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_chr_devt;
89 static struct class *nvme_class;
90 static struct class *nvme_subsys_class;
92 static int _nvme_revalidate_disk(struct gendisk *disk);
93 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
94 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
97 static void nvme_update_bdev_size(struct gendisk *disk)
99 struct block_device *bdev = bdget_disk(disk, 0);
102 bd_set_nr_sectors(bdev, get_capacity(disk));
108 * Prepare a queue for teardown.
110 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
111 * the capacity to 0 after that to avoid blocking dispatchers that may be
112 * holding bd_butex. This will end buffered writers dirtying pages that can't
115 static void nvme_set_queue_dying(struct nvme_ns *ns)
117 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
120 blk_set_queue_dying(ns->queue);
121 blk_mq_unquiesce_queue(ns->queue);
123 set_capacity(ns->disk, 0);
124 nvme_update_bdev_size(ns->disk);
127 static void nvme_queue_scan(struct nvme_ctrl *ctrl)
130 * Only new queue scan work when admin and IO queues are both alive
132 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
133 queue_work(nvme_wq, &ctrl->scan_work);
137 * Use this function to proceed with scheduling reset_work for a controller
138 * that had previously been set to the resetting state. This is intended for
139 * code paths that can't be interrupted by other reset attempts. A hot removal
140 * may prevent this from succeeding.
142 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
144 if (ctrl->state != NVME_CTRL_RESETTING)
146 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
150 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
152 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
154 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
156 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
160 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
162 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
166 ret = nvme_reset_ctrl(ctrl);
168 flush_work(&ctrl->reset_work);
169 if (ctrl->state != NVME_CTRL_LIVE)
175 EXPORT_SYMBOL_GPL(nvme_reset_ctrl_sync);
177 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
179 dev_info(ctrl->device,
180 "Removing ctrl: NQN \"%s\"\n", ctrl->opts->subsysnqn);
182 flush_work(&ctrl->reset_work);
183 nvme_stop_ctrl(ctrl);
184 nvme_remove_namespaces(ctrl);
185 ctrl->ops->delete_ctrl(ctrl);
186 nvme_uninit_ctrl(ctrl);
189 static void nvme_delete_ctrl_work(struct work_struct *work)
191 struct nvme_ctrl *ctrl =
192 container_of(work, struct nvme_ctrl, delete_work);
194 nvme_do_delete_ctrl(ctrl);
197 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
199 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
201 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
205 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
207 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
210 * Keep a reference until nvme_do_delete_ctrl() complete,
211 * since ->delete_ctrl can free the controller.
214 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
215 nvme_do_delete_ctrl(ctrl);
219 static blk_status_t nvme_error_status(u16 status)
221 switch (status & 0x7ff) {
222 case NVME_SC_SUCCESS:
224 case NVME_SC_CAP_EXCEEDED:
225 return BLK_STS_NOSPC;
226 case NVME_SC_LBA_RANGE:
227 case NVME_SC_CMD_INTERRUPTED:
228 case NVME_SC_NS_NOT_READY:
229 return BLK_STS_TARGET;
230 case NVME_SC_BAD_ATTRIBUTES:
231 case NVME_SC_ONCS_NOT_SUPPORTED:
232 case NVME_SC_INVALID_OPCODE:
233 case NVME_SC_INVALID_FIELD:
234 case NVME_SC_INVALID_NS:
235 return BLK_STS_NOTSUPP;
236 case NVME_SC_WRITE_FAULT:
237 case NVME_SC_READ_ERROR:
238 case NVME_SC_UNWRITTEN_BLOCK:
239 case NVME_SC_ACCESS_DENIED:
240 case NVME_SC_READ_ONLY:
241 case NVME_SC_COMPARE_FAILED:
242 return BLK_STS_MEDIUM;
243 case NVME_SC_GUARD_CHECK:
244 case NVME_SC_APPTAG_CHECK:
245 case NVME_SC_REFTAG_CHECK:
246 case NVME_SC_INVALID_PI:
247 return BLK_STS_PROTECTION;
248 case NVME_SC_RESERVATION_CONFLICT:
249 return BLK_STS_NEXUS;
250 case NVME_SC_HOST_PATH_ERROR:
251 return BLK_STS_TRANSPORT;
253 return BLK_STS_IOERR;
257 static void nvme_retry_req(struct request *req)
259 struct nvme_ns *ns = req->q->queuedata;
260 unsigned long delay = 0;
263 /* The mask and shift result must be <= 3 */
264 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
266 delay = ns->ctrl->crdt[crd - 1] * 100;
268 nvme_req(req)->retries++;
269 blk_mq_requeue_request(req, false);
270 blk_mq_delay_kick_requeue_list(req->q, delay);
273 enum nvme_disposition {
279 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
281 if (likely(nvme_req(req)->status == 0))
284 if (blk_noretry_request(req) ||
285 (nvme_req(req)->status & NVME_SC_DNR) ||
286 nvme_req(req)->retries >= nvme_max_retries)
289 if (req->cmd_flags & REQ_NVME_MPATH) {
290 if (nvme_is_path_error(nvme_req(req)->status) ||
291 blk_queue_dying(req->q))
294 if (blk_queue_dying(req->q))
301 static inline void nvme_end_req(struct request *req)
303 blk_status_t status = nvme_error_status(nvme_req(req)->status);
305 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
306 req_op(req) == REQ_OP_ZONE_APPEND)
307 req->__sector = nvme_lba_to_sect(req->q->queuedata,
308 le64_to_cpu(nvme_req(req)->result.u64));
310 nvme_trace_bio_complete(req, status);
311 blk_mq_end_request(req, status);
314 void nvme_complete_rq(struct request *req)
316 trace_nvme_complete_rq(req);
317 nvme_cleanup_cmd(req);
319 if (nvme_req(req)->ctrl->kas)
320 nvme_req(req)->ctrl->comp_seen = true;
322 switch (nvme_decide_disposition(req)) {
330 nvme_failover_req(req);
334 EXPORT_SYMBOL_GPL(nvme_complete_rq);
336 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
338 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
339 "Cancelling I/O %d", req->tag);
341 /* don't abort one completed request */
342 if (blk_mq_request_completed(req))
345 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
346 blk_mq_complete_request(req);
349 EXPORT_SYMBOL_GPL(nvme_cancel_request);
351 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
352 enum nvme_ctrl_state new_state)
354 enum nvme_ctrl_state old_state;
356 bool changed = false;
358 spin_lock_irqsave(&ctrl->lock, flags);
360 old_state = ctrl->state;
365 case NVME_CTRL_RESETTING:
366 case NVME_CTRL_CONNECTING:
373 case NVME_CTRL_RESETTING:
383 case NVME_CTRL_CONNECTING:
386 case NVME_CTRL_RESETTING:
393 case NVME_CTRL_DELETING:
396 case NVME_CTRL_RESETTING:
397 case NVME_CTRL_CONNECTING:
404 case NVME_CTRL_DELETING_NOIO:
406 case NVME_CTRL_DELETING:
416 case NVME_CTRL_DELETING:
428 ctrl->state = new_state;
429 wake_up_all(&ctrl->state_wq);
432 spin_unlock_irqrestore(&ctrl->lock, flags);
433 if (changed && ctrl->state == NVME_CTRL_LIVE)
434 nvme_kick_requeue_lists(ctrl);
437 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
440 * Returns true for sink states that can't ever transition back to live.
442 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
444 switch (ctrl->state) {
447 case NVME_CTRL_RESETTING:
448 case NVME_CTRL_CONNECTING:
450 case NVME_CTRL_DELETING:
451 case NVME_CTRL_DELETING_NOIO:
455 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
461 * Waits for the controller state to be resetting, or returns false if it is
462 * not possible to ever transition to that state.
