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 struct nvme_ns *ns = req->q->queuedata;
283 unsigned long delay = 0;
286 /* The mask and shift result must be <= 3 */
287 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
289 delay = ns->ctrl->crdt[crd - 1] * 100;
291 nvme_req(req)->retries++;
292 blk_mq_requeue_request(req, false);
293 blk_mq_delay_kick_requeue_list(req->q, delay);
296 enum nvme_disposition {
302 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
304 if (likely(nvme_req(req)->status == 0))
307 if (blk_noretry_request(req) ||
308 (nvme_req(req)->status & NVME_SC_DNR) ||
309 nvme_req(req)->retries >= nvme_max_retries)
312 if (req->cmd_flags & REQ_NVME_MPATH) {
313 if (nvme_is_path_error(nvme_req(req)->status) ||
314 blk_queue_dying(req->q))
317 if (blk_queue_dying(req->q))
324 static inline void nvme_end_req(struct request *req)
326 blk_status_t status = nvme_error_status(nvme_req(req)->status);
328 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
329 req_op(req) == REQ_OP_ZONE_APPEND)
330 req->__sector = nvme_lba_to_sect(req->q->queuedata,
331 le64_to_cpu(nvme_req(req)->result.u64));
333 nvme_trace_bio_complete(req);
334 blk_mq_end_request(req, status);
337 void nvme_complete_rq(struct request *req)
339 trace_nvme_complete_rq(req);
340 nvme_cleanup_cmd(req);
342 if (nvme_req(req)->ctrl->kas)
343 nvme_req(req)->ctrl->comp_seen = true;
345 switch (nvme_decide_disposition(req)) {
353 nvme_failover_req(req);
357 EXPORT_SYMBOL_GPL(nvme_complete_rq);
359 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
361 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
362 "Cancelling I/O %d", req->tag);
364 /* don't abort one completed request */
365 if (blk_mq_request_completed(req))
368 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
369 blk_mq_complete_request(req);
372 EXPORT_SYMBOL_GPL(nvme_cancel_request);
374 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
375 enum nvme_ctrl_state new_state)
377 enum nvme_ctrl_state old_state;
379 bool changed = false;
381 spin_lock_irqsave(&ctrl->lock, flags);
383 old_state = ctrl->state;
388 case NVME_CTRL_RESETTING:
389 case NVME_CTRL_CONNECTING:
396 case NVME_CTRL_RESETTING:
406 case NVME_CTRL_CONNECTING:
409 case NVME_CTRL_RESETTING:
416 case NVME_CTRL_DELETING:
419 case NVME_CTRL_RESETTING:
420 case NVME_CTRL_CONNECTING:
427 case NVME_CTRL_DELETING_NOIO:
429 case NVME_CTRL_DELETING:
439 case NVME_CTRL_DELETING:
451 ctrl->state = new_state;
452 wake_up_all(&ctrl->state_wq);
455 spin_unlock_irqrestore(&ctrl->lock, flags);
459 if (ctrl->state == NVME_CTRL_LIVE) {
460 if (old_state == NVME_CTRL_CONNECTING)
461 nvme_stop_failfast_work(ctrl);
462 nvme_kick_requeue_lists(ctrl);
463 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
464 old_state == NVME_CTRL_RESETTING) {
465 nvme_start_failfast_work(ctrl);
469 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
472 * Returns true for sink states that can't ever transition back to live.
474 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
476 switch (ctrl->state) {
479 case NVME_CTRL_RESETTING:
480 case NVME_CTRL_CONNECTING:
482 case NVME_CTRL_DELETING:
483 case NVME_CTRL_DELETING_NOIO:
487 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
493 * Waits for the controller state to be resetting, or returns false if it is
494 * not possible to ever transition to that state.
496 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
498 wait_event(ctrl->state_wq,
499 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
500 nvme_state_terminal(ctrl));
501 return ctrl->state == NVME_CTRL_RESETTING;
503 EXPORT_SYMBOL_GPL(nvme_wait_reset);
505 static void nvme_free_ns_head(struct kref *ref)
507 struct nvme_ns_head *head =
508 container_of(ref, struct nvme_ns_head, ref);
510 nvme_mpath_remove_disk(head);
511 ida_simple_remove(&head->subsys->ns_ida, head->instance);
512 cleanup_srcu_struct(&head->srcu);
513 nvme_put_subsystem(head->subsys);
517 static void nvme_put_ns_head(struct nvme_ns_head *head)
519 kref_put(&head->ref, nvme_free_ns_head);
522 static void nvme_free_ns(struct kref *kref)
524 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
527 nvme_nvm_unregister(ns);
530 nvme_put_ns_head(ns->head);
531 nvme_put_ctrl(ns->ctrl);
535 void nvme_put_ns(struct nvme_ns *ns)
537 kref_put(&ns->kref, nvme_free_ns);
539 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
541 static inline void nvme_clear_nvme_request(struct request *req)
543 if (!(req->rq_flags & RQF_DONTPREP)) {
544 nvme_req(req)->retries = 0;
545 nvme_req(req)->flags = 0;
546 req->rq_flags |= RQF_DONTPREP;
550 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
552 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
555 static inline void nvme_init_request(struct request *req,
556 struct nvme_command *cmd)
558 if (req->q->queuedata)
559 req->timeout = NVME_IO_TIMEOUT;
560 else /* no queuedata implies admin queue */
561 req->timeout = NVME_ADMIN_TIMEOUT;
563 req->cmd_flags |= REQ_FAILFAST_DRIVER;
564 nvme_clear_nvme_request(req);
565 nvme_req(req)->cmd = cmd;
568 struct request *nvme_alloc_request(struct request_queue *q,
569 struct nvme_command *cmd, blk_mq_req_flags_t flags)
573 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
575 nvme_init_request(req, cmd);
578 EXPORT_SYMBOL_GPL(nvme_alloc_request);
580 static struct request *nvme_alloc_request_qid(struct request_queue *q,
581 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
585 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
588 nvme_init_request(req, cmd);
592 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
594 struct nvme_command c;
596 memset(&c, 0, sizeof(c));
598 c.directive.opcode = nvme_admin_directive_send;
599 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
600 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
601 c.directive.dtype = NVME_DIR_IDENTIFY;
602 c.directive.tdtype = NVME_DIR_STREAMS;
603 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
605 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
608 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
610 return nvme_toggle_streams(ctrl, false);
613 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
615 return nvme_toggle_streams(ctrl, true);
618 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
619 struct streams_directive_params *s, u32 nsid)
621 struct nvme_command c;
623 memset(&c, 0, sizeof(c));
624 memset(s, 0, sizeof(*s));
626 c.directive.opcode = nvme_admin_directive_recv;
627 c.directive.nsid = cpu_to_le32(nsid);
628 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
629 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
630 c.directive.dtype = NVME_DIR_STREAMS;
632 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
635 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
637 struct streams_directive_params s;
640 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
645 ret = nvme_enable_streams(ctrl);
649 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
651 goto out_disable_stream;
653 ctrl->nssa = le16_to_cpu(s.nssa);
654 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
655 dev_info(ctrl->device, "too few streams (%u) available\n",
657 goto out_disable_stream;
660 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
661 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
665 nvme_disable_streams(ctrl);
670 * Check if 'req' has a write hint associated with it. If it does, assign
671 * a valid namespace stream to the write.
673 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
674 struct request *req, u16 *control,
677 enum rw_hint streamid = req->write_hint;
679 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
683 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
686 *control |= NVME_RW_DTYPE_STREAMS;
687 *dsmgmt |= streamid << 16;
690 if (streamid < ARRAY_SIZE(req->q->write_hints))
691 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
694 static void nvme_setup_passthrough(struct request *req,
695 struct nvme_command *cmd)
697 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
698 /* passthru commands should let the driver set the SGL flags */
699 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
702 static inline void nvme_setup_flush(struct nvme_ns *ns,
703 struct nvme_command *cmnd)
705 cmnd->common.opcode = nvme_cmd_flush;
706 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
709 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
710 struct nvme_command *cmnd)
712 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
713 struct nvme_dsm_range *range;
717 * Some devices do not consider the DSM 'Number of Ranges' field when
718 * determining how much data to DMA. Always allocate memory for maximum
719 * number of segments to prevent device reading beyond end of buffer.
721 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
723 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
726 * If we fail allocation our range, fallback to the controller
727 * discard page. If that's also busy, it's safe to return
728 * busy, as we know we can make progress once that's freed.
730 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
731 return BLK_STS_RESOURCE;
733 range = page_address(ns->ctrl->discard_page);
736 __rq_for_each_bio(bio, req) {
737 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
738 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
741 range[n].cattr = cpu_to_le32(0);
742 range[n].nlb = cpu_to_le32(nlb);
743 range[n].slba = cpu_to_le64(slba);
748 if (WARN_ON_ONCE(n != segments)) {
749 if (virt_to_page(range) == ns->ctrl->discard_page)
750 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
753 return BLK_STS_IOERR;
756 cmnd->dsm.opcode = nvme_cmd_dsm;
757 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
758 cmnd->dsm.nr = cpu_to_le32(segments - 1);
759 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
761 req->special_vec.bv_page = virt_to_page(range);
762 req->special_vec.bv_offset = offset_in_page(range);
763 req->special_vec.bv_len = alloc_size;
764 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
769 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
770 struct request *req, struct nvme_command *cmnd)
772 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
773 return nvme_setup_discard(ns, req, cmnd);
775 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
776 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
777 cmnd->write_zeroes.slba =
778 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
779 cmnd->write_zeroes.length =
780 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
781 cmnd->write_zeroes.control = 0;
785 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
786 struct request *req, struct nvme_command *cmnd,
789 struct nvme_ctrl *ctrl = ns->ctrl;
793 if (req->cmd_flags & REQ_FUA)
794 control |= NVME_RW_FUA;
795 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
796 control |= NVME_RW_LR;
798 if (req->cmd_flags & REQ_RAHEAD)
799 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
801 cmnd->rw.opcode = op;
802 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
803 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
804 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
806 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
807 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
811 * If formated with metadata, the block layer always provides a
812 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
813 * we enable the PRACT bit for protection information or set the
814 * namespace capacity to zero to prevent any I/O.
