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(q, 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(req->q, 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(rq->q, 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 gendisk *disk = ns ? ns->disk : 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,
1136 bio->bi_disk = disk;
1137 if (disk && meta_buffer && meta_len) {
1138 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1141 ret = PTR_ERR(meta);
1144 req->cmd_flags |= REQ_INTEGRITY;
1148 nvme_execute_passthru_rq(req);
1149 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1152 ret = nvme_req(req)->status;
1154 *result = le64_to_cpu(nvme_req(req)->result.u64);
1155 if (meta && !ret && !write) {
1156 if (copy_to_user(meta_buffer, meta, meta_len))
1162 blk_rq_unmap_user(bio);
1164 blk_mq_free_request(req);
1168 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1170 struct nvme_ctrl *ctrl = rq->end_io_data;
1171 unsigned long flags;
1172 bool startka = false;
1174 blk_mq_free_request(rq);
1177 dev_err(ctrl->device,
1178 "failed nvme_keep_alive_end_io error=%d\n",
1183 ctrl->comp_seen = false;
1184 spin_lock_irqsave(&ctrl->lock, flags);
1185 if (ctrl->state == NVME_CTRL_LIVE ||
1186 ctrl->state == NVME_CTRL_CONNECTING)
1188 spin_unlock_irqrestore(&ctrl->lock, flags);
1190 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1193 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1197 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1198 BLK_MQ_REQ_RESERVED);
1202 rq->timeout = ctrl->kato * HZ;
1203 rq->end_io_data = ctrl;
1205 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1210 static void nvme_keep_alive_work(struct work_struct *work)
1212 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1213 struct nvme_ctrl, ka_work);
1214 bool comp_seen = ctrl->comp_seen;
1216 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1217 dev_dbg(ctrl->device,
1218 "reschedule traffic based keep-alive timer\n");
1219 ctrl->comp_seen = false;
1220 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1224 if (nvme_keep_alive(ctrl)) {
1225 /* allocation failure, reset the controller */
1226 dev_err(ctrl->device, "keep-alive failed\n");
1227 nvme_reset_ctrl(ctrl);
1232 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1234 if (unlikely(ctrl->kato == 0))
1237 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1240 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1242 if (unlikely(ctrl->kato == 0))
1245 cancel_delayed_work_sync(&ctrl->ka_work);
1247 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1250 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1251 * flag, thus sending any new CNS opcodes has a big chance of not working.
1252 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1253 * (but not for any later version).
1255 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1257 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1258 return ctrl->vs < NVME_VS(1, 2, 0);
1259 return ctrl->vs < NVME_VS(1, 1, 0);
1262 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1264 struct nvme_command c = { };
1267 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1268 c.identify.opcode = nvme_admin_identify;
1269 c.identify.cns = NVME_ID_CNS_CTRL;
1271 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1275 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1276 sizeof(struct nvme_id_ctrl));
1282 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1284 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1287 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1288 struct nvme_ns_id_desc *cur, bool *csi_seen)
1290 const char *warn_str = "ctrl returned bogus length:";
1293 switch (cur->nidt) {
1294 case NVME_NIDT_EUI64:
1295 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1296 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1297 warn_str, cur->nidl);
1300 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1301 return NVME_NIDT_EUI64_LEN;
1302 case NVME_NIDT_NGUID:
1303 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1304 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1305 warn_str, cur->nidl);
1308 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1309 return NVME_NIDT_NGUID_LEN;
1310 case NVME_NIDT_UUID:
1311 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1312 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1313 warn_str, cur->nidl);
1316 uuid_copy(&ids->uuid, data + sizeof(*cur));
1317 return NVME_NIDT_UUID_LEN;
1319 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1320 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1321 warn_str, cur->nidl);
1324 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1326 return NVME_NIDT_CSI_LEN;
1328 /* Skip unknown types */
1333 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1334 struct nvme_ns_ids *ids)
1336 struct nvme_command c = { };
1337 bool csi_seen = false;
1338 int status, pos, len;
1341 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1343 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1346 c.identify.opcode = nvme_admin_identify;
1347 c.identify.nsid = cpu_to_le32(nsid);
1348 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1350 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1354 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1355 NVME_IDENTIFY_DATA_SIZE);
1357 dev_warn(ctrl->device,
1358 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1363 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1364 struct nvme_ns_id_desc *cur = data + pos;
1369 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1373 len += sizeof(*cur);
1376 if (nvme_multi_css(ctrl) && !csi_seen) {
1377 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1387 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1388 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1390 struct nvme_command c = { };
1393 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1394 c.identify.opcode = nvme_admin_identify;
1395 c.identify.nsid = cpu_to_le32(nsid);
1396 c.identify.cns = NVME_ID_CNS_NS;
1398 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1402 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1404 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1409 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1412 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1413 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1414 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1415 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1416 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1417 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1426 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1427 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1429 union nvme_result res = { 0 };
1430 struct nvme_command c;
1433 memset(&c, 0, sizeof(c));
1434 c.features.opcode = op;
1435 c.features.fid = cpu_to_le32(fid);
1436 c.features.dword11 = cpu_to_le32(dword11);
1438 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1439 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1440 if (ret >= 0 && result)
1441 *result = le32_to_cpu(res.u32);
1445 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1446 unsigned int dword11, void *buffer, size_t buflen,
1449 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1452 EXPORT_SYMBOL_GPL(nvme_set_features);
1454 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1455 unsigned int dword11, void *buffer, size_t buflen,
1458 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1461 EXPORT_SYMBOL_GPL(nvme_get_features);
1463 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1465 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1467 int status, nr_io_queues;
1469 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1475 * Degraded controllers might return an error when setting the queue
1476 * count. We still want to be able to bring them online and offer
1477 * access to the admin queue, as that might be only way to fix them up.
1480 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1483 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1484 *count = min(*count, nr_io_queues);
1489 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1491 #define NVME_AEN_SUPPORTED \
1492 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1493 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1495 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1497 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1500 if (!supported_aens)
1503 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1506 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1509 queue_work(nvme_wq, &ctrl->async_event_work);
1513 * Convert integer values from ioctl structures to user pointers, silently
1514 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1517 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1519 if (in_compat_syscall())
1520 ptrval = (compat_uptr_t)ptrval;
1521 return (void __user *)ptrval;
1524 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1526 struct nvme_user_io io;
1527 struct nvme_command c;
1528 unsigned length, meta_len;
1529 void __user *metadata;
1531 if (copy_from_user(&io, uio, sizeof(io)))
1536 switch (io.opcode) {
1537 case nvme_cmd_write:
1539 case nvme_cmd_compare:
1545 length = (io.nblocks + 1) << ns->lba_shift;
1547 if ((io.control & NVME_RW_PRINFO_PRACT) &&
1548 ns->ms == sizeof(struct t10_pi_tuple)) {
1550 * Protection information is stripped/inserted by the
1553 if (nvme_to_user_ptr(io.metadata))
1558 meta_len = (io.nblocks + 1) * ns->ms;
1559 metadata = nvme_to_user_ptr(io.metadata);
1562 if (ns->features & NVME_NS_EXT_LBAS) {
1565 } else if (meta_len) {
1566 if ((io.metadata & 3) || !io.metadata)
1570 memset(&c, 0, sizeof(c));
1571 c.rw.opcode = io.opcode;
1572 c.rw.flags = io.flags;
1573 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1574 c.rw.slba = cpu_to_le64(io.slba);
1575 c.rw.length = cpu_to_le16(io.nblocks);
1576 c.rw.control = cpu_to_le16(io.control);
1577 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1578 c.rw.reftag = cpu_to_le32(io.reftag);
1579 c.rw.apptag = cpu_to_le16(io.apptag);
1580 c.rw.appmask = cpu_to_le16(io.appmask);
1582 return nvme_submit_user_cmd(ns->queue, &c,
1583 nvme_to_user_ptr(io.addr), length,
1584 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1587 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1588 struct nvme_passthru_cmd __user *ucmd)
1590 struct nvme_passthru_cmd cmd;
1591 struct nvme_command c;
1592 unsigned timeout = 0;
1596 if (!capable(CAP_SYS_ADMIN))
1598 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1603 memset(&c, 0, sizeof(c));
1604 c.common.opcode = cmd.opcode;
1605 c.common.flags = cmd.flags;
1606 c.common.nsid = cpu_to_le32(cmd.nsid);
1607 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1608 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1609 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1610 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1611 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1612 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1613 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1614 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1617 timeout = msecs_to_jiffies(cmd.timeout_ms);
1619 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1620 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1621 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1622 0, &result, timeout);
1625 if (put_user(result, &ucmd->result))
1632 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1633 struct nvme_passthru_cmd64 __user *ucmd)
1635 struct nvme_passthru_cmd64 cmd;
1636 struct nvme_command c;
1637 unsigned timeout = 0;
1640 if (!capable(CAP_SYS_ADMIN))
1642 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1647 memset(&c, 0, sizeof(c));
1648 c.common.opcode = cmd.opcode;
1649 c.common.flags = cmd.flags;
1650 c.common.nsid = cpu_to_le32(cmd.nsid);
1651 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1652 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1653 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1654 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1655 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1656 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1657 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1658 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1661 timeout = msecs_to_jiffies(cmd.timeout_ms);
1663 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1664 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1665 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1666 0, &cmd.result, timeout);
1669 if (put_user(cmd.result, &ucmd->result))
1677 * Issue ioctl requests on the first available path. Note that unlike normal
1678 * block layer requests we will not retry failed request on another controller.