464 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
466 wait_event(ctrl->state_wq,
467 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
468 nvme_state_terminal(ctrl));
469 return ctrl->state == NVME_CTRL_RESETTING;
471 EXPORT_SYMBOL_GPL(nvme_wait_reset);
473 static void nvme_free_ns_head(struct kref *ref)
475 struct nvme_ns_head *head =
476 container_of(ref, struct nvme_ns_head, ref);
478 nvme_mpath_remove_disk(head);
479 ida_simple_remove(&head->subsys->ns_ida, head->instance);
480 cleanup_srcu_struct(&head->srcu);
481 nvme_put_subsystem(head->subsys);
485 static void nvme_put_ns_head(struct nvme_ns_head *head)
487 kref_put(&head->ref, nvme_free_ns_head);
490 static void nvme_free_ns(struct kref *kref)
492 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
495 nvme_nvm_unregister(ns);
498 nvme_put_ns_head(ns->head);
499 nvme_put_ctrl(ns->ctrl);
503 void nvme_put_ns(struct nvme_ns *ns)
505 kref_put(&ns->kref, nvme_free_ns);
507 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
509 static inline void nvme_clear_nvme_request(struct request *req)
511 if (!(req->rq_flags & RQF_DONTPREP)) {
512 nvme_req(req)->retries = 0;
513 nvme_req(req)->flags = 0;
514 req->rq_flags |= RQF_DONTPREP;
518 struct request *nvme_alloc_request(struct request_queue *q,
519 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
521 unsigned op = nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
524 if (qid == NVME_QID_ANY) {
525 req = blk_mq_alloc_request(q, op, flags);
527 req = blk_mq_alloc_request_hctx(q, op, flags,
533 req->cmd_flags |= REQ_FAILFAST_DRIVER;
534 nvme_clear_nvme_request(req);
535 nvme_req(req)->cmd = cmd;
539 EXPORT_SYMBOL_GPL(nvme_alloc_request);
541 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
543 struct nvme_command c;
545 memset(&c, 0, sizeof(c));
547 c.directive.opcode = nvme_admin_directive_send;
548 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
549 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
550 c.directive.dtype = NVME_DIR_IDENTIFY;
551 c.directive.tdtype = NVME_DIR_STREAMS;
552 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
554 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
557 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
559 return nvme_toggle_streams(ctrl, false);
562 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
564 return nvme_toggle_streams(ctrl, true);
567 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
568 struct streams_directive_params *s, u32 nsid)
570 struct nvme_command c;
572 memset(&c, 0, sizeof(c));
573 memset(s, 0, sizeof(*s));
575 c.directive.opcode = nvme_admin_directive_recv;
576 c.directive.nsid = cpu_to_le32(nsid);
577 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
578 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
579 c.directive.dtype = NVME_DIR_STREAMS;
581 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
584 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
586 struct streams_directive_params s;
589 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
594 ret = nvme_enable_streams(ctrl);
598 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
600 goto out_disable_stream;
602 ctrl->nssa = le16_to_cpu(s.nssa);
603 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
604 dev_info(ctrl->device, "too few streams (%u) available\n",
606 goto out_disable_stream;
609 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
610 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
614 nvme_disable_streams(ctrl);
619 * Check if 'req' has a write hint associated with it. If it does, assign
620 * a valid namespace stream to the write.
622 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
623 struct request *req, u16 *control,
626 enum rw_hint streamid = req->write_hint;
628 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
632 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
635 *control |= NVME_RW_DTYPE_STREAMS;
636 *dsmgmt |= streamid << 16;
639 if (streamid < ARRAY_SIZE(req->q->write_hints))
640 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
643 static void nvme_setup_passthrough(struct request *req,
644 struct nvme_command *cmd)
646 memcpy(cmd, nvme_req(req)->cmd, sizeof(*cmd));
647 /* passthru commands should let the driver set the SGL flags */
648 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
651 static inline void nvme_setup_flush(struct nvme_ns *ns,
652 struct nvme_command *cmnd)
654 cmnd->common.opcode = nvme_cmd_flush;
655 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
658 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
659 struct nvme_command *cmnd)
661 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
662 struct nvme_dsm_range *range;
666 * Some devices do not consider the DSM 'Number of Ranges' field when
667 * determining how much data to DMA. Always allocate memory for maximum
668 * number of segments to prevent device reading beyond end of buffer.
670 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
672 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
675 * If we fail allocation our range, fallback to the controller
676 * discard page. If that's also busy, it's safe to return
677 * busy, as we know we can make progress once that's freed.
679 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
680 return BLK_STS_RESOURCE;
682 range = page_address(ns->ctrl->discard_page);
685 __rq_for_each_bio(bio, req) {
686 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
687 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
690 range[n].cattr = cpu_to_le32(0);
691 range[n].nlb = cpu_to_le32(nlb);
692 range[n].slba = cpu_to_le64(slba);
697 if (WARN_ON_ONCE(n != segments)) {
698 if (virt_to_page(range) == ns->ctrl->discard_page)
699 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
702 return BLK_STS_IOERR;
705 cmnd->dsm.opcode = nvme_cmd_dsm;
706 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
707 cmnd->dsm.nr = cpu_to_le32(segments - 1);
708 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
710 req->special_vec.bv_page = virt_to_page(range);
711 req->special_vec.bv_offset = offset_in_page(range);
712 req->special_vec.bv_len = alloc_size;
713 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
718 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
719 struct request *req, struct nvme_command *cmnd)
721 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
722 return nvme_setup_discard(ns, req, cmnd);
724 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
725 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
726 cmnd->write_zeroes.slba =
727 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
728 cmnd->write_zeroes.length =
729 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
730 cmnd->write_zeroes.control = 0;
734 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
735 struct request *req, struct nvme_command *cmnd,
738 struct nvme_ctrl *ctrl = ns->ctrl;
742 if (req->cmd_flags & REQ_FUA)
743 control |= NVME_RW_FUA;
744 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
745 control |= NVME_RW_LR;
747 if (req->cmd_flags & REQ_RAHEAD)
748 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
750 cmnd->rw.opcode = op;
751 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
752 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
753 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
755 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
756 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
760 * If formated with metadata, the block layer always provides a
761 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
762 * we enable the PRACT bit for protection information or set the
763 * namespace capacity to zero to prevent any I/O.