816 if (!blk_integrity_rq(req)) {
817 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
818 return BLK_STS_NOTSUPP;
819 control |= NVME_RW_PRINFO_PRACT;
822 switch (ns->pi_type) {
823 case NVME_NS_DPS_PI_TYPE3:
824 control |= NVME_RW_PRINFO_PRCHK_GUARD;
826 case NVME_NS_DPS_PI_TYPE1:
827 case NVME_NS_DPS_PI_TYPE2:
828 control |= NVME_RW_PRINFO_PRCHK_GUARD |
829 NVME_RW_PRINFO_PRCHK_REF;
830 if (op == nvme_cmd_zone_append)
831 control |= NVME_RW_APPEND_PIREMAP;
832 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
837 cmnd->rw.control = cpu_to_le16(control);
838 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
842 void nvme_cleanup_cmd(struct request *req)
844 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
845 struct nvme_ns *ns = req->rq_disk->private_data;
846 struct page *page = req->special_vec.bv_page;
848 if (page == ns->ctrl->discard_page)
849 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
851 kfree(page_address(page) + req->special_vec.bv_offset);
854 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
856 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
857 struct nvme_command *cmd)
859 blk_status_t ret = BLK_STS_OK;
861 nvme_clear_nvme_request(req);
863 memset(cmd, 0, sizeof(*cmd));
864 switch (req_op(req)) {
867 nvme_setup_passthrough(req, cmd);
870 nvme_setup_flush(ns, cmd);
872 case REQ_OP_ZONE_RESET_ALL:
873 case REQ_OP_ZONE_RESET:
874 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
876 case REQ_OP_ZONE_OPEN:
877 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
879 case REQ_OP_ZONE_CLOSE:
880 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
882 case REQ_OP_ZONE_FINISH:
883 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
885 case REQ_OP_WRITE_ZEROES:
886 ret = nvme_setup_write_zeroes(ns, req, cmd);
889 ret = nvme_setup_discard(ns, req, cmd);
892 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
895 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
897 case REQ_OP_ZONE_APPEND:
898 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
902 return BLK_STS_IOERR;
905 cmd->common.command_id = req->tag;
906 trace_nvme_setup_cmd(req, cmd);
909 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
911 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
913 struct completion *waiting = rq->end_io_data;
915 rq->end_io_data = NULL;
919 static void nvme_execute_rq_polled(struct request_queue *q,
920 struct gendisk *bd_disk, struct request *rq, int at_head)
922 DECLARE_COMPLETION_ONSTACK(wait);
924 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
926 rq->cmd_flags |= REQ_HIPRI;
927 rq->end_io_data = &wait;
928 blk_execute_rq_nowait(bd_disk, rq, at_head, nvme_end_sync_rq);
930 while (!completion_done(&wait)) {
931 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
937 * Returns 0 on success. If the result is negative, it's a Linux error code;
938 * if the result is positive, it's an NVM Express status code
940 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
941 union nvme_result *result, void *buffer, unsigned bufflen,
942 unsigned timeout, int qid, int at_head,
943 blk_mq_req_flags_t flags, bool poll)
948 if (qid == NVME_QID_ANY)
949 req = nvme_alloc_request(q, cmd, flags);
951 req = nvme_alloc_request_qid(q, cmd, flags, qid);
956 req->timeout = timeout;
958 if (buffer && bufflen) {
959 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
965 nvme_execute_rq_polled(req->q, NULL, req, at_head);
967 blk_execute_rq(NULL, req, at_head);
969 *result = nvme_req(req)->result;
970 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
973 ret = nvme_req(req)->status;
975 blk_mq_free_request(req);
978 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
980 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
981 void *buffer, unsigned bufflen)
983 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
984 NVME_QID_ANY, 0, 0, false);
986 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
988 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
989 unsigned len, u32 seed, bool write)
991 struct bio_integrity_payload *bip;
995 buf = kmalloc(len, GFP_KERNEL);
1000 if (write && copy_from_user(buf, ubuf, len))
1003 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
1009 bip->bip_iter.bi_size = len;
1010 bip->bip_iter.bi_sector = seed;
1011 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
1012 offset_in_page(buf));
1019 return ERR_PTR(ret);
1022 static u32 nvme_known_admin_effects(u8 opcode)
1025 case nvme_admin_format_nvm:
1026 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1027 NVME_CMD_EFFECTS_CSE_MASK;
1028 case nvme_admin_sanitize_nvm:
1029 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1036 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1041 if (ns->head->effects)
1042 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1043 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1044 dev_warn(ctrl->device,
1045 "IO command:%02x has unhandled effects:%08x\n",
1051 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1052 effects |= nvme_known_admin_effects(opcode);
1056 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1058 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1061 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1064 * For simplicity, IO to all namespaces is quiesced even if the command
1065 * effects say only one namespace is affected.
1067 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1068 mutex_lock(&ctrl->scan_lock);
1069 mutex_lock(&ctrl->subsys->lock);
1070 nvme_mpath_start_freeze(ctrl->subsys);
1071 nvme_mpath_wait_freeze(ctrl->subsys);
1072 nvme_start_freeze(ctrl);
1073 nvme_wait_freeze(ctrl);
1078 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1080 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1081 nvme_unfreeze(ctrl);
1082 nvme_mpath_unfreeze(ctrl->subsys);
1083 mutex_unlock(&ctrl->subsys->lock);
1084 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1085 mutex_unlock(&ctrl->scan_lock);
1087 if (effects & NVME_CMD_EFFECTS_CCC)
1088 nvme_init_identify(ctrl);
1089 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1090 nvme_queue_scan(ctrl);
1091 flush_work(&ctrl->scan_work);
1095 void nvme_execute_passthru_rq(struct request *rq)
1097 struct nvme_command *cmd = nvme_req(rq)->cmd;
1098 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1099 struct nvme_ns *ns = rq->q->queuedata;
1100 struct gendisk *disk = ns ? ns->disk : NULL;
1103 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1104 blk_execute_rq(disk, rq, 0);
1105 nvme_passthru_end(ctrl, effects);
1107 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1109 static int nvme_submit_user_cmd(struct request_queue *q,
1110 struct nvme_command *cmd, void __user *ubuffer,
1111 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1112 u32 meta_seed, u64 *result, unsigned timeout)
1114 bool write = nvme_is_write(cmd);
1115 struct nvme_ns *ns = q->queuedata;
1116 struct block_device *bdev = ns ? ns->disk->part0 : NULL;
1117 struct request *req;
1118 struct bio *bio = NULL;
1122 req = nvme_alloc_request(q, cmd, 0);
1124 return PTR_ERR(req);
1127 req->timeout = timeout;
1128 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1130 if (ubuffer && bufflen) {
1131 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1137 bio_set_dev(bio, bdev);
1138 if (bdev && meta_buffer && meta_len) {
1139 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1142 ret = PTR_ERR(meta);
1145 req->cmd_flags |= REQ_INTEGRITY;
1149 nvme_execute_passthru_rq(req);
1150 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1153 ret = nvme_req(req)->status;
1155 *result = le64_to_cpu(nvme_req(req)->result.u64);
1156 if (meta && !ret && !write) {
1157 if (copy_to_user(meta_buffer, meta, meta_len))
1163 blk_rq_unmap_user(bio);
1165 blk_mq_free_request(req);
1169 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1171 struct nvme_ctrl *ctrl = rq->end_io_data;
1172 unsigned long flags;
1173 bool startka = false;
1175 blk_mq_free_request(rq);
1178 dev_err(ctrl->device,
1179 "failed nvme_keep_alive_end_io error=%d\n",
1184 ctrl->comp_seen = false;
1185 spin_lock_irqsave(&ctrl->lock, flags);
1186 if (ctrl->state == NVME_CTRL_LIVE ||
1187 ctrl->state == NVME_CTRL_CONNECTING)
1189 spin_unlock_irqrestore(&ctrl->lock, flags);
1191 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1194 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1198 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1199 BLK_MQ_REQ_RESERVED);
1203 rq->timeout = ctrl->kato * HZ;
1204 rq->end_io_data = ctrl;
1206 blk_execute_rq_nowait(NULL, rq, 0, nvme_keep_alive_end_io);
1211 static void nvme_keep_alive_work(struct work_struct *work)
1213 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1214 struct nvme_ctrl, ka_work);
1215 bool comp_seen = ctrl->comp_seen;
1217 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1218 dev_dbg(ctrl->device,
1219 "reschedule traffic based keep-alive timer\n");
1220 ctrl->comp_seen = false;
1221 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1225 if (nvme_keep_alive(ctrl)) {
1226 /* allocation failure, reset the controller */
1227 dev_err(ctrl->device, "keep-alive failed\n");
1228 nvme_reset_ctrl(ctrl);
1233 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1235 if (unlikely(ctrl->kato == 0))
1238 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1241 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1243 if (unlikely(ctrl->kato == 0))
1246 cancel_delayed_work_sync(&ctrl->ka_work);
1248 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1251 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1252 * flag, thus sending any new CNS opcodes has a big chance of not working.
1253 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1254 * (but not for any later version).
1256 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1258 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1259 return ctrl->vs < NVME_VS(1, 2, 0);
1260 return ctrl->vs < NVME_VS(1, 1, 0);
1263 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1265 struct nvme_command c = { };
1268 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1269 c.identify.opcode = nvme_admin_identify;
1270 c.identify.cns = NVME_ID_CNS_CTRL;
1272 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1276 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1277 sizeof(struct nvme_id_ctrl));
1283 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1285 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1288 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1289 struct nvme_ns_id_desc *cur, bool *csi_seen)
1291 const char *warn_str = "ctrl returned bogus length:";
1294 switch (cur->nidt) {
1295 case NVME_NIDT_EUI64:
1296 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1297 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1298 warn_str, cur->nidl);
1301 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1302 return NVME_NIDT_EUI64_LEN;
1303 case NVME_NIDT_NGUID:
1304 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1305 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1306 warn_str, cur->nidl);
1309 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1310 return NVME_NIDT_NGUID_LEN;
1311 case NVME_NIDT_UUID:
1312 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1313 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1314 warn_str, cur->nidl);
1317 uuid_copy(&ids->uuid, data + sizeof(*cur));
1318 return NVME_NIDT_UUID_LEN;
1320 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1321 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1322 warn_str, cur->nidl);
1325 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1327 return NVME_NIDT_CSI_LEN;
1329 /* Skip unknown types */
1334 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1335 struct nvme_ns_ids *ids)
1337 struct nvme_command c = { };
1338 bool csi_seen = false;
1339 int status, pos, len;
1342 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1344 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1347 c.identify.opcode = nvme_admin_identify;
1348 c.identify.nsid = cpu_to_le32(nsid);
1349 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1351 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1355 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1356 NVME_IDENTIFY_DATA_SIZE);
1358 dev_warn(ctrl->device,
1359 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1364 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1365 struct nvme_ns_id_desc *cur = data + pos;
1370 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1374 len += sizeof(*cur);
1377 if (nvme_multi_css(ctrl) && !csi_seen) {
1378 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1388 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1389 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1391 struct nvme_command c = { };
1394 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1395 c.identify.opcode = nvme_admin_identify;
1396 c.identify.nsid = cpu_to_le32(nsid);
1397 c.identify.cns = NVME_ID_CNS_NS;
1399 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1403 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1405 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1410 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1413 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1414 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1415 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1416 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1417 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1418 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1427 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1428 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1430 union nvme_result res = { 0 };
1431 struct nvme_command c;
1434 memset(&c, 0, sizeof(c));
1435 c.features.opcode = op;
1436 c.features.fid = cpu_to_le32(fid);
1437 c.features.dword11 = cpu_to_le32(dword11);
1439 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1440 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1441 if (ret >= 0 && result)
1442 *result = le32_to_cpu(res.u32);
1446 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1447 unsigned int dword11, void *buffer, size_t buflen,
1450 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1453 EXPORT_SYMBOL_GPL(nvme_set_features);
1455 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1456 unsigned int dword11, void *buffer, size_t buflen,
1459 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1462 EXPORT_SYMBOL_GPL(nvme_get_features);
1464 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1466 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1468 int status, nr_io_queues;
1470 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1476 * Degraded controllers might return an error when setting the queue
1477 * count. We still want to be able to bring them online and offer
1478 * access to the admin queue, as that might be only way to fix them up.
