 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4533 init_waitqueue_head(&ctrl->state_wq);
4535 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4536 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4537 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4538 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4540 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4542 ctrl->discard_page = alloc_page(GFP_KERNEL);
4543 if (!ctrl->discard_page) {
4548 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4551 ctrl->instance = ret;
4553 device_initialize(&ctrl->ctrl_device);
4554 ctrl->device = &ctrl->ctrl_device;
4555 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4557 ctrl->device->class = nvme_class;
4558 ctrl->device->parent = ctrl->dev;
4559 ctrl->device->groups = nvme_dev_attr_groups;
4560 ctrl->device->release = nvme_free_ctrl;
4561 dev_set_drvdata(ctrl->device, ctrl);
4562 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4564 goto out_release_instance;
4566 nvme_get_ctrl(ctrl);
4567 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4568 ctrl->cdev.owner = ops->module;
4569 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4574 * Initialize latency tolerance controls. The sysfs files won't
4575 * be visible to userspace unless the device actually supports APST.
4577 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4578 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4579 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4581 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4585 nvme_put_ctrl(ctrl);
4586 kfree_const(ctrl->device->kobj.name);
4587 out_release_instance:
4588 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4590 if (ctrl->discard_page)
4591 __free_page(ctrl->discard_page);
4594 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4597 * nvme_kill_queues(): Ends all namespace queues
4598 * @ctrl: the dead controller that needs to end
4600 * Call this function when the driver determines it is unable to get the
4601 * controller in a state capable of servicing IO.
4603 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4607 down_read(&ctrl->namespaces_rwsem);
4609 /* Forcibly unquiesce queues to avoid blocking dispatch */
4610 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4611 blk_mq_unquiesce_queue(ctrl->admin_q);
4613 list_for_each_entry(ns, &ctrl->namespaces, list)
4614 nvme_set_queue_dying(ns);
4616 up_read(&ctrl->namespaces_rwsem);
4618 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4620 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4624 down_read(&ctrl->namespaces_rwsem);
4625 list_for_each_entry(ns, &ctrl->namespaces, list)
4626 blk_mq_unfreeze_queue(ns->queue);
4627 up_read(&ctrl->namespaces_rwsem);
4629 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4631 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4635 down_read(&ctrl->namespaces_rwsem);
4636 list_for_each_entry(ns, &ctrl->namespaces, list) {
4637 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4641 up_read(&ctrl->namespaces_rwsem);
4644 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4646 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4650 down_read(&ctrl->namespaces_rwsem);
4651 list_for_each_entry(ns, &ctrl->namespaces, list)
4652 blk_mq_freeze_queue_wait(ns->queue);
4653 up_read(&ctrl->namespaces_rwsem);
4655 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4657 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4661 down_read(&ctrl->namespaces_rwsem);
4662 list_for_each_entry(ns, &ctrl->namespaces, list)
4663 blk_freeze_queue_start(ns->queue);
4664 up_read(&ctrl->namespaces_rwsem);
4666 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4668 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4672 down_read(&ctrl->namespaces_rwsem);
4673 list_for_each_entry(ns, &ctrl->namespaces, list)
4674 blk_mq_quiesce_queue(ns->queue);
4675 up_read(&ctrl->namespaces_rwsem);
4677 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4679 void nvme_start_queues(struct nvme_ctrl *ctrl)
4683 down_read(&ctrl->namespaces_rwsem);
4684 list_for_each_entry(ns, &ctrl->namespaces, list)
4685 blk_mq_unquiesce_queue(ns->queue);
4686 up_read(&ctrl->namespaces_rwsem);
4688 EXPORT_SYMBOL_GPL(nvme_start_queues);
4690 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4694 down_read(&ctrl->namespaces_rwsem);
4695 list_for_each_entry(ns, &ctrl->namespaces, list)
4696 blk_sync_queue(ns->queue);
4697 up_read(&ctrl->namespaces_rwsem);
4699 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4701 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4703 nvme_sync_io_queues(ctrl);
4705 blk_sync_queue(ctrl->admin_q);
4707 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4709 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4711 if (file->f_op != &nvme_dev_fops)
4713 return file->private_data;
4715 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4718 * Check we didn't inadvertently grow the command structure sizes:
4720 static inline void _nvme_check_size(void)
4722 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4723 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4724 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4725 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4726 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4727 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4728 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4729 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4730 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4731 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4732 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4733 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4734 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4735 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4736 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4737 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4738 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4739 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4740 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4744 static int __init nvme_core_init(void)
4746 int result = -ENOMEM;
4750 nvme_wq = alloc_workqueue("nvme-wq",
4751 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4755 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4756 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4760 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4761 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4762 if (!nvme_delete_wq)
4763 goto destroy_reset_wq;
4765 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4766 NVME_MINORS, "nvme");
4768 goto destroy_delete_wq;
4770 nvme_class = class_create(THIS_MODULE, "nvme");
4771 if (IS_ERR(nvme_class)) {
4772 result = PTR_ERR(nvme_class);
4773 goto unregister_chrdev;
4775 nvme_class->dev_uevent = nvme_class_uevent;
4777 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4778 if (IS_ERR(nvme_subsys_class)) {
4779 result = PTR_ERR(nvme_subsys_class);
4785 class_destroy(nvme_class);
4787 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4789 destroy_workqueue(nvme_delete_wq);
4791 destroy_workqueue(nvme_reset_wq);
4793 destroy_workqueue(nvme_wq);
4798 static void __exit nvme_core_exit(void)
4800 class_destroy(nvme_subsys_class);
4801 class_destroy(nvme_class);
4802 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4803 destroy_workqueue(nvme_delete_wq);
4804 destroy_workqueue(nvme_reset_wq);
4805 destroy_workqueue(nvme_wq);
4806 ida_destroy(&nvme_instance_ida);
4809 MODULE_LICENSE("GPL");
4810 MODULE_VERSION("1.0");
4811 module_init(nvme_core_init);
4812 module_exit(nvme_core_exit);