765 if (!blk_integrity_rq(req)) {
766 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
767 return BLK_STS_NOTSUPP;
768 control |= NVME_RW_PRINFO_PRACT;
771 switch (ns->pi_type) {
772 case NVME_NS_DPS_PI_TYPE3:
773 control |= NVME_RW_PRINFO_PRCHK_GUARD;
775 case NVME_NS_DPS_PI_TYPE1:
776 case NVME_NS_DPS_PI_TYPE2:
777 control |= NVME_RW_PRINFO_PRCHK_GUARD |
778 NVME_RW_PRINFO_PRCHK_REF;
779 if (op == nvme_cmd_zone_append)
780 control |= NVME_RW_APPEND_PIREMAP;
781 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
786 cmnd->rw.control = cpu_to_le16(control);
787 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
791 void nvme_cleanup_cmd(struct request *req)
793 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
794 struct nvme_ns *ns = req->rq_disk->private_data;
795 struct page *page = req->special_vec.bv_page;
797 if (page == ns->ctrl->discard_page)
798 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
800 kfree(page_address(page) + req->special_vec.bv_offset);
803 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
805 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req,
806 struct nvme_command *cmd)
808 blk_status_t ret = BLK_STS_OK;
810 nvme_clear_nvme_request(req);
812 memset(cmd, 0, sizeof(*cmd));
813 switch (req_op(req)) {
816 nvme_setup_passthrough(req, cmd);
819 nvme_setup_flush(ns, cmd);
821 case REQ_OP_ZONE_RESET_ALL:
822 case REQ_OP_ZONE_RESET:
823 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
825 case REQ_OP_ZONE_OPEN:
826 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
828 case REQ_OP_ZONE_CLOSE:
829 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
831 case REQ_OP_ZONE_FINISH:
832 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
834 case REQ_OP_WRITE_ZEROES:
835 ret = nvme_setup_write_zeroes(ns, req, cmd);
838 ret = nvme_setup_discard(ns, req, cmd);
841 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
844 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
846 case REQ_OP_ZONE_APPEND:
847 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
851 return BLK_STS_IOERR;
854 cmd->common.command_id = req->tag;
855 trace_nvme_setup_cmd(req, cmd);
858 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
860 static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
862 struct completion *waiting = rq->end_io_data;
864 rq->end_io_data = NULL;
868 static void nvme_execute_rq_polled(struct request_queue *q,
869 struct gendisk *bd_disk, struct request *rq, int at_head)
871 DECLARE_COMPLETION_ONSTACK(wait);
873 WARN_ON_ONCE(!test_bit(QUEUE_FLAG_POLL, &q->queue_flags));
875 rq->cmd_flags |= REQ_HIPRI;
876 rq->end_io_data = &wait;
877 blk_execute_rq_nowait(q, bd_disk, rq, at_head, nvme_end_sync_rq);
879 while (!completion_done(&wait)) {
880 blk_poll(q, request_to_qc_t(rq->mq_hctx, rq), true);
886 * Returns 0 on success. If the result is negative, it's a Linux error code;
887 * if the result is positive, it's an NVM Express status code
889 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
890 union nvme_result *result, void *buffer, unsigned bufflen,
891 unsigned timeout, int qid, int at_head,
892 blk_mq_req_flags_t flags, bool poll)
897 req = nvme_alloc_request(q, cmd, flags, qid);
901 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
903 if (buffer && bufflen) {
904 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
910 nvme_execute_rq_polled(req->q, NULL, req, at_head);
912 blk_execute_rq(req->q, NULL, req, at_head);
914 *result = nvme_req(req)->result;
915 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
918 ret = nvme_req(req)->status;
920 blk_mq_free_request(req);
923 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
925 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
926 void *buffer, unsigned bufflen)
928 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
929 NVME_QID_ANY, 0, 0, false);
931 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
933 static void *nvme_add_user_metadata(struct bio *bio, void __user *ubuf,
934 unsigned len, u32 seed, bool write)
936 struct bio_integrity_payload *bip;
940 buf = kmalloc(len, GFP_KERNEL);
945 if (write && copy_from_user(buf, ubuf, len))
948 bip = bio_integrity_alloc(bio, GFP_KERNEL, 1);
954 bip->bip_iter.bi_size = len;
955 bip->bip_iter.bi_sector = seed;
956 ret = bio_integrity_add_page(bio, virt_to_page(buf), len,
957 offset_in_page(buf));
967 static u32 nvme_known_admin_effects(u8 opcode)
970 case nvme_admin_format_nvm:
971 return NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC |
972 NVME_CMD_EFFECTS_CSE_MASK;
973 case nvme_admin_sanitize_nvm:
974 return NVME_CMD_EFFECTS_CSE_MASK;
981 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
986 if (ns->head->effects)
987 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
988 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
989 dev_warn(ctrl->device,
990 "IO command:%02x has unhandled effects:%08x\n",
996 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
997 effects |= nvme_known_admin_effects(opcode);
1001 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1003 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1006 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1009 * For simplicity, IO to all namespaces is quiesced even if the command
1010 * effects say only one namespace is affected.
1012 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1013 mutex_lock(&ctrl->scan_lock);
1014 mutex_lock(&ctrl->subsys->lock);
1015 nvme_mpath_start_freeze(ctrl->subsys);
1016 nvme_mpath_wait_freeze(ctrl->subsys);
1017 nvme_start_freeze(ctrl);
1018 nvme_wait_freeze(ctrl);
1023 static void nvme_update_formats(struct nvme_ctrl *ctrl, u32 *effects)
1027 down_read(&ctrl->namespaces_rwsem);
1028 list_for_each_entry(ns, &ctrl->namespaces, list)
1029 if (_nvme_revalidate_disk(ns->disk))
1030 nvme_set_queue_dying(ns);
1031 else if (blk_queue_is_zoned(ns->disk->queue)) {
1033 * IO commands are required to fully revalidate a zoned
1034 * device. Force the command effects to trigger rescan
1035 * work so report zones can run in a context with
1036 * unfrozen IO queues.
1038 *effects |= NVME_CMD_EFFECTS_NCC;
1040 up_read(&ctrl->namespaces_rwsem);
1043 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects)
1046 * Revalidate LBA changes prior to unfreezing. This is necessary to
1047 * prevent memory corruption if a logical block size was changed by
1050 if (effects & NVME_CMD_EFFECTS_LBCC)
1051 nvme_update_formats(ctrl, &effects);
1052 if (effects & (NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK)) {
1053 nvme_unfreeze(ctrl);
1054 nvme_mpath_unfreeze(ctrl->subsys);
1055 mutex_unlock(&ctrl->subsys->lock);
1056 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1057 mutex_unlock(&ctrl->scan_lock);
1059 if (effects & NVME_CMD_EFFECTS_CCC)
1060 nvme_init_identify(ctrl);
1061 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1062 nvme_queue_scan(ctrl);
1063 flush_work(&ctrl->scan_work);
1067 void nvme_execute_passthru_rq(struct request *rq)
1069 struct nvme_command *cmd = nvme_req(rq)->cmd;
1070 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1071 struct nvme_ns *ns = rq->q->queuedata;
1072 struct gendisk *disk = ns ? ns->disk : NULL;
1075 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1076 blk_execute_rq(rq->q, disk, rq, 0);
1077 nvme_passthru_end(ctrl, effects);
1079 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1081 static int nvme_submit_user_cmd(struct request_queue *q,
1082 struct nvme_command *cmd, void __user *ubuffer,
1083 unsigned bufflen, void __user *meta_buffer, unsigned meta_len,
1084 u32 meta_seed, u64 *result, unsigned timeout)
1086 bool write = nvme_is_write(cmd);
1087 struct nvme_ns *ns = q->queuedata;
1088 struct gendisk *disk = ns ? ns->disk : NULL;
1089 struct request *req;
1090 struct bio *bio = NULL;
1094 req = nvme_alloc_request(q, cmd, 0, NVME_QID_ANY);
1096 return PTR_ERR(req);
1098 req->timeout = timeout ? timeout : ADMIN_TIMEOUT;
1099 nvme_req(req)->flags |= NVME_REQ_USERCMD;
1101 if (ubuffer && bufflen) {
1102 ret = blk_rq_map_user(q, req, NULL, ubuffer, bufflen,
1107 bio->bi_disk = disk;
1108 if (disk && meta_buffer && meta_len) {
1109 meta = nvme_add_user_metadata(bio, meta_buffer, meta_len,
1112 ret = PTR_ERR(meta);
1115 req->cmd_flags |= REQ_INTEGRITY;
1119 nvme_execute_passthru_rq(req);
1120 if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
1123 ret = nvme_req(req)->status;
1125 *result = le64_to_cpu(nvme_req(req)->result.u64);
1126 if (meta && !ret && !write) {
1127 if (copy_to_user(meta_buffer, meta, meta_len))
1133 blk_rq_unmap_user(bio);
1135 blk_mq_free_request(req);
1139 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1141 struct nvme_ctrl *ctrl = rq->end_io_data;
1142 unsigned long flags;
1143 bool startka = false;
1145 blk_mq_free_request(rq);
1148 dev_err(ctrl->device,
1149 "failed nvme_keep_alive_end_io error=%d\n",
1154 ctrl->comp_seen = false;
1155 spin_lock_irqsave(&ctrl->lock, flags);
1156 if (ctrl->state == NVME_CTRL_LIVE ||
1157 ctrl->state == NVME_CTRL_CONNECTING)
1159 spin_unlock_irqrestore(&ctrl->lock, flags);
1161 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1164 static int nvme_keep_alive(struct nvme_ctrl *ctrl)
1168 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd, BLK_MQ_REQ_RESERVED,
1173 rq->timeout = ctrl->kato * HZ;
1174 rq->end_io_data = ctrl;
1176 blk_execute_rq_nowait(rq->q, NULL, rq, 0, nvme_keep_alive_end_io);
1181 static void nvme_keep_alive_work(struct work_struct *work)
1183 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1184 struct nvme_ctrl, ka_work);
1185 bool comp_seen = ctrl->comp_seen;
1187 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1188 dev_dbg(ctrl->device,
1189 "reschedule traffic based keep-alive timer\n");
1190 ctrl->comp_seen = false;
1191 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1195 if (nvme_keep_alive(ctrl)) {
1196 /* allocation failure, reset the controller */
1197 dev_err(ctrl->device, "keep-alive failed\n");
1198 nvme_reset_ctrl(ctrl);
1203 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1205 if (unlikely(ctrl->kato == 0))
1208 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ);
1211 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1213 if (unlikely(ctrl->kato == 0))
1216 cancel_delayed_work_sync(&ctrl->ka_work);
1218 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1221 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1222 * flag, thus sending any new CNS opcodes has a big chance of not working.