1680 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1681 struct nvme_ns_head **head, int *srcu_idx)
1683 #ifdef CONFIG_NVME_MULTIPATH
1684 if (disk->fops == &nvme_ns_head_ops) {
1687 *head = disk->private_data;
1688 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1689 ns = nvme_find_path(*head);
1691 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1697 return disk->private_data;
1700 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1703 srcu_read_unlock(&head->srcu, idx);
1706 static bool is_ctrl_ioctl(unsigned int cmd)
1708 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1710 if (is_sed_ioctl(cmd))
1715 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1717 struct nvme_ns_head *head,
1720 struct nvme_ctrl *ctrl = ns->ctrl;
1723 nvme_get_ctrl(ns->ctrl);
1724 nvme_put_ns_from_disk(head, srcu_idx);
1727 case NVME_IOCTL_ADMIN_CMD:
1728 ret = nvme_user_cmd(ctrl, NULL, argp);
1730 case NVME_IOCTL_ADMIN64_CMD:
1731 ret = nvme_user_cmd64(ctrl, NULL, argp);
1734 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1737 nvme_put_ctrl(ctrl);
1741 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1742 unsigned int cmd, unsigned long arg)
1744 struct nvme_ns_head *head = NULL;
1745 void __user *argp = (void __user *)arg;
1749 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1751 return -EWOULDBLOCK;
1754 * Handle ioctls that apply to the controller instead of the namespace
1755 * seperately and drop the ns SRCU reference early. This avoids a
1756 * deadlock when deleting namespaces using the passthrough interface.
1758 if (is_ctrl_ioctl(cmd))
1759 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1763 force_successful_syscall_return();
1764 ret = ns->head->ns_id;
1766 case NVME_IOCTL_IO_CMD:
1767 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1769 case NVME_IOCTL_SUBMIT_IO:
1770 ret = nvme_submit_io(ns, argp);
1772 case NVME_IOCTL_IO64_CMD:
1773 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1777 ret = nvme_nvm_ioctl(ns, cmd, arg);
1782 nvme_put_ns_from_disk(head, srcu_idx);
1786 #ifdef CONFIG_COMPAT
1787 struct nvme_user_io32 {
1800 } __attribute__((__packed__));
1802 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1804 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1805 unsigned int cmd, unsigned long arg)
1808 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1809 * between 32 bit programs and 64 bit kernel.
1810 * The cause is that the results of sizeof(struct nvme_user_io),
1811 * which is used to define NVME_IOCTL_SUBMIT_IO,
1812 * are not same between 32 bit compiler and 64 bit compiler.
1813 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1814 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1815 * Other IOCTL numbers are same between 32 bit and 64 bit.
1816 * So there is nothing to do regarding to other IOCTL numbers.
1818 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1819 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1821 return nvme_ioctl(bdev, mode, cmd, arg);
1824 #define nvme_compat_ioctl NULL
1825 #endif /* CONFIG_COMPAT */
1827 static int nvme_open(struct block_device *bdev, fmode_t mode)
1829 struct nvme_ns *ns = bdev->bd_disk->private_data;
1831 #ifdef CONFIG_NVME_MULTIPATH
1832 /* should never be called due to GENHD_FL_HIDDEN */
1833 if (WARN_ON_ONCE(ns->head->disk))
1836 if (!kref_get_unless_zero(&ns->kref))
1838 if (!try_module_get(ns->ctrl->ops->module))
1849 static void nvme_release(struct gendisk *disk, fmode_t mode)
1851 struct nvme_ns *ns = disk->private_data;
1853 module_put(ns->ctrl->ops->module);
1857 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1859 /* some standard values */
1860 geo->heads = 1 << 6;
1861 geo->sectors = 1 << 5;
1862 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1866 #ifdef CONFIG_BLK_DEV_INTEGRITY
1867 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1868 u32 max_integrity_segments)
1870 struct blk_integrity integrity;
1872 memset(&integrity, 0, sizeof(integrity));
1874 case NVME_NS_DPS_PI_TYPE3:
1875 integrity.profile = &t10_pi_type3_crc;
1876 integrity.tag_size = sizeof(u16) + sizeof(u32);
1877 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1879 case NVME_NS_DPS_PI_TYPE1:
1880 case NVME_NS_DPS_PI_TYPE2:
1881 integrity.profile = &t10_pi_type1_crc;
1882 integrity.tag_size = sizeof(u16);
1883 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1886 integrity.profile = NULL;
1889 integrity.tuple_size = ms;
1890 blk_integrity_register(disk, &integrity);
1891 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1894 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1895 u32 max_integrity_segments)
1898 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1900 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1902 struct nvme_ctrl *ctrl = ns->ctrl;
1903 struct request_queue *queue = disk->queue;
1904 u32 size = queue_logical_block_size(queue);
1906 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1907 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1911 if (ctrl->nr_streams && ns->sws && ns->sgs)
1912 size *= ns->sws * ns->sgs;
1914 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1915 NVME_DSM_MAX_RANGES);
1917 queue->limits.discard_alignment = 0;
1918 queue->limits.discard_granularity = size;
1920 /* If discard is already enabled, don't reset queue limits */
1921 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1924 blk_queue_max_discard_sectors(queue, UINT_MAX);
1925 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1927 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1928 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1931 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1935 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1936 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1939 * Even though NVMe spec explicitly states that MDTS is not
1940 * applicable to the write-zeroes:- "The restriction does not apply to
1941 * commands that do not transfer data between the host and the
1942 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1943 * In order to be more cautious use controller's max_hw_sectors value
1944 * to configure the maximum sectors for the write-zeroes which is
1945 * configured based on the controller's MDTS field in the
1946 * nvme_init_identify() if available.
1948 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1949 max_blocks = (u64)USHRT_MAX + 1;
1951 max_blocks = ns->ctrl->max_hw_sectors + 1;
1953 blk_queue_max_write_zeroes_sectors(disk->queue,
1954 nvme_lba_to_sect(ns, max_blocks));
1957 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1959 return !uuid_is_null(&ids->uuid) ||
1960 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1961 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1964 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1966 return uuid_equal(&a->uuid, &b->uuid) &&
1967 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1968 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1972 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1973 u32 *phys_bs, u32 *io_opt)
1975 struct streams_directive_params s;
1978 if (!ctrl->nr_streams)
1981 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1985 ns->sws = le32_to_cpu(s.sws);
1986 ns->sgs = le16_to_cpu(s.sgs);
1989 *phys_bs = ns->sws * (1 << ns->lba_shift);
1991 *io_opt = *phys_bs * ns->sgs;
1997 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1999 struct nvme_ctrl *ctrl = ns->ctrl;
2002 * The PI implementation requires the metadata size to be equal to the
2003 * t10 pi tuple size.
2005 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
2006 if (ns->ms == sizeof(struct t10_pi_tuple))
2007 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2011 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2012 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
2014 if (ctrl->ops->flags & NVME_F_FABRICS) {
2016 * The NVMe over Fabrics specification only supports metadata as
2017 * part of the extended data LBA. We rely on HCA/HBA support to
2018 * remap the separate metadata buffer from the block layer.
2020 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
2022 if (ctrl->max_integrity_segments)
2024 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
2027 * For PCIe controllers, we can't easily remap the separate
2028 * metadata buffer from the block layer and thus require a
2029 * separate metadata buffer for block layer metadata/PI support.
2030 * We allow extended LBAs for the passthrough interface, though.
2032 if (id->flbas & NVME_NS_FLBAS_META_EXT)
2033 ns->features |= NVME_NS_EXT_LBAS;
2035 ns->features |= NVME_NS_METADATA_SUPPORTED;
2041 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2042 struct request_queue *q)
2044 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2046 if (ctrl->max_hw_sectors) {
2048 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2050 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2051 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2052 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2054 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2055 blk_queue_dma_alignment(q, 7);
2056 blk_queue_write_cache(q, vwc, vwc);
2059 static void nvme_update_disk_info(struct gendisk *disk,
2060 struct nvme_ns *ns, struct nvme_id_ns *id)
2062 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
2063 unsigned short bs = 1 << ns->lba_shift;
2064 u32 atomic_bs, phys_bs, io_opt = 0;
2067 * The block layer can't support LBA sizes larger than the page size
2068 * yet, so catch this early and don't allow block I/O.
2070 if (ns->lba_shift > PAGE_SHIFT) {
2075 blk_integrity_unregister(disk);
2077 atomic_bs = phys_bs = bs;
2078 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
2079 if (id->nabo == 0) {
2081 * Bit 1 indicates whether NAWUPF is defined for this namespace
2082 * and whether it should be used instead of AWUPF. If NAWUPF ==
2083 * 0 then AWUPF must be used instead.
2085 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
2086 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
2088 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
2091 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
2092 /* NPWG = Namespace Preferred Write Granularity */
2093 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
2094 /* NOWS = Namespace Optimal Write Size */
2095 io_opt = bs * (1 + le16_to_cpu(id->nows));
2098 blk_queue_logical_block_size(disk->queue, bs);
2100 * Linux filesystems assume writing a single physical block is
2101 * an atomic operation. Hence limit the physical block size to the
2102 * value of the Atomic Write Unit Power Fail parameter.