1481 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1484 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1485 *count = min(*count, nr_io_queues);
1490 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1492 #define NVME_AEN_SUPPORTED \
1493 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1494 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1496 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1498 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1501 if (!supported_aens)
1504 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1507 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1510 queue_work(nvme_wq, &ctrl->async_event_work);
1514 * Convert integer values from ioctl structures to user pointers, silently
1515 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1518 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1520 if (in_compat_syscall())
1521 ptrval = (compat_uptr_t)ptrval;
1522 return (void __user *)ptrval;
1525 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1527 struct nvme_user_io io;
1528 struct nvme_command c;
1529 unsigned length, meta_len;
1530 void __user *metadata;
1532 if (copy_from_user(&io, uio, sizeof(io)))
1537 switch (io.opcode) {
1538 case nvme_cmd_write:
1540 case nvme_cmd_compare:
1546 length = (io.nblocks + 1) << ns->lba_shift;
1548 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1549 ns->ms == sizeof(struct t10_pi_tuple)) {
1551 * Protection information is stripped/inserted by the
1554 if (nvme_to_user_ptr(io.metadata))
1559 meta_len = (io.nblocks + 1) * ns->ms;
1560 metadata = nvme_to_user_ptr(io.metadata);
1563 if (ns->features & NVME_NS_EXT_LBAS) {
1566 } else if (meta_len) {
1567 if ((io.metadata & 3) || !io.metadata)
1571 memset(&c, 0, sizeof(c));
1572 c.rw.opcode = io.opcode;
1573 c.rw.flags = io.flags;
1574 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1575 c.rw.slba = cpu_to_le64(io.slba);
1576 c.rw.length = cpu_to_le16(io.nblocks);
1577 c.rw.control = cpu_to_le16(io.control);
1578 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1579 c.rw.reftag = cpu_to_le32(io.reftag);
1580 c.rw.apptag = cpu_to_le16(io.apptag);
1581 c.rw.appmask = cpu_to_le16(io.appmask);
1583 return nvme_submit_user_cmd(ns->queue, &c,
1584 nvme_to_user_ptr(io.addr), length,
1585 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1588 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1589 struct nvme_passthru_cmd __user *ucmd)
1591 struct nvme_passthru_cmd cmd;
1592 struct nvme_command c;
1593 unsigned timeout = 0;
1597 if (!capable(CAP_SYS_ADMIN))
1599 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1604 memset(&c, 0, sizeof(c));
1605 c.common.opcode = cmd.opcode;
1606 c.common.flags = cmd.flags;
1607 c.common.nsid = cpu_to_le32(cmd.nsid);
1608 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1609 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1610 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1611 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1612 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1613 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1614 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1615 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1618 timeout = msecs_to_jiffies(cmd.timeout_ms);
1620 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1621 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1622 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1623 0, &result, timeout);
1626 if (put_user(result, &ucmd->result))
1633 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1634 struct nvme_passthru_cmd64 __user *ucmd)
1636 struct nvme_passthru_cmd64 cmd;
1637 struct nvme_command c;
1638 unsigned timeout = 0;
1641 if (!capable(CAP_SYS_ADMIN))
1643 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1648 memset(&c, 0, sizeof(c));
1649 c.common.opcode = cmd.opcode;
1650 c.common.flags = cmd.flags;
1651 c.common.nsid = cpu_to_le32(cmd.nsid);
1652 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1653 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1654 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1655 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1656 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1657 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1658 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1659 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1662 timeout = msecs_to_jiffies(cmd.timeout_ms);
1664 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1665 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1666 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1667 0, &cmd.result, timeout);
1670 if (put_user(cmd.result, &ucmd->result))
1678 * Issue ioctl requests on the first available path. Note that unlike normal
1679 * block layer requests we will not retry failed request on another controller.
1681 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1682 struct nvme_ns_head **head, int *srcu_idx)
1684 #ifdef CONFIG_NVME_MULTIPATH
1685 if (disk->fops == &nvme_ns_head_ops) {
1688 *head = disk->private_data;
1689 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1690 ns = nvme_find_path(*head);
1692 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1698 return disk->private_data;
1701 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1704 srcu_read_unlock(&head->srcu, idx);
1707 static bool is_ctrl_ioctl(unsigned int cmd)
1709 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1711 if (is_sed_ioctl(cmd))
1716 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1718 struct nvme_ns_head *head,
1721 struct nvme_ctrl *ctrl = ns->ctrl;
1724 nvme_get_ctrl(ns->ctrl);
1725 nvme_put_ns_from_disk(head, srcu_idx);
1728 case NVME_IOCTL_ADMIN_CMD:
1729 ret = nvme_user_cmd(ctrl, NULL, argp);
1731 case NVME_IOCTL_ADMIN64_CMD:
1732 ret = nvme_user_cmd64(ctrl, NULL, argp);
1735 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1738 nvme_put_ctrl(ctrl);
1742 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1743 unsigned int cmd, unsigned long arg)
1745 struct nvme_ns_head *head = NULL;
1746 void __user *argp = (void __user *)arg;
1750 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1752 return -EWOULDBLOCK;
1755 * Handle ioctls that apply to the controller instead of the namespace
1756 * seperately and drop the ns SRCU reference early. This avoids a
1757 * deadlock when deleting namespaces using the passthrough interface.
1759 if (is_ctrl_ioctl(cmd))
1760 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1764 force_successful_syscall_return();
1765 ret = ns->head->ns_id;
1767 case NVME_IOCTL_IO_CMD:
1768 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1770 case NVME_IOCTL_SUBMIT_IO:
1771 ret = nvme_submit_io(ns, argp);
1773 case NVME_IOCTL_IO64_CMD:
1774 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1778 ret = nvme_nvm_ioctl(ns, cmd, arg);
1783 nvme_put_ns_from_disk(head, srcu_idx);
1787 #ifdef CONFIG_COMPAT
1788 struct nvme_user_io32 {
1801 } __attribute__((__packed__));
1803 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1805 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1806 unsigned int cmd, unsigned long arg)
1809 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1810 * between 32 bit programs and 64 bit kernel.
1811 * The cause is that the results of sizeof(struct nvme_user_io),
1812 * which is used to define NVME_IOCTL_SUBMIT_IO,
1813 * are not same between 32 bit compiler and 64 bit compiler.
1814 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1815 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1816 * Other IOCTL numbers are same between 32 bit and 64 bit.
1817 * So there is nothing to do regarding to other IOCTL numbers.
1819 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1820 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1822 return nvme_ioctl(bdev, mode, cmd, arg);
1825 #define nvme_compat_ioctl NULL
1826 #endif /* CONFIG_COMPAT */
1828 static int nvme_open(struct block_device *bdev, fmode_t mode)
1830 struct nvme_ns *ns = bdev->bd_disk->private_data;
1832 #ifdef CONFIG_NVME_MULTIPATH
1833 /* should never be called due to GENHD_FL_HIDDEN */
1834 if (WARN_ON_ONCE(ns->head->disk))
1837 if (!kref_get_unless_zero(&ns->kref))
1839 if (!try_module_get(ns->ctrl->ops->module))
1850 static void nvme_release(struct gendisk *disk, fmode_t mode)
1852 struct nvme_ns *ns = disk->private_data;
1854 module_put(ns->ctrl->ops->module);
1858 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1860 /* some standard values */
1861 geo->heads = 1 << 6;
1862 geo->sectors = 1 << 5;
1863 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1867 #ifdef CONFIG_BLK_DEV_INTEGRITY
1868 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1869 u32 max_integrity_segments)
1871 struct blk_integrity integrity;
1873 memset(&integrity, 0, sizeof(integrity));
1875 case NVME_NS_DPS_PI_TYPE3:
1876 integrity.profile = &t10_pi_type3_crc;
1877 integrity.tag_size = sizeof(u16) + sizeof(u32);
1878 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1880 case NVME_NS_DPS_PI_TYPE1:
1881 case NVME_NS_DPS_PI_TYPE2:
1882 integrity.profile = &t10_pi_type1_crc;
1883 integrity.tag_size = sizeof(u16);
1884 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1887 integrity.profile = NULL;
1890 integrity.tuple_size = ms;
1891 blk_integrity_register(disk, &integrity);
1892 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1895 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1896 u32 max_integrity_segments)
1899 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1901 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1903 struct nvme_ctrl *ctrl = ns->ctrl;
1904 struct request_queue *queue = disk->queue;
1905 u32 size = queue_logical_block_size(queue);
1907 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1908 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1912 if (ctrl->nr_streams && ns->sws && ns->sgs)
1913 size *= ns->sws * ns->sgs;
1915 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1916 NVME_DSM_MAX_RANGES);
1918 queue->limits.discard_alignment = 0;
1919 queue->limits.discard_granularity = size;
1921 /* If discard is already enabled, don't reset queue limits */
1922 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1925 blk_queue_max_discard_sectors(queue, UINT_MAX);
1926 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1928 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1929 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1932 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1936 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1937 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1940 * Even though NVMe spec explicitly states that MDTS is not
1941 * applicable to the write-zeroes:- "The restriction does not apply to
1942 * commands that do not transfer data between the host and the
1943 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1944 * In order to be more cautious use controller's max_hw_sectors value
1945 * to configure the maximum sectors for the write-zeroes which is
1946 * configured based on the controller's MDTS field in the
1947 * nvme_init_identify() if available.
1949 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1950 max_blocks = (u64)USHRT_MAX + 1;
1952 max_blocks = ns->ctrl->max_hw_sectors + 1;
1954 blk_queue_max_write_zeroes_sectors(disk->queue,
1955 nvme_lba_to_sect(ns, max_blocks));
1958 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1960 return !uuid_is_null(&ids->uuid) ||
1961 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1962 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1965 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1967 return uuid_equal(&a->uuid, &b->uuid) &&
1968 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1969 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1973 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1974 u32 *phys_bs, u32 *io_opt)
1976 struct streams_directive_params s;
1979 if (!ctrl->nr_streams)
1982 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1986 ns->sws = le32_to_cpu(s.sws);
1987 ns->sgs = le16_to_cpu(s.sgs);
1990 *phys_bs = ns->sws * (1 << ns->lba_shift);
1992 *io_opt = *phys_bs * ns->sgs;
1998 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
2000 struct nvme_ctrl *ctrl = ns->ctrl;
2003 * The PI implementation requires the metadata size to be equal to the
2004 * t10 pi tuple size.
2006 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2007 if (ns->ms == sizeof(struct t10_pi_tuple))
2008 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2012 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2013 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2015 if (ctrl->ops->flags & NVME_F_FABRICS) {
2017 * The NVMe over Fabrics specification only supports metadata as
2018 * part of the extended data LBA. We rely on HCA/HBA support to
2019 * remap the separate metadata buffer from the block layer.
2021 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2023 if (ctrl->max_integrity_segments)
2025 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2028 * For PCIe controllers, we can't easily remap the separate
2029 * metadata buffer from the block layer and thus require a
2030 * separate metadata buffer for block layer metadata/PI support.
2031 * We allow extended LBAs for the passthrough interface, though.
2033 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2034 ns->features |= NVME_NS_EXT_LBAS;
2036 ns->features |= NVME_NS_METADATA_SUPPORTED;
2042 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2043 struct request_queue *q)
2045 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2047 if (ctrl->max_hw_sectors) {
2049 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2051 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2052 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2053 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2055 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2056 blk_queue_dma_alignment(q, 7);
2057 blk_queue_write_cache(q, vwc, vwc);
2060 static void nvme_update_disk_info(struct gendisk *disk,
2061 struct nvme_ns *ns, struct nvme_id_ns *id)
2063 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2064 unsigned short bs = 1 << ns->lba_shift;
2065 u32 atomic_bs, phys_bs, io_opt = 0;
2068 * The block layer can't support LBA sizes larger than the page size
2069 * yet, so catch this early and don't allow block I/O.