1223 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1224 * (but not for any later version).
1226 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1228 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1229 return ctrl->vs < NVME_VS(1, 2, 0);
1230 return ctrl->vs < NVME_VS(1, 1, 0);
1233 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1235 struct nvme_command c = { };
1238 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1239 c.identify.opcode = nvme_admin_identify;
1240 c.identify.cns = NVME_ID_CNS_CTRL;
1242 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1246 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1247 sizeof(struct nvme_id_ctrl));
1253 static bool nvme_multi_css(struct nvme_ctrl *ctrl)
1255 return (ctrl->ctrl_config & NVME_CC_CSS_MASK) == NVME_CC_CSS_CSI;
1258 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1259 struct nvme_ns_id_desc *cur, bool *csi_seen)
1261 const char *warn_str = "ctrl returned bogus length:";
1264 switch (cur->nidt) {
1265 case NVME_NIDT_EUI64:
1266 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1267 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1268 warn_str, cur->nidl);
1271 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1272 return NVME_NIDT_EUI64_LEN;
1273 case NVME_NIDT_NGUID:
1274 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1275 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1276 warn_str, cur->nidl);
1279 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1280 return NVME_NIDT_NGUID_LEN;
1281 case NVME_NIDT_UUID:
1282 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1283 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1284 warn_str, cur->nidl);
1287 uuid_copy(&ids->uuid, data + sizeof(*cur));
1288 return NVME_NIDT_UUID_LEN;
1290 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1291 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1292 warn_str, cur->nidl);
1295 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1297 return NVME_NIDT_CSI_LEN;
1299 /* Skip unknown types */
1304 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1305 struct nvme_ns_ids *ids)
1307 struct nvme_command c = { };
1308 bool csi_seen = false;
1309 int status, pos, len;
1312 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1315 c.identify.opcode = nvme_admin_identify;
1316 c.identify.nsid = cpu_to_le32(nsid);
1317 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1319 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1323 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1324 NVME_IDENTIFY_DATA_SIZE);
1326 dev_warn(ctrl->device,
1327 "Identify Descriptors failed (%d)\n", status);
1331 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1332 struct nvme_ns_id_desc *cur = data + pos;
1337 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1341 len += sizeof(*cur);
1344 if (nvme_multi_css(ctrl) && !csi_seen) {
1345 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1355 static int nvme_identify_ns_list(struct nvme_ctrl *dev, unsigned nsid, __le32 *ns_list)
1357 struct nvme_command c = { };
1359 c.identify.opcode = nvme_admin_identify;
1360 c.identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST;
1361 c.identify.nsid = cpu_to_le32(nsid);
1362 return nvme_submit_sync_cmd(dev->admin_q, &c, ns_list,
1363 NVME_IDENTIFY_DATA_SIZE);
1366 static int nvme_identify_ns(struct nvme_ctrl *ctrl,
1367 unsigned nsid, struct nvme_id_ns **id)
1369 struct nvme_command c = { };
1372 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1373 c.identify.opcode = nvme_admin_identify;
1374 c.identify.nsid = cpu_to_le32(nsid);
1375 c.identify.cns = NVME_ID_CNS_NS;
1377 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1381 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1383 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1390 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1391 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1393 union nvme_result res = { 0 };
1394 struct nvme_command c;
1397 memset(&c, 0, sizeof(c));
1398 c.features.opcode = op;
1399 c.features.fid = cpu_to_le32(fid);
1400 c.features.dword11 = cpu_to_le32(dword11);
1402 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1403 buffer, buflen, 0, NVME_QID_ANY, 0, 0, false);
1404 if (ret >= 0 && result)
1405 *result = le32_to_cpu(res.u32);
1409 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1410 unsigned int dword11, void *buffer, size_t buflen,
1413 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1416 EXPORT_SYMBOL_GPL(nvme_set_features);
1418 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1419 unsigned int dword11, void *buffer, size_t buflen,
1422 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1425 EXPORT_SYMBOL_GPL(nvme_get_features);
1427 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1429 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1431 int status, nr_io_queues;
1433 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1439 * Degraded controllers might return an error when setting the queue
1440 * count. We still want to be able to bring them online and offer
1441 * access to the admin queue, as that might be only way to fix them up.