2104 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2105 blk_queue_io_min(disk->queue, phys_bs);
2106 blk_queue_io_opt(disk->queue, io_opt);
2109 * Register a metadata profile for PI, or the plain non-integrity NVMe
2110 * metadata masquerading as Type 0 if supported, otherwise reject block
2111 * I/O to namespaces with metadata except when the namespace supports
2112 * PI, as it can strip/insert in that case.
2115 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2116 (ns->features & NVME_NS_METADATA_SUPPORTED))
2117 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2118 ns->ctrl->max_integrity_segments);
2119 else if (!nvme_ns_has_pi(ns))
2123 set_capacity_and_notify(disk, capacity);
2125 nvme_config_discard(disk, ns);
2126 nvme_config_write_zeroes(disk, ns);
2128 if ((id->nsattr & NVME_NS_ATTR_RO) ||
2129 test_bit(NVME_NS_FORCE_RO, &ns->flags))
2130 set_disk_ro(disk, true);
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 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2180 ret = nvme_update_zone_info(ns, lbaf);
2185 ret = nvme_configure_metadata(ns, id);
2188 nvme_set_chunk_sectors(ns, id);
2189 nvme_update_disk_info(ns->disk, ns, id);
2190 blk_mq_unfreeze_queue(ns->disk->queue);
2192 if (blk_queue_is_zoned(ns->queue)) {
2193 ret = nvme_revalidate_zones(ns);
2194 if (ret && !nvme_first_scan(ns->disk))
2198 #ifdef CONFIG_NVME_MULTIPATH
2199 if (ns->head->disk) {
2200 blk_mq_freeze_queue(ns->head->disk->queue);
2201 nvme_update_disk_info(ns->head->disk, ns, id);
2202 blk_stack_limits(&ns->head->disk->queue->limits,
2203 &ns->queue->limits, 0);
2204 blk_queue_update_readahead(ns->head->disk->queue);
2205 blk_mq_unfreeze_queue(ns->head->disk->queue);
2211 blk_mq_unfreeze_queue(ns->disk->queue);
2215 static char nvme_pr_type(enum pr_type type)
2218 case PR_WRITE_EXCLUSIVE:
2220 case PR_EXCLUSIVE_ACCESS:
2222 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2224 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2226 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2228 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2235 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2236 u64 key, u64 sa_key, u8 op)
2238 struct nvme_ns_head *head = NULL;
2240 struct nvme_command c;
2242 u8 data[16] = { 0, };
2244 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2246 return -EWOULDBLOCK;
2248 put_unaligned_le64(key, &data[0]);
2249 put_unaligned_le64(sa_key, &data[8]);
2251 memset(&c, 0, sizeof(c));
2252 c.common.opcode = op;
2253 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2254 c.common.cdw10 = cpu_to_le32(cdw10);
2256 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2257 nvme_put_ns_from_disk(head, srcu_idx);
2261 static int nvme_pr_register(struct block_device *bdev, u64 old,
2262 u64 new, unsigned flags)
2266 if (flags & ~PR_FL_IGNORE_KEY)
2269 cdw10 = old ? 2 : 0;
2270 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2271 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2272 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2275 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2276 enum pr_type type, unsigned flags)
2280 if (flags & ~PR_FL_IGNORE_KEY)
2283 cdw10 = nvme_pr_type(type) << 8;
2284 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2285 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2288 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2289 enum pr_type type, bool abort)
2291 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2292 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2295 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2297 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2298 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2301 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2303 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2304 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2307 static const struct pr_ops nvme_pr_ops = {
2308 .pr_register = nvme_pr_register,
2309 .pr_reserve = nvme_pr_reserve,
2310 .pr_release = nvme_pr_release,
2311 .pr_preempt = nvme_pr_preempt,
2312 .pr_clear = nvme_pr_clear,
2315 #ifdef CONFIG_BLK_SED_OPAL
2316 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2319 struct nvme_ctrl *ctrl = data;
2320 struct nvme_command cmd;
2322 memset(&cmd, 0, sizeof(cmd));
2324 cmd.common.opcode = nvme_admin_security_send;
2326 cmd.common.opcode = nvme_admin_security_recv;
2327 cmd.common.nsid = 0;
2328 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2329 cmd.common.cdw11 = cpu_to_le32(len);
2331 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2332 NVME_QID_ANY, 1, 0, false);
2334 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2335 #endif /* CONFIG_BLK_SED_OPAL */
2337 static const struct block_device_operations nvme_bdev_ops = {
2338 .owner = THIS_MODULE,
2339 .ioctl = nvme_ioctl,
2340 .compat_ioctl = nvme_compat_ioctl,
2342 .release = nvme_release,
2343 .getgeo = nvme_getgeo,
2344 .report_zones = nvme_report_zones,
2345 .pr_ops = &nvme_pr_ops,
2348 #ifdef CONFIG_NVME_MULTIPATH
2349 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2351 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2353 if (!kref_get_unless_zero(&head->ref))
2358 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2360 nvme_put_ns_head(disk->private_data);
2363 const struct block_device_operations nvme_ns_head_ops = {
2364 .owner = THIS_MODULE,
2365 .submit_bio = nvme_ns_head_submit_bio,
2366 .open = nvme_ns_head_open,
2367 .release = nvme_ns_head_release,
2368 .ioctl = nvme_ioctl,
2369 .compat_ioctl = nvme_compat_ioctl,
2370 .getgeo = nvme_getgeo,
2371 .report_zones = nvme_report_zones,
2372 .pr_ops = &nvme_pr_ops,
2374 #endif /* CONFIG_NVME_MULTIPATH */
2376 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2378 unsigned long timeout =
2379 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2380 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2383 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2386 if ((csts & NVME_CSTS_RDY) == bit)
2389 usleep_range(1000, 2000);
2390 if (fatal_signal_pending(current))
2392 if (time_after(jiffies, timeout)) {
2393 dev_err(ctrl->device,
2394 "Device not ready; aborting %s, CSTS=0x%x\n",
2395 enabled ? "initialisation" : "reset", csts);
2404 * If the device has been passed off to us in an enabled state, just clear
2405 * the enabled bit. The spec says we should set the 'shutdown notification
2406 * bits', but doing so may cause the device to complete commands to the
2407 * admin queue ... and we don't know what memory that might be pointing at!
2409 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2413 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2414 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2416 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2420 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2421 msleep(NVME_QUIRK_DELAY_AMOUNT);
2423 return nvme_wait_ready(ctrl, ctrl->cap, false);
2425 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2427 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2429 unsigned dev_page_min;
2432 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2434 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2437 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2439 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2440 dev_err(ctrl->device,
2441 "Minimum device page size %u too large for host (%u)\n",
2442 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2446 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2447 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2449 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2450 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2451 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2452 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2453 ctrl->ctrl_config |= NVME_CC_ENABLE;
2455 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2458 return nvme_wait_ready(ctrl, ctrl->cap, true);
2460 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2462 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2464 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2468 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2469 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2471 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2475 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2476 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2480 if (fatal_signal_pending(current))
2482 if (time_after(jiffies, timeout)) {
2483 dev_err(ctrl->device,
2484 "Device shutdown incomplete; abort shutdown\n");
2491 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2493 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2498 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2501 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2502 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2505 dev_warn_once(ctrl->device,
2506 "could not set timestamp (%d)\n", ret);
2510 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2512 struct nvme_feat_host_behavior *host;
2515 /* Don't bother enabling the feature if retry delay is not reported */
2519 host = kzalloc(sizeof(*host), GFP_KERNEL);
2523 host->acre = NVME_ENABLE_ACRE;
2524 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2525 host, sizeof(*host), NULL);
2530 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2533 * APST (Autonomous Power State Transition) lets us program a
2534 * table of power state transitions that the controller will
2535 * perform automatically. We configure it with a simple
2536 * heuristic: we are willing to spend at most 2% of the time
2537 * transitioning between power states. Therefore, when running
2538 * in any given state, we will enter the next lower-power
2539 * non-operational state after waiting 50 * (enlat + exlat)
2540 * microseconds, as long as that state's exit latency is under
2541 * the requested maximum latency.
2543 * We will not autonomously enter any non-operational state for
2544 * which the total latency exceeds ps_max_latency_us. Users
2545 * can set ps_max_latency_us to zero to turn off APST.
2549 struct nvme_feat_auto_pst *table;
2555 * If APST isn't supported or if we haven't been initialized yet,
2556 * then don't do anything.
2561 if (ctrl->npss > 31) {
2562 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2566 table = kzalloc(sizeof(*table), GFP_KERNEL);
2570 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2571 /* Turn off APST. */
2573 dev_dbg(ctrl->device, "APST disabled\n");
2575 __le64 target = cpu_to_le64(0);
2579 * Walk through all states from lowest- to highest-power.
2580 * According to the spec, lower-numbered states use more
2581 * power. NPSS, despite the name, is the index of the
2582 * lowest-power state, not the number of states.
2584 for (state = (int)ctrl->npss; state >= 0; state--) {
2585 u64 total_latency_us, exit_latency_us, transition_ms;
2588 table->entries[state] = target;
2591 * Don't allow transitions to the deepest state
2592 * if it's quirked off.