2071 if (ns->lba_shift > PAGE_SHIFT) {
2076 blk_integrity_unregister(disk);
2078 atomic_bs = phys_bs = bs;
2079 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2080 if (id->nabo == 0) {
2082 * Bit 1 indicates whether NAWUPF is defined for this namespace
2083 * and whether it should be used instead of AWUPF. If NAWUPF ==
2084 * 0 then AWUPF must be used instead.
2086 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2087 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2089 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2092 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2093 /* NPWG = Namespace Preferred Write Granularity */
2094 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2095 /* NOWS = Namespace Optimal Write Size */
2096 io_opt = bs * (1 + le16_to_cpu(id->nows));
2099 blk_queue_logical_block_size(disk->queue, bs);
2101 * Linux filesystems assume writing a single physical block is
2102 * an atomic operation. Hence limit the physical block size to the
2103 * value of the Atomic Write Unit Power Fail parameter.
2105 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2106 blk_queue_io_min(disk->queue, phys_bs);
2107 blk_queue_io_opt(disk->queue, io_opt);
2110 * Register a metadata profile for PI, or the plain non-integrity NVMe
2111 * metadata masquerading as Type 0 if supported, otherwise reject block
2112 * I/O to namespaces with metadata except when the namespace supports
2113 * PI, as it can strip/insert in that case.
2116 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2117 (ns->features & NVME_NS_METADATA_SUPPORTED))
2118 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2119 ns->ctrl->max_integrity_segments);
2120 else if (!nvme_ns_has_pi(ns))
2124 set_capacity_and_notify(disk, capacity);
2126 nvme_config_discard(disk, ns);
2127 nvme_config_write_zeroes(disk, ns);
2129 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
2130 test_bit(NVME_NS_FORCE_RO, &ns->flags));
2133 static inline bool nvme_first_scan(struct gendisk *disk)
2135 /* nvme_alloc_ns() scans the disk prior to adding it */
2136 return !(disk->flags & GENHD_FL_UP);
2139 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2141 struct nvme_ctrl *ctrl = ns->ctrl;
2144 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2145 is_power_of_2(ctrl->max_hw_sectors))
2146 iob = ctrl->max_hw_sectors;
2148 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2153 if (!is_power_of_2(iob)) {
2154 if (nvme_first_scan(ns->disk))
2155 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2156 ns->disk->disk_name, iob);
2160 if (blk_queue_is_zoned(ns->disk->queue)) {
2161 if (nvme_first_scan(ns->disk))
2162 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2163 ns->disk->disk_name);
2167 blk_queue_chunk_sectors(ns->queue, iob);
2170 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
2172 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2175 blk_mq_freeze_queue(ns->disk->queue);
2176 ns->lba_shift = id->lbaf[lbaf].ds;
2177 nvme_set_queue_limits(ns->ctrl, ns->queue);
2179 ret = nvme_configure_metadata(ns, id);
2182 nvme_set_chunk_sectors(ns, id);
2183 nvme_update_disk_info(ns->disk, ns, id);
2185 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2186 ret = nvme_update_zone_info(ns, lbaf);
2191 blk_mq_unfreeze_queue(ns->disk->queue);
2193 if (blk_queue_is_zoned(ns->queue)) {
2194 ret = nvme_revalidate_zones(ns);
2195 if (ret && !nvme_first_scan(ns->disk))
2199 #ifdef CONFIG_NVME_MULTIPATH
2200 if (ns->head->disk) {
2201 blk_mq_freeze_queue(ns->head->disk->queue);
2202 nvme_update_disk_info(ns->head->disk, ns, id);
2203 blk_stack_limits(&ns->head->disk->queue->limits,
2204 &ns->queue->limits, 0);
2205 blk_queue_update_readahead(ns->head->disk->queue);
2206 blk_mq_unfreeze_queue(ns->head->disk->queue);
2212 blk_mq_unfreeze_queue(ns->disk->queue);
2216 static char nvme_pr_type(enum pr_type type)
2219 case PR_WRITE_EXCLUSIVE:
2221 case PR_EXCLUSIVE_ACCESS:
2223 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2225 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2227 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2229 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2236 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2237 u64 key, u64 sa_key, u8 op)
2239 struct nvme_ns_head *head = NULL;
2241 struct nvme_command c;
2243 u8 data[16] = { 0, };
2245 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2247 return -EWOULDBLOCK;
2249 put_unaligned_le64(key, &data[0]);
2250 put_unaligned_le64(sa_key, &data[8]);
2252 memset(&c, 0, sizeof(c));
2253 c.common.opcode = op;
2254 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2255 c.common.cdw10 = cpu_to_le32(cdw10);
2257 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2258 nvme_put_ns_from_disk(head, srcu_idx);
2262 static int nvme_pr_register(struct block_device *bdev, u64 old,
2263 u64 new, unsigned flags)
2267 if (flags & ~PR_FL_IGNORE_KEY)
2270 cdw10 = old ? 2 : 0;
2271 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2272 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2273 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2276 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2277 enum pr_type type, unsigned flags)
2281 if (flags & ~PR_FL_IGNORE_KEY)
2284 cdw10 = nvme_pr_type(type) << 8;
2285 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2286 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2289 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2290 enum pr_type type, bool abort)
2292 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2293 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2296 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2298 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2299 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2302 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2304 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2305 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2308 static const struct pr_ops nvme_pr_ops = {
2309 .pr_register = nvme_pr_register,
2310 .pr_reserve = nvme_pr_reserve,
2311 .pr_release = nvme_pr_release,
2312 .pr_preempt = nvme_pr_preempt,
2313 .pr_clear = nvme_pr_clear,
2316 #ifdef CONFIG_BLK_SED_OPAL
2317 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2320 struct nvme_ctrl *ctrl = data;
2321 struct nvme_command cmd;
2323 memset(&cmd, 0, sizeof(cmd));
2325 cmd.common.opcode = nvme_admin_security_send;
2327 cmd.common.opcode = nvme_admin_security_recv;
2328 cmd.common.nsid = 0;
2329 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2330 cmd.common.cdw11 = cpu_to_le32(len);
2332 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2333 NVME_QID_ANY, 1, 0, false);
2335 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2336 #endif /* CONFIG_BLK_SED_OPAL */
2338 static const struct block_device_operations nvme_bdev_ops = {
2339 .owner = THIS_MODULE,
2340 .ioctl = nvme_ioctl,
2341 .compat_ioctl = nvme_compat_ioctl,
2343 .release = nvme_release,
2344 .getgeo = nvme_getgeo,
2345 .report_zones = nvme_report_zones,
2346 .pr_ops = &nvme_pr_ops,
2349 #ifdef CONFIG_NVME_MULTIPATH
2350 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2352 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2354 if (!kref_get_unless_zero(&head->ref))
2359 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2361 nvme_put_ns_head(disk->private_data);
2364 const struct block_device_operations nvme_ns_head_ops = {
2365 .owner = THIS_MODULE,
2366 .submit_bio = nvme_ns_head_submit_bio,
2367 .open = nvme_ns_head_open,
2368 .release = nvme_ns_head_release,
2369 .ioctl = nvme_ioctl,
2370 .compat_ioctl = nvme_compat_ioctl,
2371 .getgeo = nvme_getgeo,
2372 .report_zones = nvme_report_zones,
2373 .pr_ops = &nvme_pr_ops,
2375 #endif /* CONFIG_NVME_MULTIPATH */
2377 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2379 unsigned long timeout =
2380 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2381 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2384 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2387 if ((csts & NVME_CSTS_RDY) == bit)
2390 usleep_range(1000, 2000);
2391 if (fatal_signal_pending(current))
2393 if (time_after(jiffies, timeout)) {
2394 dev_err(ctrl->device,
2395 "Device not ready; aborting %s, CSTS=0x%x\n",
2396 enabled ? "initialisation" : "reset", csts);
2405 * If the device has been passed off to us in an enabled state, just clear
2406 * the enabled bit. The spec says we should set the 'shutdown notification
2407 * bits', but doing so may cause the device to complete commands to the
2408 * admin queue ... and we don't know what memory that might be pointing at!
2410 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2414 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2415 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2417 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2421 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2422 msleep(NVME_QUIRK_DELAY_AMOUNT);
2424 return nvme_wait_ready(ctrl, ctrl->cap, false);
2426 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2428 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2430 unsigned dev_page_min;
2433 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2435 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2438 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2440 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2441 dev_err(ctrl->device,
2442 "Minimum device page size %u too large for host (%u)\n",
2443 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2447 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2448 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2450 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2451 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2452 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2453 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2454 ctrl->ctrl_config |= NVME_CC_ENABLE;
2456 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2459 return nvme_wait_ready(ctrl, ctrl->cap, true);
2461 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2463 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2465 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2469 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2470 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2472 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2476 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2477 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2481 if (fatal_signal_pending(current))
2483 if (time_after(jiffies, timeout)) {
2484 dev_err(ctrl->device,
2485 "Device shutdown incomplete; abort shutdown\n");
2492 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2494 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2499 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2502 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2503 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2506 dev_warn_once(ctrl->device,
2507 "could not set timestamp (%d)\n", ret);
2511 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2513 struct nvme_feat_host_behavior *host;
2516 /* Don't bother enabling the feature if retry delay is not reported */
2520 host = kzalloc(sizeof(*host), GFP_KERNEL);
2524 host->acre = NVME_ENABLE_ACRE;
2525 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2526 host, sizeof(*host), NULL);
2531 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2534 * APST (Autonomous Power State Transition) lets us program a
2535 * table of power state transitions that the controller will
2536 * perform automatically. We configure it with a simple
2537 * heuristic: we are willing to spend at most 2% of the time
2538 * transitioning between power states. Therefore, when running
2539 * in any given state, we will enter the next lower-power
2540 * non-operational state after waiting 50 * (enlat + exlat)
2541 * microseconds, as long as that state's exit latency is under
2542 * the requested maximum latency.
2544 * We will not autonomously enter any non-operational state for
2545 * which the total latency exceeds ps_max_latency_us. Users
2546 * can set ps_max_latency_us to zero to turn off APST.
2550 struct nvme_feat_auto_pst *table;
2556 * If APST isn't supported or if we haven't been initialized yet,
2557 * then don't do anything.
2562 if (ctrl->npss > 31) {
2563 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2567 table = kzalloc(sizeof(*table), GFP_KERNEL);
2571 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2572 /* Turn off APST. */
2574 dev_dbg(ctrl->device, "APST disabled\n");
2576 __le64 target = cpu_to_le64(0);
2580 * Walk through all states from lowest- to highest-power.
2581 * According to the spec, lower-numbered states use more
2582 * power. NPSS, despite the name, is the index of the
2583 * lowest-power state, not the number of states.
2585 for (state = (int)ctrl->npss; state >= 0; state--) {
2586 u64 total_latency_us, exit_latency_us, transition_ms;
2589 table->entries[state] = target;
2592 * Don't allow transitions to the deepest state
2593 * if it's quirked off.
2595 if (state == ctrl->npss &&
2596 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2600 * Is this state a useful non-operational state for
2601 * higher-power states to autonomously transition to?
2603 if (!(ctrl->psd[state].flags &
2604 NVME_PS_FLAGS_NON_OP_STATE))
2608 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2609 if (exit_latency_us > ctrl->ps_max_latency_us)
2614 le32_to_cpu(ctrl->psd[state].entry_lat);
2617 * This state is good. Use it as the APST idle
2618 * target for higher power states.