1444 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1447 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1448 *count = min(*count, nr_io_queues);
1453 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1455 #define NVME_AEN_SUPPORTED \
1456 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1457 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1459 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1461 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1464 if (!supported_aens)
1467 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1470 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1473 queue_work(nvme_wq, &ctrl->async_event_work);
1477 * Convert integer values from ioctl structures to user pointers, silently
1478 * ignoring the upper bits in the compat case to match behaviour of 32-bit
1481 static void __user *nvme_to_user_ptr(uintptr_t ptrval)
1483 if (in_compat_syscall())
1484 ptrval = (compat_uptr_t)ptrval;
1485 return (void __user *)ptrval;
1488 static int nvme_submit_io(struct nvme_ns *ns, struct nvme_user_io __user *uio)
1490 struct nvme_user_io io;
1491 struct nvme_command c;
1492 unsigned length, meta_len;
1493 void __user *metadata;
1495 if (copy_from_user(&io, uio, sizeof(io)))
1500 switch (io.opcode) {
1501 case nvme_cmd_write:
1503 case nvme_cmd_compare:
1509 length = (io.nblocks + 1) << ns->lba_shift;
1510 meta_len = (io.nblocks + 1) * ns->ms;
1511 metadata = nvme_to_user_ptr(io.metadata);
1513 if (ns->features & NVME_NS_EXT_LBAS) {
1516 } else if (meta_len) {
1517 if ((io.metadata & 3) || !io.metadata)
1521 memset(&c, 0, sizeof(c));
1522 c.rw.opcode = io.opcode;
1523 c.rw.flags = io.flags;
1524 c.rw.nsid = cpu_to_le32(ns->head->ns_id);
1525 c.rw.slba = cpu_to_le64(io.slba);
1526 c.rw.length = cpu_to_le16(io.nblocks);
1527 c.rw.control = cpu_to_le16(io.control);
1528 c.rw.dsmgmt = cpu_to_le32(io.dsmgmt);
1529 c.rw.reftag = cpu_to_le32(io.reftag);
1530 c.rw.apptag = cpu_to_le16(io.apptag);
1531 c.rw.appmask = cpu_to_le16(io.appmask);
1533 return nvme_submit_user_cmd(ns->queue, &c,
1534 nvme_to_user_ptr(io.addr), length,
1535 metadata, meta_len, lower_32_bits(io.slba), NULL, 0);
1538 static int nvme_user_cmd(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1539 struct nvme_passthru_cmd __user *ucmd)
1541 struct nvme_passthru_cmd cmd;
1542 struct nvme_command c;
1543 unsigned timeout = 0;
1547 if (!capable(CAP_SYS_ADMIN))
1549 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1554 memset(&c, 0, sizeof(c));
1555 c.common.opcode = cmd.opcode;
1556 c.common.flags = cmd.flags;
1557 c.common.nsid = cpu_to_le32(cmd.nsid);
1558 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1559 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1560 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1561 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1562 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1563 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1564 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1565 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1568 timeout = msecs_to_jiffies(cmd.timeout_ms);
1570 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1571 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1572 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1573 0, &result, timeout);
1576 if (put_user(result, &ucmd->result))
1583 static int nvme_user_cmd64(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1584 struct nvme_passthru_cmd64 __user *ucmd)
1586 struct nvme_passthru_cmd64 cmd;
1587 struct nvme_command c;
1588 unsigned timeout = 0;
1591 if (!capable(CAP_SYS_ADMIN))
1593 if (copy_from_user(&cmd, ucmd, sizeof(cmd)))
1598 memset(&c, 0, sizeof(c));
1599 c.common.opcode = cmd.opcode;
1600 c.common.flags = cmd.flags;
1601 c.common.nsid = cpu_to_le32(cmd.nsid);
1602 c.common.cdw2[0] = cpu_to_le32(cmd.cdw2);
1603 c.common.cdw2[1] = cpu_to_le32(cmd.cdw3);
1604 c.common.cdw10 = cpu_to_le32(cmd.cdw10);
1605 c.common.cdw11 = cpu_to_le32(cmd.cdw11);
1606 c.common.cdw12 = cpu_to_le32(cmd.cdw12);
1607 c.common.cdw13 = cpu_to_le32(cmd.cdw13);
1608 c.common.cdw14 = cpu_to_le32(cmd.cdw14);
1609 c.common.cdw15 = cpu_to_le32(cmd.cdw15);
1612 timeout = msecs_to_jiffies(cmd.timeout_ms);
1614 status = nvme_submit_user_cmd(ns ? ns->queue : ctrl->admin_q, &c,
1615 nvme_to_user_ptr(cmd.addr), cmd.data_len,
1616 nvme_to_user_ptr(cmd.metadata), cmd.metadata_len,
1617 0, &cmd.result, timeout);
1620 if (put_user(cmd.result, &ucmd->result))
1628 * Issue ioctl requests on the first available path. Note that unlike normal
1629 * block layer requests we will not retry failed request on another controller.
1631 struct nvme_ns *nvme_get_ns_from_disk(struct gendisk *disk,
1632 struct nvme_ns_head **head, int *srcu_idx)
1634 #ifdef CONFIG_NVME_MULTIPATH
1635 if (disk->fops == &nvme_ns_head_ops) {
1638 *head = disk->private_data;
1639 *srcu_idx = srcu_read_lock(&(*head)->srcu);
1640 ns = nvme_find_path(*head);
1642 srcu_read_unlock(&(*head)->srcu, *srcu_idx);
1648 return disk->private_data;
1651 void nvme_put_ns_from_disk(struct nvme_ns_head *head, int idx)
1654 srcu_read_unlock(&head->srcu, idx);
1657 static bool is_ctrl_ioctl(unsigned int cmd)
1659 if (cmd == NVME_IOCTL_ADMIN_CMD || cmd == NVME_IOCTL_ADMIN64_CMD)
1661 if (is_sed_ioctl(cmd))
1666 static int nvme_handle_ctrl_ioctl(struct nvme_ns *ns, unsigned int cmd,
1668 struct nvme_ns_head *head,
1671 struct nvme_ctrl *ctrl = ns->ctrl;
1674 nvme_get_ctrl(ns->ctrl);
1675 nvme_put_ns_from_disk(head, srcu_idx);
1678 case NVME_IOCTL_ADMIN_CMD:
1679 ret = nvme_user_cmd(ctrl, NULL, argp);
1681 case NVME_IOCTL_ADMIN64_CMD:
1682 ret = nvme_user_cmd64(ctrl, NULL, argp);
1685 ret = sed_ioctl(ctrl->opal_dev, cmd, argp);
1688 nvme_put_ctrl(ctrl);
1692 static int nvme_ioctl(struct block_device *bdev, fmode_t mode,
1693 unsigned int cmd, unsigned long arg)
1695 struct nvme_ns_head *head = NULL;
1696 void __user *argp = (void __user *)arg;
1700 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
1702 return -EWOULDBLOCK;
1705 * Handle ioctls that apply to the controller instead of the namespace
1706 * seperately and drop the ns SRCU reference early. This avoids a
1707 * deadlock when deleting namespaces using the passthrough interface.
1709 if (is_ctrl_ioctl(cmd))
1710 return nvme_handle_ctrl_ioctl(ns, cmd, argp, head, srcu_idx);
1714 force_successful_syscall_return();
1715 ret = ns->head->ns_id;
1717 case NVME_IOCTL_IO_CMD:
1718 ret = nvme_user_cmd(ns->ctrl, ns, argp);
1720 case NVME_IOCTL_SUBMIT_IO:
1721 ret = nvme_submit_io(ns, argp);
1723 case NVME_IOCTL_IO64_CMD:
1724 ret = nvme_user_cmd64(ns->ctrl, ns, argp);
1728 ret = nvme_nvm_ioctl(ns, cmd, arg);
1733 nvme_put_ns_from_disk(head, srcu_idx);
1737 #ifdef CONFIG_COMPAT
1738 struct nvme_user_io32 {
1751 } __attribute__((__packed__));
1753 #define NVME_IOCTL_SUBMIT_IO32 _IOW('N', 0x42, struct nvme_user_io32)
1755 static int nvme_compat_ioctl(struct block_device *bdev, fmode_t mode,
1756 unsigned int cmd, unsigned long arg)
1759 * Corresponds to the difference of NVME_IOCTL_SUBMIT_IO
1760 * between 32 bit programs and 64 bit kernel.
1761 * The cause is that the results of sizeof(struct nvme_user_io),
1762 * which is used to define NVME_IOCTL_SUBMIT_IO,
1763 * are not same between 32 bit compiler and 64 bit compiler.
1764 * NVME_IOCTL_SUBMIT_IO32 is for 64 bit kernel handling
1765 * NVME_IOCTL_SUBMIT_IO issued from 32 bit programs.
1766 * Other IOCTL numbers are same between 32 bit and 64 bit.
1767 * So there is nothing to do regarding to other IOCTL numbers.