2594 if (state == ctrl->npss &&
2595 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2599 * Is this state a useful non-operational state for
2600 * higher-power states to autonomously transition to?
2602 if (!(ctrl->psd[state].flags &
2603 NVME_PS_FLAGS_NON_OP_STATE))
2607 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2608 if (exit_latency_us > ctrl->ps_max_latency_us)
2613 le32_to_cpu(ctrl->psd[state].entry_lat);
2616 * This state is good. Use it as the APST idle
2617 * target for higher power states.
2619 transition_ms = total_latency_us + 19;
2620 do_div(transition_ms, 20);
2621 if (transition_ms > (1 << 24) - 1)
2622 transition_ms = (1 << 24) - 1;
2624 target = cpu_to_le64((state << 3) |
2625 (transition_ms << 8));
2630 if (total_latency_us > max_lat_us)
2631 max_lat_us = total_latency_us;
2637 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2639 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2640 max_ps, max_lat_us, (int)sizeof(*table), table);
2644 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2645 table, sizeof(*table), NULL);
2647 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2653 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2655 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2659 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2660 case PM_QOS_LATENCY_ANY:
2668 if (ctrl->ps_max_latency_us != latency) {
2669 ctrl->ps_max_latency_us = latency;
2670 nvme_configure_apst(ctrl);
2674 struct nvme_core_quirk_entry {
2676 * NVMe model and firmware strings are padded with spaces. For
2677 * simplicity, strings in the quirk table are padded with NULLs
2683 unsigned long quirks;
2686 static const struct nvme_core_quirk_entry core_quirks[] = {
2689 * This Toshiba device seems to die using any APST states. See:
2690 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2693 .mn = "THNSF5256GPUK TOSHIBA",
2694 .quirks = NVME_QUIRK_NO_APST,
2698 * This LiteON CL1-3D*-Q11 firmware version has a race
2699 * condition associated with actions related to suspend to idle
2700 * LiteON has resolved the problem in future firmware
2704 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2708 /* match is null-terminated but idstr is space-padded. */
2709 static bool string_matches(const char *idstr, const char *match, size_t len)
2716 matchlen = strlen(match);
2717 WARN_ON_ONCE(matchlen > len);
2719 if (memcmp(idstr, match, matchlen))
2722 for (; matchlen < len; matchlen++)
2723 if (idstr[matchlen] != ' ')
2729 static bool quirk_matches(const struct nvme_id_ctrl *id,
2730 const struct nvme_core_quirk_entry *q)
2732 return q->vid == le16_to_cpu(id->vid) &&
2733 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2734 string_matches(id->fr, q->fr, sizeof(id->fr));
2737 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2738 struct nvme_id_ctrl *id)
2743 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2744 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2745 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2746 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2750 if (ctrl->vs >= NVME_VS(1, 2, 1))
2751 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2754 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2755 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2756 "nqn.2014.08.org.nvmexpress:%04x%04x",
2757 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2758 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2759 off += sizeof(id->sn);
2760 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2761 off += sizeof(id->mn);
2762 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2765 static void nvme_release_subsystem(struct device *dev)
2767 struct nvme_subsystem *subsys =
2768 container_of(dev, struct nvme_subsystem, dev);
2770 if (subsys->instance >= 0)
2771 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2775 static void nvme_destroy_subsystem(struct kref *ref)
2777 struct nvme_subsystem *subsys =
2778 container_of(ref, struct nvme_subsystem, ref);
2780 mutex_lock(&nvme_subsystems_lock);
2781 list_del(&subsys->entry);
2782 mutex_unlock(&nvme_subsystems_lock);
2784 ida_destroy(&subsys->ns_ida);
2785 device_del(&subsys->dev);
2786 put_device(&subsys->dev);
2789 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2791 kref_put(&subsys->ref, nvme_destroy_subsystem);
2794 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2796 struct nvme_subsystem *subsys;
2798 lockdep_assert_held(&nvme_subsystems_lock);
2801 * Fail matches for discovery subsystems. This results
2802 * in each discovery controller bound to a unique subsystem.
2803 * This avoids issues with validating controller values
2804 * that can only be true when there is a single unique subsystem.
2805 * There may be multiple and completely independent entities
2806 * that provide discovery controllers.
2808 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2811 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2812 if (strcmp(subsys->subnqn, subsysnqn))
2814 if (!kref_get_unless_zero(&subsys->ref))
2822 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2823 struct device_attribute subsys_attr_##_name = \
2824 __ATTR(_name, _mode, _show, NULL)
2826 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2827 struct device_attribute *attr,
2830 struct nvme_subsystem *subsys =
2831 container_of(dev, struct nvme_subsystem, dev);
2833 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2835 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2837 #define nvme_subsys_show_str_function(field) \
2838 static ssize_t subsys_##field##_show(struct device *dev, \
2839 struct device_attribute *attr, char *buf) \
2841 struct nvme_subsystem *subsys = \
2842 container_of(dev, struct nvme_subsystem, dev); \
2843 return sprintf(buf, "%.*s\n", \
2844 (int)sizeof(subsys->field), subsys->field); \
2846 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2848 nvme_subsys_show_str_function(model);
2849 nvme_subsys_show_str_function(serial);
2850 nvme_subsys_show_str_function(firmware_rev);
2852 static struct attribute *nvme_subsys_attrs[] = {
2853 &subsys_attr_model.attr,
2854 &subsys_attr_serial.attr,
2855 &subsys_attr_firmware_rev.attr,
2856 &subsys_attr_subsysnqn.attr,
2857 #ifdef CONFIG_NVME_MULTIPATH
2858 &subsys_attr_iopolicy.attr,
2863 static struct attribute_group nvme_subsys_attrs_group = {
2864 .attrs = nvme_subsys_attrs,
2867 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2868 &nvme_subsys_attrs_group,
2872 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2874 return ctrl->opts && ctrl->opts->discovery_nqn;
2877 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2878 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2880 struct nvme_ctrl *tmp;
2882 lockdep_assert_held(&nvme_subsystems_lock);
2884 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2885 if (nvme_state_terminal(tmp))
2888 if (tmp->cntlid == ctrl->cntlid) {
2889 dev_err(ctrl->device,
2890 "Duplicate cntlid %u with %s, rejecting\n",
2891 ctrl->cntlid, dev_name(tmp->device));
2895 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2896 nvme_discovery_ctrl(ctrl))
2899 dev_err(ctrl->device,
2900 "Subsystem does not support multiple controllers\n");
2907 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2909 struct nvme_subsystem *subsys, *found;
2912 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2916 subsys->instance = -1;
2917 mutex_init(&subsys->lock);
2918 kref_init(&subsys->ref);
2919 INIT_LIST_HEAD(&subsys->ctrls);
2920 INIT_LIST_HEAD(&subsys->nsheads);
2921 nvme_init_subnqn(subsys, ctrl, id);
2922 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2923 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2924 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2925 subsys->vendor_id = le16_to_cpu(id->vid);
2926 subsys->cmic = id->cmic;
2927 subsys->awupf = le16_to_cpu(id->awupf);
2928 #ifdef CONFIG_NVME_MULTIPATH
2929 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2932 subsys->dev.class = nvme_subsys_class;
2933 subsys->dev.release = nvme_release_subsystem;
2934 subsys->dev.groups = nvme_subsys_attrs_groups;
2935 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2936 device_initialize(&subsys->dev);
2938 mutex_lock(&nvme_subsystems_lock);
2939 found = __nvme_find_get_subsystem(subsys->subnqn);
2941 put_device(&subsys->dev);
2944 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2946 goto out_put_subsystem;
2949 ret = device_add(&subsys->dev);
2951 dev_err(ctrl->device,
2952 "failed to register subsystem device.\n");
2953 put_device(&subsys->dev);
2956 ida_init(&subsys->ns_ida);
2957 list_add_tail(&subsys->entry, &nvme_subsystems);
2960 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2961 dev_name(ctrl->device));
2963 dev_err(ctrl->device,
2964 "failed to create sysfs link from subsystem.\n");
2965 goto out_put_subsystem;
2969 subsys->instance = ctrl->instance;
2970 ctrl->subsys = subsys;
2971 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2972 mutex_unlock(&nvme_subsystems_lock);
2976 nvme_put_subsystem(subsys);
2978 mutex_unlock(&nvme_subsystems_lock);
2982 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2983 void *log, size_t size, u64 offset)
2985 struct nvme_command c = { };
2986 u32 dwlen = nvme_bytes_to_numd(size);
2988 c.get_log_page.opcode = nvme_admin_get_log_page;
2989 c.get_log_page.nsid = cpu_to_le32(nsid);
2990 c.get_log_page.lid = log_page;
2991 c.get_log_page.lsp = lsp;
2992 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2993 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2994 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2995 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2996 c.get_log_page.csi = csi;
2998 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3001 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3002 struct nvme_effects_log **log)
3004 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3010 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3014 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3015 cel, sizeof(*cel), 0);
3021 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3028 * Initialize the cached copies of the Identify data and various controller
3029 * register in our nvme_ctrl structure. This should be called as soon as
3030 * the admin queue is fully up and running.