2620 transition_ms = total_latency_us + 19;
2621 do_div(transition_ms, 20);
2622 if (transition_ms > (1 << 24) - 1)
2623 transition_ms = (1 << 24) - 1;
2625 target = cpu_to_le64((state << 3) |
2626 (transition_ms << 8));
2631 if (total_latency_us > max_lat_us)
2632 max_lat_us = total_latency_us;
2638 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2640 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2641 max_ps, max_lat_us, (int)sizeof(*table), table);
2645 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2646 table, sizeof(*table), NULL);
2648 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2654 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2656 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2660 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2661 case PM_QOS_LATENCY_ANY:
2669 if (ctrl->ps_max_latency_us != latency) {
2670 ctrl->ps_max_latency_us = latency;
2671 nvme_configure_apst(ctrl);
2675 struct nvme_core_quirk_entry {
2677 * NVMe model and firmware strings are padded with spaces. For
2678 * simplicity, strings in the quirk table are padded with NULLs
2684 unsigned long quirks;
2687 static const struct nvme_core_quirk_entry core_quirks[] = {
2690 * This Toshiba device seems to die using any APST states. See:
2691 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2694 .mn = "THNSF5256GPUK TOSHIBA",
2695 .quirks = NVME_QUIRK_NO_APST,
2699 * This LiteON CL1-3D*-Q11 firmware version has a race
2700 * condition associated with actions related to suspend to idle
2701 * LiteON has resolved the problem in future firmware
2705 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2709 /* match is null-terminated but idstr is space-padded. */
2710 static bool string_matches(const char *idstr, const char *match, size_t len)
2717 matchlen = strlen(match);
2718 WARN_ON_ONCE(matchlen > len);
2720 if (memcmp(idstr, match, matchlen))
2723 for (; matchlen < len; matchlen++)
2724 if (idstr[matchlen] != ' ')
2730 static bool quirk_matches(const struct nvme_id_ctrl *id,
2731 const struct nvme_core_quirk_entry *q)
2733 return q->vid == le16_to_cpu(id->vid) &&
2734 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2735 string_matches(id->fr, q->fr, sizeof(id->fr));
2738 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2739 struct nvme_id_ctrl *id)
2744 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2745 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2746 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2747 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2751 if (ctrl->vs >= NVME_VS(1, 2, 1))
2752 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2755 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2756 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2757 "nqn.2014.08.org.nvmexpress:%04x%04x",
2758 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2759 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2760 off += sizeof(id->sn);
2761 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2762 off += sizeof(id->mn);
2763 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2766 static void nvme_release_subsystem(struct device *dev)
2768 struct nvme_subsystem *subsys =
2769 container_of(dev, struct nvme_subsystem, dev);
2771 if (subsys->instance >= 0)
2772 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2776 static void nvme_destroy_subsystem(struct kref *ref)
2778 struct nvme_subsystem *subsys =
2779 container_of(ref, struct nvme_subsystem, ref);
2781 mutex_lock(&nvme_subsystems_lock);
2782 list_del(&subsys->entry);
2783 mutex_unlock(&nvme_subsystems_lock);
2785 ida_destroy(&subsys->ns_ida);
2786 device_del(&subsys->dev);
2787 put_device(&subsys->dev);
2790 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2792 kref_put(&subsys->ref, nvme_destroy_subsystem);
2795 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2797 struct nvme_subsystem *subsys;
2799 lockdep_assert_held(&nvme_subsystems_lock);
2802 * Fail matches for discovery subsystems. This results
2803 * in each discovery controller bound to a unique subsystem.
2804 * This avoids issues with validating controller values
2805 * that can only be true when there is a single unique subsystem.
2806 * There may be multiple and completely independent entities
2807 * that provide discovery controllers.
2809 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2812 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2813 if (strcmp(subsys->subnqn, subsysnqn))
2815 if (!kref_get_unless_zero(&subsys->ref))
2823 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2824 struct device_attribute subsys_attr_##_name = \
2825 __ATTR(_name, _mode, _show, NULL)
2827 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2828 struct device_attribute *attr,
2831 struct nvme_subsystem *subsys =
2832 container_of(dev, struct nvme_subsystem, dev);
2834 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2836 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2838 #define nvme_subsys_show_str_function(field) \
2839 static ssize_t subsys_##field##_show(struct device *dev, \
2840 struct device_attribute *attr, char *buf) \
2842 struct nvme_subsystem *subsys = \
2843 container_of(dev, struct nvme_subsystem, dev); \
2844 return sprintf(buf, "%.*s\n", \
2845 (int)sizeof(subsys->field), subsys->field); \
2847 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2849 nvme_subsys_show_str_function(model);
2850 nvme_subsys_show_str_function(serial);
2851 nvme_subsys_show_str_function(firmware_rev);
2853 static struct attribute *nvme_subsys_attrs[] = {
2854 &subsys_attr_model.attr,
2855 &subsys_attr_serial.attr,
2856 &subsys_attr_firmware_rev.attr,
2857 &subsys_attr_subsysnqn.attr,
2858 #ifdef CONFIG_NVME_MULTIPATH
2859 &subsys_attr_iopolicy.attr,
2864 static struct attribute_group nvme_subsys_attrs_group = {
2865 .attrs = nvme_subsys_attrs,
2868 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2869 &nvme_subsys_attrs_group,
2873 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2875 return ctrl->opts && ctrl->opts->discovery_nqn;
2878 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2879 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2881 struct nvme_ctrl *tmp;
2883 lockdep_assert_held(&nvme_subsystems_lock);
2885 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2886 if (nvme_state_terminal(tmp))
2889 if (tmp->cntlid == ctrl->cntlid) {
2890 dev_err(ctrl->device,
2891 "Duplicate cntlid %u with %s, rejecting\n",
2892 ctrl->cntlid, dev_name(tmp->device));
2896 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2897 nvme_discovery_ctrl(ctrl))
2900 dev_err(ctrl->device,
2901 "Subsystem does not support multiple controllers\n");
2908 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2910 struct nvme_subsystem *subsys, *found;
2913 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2917 subsys->instance = -1;
2918 mutex_init(&subsys->lock);
2919 kref_init(&subsys->ref);
2920 INIT_LIST_HEAD(&subsys->ctrls);
2921 INIT_LIST_HEAD(&subsys->nsheads);
2922 nvme_init_subnqn(subsys, ctrl, id);
2923 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2924 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2925 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2926 subsys->vendor_id = le16_to_cpu(id->vid);
2927 subsys->cmic = id->cmic;
2928 subsys->awupf = le16_to_cpu(id->awupf);
2929 #ifdef CONFIG_NVME_MULTIPATH
2930 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2933 subsys->dev.class = nvme_subsys_class;
2934 subsys->dev.release = nvme_release_subsystem;
2935 subsys->dev.groups = nvme_subsys_attrs_groups;
2936 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2937 device_initialize(&subsys->dev);
2939 mutex_lock(&nvme_subsystems_lock);
2940 found = __nvme_find_get_subsystem(subsys->subnqn);
2942 put_device(&subsys->dev);
2945 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2947 goto out_put_subsystem;
2950 ret = device_add(&subsys->dev);
2952 dev_err(ctrl->device,
2953 "failed to register subsystem device.\n");
2954 put_device(&subsys->dev);
2957 ida_init(&subsys->ns_ida);
2958 list_add_tail(&subsys->entry, &nvme_subsystems);
2961 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2962 dev_name(ctrl->device));
2964 dev_err(ctrl->device,
2965 "failed to create sysfs link from subsystem.\n");
2966 goto out_put_subsystem;
2970 subsys->instance = ctrl->instance;
2971 ctrl->subsys = subsys;
2972 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2973 mutex_unlock(&nvme_subsystems_lock);
2977 nvme_put_subsystem(subsys);
2979 mutex_unlock(&nvme_subsystems_lock);
2983 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2984 void *log, size_t size, u64 offset)
2986 struct nvme_command c = { };
2987 u32 dwlen = nvme_bytes_to_numd(size);
2989 c.get_log_page.opcode = nvme_admin_get_log_page;
2990 c.get_log_page.nsid = cpu_to_le32(nsid);
2991 c.get_log_page.lid = log_page;
2992 c.get_log_page.lsp = lsp;
2993 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2994 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2995 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2996 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2997 c.get_log_page.csi = csi;
2999 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3002 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3003 struct nvme_effects_log **log)
3005 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3011 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3015 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3016 cel, sizeof(*cel), 0);
3022 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3029 * Initialize the cached copies of the Identify data and various controller
3030 * register in our nvme_ctrl structure. This should be called as soon as
3031 * the admin queue is fully up and running.
3033 int nvme_init_identify(struct nvme_ctrl *ctrl)
3035 struct nvme_id_ctrl *id;
3036 int ret, page_shift;
3038 bool prev_apst_enabled;
3040 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3042 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3045 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3046 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3048 if (ctrl->vs >= NVME_VS(1, 1, 0))
3049 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3051 ret = nvme_identify_ctrl(ctrl, &id);
3053 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3057 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3058 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3063 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3064 ctrl->cntlid = le16_to_cpu(id->cntlid);
3066 if (!ctrl->identified) {
3069 ret = nvme_init_subsystem(ctrl, id);
3074 * Check for quirks. Quirk can depend on firmware version,
3075 * so, in principle, the set of quirks present can change
3076 * across a reset. As a possible future enhancement, we
3077 * could re-scan for quirks every time we reinitialize
3078 * the device, but we'd have to make sure that the driver
3079 * behaves intelligently if the quirks change.