1769 if (cmd == NVME_IOCTL_SUBMIT_IO32)
1770 return nvme_ioctl(bdev, mode, NVME_IOCTL_SUBMIT_IO, arg);
1772 return nvme_ioctl(bdev, mode, cmd, arg);
1775 #define nvme_compat_ioctl NULL
1776 #endif /* CONFIG_COMPAT */
1778 static int nvme_open(struct block_device *bdev, fmode_t mode)
1780 struct nvme_ns *ns = bdev->bd_disk->private_data;
1782 #ifdef CONFIG_NVME_MULTIPATH
1783 /* should never be called due to GENHD_FL_HIDDEN */
1784 if (WARN_ON_ONCE(ns->head->disk))
1787 if (!kref_get_unless_zero(&ns->kref))
1789 if (!try_module_get(ns->ctrl->ops->module))
1800 static void nvme_release(struct gendisk *disk, fmode_t mode)
1802 struct nvme_ns *ns = disk->private_data;
1804 module_put(ns->ctrl->ops->module);
1808 static int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1810 /* some standard values */
1811 geo->heads = 1 << 6;
1812 geo->sectors = 1 << 5;
1813 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1817 #ifdef CONFIG_BLK_DEV_INTEGRITY
1818 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1819 u32 max_integrity_segments)
1821 struct blk_integrity integrity;
1823 memset(&integrity, 0, sizeof(integrity));
1825 case NVME_NS_DPS_PI_TYPE3:
1826 integrity.profile = &t10_pi_type3_crc;
1827 integrity.tag_size = sizeof(u16) + sizeof(u32);
1828 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1830 case NVME_NS_DPS_PI_TYPE1:
1831 case NVME_NS_DPS_PI_TYPE2:
1832 integrity.profile = &t10_pi_type1_crc;
1833 integrity.tag_size = sizeof(u16);
1834 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1837 integrity.profile = NULL;
1840 integrity.tuple_size = ms;
1841 blk_integrity_register(disk, &integrity);
1842 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1845 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1846 u32 max_integrity_segments)
1849 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1851 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1853 struct nvme_ctrl *ctrl = ns->ctrl;
1854 struct request_queue *queue = disk->queue;
1855 u32 size = queue_logical_block_size(queue);
1857 if (!(ctrl->oncs & NVME_CTRL_ONCS_DSM)) {
1858 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1862 if (ctrl->nr_streams && ns->sws && ns->sgs)
1863 size *= ns->sws * ns->sgs;
1865 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1866 NVME_DSM_MAX_RANGES);
1868 queue->limits.discard_alignment = 0;
1869 queue->limits.discard_granularity = size;
1871 /* If discard is already enabled, don't reset queue limits */
1872 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1875 blk_queue_max_discard_sectors(queue, UINT_MAX);
1876 blk_queue_max_discard_segments(queue, NVME_DSM_MAX_RANGES);
1878 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1879 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1882 static void nvme_config_write_zeroes(struct gendisk *disk, struct nvme_ns *ns)
1886 if (!(ns->ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) ||
1887 (ns->ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
1890 * Even though NVMe spec explicitly states that MDTS is not
1891 * applicable to the write-zeroes:- "The restriction does not apply to
1892 * commands that do not transfer data between the host and the
1893 * controller (e.g., Write Uncorrectable ro Write Zeroes command).".
1894 * In order to be more cautious use controller's max_hw_sectors value
1895 * to configure the maximum sectors for the write-zeroes which is
1896 * configured based on the controller's MDTS field in the
1897 * nvme_init_identify() if available.
1899 if (ns->ctrl->max_hw_sectors == UINT_MAX)
1900 max_blocks = (u64)USHRT_MAX + 1;
1902 max_blocks = ns->ctrl->max_hw_sectors + 1;
1904 blk_queue_max_write_zeroes_sectors(disk->queue,
1905 nvme_lba_to_sect(ns, max_blocks));
1908 static int nvme_report_ns_ids(struct nvme_ctrl *ctrl, unsigned int nsid,
1909 struct nvme_id_ns *id, struct nvme_ns_ids *ids)
1911 memset(ids, 0, sizeof(*ids));
1913 if (ctrl->vs >= NVME_VS(1, 1, 0))
1914 memcpy(ids->eui64, id->eui64, sizeof(id->eui64));
1915 if (ctrl->vs >= NVME_VS(1, 2, 0))
1916 memcpy(ids->nguid, id->nguid, sizeof(id->nguid));
1917 if (ctrl->vs >= NVME_VS(1, 3, 0) || nvme_multi_css(ctrl))
1918 return nvme_identify_ns_descs(ctrl, nsid, ids);
1922 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1924 return !uuid_is_null(&ids->uuid) ||
1925 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1926 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1929 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1931 return uuid_equal(&a->uuid, &b->uuid) &&
1932 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1933 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1937 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1938 u32 *phys_bs, u32 *io_opt)
1940 struct streams_directive_params s;
1943 if (!ctrl->nr_streams)
1946 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1950 ns->sws = le32_to_cpu(s.sws);
1951 ns->sgs = le16_to_cpu(s.sgs);
1954 *phys_bs = ns->sws * (1 << ns->lba_shift);
1956 *io_opt = *phys_bs * ns->sgs;
1962 static void nvme_update_disk_info(struct gendisk *disk,
1963 struct nvme_ns *ns, struct nvme_id_ns *id)
1965 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1966 unsigned short bs = 1 << ns->lba_shift;
1967 u32 atomic_bs, phys_bs, io_opt = 0;
1969 if (ns->lba_shift > PAGE_SHIFT) {
1970 /* unsupported block size, set capacity to 0 later */
1973 blk_mq_freeze_queue(disk->queue);
1974 blk_integrity_unregister(disk);
1976 atomic_bs = phys_bs = bs;
1977 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1978 if (id->nabo == 0) {
1980 * Bit 1 indicates whether NAWUPF is defined for this namespace
1981 * and whether it should be used instead of AWUPF. If NAWUPF ==
1982 * 0 then AWUPF must be used instead.
1984 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1985 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1987 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1990 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1991 /* NPWG = Namespace Preferred Write Granularity */
1992 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1993 /* NOWS = Namespace Optimal Write Size */
1994 io_opt = bs * (1 + le16_to_cpu(id->nows));
1997 blk_queue_logical_block_size(disk->queue, bs);
1999 * Linux filesystems assume writing a single physical block is
2000 * an atomic operation. Hence limit the physical block size to the
2001 * value of the Atomic Write Unit Power Fail parameter.
2003 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
2004 blk_queue_io_min(disk->queue, phys_bs);
2005 blk_queue_io_opt(disk->queue, io_opt);
2008 * The block layer can't support LBA sizes larger than the page size
2009 * yet, so catch this early and don't allow block I/O.
2011 if (ns->lba_shift > PAGE_SHIFT)
2015 * Register a metadata profile for PI, or the plain non-integrity NVMe
2016 * metadata masquerading as Type 0 if supported, otherwise reject block
2017 * I/O to namespaces with metadata except when the namespace supports
2018 * PI, as it can strip/insert in that case.
2021 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
2022 (ns->features & NVME_NS_METADATA_SUPPORTED))
2023 nvme_init_integrity(disk, ns->ms, ns->pi_type,
2024 ns->ctrl->max_integrity_segments);
2025 else if (!nvme_ns_has_pi(ns))
2029 set_capacity_revalidate_and_notify(disk, capacity, false);
2031 nvme_config_discard(disk, ns);
2032 nvme_config_write_zeroes(disk, ns);
2034 if (id->nsattr & NVME_NS_ATTR_RO)
2035 set_disk_ro(disk, true);
2037 set_disk_ro(disk, false);
2039 blk_mq_unfreeze_queue(disk->queue);
2042 static inline bool nvme_first_scan(struct gendisk *disk)
2044 /* nvme_alloc_ns() scans the disk prior to adding it */
2045 return !(disk->flags & GENHD_FL_UP);
2048 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
2050 struct nvme_ctrl *ctrl = ns->ctrl;
2053 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2054 is_power_of_2(ctrl->max_hw_sectors))
2055 iob = ctrl->max_hw_sectors;
2057 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
2062 if (!is_power_of_2(iob)) {
2063 if (nvme_first_scan(ns->disk))
2064 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2065 ns->disk->disk_name, iob);
2069 if (blk_queue_is_zoned(ns->disk->queue)) {
2070 if (nvme_first_scan(ns->disk))
2071 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2072 ns->disk->disk_name);
2076 blk_queue_chunk_sectors(ns->queue, iob);
2079 static int __nvme_revalidate_disk(struct gendisk *disk, struct nvme_id_ns *id)
2081 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
2082 struct nvme_ns *ns = disk->private_data;
2083 struct nvme_ctrl *ctrl = ns->ctrl;
2087 * If identify namespace failed, use default 512 byte block size so
2088 * block layer can use before failing read/write for 0 capacity.