3032 int nvme_init_identify(struct nvme_ctrl *ctrl)
3034 struct nvme_id_ctrl *id;
3035 int ret, page_shift;
3037 bool prev_apst_enabled;
3039 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3041 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3044 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3045 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3047 if (ctrl->vs >= NVME_VS(1, 1, 0))
3048 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3050 ret = nvme_identify_ctrl(ctrl, &id);
3052 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3056 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3057 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3062 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3063 ctrl->cntlid = le16_to_cpu(id->cntlid);
3065 if (!ctrl->identified) {
3068 ret = nvme_init_subsystem(ctrl, id);
3073 * Check for quirks. Quirk can depend on firmware version,
3074 * so, in principle, the set of quirks present can change
3075 * across a reset. As a possible future enhancement, we
3076 * could re-scan for quirks every time we reinitialize
3077 * the device, but we'd have to make sure that the driver
3078 * behaves intelligently if the quirks change.
3080 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3081 if (quirk_matches(id, &core_quirks[i]))
3082 ctrl->quirks |= core_quirks[i].quirks;
3086 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3087 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3088 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3091 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3092 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3093 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3095 ctrl->oacs = le16_to_cpu(id->oacs);
3096 ctrl->oncs = le16_to_cpu(id->oncs);
3097 ctrl->mtfa = le16_to_cpu(id->mtfa);
3098 ctrl->oaes = le32_to_cpu(id->oaes);
3099 ctrl->wctemp = le16_to_cpu(id->wctemp);
3100 ctrl->cctemp = le16_to_cpu(id->cctemp);
3102 atomic_set(&ctrl->abort_limit, id->acl + 1);
3103 ctrl->vwc = id->vwc;
3105 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3107 max_hw_sectors = UINT_MAX;
3108 ctrl->max_hw_sectors =
3109 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3111 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3112 ctrl->sgls = le32_to_cpu(id->sgls);
3113 ctrl->kas = le16_to_cpu(id->kas);
3114 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3115 ctrl->ctratt = le32_to_cpu(id->ctratt);
3119 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3121 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3122 shutdown_timeout, 60);
3124 if (ctrl->shutdown_timeout != shutdown_timeout)
3125 dev_info(ctrl->device,
3126 "Shutdown timeout set to %u seconds\n",
3127 ctrl->shutdown_timeout);
3129 ctrl->shutdown_timeout = shutdown_timeout;
3131 ctrl->npss = id->npss;
3132 ctrl->apsta = id->apsta;
3133 prev_apst_enabled = ctrl->apst_enabled;
3134 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3135 if (force_apst && id->apsta) {
3136 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3137 ctrl->apst_enabled = true;
3139 ctrl->apst_enabled = false;
3142 ctrl->apst_enabled = id->apsta;
3144 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3146 if (ctrl->ops->flags & NVME_F_FABRICS) {
3147 ctrl->icdoff = le16_to_cpu(id->icdoff);
3148 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3149 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3150 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3153 * In fabrics we need to verify the cntlid matches the
3156 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3157 dev_err(ctrl->device,
3158 "Mismatching cntlid: Connect %u vs Identify "
3160 ctrl->cntlid, le16_to_cpu(id->cntlid));
3165 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3166 dev_err(ctrl->device,
3167 "keep-alive support is mandatory for fabrics\n");
3172 ctrl->hmpre = le32_to_cpu(id->hmpre);
3173 ctrl->hmmin = le32_to_cpu(id->hmmin);
3174 ctrl->hmminds = le32_to_cpu(id->hmminds);
3175 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3178 ret = nvme_mpath_init(ctrl, id);
3184 if (ctrl->apst_enabled && !prev_apst_enabled)
3185 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3186 else if (!ctrl->apst_enabled && prev_apst_enabled)
3187 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3189 ret = nvme_configure_apst(ctrl);
3193 ret = nvme_configure_timestamp(ctrl);
3197 ret = nvme_configure_directives(ctrl);
3201 ret = nvme_configure_acre(ctrl);
3205 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3206 ret = nvme_hwmon_init(ctrl);
3211 ctrl->identified = true;
3219 EXPORT_SYMBOL_GPL(nvme_init_identify);
3221 static int nvme_dev_open(struct inode *inode, struct file *file)
3223 struct nvme_ctrl *ctrl =
3224 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3226 switch (ctrl->state) {
3227 case NVME_CTRL_LIVE:
3230 return -EWOULDBLOCK;
3233 nvme_get_ctrl(ctrl);
3234 if (!try_module_get(ctrl->ops->module)) {
3235 nvme_put_ctrl(ctrl);
3239 file->private_data = ctrl;
3243 static int nvme_dev_release(struct inode *inode, struct file *file)
3245 struct nvme_ctrl *ctrl =
3246 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3248 module_put(ctrl->ops->module);
3249 nvme_put_ctrl(ctrl);
3253 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3258 down_read(&ctrl->namespaces_rwsem);
3259 if (list_empty(&ctrl->namespaces)) {
3264 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3265 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3266 dev_warn(ctrl->device,
3267 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3272 dev_warn(ctrl->device,
3273 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3274 kref_get(&ns->kref);
3275 up_read(&ctrl->namespaces_rwsem);
3277 ret = nvme_user_cmd(ctrl, ns, argp);
3282 up_read(&ctrl->namespaces_rwsem);
3286 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3289 struct nvme_ctrl *ctrl = file->private_data;
3290 void __user *argp = (void __user *)arg;
3293 case NVME_IOCTL_ADMIN_CMD:
3294 return nvme_user_cmd(ctrl, NULL, argp);
3295 case NVME_IOCTL_ADMIN64_CMD:
3296 return nvme_user_cmd64(ctrl, NULL, argp);
3297 case NVME_IOCTL_IO_CMD:
3298 return nvme_dev_user_cmd(ctrl, argp);
3299 case NVME_IOCTL_RESET:
3300 dev_warn(ctrl->device, "resetting controller\n");
3301 return nvme_reset_ctrl_sync(ctrl);
3302 case NVME_IOCTL_SUBSYS_RESET:
3303 return nvme_reset_subsystem(ctrl);
3304 case NVME_IOCTL_RESCAN:
3305 nvme_queue_scan(ctrl);
3312 static const struct file_operations nvme_dev_fops = {
3313 .owner = THIS_MODULE,
3314 .open = nvme_dev_open,
3315 .release = nvme_dev_release,
3316 .unlocked_ioctl = nvme_dev_ioctl,
3317 .compat_ioctl = compat_ptr_ioctl,
3320 static ssize_t nvme_sysfs_reset(struct device *dev,
3321 struct device_attribute *attr, const char *buf,
3324 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3327 ret = nvme_reset_ctrl_sync(ctrl);
3332 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3334 static ssize_t nvme_sysfs_rescan(struct device *dev,
3335 struct device_attribute *attr, const char *buf,
3338 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3340 nvme_queue_scan(ctrl);
3343 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3345 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3347 struct gendisk *disk = dev_to_disk(dev);
3349 if (disk->fops == &nvme_bdev_ops)
3350 return nvme_get_ns_from_dev(dev)->head;
3352 return disk->private_data;
3355 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3358 struct nvme_ns_head *head = dev_to_ns_head(dev);
3359 struct nvme_ns_ids *ids = &head->ids;
3360 struct nvme_subsystem *subsys = head->subsys;
3361 int serial_len = sizeof(subsys->serial);
3362 int model_len = sizeof(subsys->model);
3364 if (!uuid_is_null(&ids->uuid))
3365 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3367 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3368 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3370 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3371 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3373 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3374 subsys->serial[serial_len - 1] == '\0'))
3376 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3377 subsys->model[model_len - 1] == '\0'))
3380 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3381 serial_len, subsys->serial, model_len, subsys->model,
3384 static DEVICE_ATTR_RO(wwid);
3386 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3389 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3391 static DEVICE_ATTR_RO(nguid);
3393 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3396 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3398 /* For backward compatibility expose the NGUID to userspace if
3399 * we have no UUID set
3401 if (uuid_is_null(&ids->uuid)) {
3402 printk_ratelimited(KERN_WARNING
3403 "No UUID available providing old NGUID\n");
3404 return sprintf(buf, "%pU\n", ids->nguid);
3406 return sprintf(buf, "%pU\n", &ids->uuid);
3408 static DEVICE_ATTR_RO(uuid);
3410 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3413 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3415 static DEVICE_ATTR_RO(eui);
3417 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3420 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3422 static DEVICE_ATTR_RO(nsid);
3424 static struct attribute *nvme_ns_id_attrs[] = {
3425 &dev_attr_wwid.attr,
3426 &dev_attr_uuid.attr,
3427 &dev_attr_nguid.attr,
3429 &dev_attr_nsid.attr,
3430 #ifdef CONFIG_NVME_MULTIPATH
3431 &dev_attr_ana_grpid.attr,
3432 &dev_attr_ana_state.attr,
3437 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3438 struct attribute *a, int n)
3440 struct device *dev = container_of(kobj, struct device, kobj);
3441 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3443 if (a == &dev_attr_uuid.attr) {
3444 if (uuid_is_null(&ids->uuid) &&
3445 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3448 if (a == &dev_attr_nguid.attr) {
3449 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3452 if (a == &dev_attr_eui.attr) {
3453 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3456 #ifdef CONFIG_NVME_MULTIPATH
3457 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3458 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3460 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3467 static const struct attribute_group nvme_ns_id_attr_group = {
3468 .attrs = nvme_ns_id_attrs,
3469 .is_visible = nvme_ns_id_attrs_are_visible,
3472 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3473 &nvme_ns_id_attr_group,
3475 &nvme_nvm_attr_group,
3480 #define nvme_show_str_function(field) \
3481 static ssize_t field##_show(struct device *dev, \
3482 struct device_attribute *attr, char *buf) \
3484 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3485 return sprintf(buf, "%.*s\n", \
3486 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3488 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3490 nvme_show_str_function(model);
3491 nvme_show_str_function(serial);
3492 nvme_show_str_function(firmware_rev);
3494 #define nvme_show_int_function(field) \
3495 static ssize_t field##_show(struct device *dev, \
3496 struct device_attribute *attr, char *buf) \
3498 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3499 return sprintf(buf, "%d\n", ctrl->field); \
3501 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3503 nvme_show_int_function(cntlid);
3504 nvme_show_int_function(numa_node);
3505 nvme_show_int_function(queue_count);
3506 nvme_show_int_function(sqsize);
3508 static ssize_t nvme_sysfs_delete(struct device *dev,
3509 struct device_attribute *attr, const char *buf,
3512 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3514 if (device_remove_file_self(dev, attr))
3515 nvme_delete_ctrl_sync(ctrl);