3081 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3082 if (quirk_matches(id, &core_quirks[i]))
3083 ctrl->quirks |= core_quirks[i].quirks;
3087 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3088 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3089 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3092 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3093 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3094 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3096 ctrl->oacs = le16_to_cpu(id->oacs);
3097 ctrl->oncs = le16_to_cpu(id->oncs);
3098 ctrl->mtfa = le16_to_cpu(id->mtfa);
3099 ctrl->oaes = le32_to_cpu(id->oaes);
3100 ctrl->wctemp = le16_to_cpu(id->wctemp);
3101 ctrl->cctemp = le16_to_cpu(id->cctemp);
3103 atomic_set(&ctrl->abort_limit, id->acl + 1);
3104 ctrl->vwc = id->vwc;
3106 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3108 max_hw_sectors = UINT_MAX;
3109 ctrl->max_hw_sectors =
3110 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3112 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3113 ctrl->sgls = le32_to_cpu(id->sgls);
3114 ctrl->kas = le16_to_cpu(id->kas);
3115 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3116 ctrl->ctratt = le32_to_cpu(id->ctratt);
3120 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3122 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3123 shutdown_timeout, 60);
3125 if (ctrl->shutdown_timeout != shutdown_timeout)
3126 dev_info(ctrl->device,
3127 "Shutdown timeout set to %u seconds\n",
3128 ctrl->shutdown_timeout);
3130 ctrl->shutdown_timeout = shutdown_timeout;
3132 ctrl->npss = id->npss;
3133 ctrl->apsta = id->apsta;
3134 prev_apst_enabled = ctrl->apst_enabled;
3135 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3136 if (force_apst && id->apsta) {
3137 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3138 ctrl->apst_enabled = true;
3140 ctrl->apst_enabled = false;
3143 ctrl->apst_enabled = id->apsta;
3145 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3147 if (ctrl->ops->flags & NVME_F_FABRICS) {
3148 ctrl->icdoff = le16_to_cpu(id->icdoff);
3149 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3150 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3151 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3154 * In fabrics we need to verify the cntlid matches the
3157 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3158 dev_err(ctrl->device,
3159 "Mismatching cntlid: Connect %u vs Identify "
3161 ctrl->cntlid, le16_to_cpu(id->cntlid));
3166 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3167 dev_err(ctrl->device,
3168 "keep-alive support is mandatory for fabrics\n");
3173 ctrl->hmpre = le32_to_cpu(id->hmpre);
3174 ctrl->hmmin = le32_to_cpu(id->hmmin);
3175 ctrl->hmminds = le32_to_cpu(id->hmminds);
3176 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3179 ret = nvme_mpath_init(ctrl, id);
3185 if (ctrl->apst_enabled && !prev_apst_enabled)
3186 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3187 else if (!ctrl->apst_enabled && prev_apst_enabled)
3188 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3190 ret = nvme_configure_apst(ctrl);
3194 ret = nvme_configure_timestamp(ctrl);
3198 ret = nvme_configure_directives(ctrl);
3202 ret = nvme_configure_acre(ctrl);
3206 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3207 ret = nvme_hwmon_init(ctrl);
3212 ctrl->identified = true;
3220 EXPORT_SYMBOL_GPL(nvme_init_identify);
3222 static int nvme_dev_open(struct inode *inode, struct file *file)
3224 struct nvme_ctrl *ctrl =
3225 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3227 switch (ctrl->state) {
3228 case NVME_CTRL_LIVE:
3231 return -EWOULDBLOCK;
3234 nvme_get_ctrl(ctrl);
3235 if (!try_module_get(ctrl->ops->module)) {
3236 nvme_put_ctrl(ctrl);
3240 file->private_data = ctrl;
3244 static int nvme_dev_release(struct inode *inode, struct file *file)
3246 struct nvme_ctrl *ctrl =
3247 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3249 module_put(ctrl->ops->module);
3250 nvme_put_ctrl(ctrl);
3254 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3259 down_read(&ctrl->namespaces_rwsem);
3260 if (list_empty(&ctrl->namespaces)) {
3265 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3266 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3267 dev_warn(ctrl->device,
3268 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3273 dev_warn(ctrl->device,
3274 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3275 kref_get(&ns->kref);
3276 up_read(&ctrl->namespaces_rwsem);
3278 ret = nvme_user_cmd(ctrl, ns, argp);
3283 up_read(&ctrl->namespaces_rwsem);
3287 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3290 struct nvme_ctrl *ctrl = file->private_data;
3291 void __user *argp = (void __user *)arg;
3294 case NVME_IOCTL_ADMIN_CMD:
3295 return nvme_user_cmd(ctrl, NULL, argp);
3296 case NVME_IOCTL_ADMIN64_CMD:
3297 return nvme_user_cmd64(ctrl, NULL, argp);
3298 case NVME_IOCTL_IO_CMD:
3299 return nvme_dev_user_cmd(ctrl, argp);
3300 case NVME_IOCTL_RESET:
3301 dev_warn(ctrl->device, "resetting controller\n");
3302 return nvme_reset_ctrl_sync(ctrl);
3303 case NVME_IOCTL_SUBSYS_RESET:
3304 return nvme_reset_subsystem(ctrl);
3305 case NVME_IOCTL_RESCAN:
3306 nvme_queue_scan(ctrl);
3313 static const struct file_operations nvme_dev_fops = {
3314 .owner = THIS_MODULE,
3315 .open = nvme_dev_open,
3316 .release = nvme_dev_release,
3317 .unlocked_ioctl = nvme_dev_ioctl,
3318 .compat_ioctl = compat_ptr_ioctl,
3321 static ssize_t nvme_sysfs_reset(struct device *dev,
3322 struct device_attribute *attr, const char *buf,
3325 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3328 ret = nvme_reset_ctrl_sync(ctrl);
3333 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3335 static ssize_t nvme_sysfs_rescan(struct device *dev,
3336 struct device_attribute *attr, const char *buf,
3339 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3341 nvme_queue_scan(ctrl);
3344 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3346 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3348 struct gendisk *disk = dev_to_disk(dev);
3350 if (disk->fops == &nvme_bdev_ops)
3351 return nvme_get_ns_from_dev(dev)->head;
3353 return disk->private_data;
3356 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3359 struct nvme_ns_head *head = dev_to_ns_head(dev);
3360 struct nvme_ns_ids *ids = &head->ids;
3361 struct nvme_subsystem *subsys = head->subsys;
3362 int serial_len = sizeof(subsys->serial);
3363 int model_len = sizeof(subsys->model);
3365 if (!uuid_is_null(&ids->uuid))
3366 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3368 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3369 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3371 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3372 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3374 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3375 subsys->serial[serial_len - 1] == '\0'))
3377 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3378 subsys->model[model_len - 1] == '\0'))
3381 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3382 serial_len, subsys->serial, model_len, subsys->model,
3385 static DEVICE_ATTR_RO(wwid);
3387 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3390 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3392 static DEVICE_ATTR_RO(nguid);
3394 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3397 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3399 /* For backward compatibility expose the NGUID to userspace if
3400 * we have no UUID set
3402 if (uuid_is_null(&ids->uuid)) {
3403 printk_ratelimited(KERN_WARNING
3404 "No UUID available providing old NGUID\n");
3405 return sprintf(buf, "%pU\n", ids->nguid);
3407 return sprintf(buf, "%pU\n", &ids->uuid);
3409 static DEVICE_ATTR_RO(uuid);
3411 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3414 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3416 static DEVICE_ATTR_RO(eui);
3418 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3421 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3423 static DEVICE_ATTR_RO(nsid);
3425 static struct attribute *nvme_ns_id_attrs[] = {
3426 &dev_attr_wwid.attr,
3427 &dev_attr_uuid.attr,
3428 &dev_attr_nguid.attr,
3430 &dev_attr_nsid.attr,
3431 #ifdef CONFIG_NVME_MULTIPATH
3432 &dev_attr_ana_grpid.attr,
3433 &dev_attr_ana_state.attr,
3438 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3439 struct attribute *a, int n)
3441 struct device *dev = container_of(kobj, struct device, kobj);
3442 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3444 if (a == &dev_attr_uuid.attr) {
3445 if (uuid_is_null(&ids->uuid) &&
3446 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3449 if (a == &dev_attr_nguid.attr) {
3450 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3453 if (a == &dev_attr_eui.attr) {
3454 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3457 #ifdef CONFIG_NVME_MULTIPATH
3458 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3459 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3461 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3468 static const struct attribute_group nvme_ns_id_attr_group = {
3469 .attrs = nvme_ns_id_attrs,
3470 .is_visible = nvme_ns_id_attrs_are_visible,
3473 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3474 &nvme_ns_id_attr_group,
3476 &nvme_nvm_attr_group,
3481 #define nvme_show_str_function(field) \
3482 static ssize_t field##_show(struct device *dev, \
3483 struct device_attribute *attr, char *buf) \
3485 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3486 return sprintf(buf, "%.*s\n", \
3487 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3489 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3491 nvme_show_str_function(model);
3492 nvme_show_str_function(serial);
3493 nvme_show_str_function(firmware_rev);
3495 #define nvme_show_int_function(field) \
3496 static ssize_t field##_show(struct device *dev, \
3497 struct device_attribute *attr, char *buf) \
3499 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3500 return sprintf(buf, "%d\n", ctrl->field); \
3502 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3504 nvme_show_int_function(cntlid);
3505 nvme_show_int_function(numa_node);
3506 nvme_show_int_function(queue_count);
3507 nvme_show_int_function(sqsize);
3509 static ssize_t nvme_sysfs_delete(struct device *dev,
3510 struct device_attribute *attr, const char *buf,
3513 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3515 if (device_remove_file_self(dev, attr))
3516 nvme_delete_ctrl_sync(ctrl);
3519 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3521 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3522 struct device_attribute *attr,
3525 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3527 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3529 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3531 static ssize_t nvme_sysfs_show_state(struct device *dev,
3532 struct device_attribute *attr,
3535 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3536 static const char *const state_name[] = {
3537 [NVME_CTRL_NEW] = "new",
3538 [NVME_CTRL_LIVE] = "live",
3539 [NVME_CTRL_RESETTING] = "resetting",
3540 [NVME_CTRL_CONNECTING] = "connecting",
3541 [NVME_CTRL_DELETING] = "deleting",
3542 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3543 [NVME_CTRL_DEAD] = "dead",
3546 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3547 state_name[ctrl->state])
3548 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3550 return sprintf(buf, "unknown state\n");
3553 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3555 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3556 struct device_attribute *attr,
3559 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3561 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3563 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3565 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3566 struct device_attribute *attr,
3569 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3571 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3573 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3575 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3576 struct device_attribute *attr,
3579 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3581 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3583 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3585 static ssize_t nvme_sysfs_show_address(struct device *dev,
3586 struct device_attribute *attr,
3589 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3591 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3593 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3595 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3596 struct device_attribute *attr, char *buf)
3598 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3599 struct nvmf_ctrl_options *opts = ctrl->opts;
3601 if (ctrl->opts->max_reconnects == -1)
3602 return sprintf(buf, "off\n");
3603 return sprintf(buf, "%d\n",
3604 opts->max_reconnects * opts->reconnect_delay);
3607 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3608 struct device_attribute *attr, const char *buf, size_t count)
3610 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3611 struct nvmf_ctrl_options *opts = ctrl->opts;
3612 int ctrl_loss_tmo, err;