2090 ns->lba_shift = id->lbaf[lbaf].ds;
2091 if (ns->lba_shift == 0)
2094 switch (ns->head->ids.csi) {
2098 ret = nvme_update_zone_info(disk, ns, lbaf);
2100 dev_warn(ctrl->device,
2101 "failed to add zoned namespace:%u ret:%d\n",
2102 ns->head->ns_id, ret);
2107 dev_warn(ctrl->device, "unknown csi:%u ns:%u\n",
2108 ns->head->ids.csi, ns->head->ns_id);
2113 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
2114 /* the PI implementation requires metadata equal t10 pi tuple size */
2115 if (ns->ms == sizeof(struct t10_pi_tuple))
2116 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
2122 * For PCIe only the separate metadata pointer is supported,
2123 * as the block layer supplies metadata in a separate bio_vec
2124 * chain. For Fabrics, only metadata as part of extended data
2125 * LBA is supported on the wire per the Fabrics specification,
2126 * but the HBA/HCA will do the remapping from the separate
2127 * metadata buffers for us.
2129 if (id->flbas & NVME_NS_FLBAS_META_EXT) {
2130 ns->features |= NVME_NS_EXT_LBAS;
2131 if ((ctrl->ops->flags & NVME_F_FABRICS) &&
2132 (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED) &&
2133 ctrl->max_integrity_segments)
2134 ns->features |= NVME_NS_METADATA_SUPPORTED;
2136 if (WARN_ON_ONCE(ctrl->ops->flags & NVME_F_FABRICS))
2138 if (ctrl->ops->flags & NVME_F_METADATA_SUPPORTED)
2139 ns->features |= NVME_NS_METADATA_SUPPORTED;
2143 nvme_set_chunk_sectors(ns, id);
2144 nvme_update_disk_info(disk, ns, id);
2145 #ifdef CONFIG_NVME_MULTIPATH
2146 if (ns->head->disk) {
2147 nvme_update_disk_info(ns->head->disk, ns, id);
2148 blk_stack_limits(&ns->head->disk->queue->limits,
2149 &ns->queue->limits, 0);
2150 blk_queue_update_readahead(ns->head->disk->queue);
2151 nvme_update_bdev_size(ns->head->disk);
2157 static int _nvme_revalidate_disk(struct gendisk *disk)
2159 struct nvme_ns *ns = disk->private_data;
2160 struct nvme_ctrl *ctrl = ns->ctrl;
2161 struct nvme_id_ns *id;
2162 struct nvme_ns_ids ids;
2165 if (test_bit(NVME_NS_DEAD, &ns->flags)) {
2166 set_capacity(disk, 0);
2170 ret = nvme_identify_ns(ctrl, ns->head->ns_id, &id);
2174 if (id->ncap == 0) {
2179 ret = nvme_report_ns_ids(ctrl, ns->head->ns_id, id, &ids);
2183 if (!nvme_ns_ids_equal(&ns->head->ids, &ids)) {
2184 dev_err(ctrl->device,
2185 "identifiers changed for nsid %d\n", ns->head->ns_id);
2190 ret = __nvme_revalidate_disk(disk, id);
2195 * Only fail the function if we got a fatal error back from the
2196 * device, otherwise ignore the error and just move on.
2198 if (ret == -ENOMEM || (ret > 0 && !(ret & NVME_SC_DNR)))
2201 ret = blk_status_to_errno(nvme_error_status(ret));
2205 static int nvme_revalidate_disk(struct gendisk *disk)
2209 ret = _nvme_revalidate_disk(disk);
2213 #ifdef CONFIG_BLK_DEV_ZONED
2214 if (blk_queue_is_zoned(disk->queue)) {
2215 struct nvme_ns *ns = disk->private_data;
2216 struct nvme_ctrl *ctrl = ns->ctrl;
2218 ret = blk_revalidate_disk_zones(disk, NULL);
2220 blk_queue_max_zone_append_sectors(disk->queue,
2221 ctrl->max_zone_append);
2227 static char nvme_pr_type(enum pr_type type)
2230 case PR_WRITE_EXCLUSIVE:
2232 case PR_EXCLUSIVE_ACCESS:
2234 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2236 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2238 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2240 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2247 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2248 u64 key, u64 sa_key, u8 op)
2250 struct nvme_ns_head *head = NULL;
2252 struct nvme_command c;
2254 u8 data[16] = { 0, };
2256 ns = nvme_get_ns_from_disk(bdev->bd_disk, &head, &srcu_idx);
2258 return -EWOULDBLOCK;
2260 put_unaligned_le64(key, &data[0]);
2261 put_unaligned_le64(sa_key, &data[8]);
2263 memset(&c, 0, sizeof(c));
2264 c.common.opcode = op;
2265 c.common.nsid = cpu_to_le32(ns->head->ns_id);
2266 c.common.cdw10 = cpu_to_le32(cdw10);
2268 ret = nvme_submit_sync_cmd(ns->queue, &c, data, 16);
2269 nvme_put_ns_from_disk(head, srcu_idx);
2273 static int nvme_pr_register(struct block_device *bdev, u64 old,
2274 u64 new, unsigned flags)
2278 if (flags & ~PR_FL_IGNORE_KEY)
2281 cdw10 = old ? 2 : 0;
2282 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2283 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2284 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2287 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2288 enum pr_type type, unsigned flags)
2292 if (flags & ~PR_FL_IGNORE_KEY)
2295 cdw10 = nvme_pr_type(type) << 8;
2296 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2297 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2300 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2301 enum pr_type type, bool abort)
2303 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2304 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2307 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2309 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2310 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2313 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2315 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2316 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2319 static const struct pr_ops nvme_pr_ops = {
2320 .pr_register = nvme_pr_register,
2321 .pr_reserve = nvme_pr_reserve,
2322 .pr_release = nvme_pr_release,
2323 .pr_preempt = nvme_pr_preempt,
2324 .pr_clear = nvme_pr_clear,
2327 #ifdef CONFIG_BLK_SED_OPAL
2328 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2331 struct nvme_ctrl *ctrl = data;
2332 struct nvme_command cmd;
2334 memset(&cmd, 0, sizeof(cmd));
2336 cmd.common.opcode = nvme_admin_security_send;
2338 cmd.common.opcode = nvme_admin_security_recv;
2339 cmd.common.nsid = 0;
2340 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2341 cmd.common.cdw11 = cpu_to_le32(len);
2343 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2344 ADMIN_TIMEOUT, NVME_QID_ANY, 1, 0, false);
2346 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2347 #endif /* CONFIG_BLK_SED_OPAL */
2349 static const struct block_device_operations nvme_fops = {
2350 .owner = THIS_MODULE,
2351 .ioctl = nvme_ioctl,
2352 .compat_ioctl = nvme_compat_ioctl,
2354 .release = nvme_release,
2355 .getgeo = nvme_getgeo,
2356 .report_zones = nvme_report_zones,
2357 .pr_ops = &nvme_pr_ops,
2360 #ifdef CONFIG_NVME_MULTIPATH
2361 static int nvme_ns_head_open(struct block_device *bdev, fmode_t mode)
2363 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2365 if (!kref_get_unless_zero(&head->ref))
2370 static void nvme_ns_head_release(struct gendisk *disk, fmode_t mode)
2372 nvme_put_ns_head(disk->private_data);
2375 const struct block_device_operations nvme_ns_head_ops = {
2376 .owner = THIS_MODULE,
2377 .submit_bio = nvme_ns_head_submit_bio,
2378 .open = nvme_ns_head_open,
2379 .release = nvme_ns_head_release,
2380 .ioctl = nvme_ioctl,
2381 .compat_ioctl = nvme_compat_ioctl,
2382 .getgeo = nvme_getgeo,
2383 .report_zones = nvme_report_zones,
2384 .pr_ops = &nvme_pr_ops,
2386 #endif /* CONFIG_NVME_MULTIPATH */
2388 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2390 unsigned long timeout =
2391 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2392 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2395 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2398 if ((csts & NVME_CSTS_RDY) == bit)
2401 usleep_range(1000, 2000);
2402 if (fatal_signal_pending(current))
2404 if (time_after(jiffies, timeout)) {
2405 dev_err(ctrl->device,
2406 "Device not ready; aborting %s, CSTS=0x%x\n",
2407 enabled ? "initialisation" : "reset", csts);
2416 * If the device has been passed off to us in an enabled state, just clear
2417 * the enabled bit. The spec says we should set the 'shutdown notification
2418 * bits', but doing so may cause the device to complete commands to the
2419 * admin queue ... and we don't know what memory that might be pointing at!