3518 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3520 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3521 struct device_attribute *attr,
3524 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3526 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3528 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3530 static ssize_t nvme_sysfs_show_state(struct device *dev,
3531 struct device_attribute *attr,
3534 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3535 static const char *const state_name[] = {
3536 [NVME_CTRL_NEW] = "new",
3537 [NVME_CTRL_LIVE] = "live",
3538 [NVME_CTRL_RESETTING] = "resetting",
3539 [NVME_CTRL_CONNECTING] = "connecting",
3540 [NVME_CTRL_DELETING] = "deleting",
3541 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3542 [NVME_CTRL_DEAD] = "dead",
3545 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3546 state_name[ctrl->state])
3547 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3549 return sprintf(buf, "unknown state\n");
3552 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3554 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3555 struct device_attribute *attr,
3558 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3560 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3562 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3564 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3565 struct device_attribute *attr,
3568 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3570 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3572 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3574 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3575 struct device_attribute *attr,
3578 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3580 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3582 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3584 static ssize_t nvme_sysfs_show_address(struct device *dev,
3585 struct device_attribute *attr,
3588 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3590 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3592 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3594 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3595 struct device_attribute *attr, char *buf)
3597 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3598 struct nvmf_ctrl_options *opts = ctrl->opts;
3600 if (ctrl->opts->max_reconnects == -1)
3601 return sprintf(buf, "off\n");
3602 return sprintf(buf, "%d\n",
3603 opts->max_reconnects * opts->reconnect_delay);
3606 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3607 struct device_attribute *attr, const char *buf, size_t count)
3609 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3610 struct nvmf_ctrl_options *opts = ctrl->opts;
3611 int ctrl_loss_tmo, err;
3613 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3617 else if (ctrl_loss_tmo < 0)
3618 opts->max_reconnects = -1;
3620 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3621 opts->reconnect_delay);
3624 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3625 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3627 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3628 struct device_attribute *attr, char *buf)
3630 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3632 if (ctrl->opts->reconnect_delay == -1)
3633 return sprintf(buf, "off\n");
3634 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3637 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3638 struct device_attribute *attr, const char *buf, size_t count)
3640 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3644 err = kstrtou32(buf, 10, &v);
3648 ctrl->opts->reconnect_delay = v;
3651 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3652 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3654 static struct attribute *nvme_dev_attrs[] = {
3655 &dev_attr_reset_controller.attr,
3656 &dev_attr_rescan_controller.attr,
3657 &dev_attr_model.attr,
3658 &dev_attr_serial.attr,
3659 &dev_attr_firmware_rev.attr,
3660 &dev_attr_cntlid.attr,
3661 &dev_attr_delete_controller.attr,
3662 &dev_attr_transport.attr,
3663 &dev_attr_subsysnqn.attr,
3664 &dev_attr_address.attr,
3665 &dev_attr_state.attr,
3666 &dev_attr_numa_node.attr,
3667 &dev_attr_queue_count.attr,
3668 &dev_attr_sqsize.attr,
3669 &dev_attr_hostnqn.attr,
3670 &dev_attr_hostid.attr,
3671 &dev_attr_ctrl_loss_tmo.attr,
3672 &dev_attr_reconnect_delay.attr,
3676 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3677 struct attribute *a, int n)
3679 struct device *dev = container_of(kobj, struct device, kobj);
3680 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3682 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3684 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3686 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3688 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3690 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3692 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3698 static struct attribute_group nvme_dev_attrs_group = {
3699 .attrs = nvme_dev_attrs,
3700 .is_visible = nvme_dev_attrs_are_visible,
3703 static const struct attribute_group *nvme_dev_attr_groups[] = {
3704 &nvme_dev_attrs_group,
3708 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3711 struct nvme_ns_head *h;
3713 lockdep_assert_held(&subsys->lock);
3715 list_for_each_entry(h, &subsys->nsheads, entry) {
3716 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3723 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3724 struct nvme_ns_head *new)
3726 struct nvme_ns_head *h;
3728 lockdep_assert_held(&subsys->lock);
3730 list_for_each_entry(h, &subsys->nsheads, entry) {
3731 if (nvme_ns_ids_valid(&new->ids) &&
3732 nvme_ns_ids_equal(&new->ids, &h->ids))
3739 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3740 unsigned nsid, struct nvme_ns_ids *ids)
3742 struct nvme_ns_head *head;
3743 size_t size = sizeof(*head);
3746 #ifdef CONFIG_NVME_MULTIPATH
3747 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3750 head = kzalloc(size, GFP_KERNEL);
3753 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3756 head->instance = ret;
3757 INIT_LIST_HEAD(&head->list);
3758 ret = init_srcu_struct(&head->srcu);
3760 goto out_ida_remove;
3761 head->subsys = ctrl->subsys;
3764 kref_init(&head->ref);
3766 ret = __nvme_check_ids(ctrl->subsys, head);
3768 dev_err(ctrl->device,
3769 "duplicate IDs for nsid %d\n", nsid);
3770 goto out_cleanup_srcu;
3773 if (head->ids.csi) {
3774 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3776 goto out_cleanup_srcu;
3778 head->effects = ctrl->effects;
3780 ret = nvme_mpath_alloc_disk(ctrl, head);
3782 goto out_cleanup_srcu;
3784 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3786 kref_get(&ctrl->subsys->ref);
3790 cleanup_srcu_struct(&head->srcu);
3792 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3797 ret = blk_status_to_errno(nvme_error_status(ret));
3798 return ERR_PTR(ret);
3801 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3802 struct nvme_ns_ids *ids, bool is_shared)
3804 struct nvme_ctrl *ctrl = ns->ctrl;
3805 struct nvme_ns_head *head = NULL;
3808 mutex_lock(&ctrl->subsys->lock);
3809 head = nvme_find_ns_head(ctrl->subsys, nsid);
3811 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3813 ret = PTR_ERR(head);
3816 head->shared = is_shared;
3819 if (!is_shared || !head->shared) {
3820 dev_err(ctrl->device,
3821 "Duplicate unshared namespace %d\n", nsid);
3822 goto out_put_ns_head;
3824 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3825 dev_err(ctrl->device,
3826 "IDs don't match for shared namespace %d\n",
3828 goto out_put_ns_head;
3832 list_add_tail_rcu(&ns->siblings, &head->list);
3834 mutex_unlock(&ctrl->subsys->lock);
3838 nvme_put_ns_head(head);
3840 mutex_unlock(&ctrl->subsys->lock);
3844 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3846 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3847 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3849 return nsa->head->ns_id - nsb->head->ns_id;
3852 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3854 struct nvme_ns *ns, *ret = NULL;
3856 down_read(&ctrl->namespaces_rwsem);
3857 list_for_each_entry(ns, &ctrl->namespaces, list) {
3858 if (ns->head->ns_id == nsid) {
3859 if (!kref_get_unless_zero(&ns->kref))
3864 if (ns->head->ns_id > nsid)
3867 up_read(&ctrl->namespaces_rwsem);
3870 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3872 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3873 struct nvme_ns_ids *ids)
3876 struct gendisk *disk;
3877 struct nvme_id_ns *id;
3878 char disk_name[DISK_NAME_LEN];
3879 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT;
3881 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3884 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3888 ns->queue = blk_mq_init_queue(ctrl->tagset);
3889 if (IS_ERR(ns->queue))
3892 if (ctrl->opts && ctrl->opts->data_digest)
3893 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3895 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3896 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3897 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3899 ns->queue->queuedata = ns;
3901 kref_init(&ns->kref);
3903 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3904 goto out_free_queue;
3905 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3907 disk = alloc_disk_node(0, node);
3911 disk->fops = &nvme_bdev_ops;
3912 disk->private_data = ns;
3913 disk->queue = ns->queue;
3914 disk->flags = flags;
3915 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3918 if (nvme_update_ns_info(ns, id))
3921 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3922 if (nvme_nvm_register(ns, disk_name, node)) {
3923 dev_warn(ctrl->device, "LightNVM init failure\n");
3928 down_write(&ctrl->namespaces_rwsem);
3929 list_add_tail(&ns->list, &ctrl->namespaces);
3930 up_write(&ctrl->namespaces_rwsem);
3932 nvme_get_ctrl(ctrl);
3934 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3936 nvme_mpath_add_disk(ns, id);
3937 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3942 /* prevent double queue cleanup */
3943 ns->disk->queue = NULL;
3946 mutex_lock(&ctrl->subsys->lock);
3947 list_del_rcu(&ns->siblings);
3948 if (list_empty(&ns->head->list))
3949 list_del_init(&ns->head->entry);
3950 mutex_unlock(&ctrl->subsys->lock);
3951 nvme_put_ns_head(ns->head);
3953 blk_cleanup_queue(ns->queue);
3960 static void nvme_ns_remove(struct nvme_ns *ns)
3962 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3965 set_capacity(ns->disk, 0);
3966 nvme_fault_inject_fini(&ns->fault_inject);
3968 mutex_lock(&ns->ctrl->subsys->lock);
3969 list_del_rcu(&ns->siblings);
3970 if (list_empty(&ns->head->list))
3971 list_del_init(&ns->head->entry);
3972 mutex_unlock(&ns->ctrl->subsys->lock);
3974 synchronize_rcu(); /* guarantee not available in head->list */
3975 nvme_mpath_clear_current_path(ns);
3976 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
3978 if (ns->disk->flags & GENHD_FL_UP) {
3979 del_gendisk(ns->disk);
3980 blk_cleanup_queue(ns->queue);
3981 if (blk_get_integrity(ns->disk))
3982 blk_integrity_unregister(ns->disk);
3985 down_write(&ns->ctrl->namespaces_rwsem);
3986 list_del_init(&ns->list);
3987 up_write(&ns->ctrl->namespaces_rwsem);
3989 nvme_mpath_check_last_path(ns);
3993 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3995 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4003 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4005 struct nvme_id_ns *id;
4008 if (test_bit(NVME_NS_DEAD, &ns->flags))
4011 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4016 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4017 dev_err(ns->ctrl->device,
4018 "identifiers changed for nsid %d\n", ns->head->ns_id);
4022 ret = nvme_update_ns_info(ns, id);
4028 * Only remove the namespace if we got a fatal error back from the
4029 * device, otherwise ignore the error and just move on.