3614 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3618 else if (ctrl_loss_tmo < 0)
3619 opts->max_reconnects = -1;
3621 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3622 opts->reconnect_delay);
3625 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3626 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3628 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3629 struct device_attribute *attr, char *buf)
3631 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3633 if (ctrl->opts->reconnect_delay == -1)
3634 return sprintf(buf, "off\n");
3635 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3638 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3639 struct device_attribute *attr, const char *buf, size_t count)
3641 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3645 err = kstrtou32(buf, 10, &v);
3649 ctrl->opts->reconnect_delay = v;
3652 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3653 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3655 static struct attribute *nvme_dev_attrs[] = {
3656 &dev_attr_reset_controller.attr,
3657 &dev_attr_rescan_controller.attr,
3658 &dev_attr_model.attr,
3659 &dev_attr_serial.attr,
3660 &dev_attr_firmware_rev.attr,
3661 &dev_attr_cntlid.attr,
3662 &dev_attr_delete_controller.attr,
3663 &dev_attr_transport.attr,
3664 &dev_attr_subsysnqn.attr,
3665 &dev_attr_address.attr,
3666 &dev_attr_state.attr,
3667 &dev_attr_numa_node.attr,
3668 &dev_attr_queue_count.attr,
3669 &dev_attr_sqsize.attr,
3670 &dev_attr_hostnqn.attr,
3671 &dev_attr_hostid.attr,
3672 &dev_attr_ctrl_loss_tmo.attr,
3673 &dev_attr_reconnect_delay.attr,
3677 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3678 struct attribute *a, int n)
3680 struct device *dev = container_of(kobj, struct device, kobj);
3681 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3683 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3685 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3687 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3689 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3691 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3693 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3699 static struct attribute_group nvme_dev_attrs_group = {
3700 .attrs = nvme_dev_attrs,
3701 .is_visible = nvme_dev_attrs_are_visible,
3704 static const struct attribute_group *nvme_dev_attr_groups[] = {
3705 &nvme_dev_attrs_group,
3709 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3712 struct nvme_ns_head *h;
3714 lockdep_assert_held(&subsys->lock);
3716 list_for_each_entry(h, &subsys->nsheads, entry) {
3717 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3724 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3725 struct nvme_ns_head *new)
3727 struct nvme_ns_head *h;
3729 lockdep_assert_held(&subsys->lock);
3731 list_for_each_entry(h, &subsys->nsheads, entry) {
3732 if (nvme_ns_ids_valid(&new->ids) &&
3733 nvme_ns_ids_equal(&new->ids, &h->ids))
3740 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3741 unsigned nsid, struct nvme_ns_ids *ids)
3743 struct nvme_ns_head *head;
3744 size_t size = sizeof(*head);
3747 #ifdef CONFIG_NVME_MULTIPATH
3748 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3751 head = kzalloc(size, GFP_KERNEL);
3754 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3757 head->instance = ret;
3758 INIT_LIST_HEAD(&head->list);
3759 ret = init_srcu_struct(&head->srcu);
3761 goto out_ida_remove;
3762 head->subsys = ctrl->subsys;
3765 kref_init(&head->ref);
3767 ret = __nvme_check_ids(ctrl->subsys, head);
3769 dev_err(ctrl->device,
3770 "duplicate IDs for nsid %d\n", nsid);
3771 goto out_cleanup_srcu;
3774 if (head->ids.csi) {
3775 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3777 goto out_cleanup_srcu;
3779 head->effects = ctrl->effects;
3781 ret = nvme_mpath_alloc_disk(ctrl, head);
3783 goto out_cleanup_srcu;
3785 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3787 kref_get(&ctrl->subsys->ref);
3791 cleanup_srcu_struct(&head->srcu);
3793 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3798 ret = blk_status_to_errno(nvme_error_status(ret));
3799 return ERR_PTR(ret);
3802 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3803 struct nvme_ns_ids *ids, bool is_shared)
3805 struct nvme_ctrl *ctrl = ns->ctrl;
3806 struct nvme_ns_head *head = NULL;
3809 mutex_lock(&ctrl->subsys->lock);
3810 head = nvme_find_ns_head(ctrl->subsys, nsid);
3812 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3814 ret = PTR_ERR(head);
3817 head->shared = is_shared;
3820 if (!is_shared || !head->shared) {
3821 dev_err(ctrl->device,
3822 "Duplicate unshared namespace %d\n", nsid);
3823 goto out_put_ns_head;
3825 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3826 dev_err(ctrl->device,
3827 "IDs don't match for shared namespace %d\n",
3829 goto out_put_ns_head;
3833 list_add_tail_rcu(&ns->siblings, &head->list);
3835 mutex_unlock(&ctrl->subsys->lock);
3839 nvme_put_ns_head(head);
3841 mutex_unlock(&ctrl->subsys->lock);
3845 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3847 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3848 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3850 return nsa->head->ns_id - nsb->head->ns_id;
3853 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3855 struct nvme_ns *ns, *ret = NULL;
3857 down_read(&ctrl->namespaces_rwsem);
3858 list_for_each_entry(ns, &ctrl->namespaces, list) {
3859 if (ns->head->ns_id == nsid) {
3860 if (!kref_get_unless_zero(&ns->kref))
3865 if (ns->head->ns_id > nsid)
3868 up_read(&ctrl->namespaces_rwsem);
3871 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3873 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3874 struct nvme_ns_ids *ids)
3877 struct gendisk *disk;
3878 struct nvme_id_ns *id;
3879 char disk_name[DISK_NAME_LEN];
3880 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3882 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3885 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3889 ns->queue = blk_mq_init_queue(ctrl->tagset);
3890 if (IS_ERR(ns->queue))
3893 if (ctrl->opts && ctrl->opts->data_digest)
3894 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3896 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3897 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3898 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3900 ns->queue->queuedata = ns;
3902 kref_init(&ns->kref);
3904 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3905 goto out_free_queue;
3906 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3908 disk = alloc_disk_node(0, node);
3912 disk->fops = &nvme_bdev_ops;
3913 disk->private_data = ns;
3914 disk->queue = ns->queue;
3915 disk->flags = flags;
3916 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3919 if (nvme_update_ns_info(ns, id))
3922 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3923 if (nvme_nvm_register(ns, disk_name, node)) {
3924 dev_warn(ctrl->device, "LightNVM init failure\n");
3929 down_write(&ctrl->namespaces_rwsem);
3930 list_add_tail(&ns->list, &ctrl->namespaces);
3931 up_write(&ctrl->namespaces_rwsem);
3933 nvme_get_ctrl(ctrl);
3935 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3937 nvme_mpath_add_disk(ns, id);
3938 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3943 /* prevent double queue cleanup */
3944 ns->disk->queue = NULL;
3947 mutex_lock(&ctrl->subsys->lock);
3948 list_del_rcu(&ns->siblings);
3949 if (list_empty(&ns->head->list))
3950 list_del_init(&ns->head->entry);
3951 mutex_unlock(&ctrl->subsys->lock);
3952 nvme_put_ns_head(ns->head);
3954 blk_cleanup_queue(ns->queue);
3961 static void nvme_ns_remove(struct nvme_ns *ns)
3963 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3966 set_capacity(ns->disk, 0);
3967 nvme_fault_inject_fini(&ns->fault_inject);
3969 mutex_lock(&ns->ctrl->subsys->lock);
3970 list_del_rcu(&ns->siblings);
3971 if (list_empty(&ns->head->list))
3972 list_del_init(&ns->head->entry);
3973 mutex_unlock(&ns->ctrl->subsys->lock);
3975 synchronize_rcu(); /* guarantee not available in head->list */
3976 nvme_mpath_clear_current_path(ns);
3977 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3979 if (ns->disk->flags & GENHD_FL_UP) {
3980 del_gendisk(ns->disk);
3981 blk_cleanup_queue(ns->queue);
3982 if (blk_get_integrity(ns->disk))
3983 blk_integrity_unregister(ns->disk);
3986 down_write(&ns->ctrl->namespaces_rwsem);
3987 list_del_init(&ns->list);
3988 up_write(&ns->ctrl->namespaces_rwsem);
3990 nvme_mpath_check_last_path(ns);
3994 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3996 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4004 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4006 struct nvme_id_ns *id;
4009 if (test_bit(NVME_NS_DEAD, &ns->flags))
4012 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4017 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4018 dev_err(ns->ctrl->device,
4019 "identifiers changed for nsid %d\n", ns->head->ns_id);
4023 ret = nvme_update_ns_info(ns, id);
4029 * Only remove the namespace if we got a fatal error back from the
4030 * device, otherwise ignore the error and just move on.
4032 * TODO: we should probably schedule a delayed retry here.
4034 if (ret && ret != -ENOMEM && !(ret > 0 && !(ret & NVME_SC_DNR)))
4038 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4040 struct nvme_ns_ids ids = { };
4043 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4046 ns = nvme_find_get_ns(ctrl, nsid);
4048 nvme_validate_ns(ns, &ids);
4055 nvme_alloc_ns(ctrl, nsid, &ids);
4058 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4059 dev_warn(ctrl->device,
4060 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4064 nvme_alloc_ns(ctrl, nsid, &ids);
4067 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4073 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4076 struct nvme_ns *ns, *next;
4079 down_write(&ctrl->namespaces_rwsem);
4080 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4081 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4082 list_move_tail(&ns->list, &rm_list);
4084 up_write(&ctrl->namespaces_rwsem);
4086 list_for_each_entry_safe(ns, next, &rm_list, list)
4091 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4093 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4098 if (nvme_ctrl_limited_cns(ctrl))
4101 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4106 struct nvme_command cmd = {
4107 .identify.opcode = nvme_admin_identify,
4108 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4109 .identify.nsid = cpu_to_le32(prev),
4112 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4113 NVME_IDENTIFY_DATA_SIZE);
4115 dev_warn(ctrl->device,
4116 "Identify NS List failed (status=0x%x)\n", ret);
4120 for (i = 0; i < nr_entries; i++) {
4121 u32 nsid = le32_to_cpu(ns_list[i]);
4123 if (!nsid) /* end of the list? */
4125 nvme_validate_or_alloc_ns(ctrl, nsid);
4126 while (++prev < nsid)
4127 nvme_ns_remove_by_nsid(ctrl, prev);
4131 nvme_remove_invalid_namespaces(ctrl, prev);
4137 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4139 struct nvme_id_ctrl *id;
4142 if (nvme_identify_ctrl(ctrl, &id))
4144 nn = le32_to_cpu(id->nn);
4147 for (i = 1; i <= nn; i++)
4148 nvme_validate_or_alloc_ns(ctrl, i);
4150 nvme_remove_invalid_namespaces(ctrl, nn);
4153 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4155 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4159 log = kzalloc(log_size, GFP_KERNEL);
4164 * We need to read the log to clear the AEN, but we don't want to rely
4165 * on it for the changed namespace information as userspace could have
4166 * raced with us in reading the log page, which could cause us to miss
4169 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4170 NVME_CSI_NVM, log, log_size, 0);
4172 dev_warn(ctrl->device,
4173 "reading changed ns log failed: %d\n", error);
4178 static void nvme_scan_work(struct work_struct *work)
4180 struct nvme_ctrl *ctrl =
4181 container_of(work, struct nvme_ctrl, scan_work);
4183 /* No tagset on a live ctrl means IO queues could not created */
4184 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4187 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4188 dev_info(ctrl->device, "rescanning namespaces.\n");
4189 nvme_clear_changed_ns_log(ctrl);
4192 mutex_lock(&ctrl->scan_lock);
4193 if (nvme_scan_ns_list(ctrl) != 0)
4194 nvme_scan_ns_sequential(ctrl);
4195 mutex_unlock(&ctrl->scan_lock);
4197 down_write(&ctrl->namespaces_rwsem);
4198 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4199 up_write(&ctrl->namespaces_rwsem);
4203 * This function iterates the namespace list unlocked to allow recovery from
4204 * controller failure. It is up to the caller to ensure the namespace list is
4205 * not modified by scan work while this function is executing.
4207 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4209 struct nvme_ns *ns, *next;
4213 * make sure to requeue I/O to all namespaces as these
4214 * might result from the scan itself and must complete
4215 * for the scan_work to make progress
4217 nvme_mpath_clear_ctrl_paths(ctrl);
4219 /* prevent racing with ns scanning */
4220 flush_work(&ctrl->scan_work);
4223 * The dead states indicates the controller was not gracefully
4224 * disconnected. In that case, we won't be able to flush any data while
4225 * removing the namespaces' disks; fail all the queues now to avoid
4226 * potentially having to clean up the failed sync later.