4031 * TODO: we should probably schedule a delayed retry here.
4033 if (ret && ret != -ENOMEM && !(ret > 0 && !(ret & NVME_SC_DNR)))
4037 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4039 struct nvme_ns_ids ids = { };
4042 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4045 ns = nvme_find_get_ns(ctrl, nsid);
4047 nvme_validate_ns(ns, &ids);
4054 nvme_alloc_ns(ctrl, nsid, &ids);
4057 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4058 dev_warn(ctrl->device,
4059 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4063 nvme_alloc_ns(ctrl, nsid, &ids);
4066 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4072 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4075 struct nvme_ns *ns, *next;
4078 down_write(&ctrl->namespaces_rwsem);
4079 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4080 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4081 list_move_tail(&ns->list, &rm_list);
4083 up_write(&ctrl->namespaces_rwsem);
4085 list_for_each_entry_safe(ns, next, &rm_list, list)
4090 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4092 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4097 if (nvme_ctrl_limited_cns(ctrl))
4100 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4105 struct nvme_command cmd = {
4106 .identify.opcode = nvme_admin_identify,
4107 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4108 .identify.nsid = cpu_to_le32(prev),
4111 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4112 NVME_IDENTIFY_DATA_SIZE);
4114 dev_warn(ctrl->device,
4115 "Identify NS List failed (status=0x%x)\n", ret);
4119 for (i = 0; i < nr_entries; i++) {
4120 u32 nsid = le32_to_cpu(ns_list[i]);
4122 if (!nsid) /* end of the list? */
4124 nvme_validate_or_alloc_ns(ctrl, nsid);
4125 while (++prev < nsid)
4126 nvme_ns_remove_by_nsid(ctrl, prev);
4130 nvme_remove_invalid_namespaces(ctrl, prev);
4136 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4138 struct nvme_id_ctrl *id;
4141 if (nvme_identify_ctrl(ctrl, &id))
4143 nn = le32_to_cpu(id->nn);
4146 for (i = 1; i <= nn; i++)
4147 nvme_validate_or_alloc_ns(ctrl, i);
4149 nvme_remove_invalid_namespaces(ctrl, nn);
4152 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4154 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4158 log = kzalloc(log_size, GFP_KERNEL);
4163 * We need to read the log to clear the AEN, but we don't want to rely
4164 * on it for the changed namespace information as userspace could have
4165 * raced with us in reading the log page, which could cause us to miss
4168 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4169 NVME_CSI_NVM, log, log_size, 0);
4171 dev_warn(ctrl->device,
4172 "reading changed ns log failed: %d\n", error);
4177 static void nvme_scan_work(struct work_struct *work)
4179 struct nvme_ctrl *ctrl =
4180 container_of(work, struct nvme_ctrl, scan_work);
4182 /* No tagset on a live ctrl means IO queues could not created */
4183 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4186 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4187 dev_info(ctrl->device, "rescanning namespaces.\n");
4188 nvme_clear_changed_ns_log(ctrl);
4191 mutex_lock(&ctrl->scan_lock);
4192 if (nvme_scan_ns_list(ctrl) != 0)
4193 nvme_scan_ns_sequential(ctrl);
4194 mutex_unlock(&ctrl->scan_lock);
4196 down_write(&ctrl->namespaces_rwsem);
4197 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4198 up_write(&ctrl->namespaces_rwsem);
4202 * This function iterates the namespace list unlocked to allow recovery from
4203 * controller failure. It is up to the caller to ensure the namespace list is
4204 * not modified by scan work while this function is executing.
4206 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4208 struct nvme_ns *ns, *next;
4212 * make sure to requeue I/O to all namespaces as these
4213 * might result from the scan itself and must complete
4214 * for the scan_work to make progress
4216 nvme_mpath_clear_ctrl_paths(ctrl);
4218 /* prevent racing with ns scanning */
4219 flush_work(&ctrl->scan_work);
4222 * The dead states indicates the controller was not gracefully
4223 * disconnected. In that case, we won't be able to flush any data while
4224 * removing the namespaces' disks; fail all the queues now to avoid
4225 * potentially having to clean up the failed sync later.
4227 if (ctrl->state == NVME_CTRL_DEAD)
4228 nvme_kill_queues(ctrl);
4230 /* this is a no-op when called from the controller reset handler */
4231 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4233 down_write(&ctrl->namespaces_rwsem);
4234 list_splice_init(&ctrl->namespaces, &ns_list);
4235 up_write(&ctrl->namespaces_rwsem);
4237 list_for_each_entry_safe(ns, next, &ns_list, list)
4240 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4242 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4244 struct nvme_ctrl *ctrl =
4245 container_of(dev, struct nvme_ctrl, ctrl_device);
4246 struct nvmf_ctrl_options *opts = ctrl->opts;
4249 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4254 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4258 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4259 opts->trsvcid ?: "none");
4263 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4264 opts->host_traddr ?: "none");
4269 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4271 char *envp[2] = { NULL, NULL };
4272 u32 aen_result = ctrl->aen_result;
4274 ctrl->aen_result = 0;
4278 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4281 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4285 static void nvme_async_event_work(struct work_struct *work)
4287 struct nvme_ctrl *ctrl =
4288 container_of(work, struct nvme_ctrl, async_event_work);
4290 nvme_aen_uevent(ctrl);
4291 ctrl->ops->submit_async_event(ctrl);
4294 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4299 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4305 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4308 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4310 struct nvme_fw_slot_info_log *log;
4312 log = kmalloc(sizeof(*log), GFP_KERNEL);
4316 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4317 log, sizeof(*log), 0))
4318 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4322 static void nvme_fw_act_work(struct work_struct *work)
4324 struct nvme_ctrl *ctrl = container_of(work,
4325 struct nvme_ctrl, fw_act_work);
4326 unsigned long fw_act_timeout;
4329 fw_act_timeout = jiffies +
4330 msecs_to_jiffies(ctrl->mtfa * 100);
4332 fw_act_timeout = jiffies +
4333 msecs_to_jiffies(admin_timeout * 1000);
4335 nvme_stop_queues(ctrl);
4336 while (nvme_ctrl_pp_status(ctrl)) {
4337 if (time_after(jiffies, fw_act_timeout)) {
4338 dev_warn(ctrl->device,
4339 "Fw activation timeout, reset controller\n");
4340 nvme_try_sched_reset(ctrl);
4346 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4349 nvme_start_queues(ctrl);
4350 /* read FW slot information to clear the AER */
4351 nvme_get_fw_slot_info(ctrl);
4354 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4356 u32 aer_notice_type = (result & 0xff00) >> 8;
4358 trace_nvme_async_event(ctrl, aer_notice_type);
4360 switch (aer_notice_type) {
4361 case NVME_AER_NOTICE_NS_CHANGED:
4362 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4363 nvme_queue_scan(ctrl);
4365 case NVME_AER_NOTICE_FW_ACT_STARTING:
4367 * We are (ab)using the RESETTING state to prevent subsequent
4368 * recovery actions from interfering with the controller's
4369 * firmware activation.
4371 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4372 queue_work(nvme_wq, &ctrl->fw_act_work);
4374 #ifdef CONFIG_NVME_MULTIPATH
4375 case NVME_AER_NOTICE_ANA:
4376 if (!ctrl->ana_log_buf)
4378 queue_work(nvme_wq, &ctrl->ana_work);
4381 case NVME_AER_NOTICE_DISC_CHANGED:
4382 ctrl->aen_result = result;
4385 dev_warn(ctrl->device, "async event result %08x\n", result);
4389 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4390 volatile union nvme_result *res)
4392 u32 result = le32_to_cpu(res->u32);
4393 u32 aer_type = result & 0x07;
4395 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4399 case NVME_AER_NOTICE:
4400 nvme_handle_aen_notice(ctrl, result);
4402 case NVME_AER_ERROR:
4403 case NVME_AER_SMART:
4406 trace_nvme_async_event(ctrl, aer_type);
4407 ctrl->aen_result = result;
4412 queue_work(nvme_wq, &ctrl->async_event_work);
4414 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4416 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4418 nvme_mpath_stop(ctrl);
4419 nvme_stop_keep_alive(ctrl);
4420 nvme_stop_failfast_work(ctrl);
4421 flush_work(&ctrl->async_event_work);
4422 cancel_work_sync(&ctrl->fw_act_work);
4424 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4426 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4428 nvme_start_keep_alive(ctrl);
4430 nvme_enable_aen(ctrl);
4432 if (ctrl->queue_count > 1) {
4433 nvme_queue_scan(ctrl);
4434 nvme_start_queues(ctrl);
4437 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4439 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4441 nvme_fault_inject_fini(&ctrl->fault_inject);
4442 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4443 cdev_device_del(&ctrl->cdev, ctrl->device);
4444 nvme_put_ctrl(ctrl);
4446 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4448 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4450 struct nvme_effects_log *cel;
4453 xa_for_each (&ctrl->cels, i, cel) {
4454 xa_erase(&ctrl->cels, i);
4458 xa_destroy(&ctrl->cels);
4461 static void nvme_free_ctrl(struct device *dev)
4463 struct nvme_ctrl *ctrl =
4464 container_of(dev, struct nvme_ctrl, ctrl_device);
4465 struct nvme_subsystem *subsys = ctrl->subsys;
4467 if (!subsys || ctrl->instance != subsys->instance)
4468 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4470 nvme_free_cels(ctrl);
4471 nvme_mpath_uninit(ctrl);
4472 __free_page(ctrl->discard_page);
4475 mutex_lock(&nvme_subsystems_lock);
4476 list_del(&ctrl->subsys_entry);
4477 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4478 mutex_unlock(&nvme_subsystems_lock);
4481 ctrl->ops->free_ctrl(ctrl);
4484 nvme_put_subsystem(subsys);
4488 * Initialize a NVMe controller structures. This needs to be called during
4489 * earliest initialization so that we have the initialized structured around
4492 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4493 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4497 ctrl->state = NVME_CTRL_NEW;
4498 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4499 spin_lock_init(&ctrl->lock);
4500 mutex_init(&ctrl->scan_lock);
4501 INIT_LIST_HEAD(&ctrl->namespaces);
4502 xa_init(&ctrl->cels);
4503 init_rwsem(&ctrl->namespaces_rwsem);
4506 ctrl->quirks = quirks;
4507 ctrl->numa_node = NUMA_NO_NODE;
4508 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4509 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4510 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4511 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4512 init_waitqueue_head(&ctrl->state_wq);
4514 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4515 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4516 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4517 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4519 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4521 ctrl->discard_page = alloc_page(GFP_KERNEL);
4522 if (!ctrl->discard_page) {
4527 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4530 ctrl->instance = ret;
4532 device_initialize(&ctrl->ctrl_device);
4533 ctrl->device = &ctrl->ctrl_device;
4534 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4536 ctrl->device->class = nvme_class;
4537 ctrl->device->parent = ctrl->dev;
4538 ctrl->device->groups = nvme_dev_attr_groups;
4539 ctrl->device->release = nvme_free_ctrl;
4540 dev_set_drvdata(ctrl->device, ctrl);
4541 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4543 goto out_release_instance;
4545 nvme_get_ctrl(ctrl);
4546 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4547 ctrl->cdev.owner = ops->module;
4548 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4553 * Initialize latency tolerance controls. The sysfs files won't
4554 * be visible to userspace unless the device actually supports APST.
4556 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4557 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4558 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4560 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4564 nvme_put_ctrl(ctrl);
4565 kfree_const(ctrl->device->kobj.name);
4566 out_release_instance:
4567 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4569 if (ctrl->discard_page)
4570 __free_page(ctrl->discard_page);
4573 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4576 * nvme_kill_queues(): Ends all namespace queues
4577 * @ctrl: the dead controller that needs to end
4579 * Call this function when the driver determines it is unable to get the
4580 * controller in a state capable of servicing IO.
4582 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4586 down_read(&ctrl->namespaces_rwsem);
4588 /* Forcibly unquiesce queues to avoid blocking dispatch */
4589 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4590 blk_mq_unquiesce_queue(ctrl->admin_q);
4592 list_for_each_entry(ns, &ctrl->namespaces, list)
4593 nvme_set_queue_dying(ns);
4595 up_read(&ctrl->namespaces_rwsem);
4597 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4599 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4603 down_read(&ctrl->namespaces_rwsem);
4604 list_for_each_entry(ns, &ctrl->namespaces, list)
4605 blk_mq_unfreeze_queue(ns->queue);
4606 up_read(&ctrl->namespaces_rwsem);
4608 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4610 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4614 down_read(&ctrl->namespaces_rwsem);
4615 list_for_each_entry(ns, &ctrl->namespaces, list) {
4616 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4620 up_read(&ctrl->namespaces_rwsem);
4623 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4625 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4629 down_read(&ctrl->namespaces_rwsem);
4630 list_for_each_entry(ns, &ctrl->namespaces, list)
4631 blk_mq_freeze_queue_wait(ns->queue);
4632 up_read(&ctrl->namespaces_rwsem);
4634 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4636 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4640 down_read(&ctrl->namespaces_rwsem);
4641 list_for_each_entry(ns, &ctrl->namespaces, list)
4642 blk_freeze_queue_start(ns->queue);
4643 up_read(&ctrl->namespaces_rwsem);
4645 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4647 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4651 down_read(&ctrl->namespaces_rwsem);
4652 list_for_each_entry(ns, &ctrl->namespaces, list)
4653 blk_mq_quiesce_queue(ns->queue);
4654 up_read(&ctrl->namespaces_rwsem);
4656 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4658 void nvme_start_queues(struct nvme_ctrl *ctrl)
4662 down_read(&ctrl->namespaces_rwsem);
4663 list_for_each_entry(ns, &ctrl->namespaces, list)
4664 blk_mq_unquiesce_queue(ns->queue);
4665 up_read(&ctrl->namespaces_rwsem);
4667 EXPORT_SYMBOL_GPL(nvme_start_queues);
4669 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4673 down_read(&ctrl->namespaces_rwsem);
4674 list_for_each_entry(ns, &ctrl->namespaces, list)
4675 blk_sync_queue(ns->queue);
4676 up_read(&ctrl->namespaces_rwsem);
4678 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4680 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4682 nvme_sync_io_queues(ctrl);
4684 blk_sync_queue(ctrl->admin_q);
4686 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4688 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4690 if (file->f_op != &nvme_dev_fops)
4692 return file->private_data;
4694 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4697 * Check we didn't inadvertently grow the command structure sizes:
4699 static inline void _nvme_check_size(void)
4701 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4702 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4703 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4704 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4705 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4706 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4707 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4708 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4709 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4710 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4711 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4712 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4713 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4714 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4715 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4716 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4717 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4718 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4719 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4723 static int __init nvme_core_init(void)
4725 int result = -ENOMEM;
4729 nvme_wq = alloc_workqueue("nvme-wq",
4730 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4734 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4735 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4739 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4740 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4741 if (!nvme_delete_wq)
4742 goto destroy_reset_wq;
4744 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4745 NVME_MINORS, "nvme");
4747 goto destroy_delete_wq;
4749 nvme_class = class_create(THIS_MODULE, "nvme");
4750 if (IS_ERR(nvme_class)) {
4751 result = PTR_ERR(nvme_class);
4752 goto unregister_chrdev;
4754 nvme_class->dev_uevent = nvme_class_uevent;
4756 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4757 if (IS_ERR(nvme_subsys_class)) {
4758 result = PTR_ERR(nvme_subsys_class);
4764 class_destroy(nvme_class);
4766 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4768 destroy_workqueue(nvme_delete_wq);
4770 destroy_workqueue(nvme_reset_wq);
4772 destroy_workqueue(nvme_wq);
4777 static void __exit nvme_core_exit(void)
4779 class_destroy(nvme_subsys_class);
4780 class_destroy(nvme_class);
4781 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4782 destroy_workqueue(nvme_delete_wq);
4783 destroy_workqueue(nvme_reset_wq);
4784 destroy_workqueue(nvme_wq);
4785 ida_destroy(&nvme_instance_ida);
4788 MODULE_LICENSE("GPL");
4789 MODULE_VERSION("1.0");
4790 module_init(nvme_core_init);
4791 module_exit(nvme_core_exit);