4228 if (ctrl->state == NVME_CTRL_DEAD)
4229 nvme_kill_queues(ctrl);
4231 /* this is a no-op when called from the controller reset handler */
4232 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4234 down_write(&ctrl->namespaces_rwsem);
4235 list_splice_init(&ctrl->namespaces, &ns_list);
4236 up_write(&ctrl->namespaces_rwsem);
4238 list_for_each_entry_safe(ns, next, &ns_list, list)
4241 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4243 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4245 struct nvme_ctrl *ctrl =
4246 container_of(dev, struct nvme_ctrl, ctrl_device);
4247 struct nvmf_ctrl_options *opts = ctrl->opts;
4250 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4255 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4259 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4260 opts->trsvcid ?: "none");
4264 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4265 opts->host_traddr ?: "none");
4270 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4272 char *envp[2] = { NULL, NULL };
4273 u32 aen_result = ctrl->aen_result;
4275 ctrl->aen_result = 0;
4279 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4282 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4286 static void nvme_async_event_work(struct work_struct *work)
4288 struct nvme_ctrl *ctrl =
4289 container_of(work, struct nvme_ctrl, async_event_work);
4291 nvme_aen_uevent(ctrl);
4292 ctrl->ops->submit_async_event(ctrl);
4295 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4300 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4306 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4309 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4311 struct nvme_fw_slot_info_log *log;
4313 log = kmalloc(sizeof(*log), GFP_KERNEL);
4317 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4318 log, sizeof(*log), 0))
4319 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4323 static void nvme_fw_act_work(struct work_struct *work)
4325 struct nvme_ctrl *ctrl = container_of(work,
4326 struct nvme_ctrl, fw_act_work);
4327 unsigned long fw_act_timeout;
4330 fw_act_timeout = jiffies +
4331 msecs_to_jiffies(ctrl->mtfa * 100);
4333 fw_act_timeout = jiffies +
4334 msecs_to_jiffies(admin_timeout * 1000);
4336 nvme_stop_queues(ctrl);
4337 while (nvme_ctrl_pp_status(ctrl)) {
4338 if (time_after(jiffies, fw_act_timeout)) {
4339 dev_warn(ctrl->device,
4340 "Fw activation timeout, reset controller\n");
4341 nvme_try_sched_reset(ctrl);
4347 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4350 nvme_start_queues(ctrl);
4351 /* read FW slot information to clear the AER */
4352 nvme_get_fw_slot_info(ctrl);
4355 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4357 u32 aer_notice_type = (result & 0xff00) >> 8;
4359 trace_nvme_async_event(ctrl, aer_notice_type);
4361 switch (aer_notice_type) {
4362 case NVME_AER_NOTICE_NS_CHANGED:
4363 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4364 nvme_queue_scan(ctrl);
4366 case NVME_AER_NOTICE_FW_ACT_STARTING:
4368 * We are (ab)using the RESETTING state to prevent subsequent
4369 * recovery actions from interfering with the controller's
4370 * firmware activation.
4372 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4373 queue_work(nvme_wq, &ctrl->fw_act_work);
4375 #ifdef CONFIG_NVME_MULTIPATH
4376 case NVME_AER_NOTICE_ANA:
4377 if (!ctrl->ana_log_buf)
4379 queue_work(nvme_wq, &ctrl->ana_work);
4382 case NVME_AER_NOTICE_DISC_CHANGED:
4383 ctrl->aen_result = result;
4386 dev_warn(ctrl->device, "async event result %08x\n", result);
4390 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4391 volatile union nvme_result *res)
4393 u32 result = le32_to_cpu(res->u32);
4394 u32 aer_type = result & 0x07;
4396 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4400 case NVME_AER_NOTICE:
4401 nvme_handle_aen_notice(ctrl, result);
4403 case NVME_AER_ERROR:
4404 case NVME_AER_SMART:
4407 trace_nvme_async_event(ctrl, aer_type);
4408 ctrl->aen_result = result;
4413 queue_work(nvme_wq, &ctrl->async_event_work);
4415 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4417 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4419 nvme_mpath_stop(ctrl);
4420 nvme_stop_keep_alive(ctrl);
4421 nvme_stop_failfast_work(ctrl);
4422 flush_work(&ctrl->async_event_work);
4423 cancel_work_sync(&ctrl->fw_act_work);
4425 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4427 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4429 nvme_start_keep_alive(ctrl);
4431 nvme_enable_aen(ctrl);
4433 if (ctrl->queue_count > 1) {
4434 nvme_queue_scan(ctrl);
4435 nvme_start_queues(ctrl);
4438 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4440 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4442 nvme_fault_inject_fini(&ctrl->fault_inject);
4443 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4444 cdev_device_del(&ctrl->cdev, ctrl->device);
4445 nvme_put_ctrl(ctrl);
4447 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4449 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4451 struct nvme_effects_log *cel;
4454 xa_for_each (&ctrl->cels, i, cel) {
4455 xa_erase(&ctrl->cels, i);
4459 xa_destroy(&ctrl->cels);
4462 static void nvme_free_ctrl(struct device *dev)
4464 struct nvme_ctrl *ctrl =
4465 container_of(dev, struct nvme_ctrl, ctrl_device);
4466 struct nvme_subsystem *subsys = ctrl->subsys;
4468 if (!subsys || ctrl->instance != subsys->instance)
4469 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4471 nvme_free_cels(ctrl);
4472 nvme_mpath_uninit(ctrl);
4473 __free_page(ctrl->discard_page);
4476 mutex_lock(&nvme_subsystems_lock);
4477 list_del(&ctrl->subsys_entry);
4478 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4479 mutex_unlock(&nvme_subsystems_lock);
4482 ctrl->ops->free_ctrl(ctrl);
4485 nvme_put_subsystem(subsys);
4489 * Initialize a NVMe controller structures. This needs to be called during
4490 * earliest initialization so that we have the initialized structured around
4493 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4494 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4498 ctrl->state = NVME_CTRL_NEW;
4499 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4500 spin_lock_init(&ctrl->lock);
4501 mutex_init(&ctrl->scan_lock);
4502 INIT_LIST_HEAD(&ctrl->namespaces);
4503 xa_init(&ctrl->cels);
4504 init_rwsem(&ctrl->namespaces_rwsem);
4507 ctrl->quirks = quirks;
4508 ctrl->numa_node = NUMA_NO_NODE;
4509 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4510 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4511 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4512 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4513 init_waitqueue_head(&ctrl->state_wq);
4515 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4516 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4517 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4518 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4520 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4522 ctrl->discard_page = alloc_page(GFP_KERNEL);
4523 if (!ctrl->discard_page) {
4528 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4531 ctrl->instance = ret;
4533 device_initialize(&ctrl->ctrl_device);
4534 ctrl->device = &ctrl->ctrl_device;
4535 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4537 ctrl->device->class = nvme_class;
4538 ctrl->device->parent = ctrl->dev;
4539 ctrl->device->groups = nvme_dev_attr_groups;
4540 ctrl->device->release = nvme_free_ctrl;
4541 dev_set_drvdata(ctrl->device, ctrl);
4542 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4544 goto out_release_instance;
4546 nvme_get_ctrl(ctrl);
4547 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4548 ctrl->cdev.owner = ops->module;
4549 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4554 * Initialize latency tolerance controls. The sysfs files won't
4555 * be visible to userspace unless the device actually supports APST.
4557 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4558 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4559 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4561 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4565 nvme_put_ctrl(ctrl);
4566 kfree_const(ctrl->device->kobj.name);
4567 out_release_instance:
4568 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4570 if (ctrl->discard_page)
4571 __free_page(ctrl->discard_page);
4574 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4577 * nvme_kill_queues(): Ends all namespace queues
4578 * @ctrl: the dead controller that needs to end
4580 * Call this function when the driver determines it is unable to get the
4581 * controller in a state capable of servicing IO.
4583 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4587 down_read(&ctrl->namespaces_rwsem);
4589 /* Forcibly unquiesce queues to avoid blocking dispatch */
4590 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4591 blk_mq_unquiesce_queue(ctrl->admin_q);
4593 list_for_each_entry(ns, &ctrl->namespaces, list)
4594 nvme_set_queue_dying(ns);
4596 up_read(&ctrl->namespaces_rwsem);
4598 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4600 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4604 down_read(&ctrl->namespaces_rwsem);
4605 list_for_each_entry(ns, &ctrl->namespaces, list)
4606 blk_mq_unfreeze_queue(ns->queue);
4607 up_read(&ctrl->namespaces_rwsem);
4609 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4611 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4615 down_read(&ctrl->namespaces_rwsem);
4616 list_for_each_entry(ns, &ctrl->namespaces, list) {
4617 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4621 up_read(&ctrl->namespaces_rwsem);
4624 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4626 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4630 down_read(&ctrl->namespaces_rwsem);
4631 list_for_each_entry(ns, &ctrl->namespaces, list)
4632 blk_mq_freeze_queue_wait(ns->queue);
4633 up_read(&ctrl->namespaces_rwsem);
4635 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4637 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4641 down_read(&ctrl->namespaces_rwsem);
4642 list_for_each_entry(ns, &ctrl->namespaces, list)
4643 blk_freeze_queue_start(ns->queue);
4644 up_read(&ctrl->namespaces_rwsem);
4646 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4648 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4652 down_read(&ctrl->namespaces_rwsem);
4653 list_for_each_entry(ns, &ctrl->namespaces, list)
4654 blk_mq_quiesce_queue(ns->queue);
4655 up_read(&ctrl->namespaces_rwsem);
4657 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4659 void nvme_start_queues(struct nvme_ctrl *ctrl)
4663 down_read(&ctrl->namespaces_rwsem);
4664 list_for_each_entry(ns, &ctrl->namespaces, list)
4665 blk_mq_unquiesce_queue(ns->queue);
4666 up_read(&ctrl->namespaces_rwsem);
4668 EXPORT_SYMBOL_GPL(nvme_start_queues);
4670 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4674 down_read(&ctrl->namespaces_rwsem);
4675 list_for_each_entry(ns, &ctrl->namespaces, list)
4676 blk_sync_queue(ns->queue);
4677 up_read(&ctrl->namespaces_rwsem);
4679 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4681 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4683 nvme_sync_io_queues(ctrl);
4685 blk_sync_queue(ctrl->admin_q);
4687 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4689 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4691 if (file->f_op != &nvme_dev_fops)
4693 return file->private_data;
4695 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4698 * Check we didn't inadvertently grow the command structure sizes:
4700 static inline void _nvme_check_size(void)
4702 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4703 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4704 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4705 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4706 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4707 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4708 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4709 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4710 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4711 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4712 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4713 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4714 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4715 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4716 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4717 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4718 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4719 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4720 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4724 static int __init nvme_core_init(void)
4726 int result = -ENOMEM;
4730 nvme_wq = alloc_workqueue("nvme-wq",
4731 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4735 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4736 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4740 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4741 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4742 if (!nvme_delete_wq)
4743 goto destroy_reset_wq;
4745 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4746 NVME_MINORS, "nvme");
4748 goto destroy_delete_wq;
4750 nvme_class = class_create(THIS_MODULE, "nvme");
4751 if (IS_ERR(nvme_class)) {
4752 result = PTR_ERR(nvme_class);
4753 goto unregister_chrdev;
4755 nvme_class->dev_uevent = nvme_class_uevent;
4757 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4758 if (IS_ERR(nvme_subsys_class)) {
4759 result = PTR_ERR(nvme_subsys_class);
4765 class_destroy(nvme_class);
4767 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4769 destroy_workqueue(nvme_delete_wq);
4771 destroy_workqueue(nvme_reset_wq);
4773 destroy_workqueue(nvme_wq);
4778 static void __exit nvme_core_exit(void)
4780 class_destroy(nvme_subsys_class);
4781 class_destroy(nvme_class);
4782 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4783 destroy_workqueue(nvme_delete_wq);
4784 destroy_workqueue(nvme_reset_wq);
4785 destroy_workqueue(nvme_wq);
4786 ida_destroy(&nvme_instance_ida);
4789 MODULE_LICENSE("GPL");
4790 MODULE_VERSION("1.0");
4791 module_init(nvme_core_init);
4792 module_exit(nvme_core_exit);