2421 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2425 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2426 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2428 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2432 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2433 msleep(NVME_QUIRK_DELAY_AMOUNT);
2435 return nvme_wait_ready(ctrl, ctrl->cap, false);
2437 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2439 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2441 unsigned dev_page_min;
2444 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2446 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2449 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2451 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2452 dev_err(ctrl->device,
2453 "Minimum device page size %u too large for host (%u)\n",
2454 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2458 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2459 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2461 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2462 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2463 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2464 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2465 ctrl->ctrl_config |= NVME_CC_ENABLE;
2467 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2470 return nvme_wait_ready(ctrl, ctrl->cap, true);
2472 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2474 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2476 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2480 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2481 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2483 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2487 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2488 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2492 if (fatal_signal_pending(current))
2494 if (time_after(jiffies, timeout)) {
2495 dev_err(ctrl->device,
2496 "Device shutdown incomplete; abort shutdown\n");
2503 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2505 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
2506 struct request_queue *q)
2510 if (ctrl->max_hw_sectors) {
2512 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
2514 max_segments = min_not_zero(max_segments, ctrl->max_segments);
2515 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
2516 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
2518 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
2519 blk_queue_dma_alignment(q, 7);
2520 if (ctrl->vwc & NVME_CTRL_VWC_PRESENT)
2522 blk_queue_write_cache(q, vwc, vwc);
2525 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2530 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2533 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2534 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2537 dev_warn_once(ctrl->device,
2538 "could not set timestamp (%d)\n", ret);
2542 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2544 struct nvme_feat_host_behavior *host;
2547 /* Don't bother enabling the feature if retry delay is not reported */
2551 host = kzalloc(sizeof(*host), GFP_KERNEL);
2555 host->acre = NVME_ENABLE_ACRE;
2556 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2557 host, sizeof(*host), NULL);
2562 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2565 * APST (Autonomous Power State Transition) lets us program a
2566 * table of power state transitions that the controller will
2567 * perform automatically. We configure it with a simple
2568 * heuristic: we are willing to spend at most 2% of the time
2569 * transitioning between power states. Therefore, when running
2570 * in any given state, we will enter the next lower-power
2571 * non-operational state after waiting 50 * (enlat + exlat)
2572 * microseconds, as long as that state's exit latency is under
2573 * the requested maximum latency.
2575 * We will not autonomously enter any non-operational state for
2576 * which the total latency exceeds ps_max_latency_us. Users
2577 * can set ps_max_latency_us to zero to turn off APST.
2581 struct nvme_feat_auto_pst *table;
2587 * If APST isn't supported or if we haven't been initialized yet,
2588 * then don't do anything.
2593 if (ctrl->npss > 31) {
2594 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2598 table = kzalloc(sizeof(*table), GFP_KERNEL);
2602 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2603 /* Turn off APST. */
2605 dev_dbg(ctrl->device, "APST disabled\n");
2607 __le64 target = cpu_to_le64(0);
2611 * Walk through all states from lowest- to highest-power.
2612 * According to the spec, lower-numbered states use more
2613 * power. NPSS, despite the name, is the index of the
2614 * lowest-power state, not the number of states.
2616 for (state = (int)ctrl->npss; state >= 0; state--) {
2617 u64 total_latency_us, exit_latency_us, transition_ms;
2620 table->entries[state] = target;
2623 * Don't allow transitions to the deepest state
2624 * if it's quirked off.
2626 if (state == ctrl->npss &&
2627 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2631 * Is this state a useful non-operational state for
2632 * higher-power states to autonomously transition to?
2634 if (!(ctrl->psd[state].flags &
2635 NVME_PS_FLAGS_NON_OP_STATE))
2639 (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2640 if (exit_latency_us > ctrl->ps_max_latency_us)
2645 le32_to_cpu(ctrl->psd[state].entry_lat);
2648 * This state is good. Use it as the APST idle
2649 * target for higher power states.
2651 transition_ms = total_latency_us + 19;
2652 do_div(transition_ms, 20);
2653 if (transition_ms > (1 << 24) - 1)
2654 transition_ms = (1 << 24) - 1;
2656 target = cpu_to_le64((state << 3) |
2657 (transition_ms << 8));
2662 if (total_latency_us > max_lat_us)
2663 max_lat_us = total_latency_us;
2669 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2671 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2672 max_ps, max_lat_us, (int)sizeof(*table), table);
2676 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2677 table, sizeof(*table), NULL);
2679 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2685 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2687 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2691 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2692 case PM_QOS_LATENCY_ANY:
2700 if (ctrl->ps_max_latency_us != latency) {
2701 ctrl->ps_max_latency_us = latency;
2702 nvme_configure_apst(ctrl);
2706 struct nvme_core_quirk_entry {
2708 * NVMe model and firmware strings are padded with spaces. For
2709 * simplicity, strings in the quirk table are padded with NULLs
2715 unsigned long quirks;
2718 static const struct nvme_core_quirk_entry core_quirks[] = {
2721 * This Toshiba device seems to die using any APST states. See:
2722 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2725 .mn = "THNSF5256GPUK TOSHIBA",
2726 .quirks = NVME_QUIRK_NO_APST,
2730 * This LiteON CL1-3D*-Q11 firmware version has a race
2731 * condition associated with actions related to suspend to idle
2732 * LiteON has resolved the problem in future firmware
2736 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2740 /* match is null-terminated but idstr is space-padded. */
2741 static bool string_matches(const char *idstr, const char *match, size_t len)
2748 matchlen = strlen(match);
2749 WARN_ON_ONCE(matchlen > len);
2751 if (memcmp(idstr, match, matchlen))
2754 for (; matchlen < len; matchlen++)
2755 if (idstr[matchlen] != ' ')
2761 static bool quirk_matches(const struct nvme_id_ctrl *id,
2762 const struct nvme_core_quirk_entry *q)
2764 return q->vid == le16_to_cpu(id->vid) &&
2765 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2766 string_matches(id->fr, q->fr, sizeof(id->fr));
2769 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2770 struct nvme_id_ctrl *id)
2775 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2776 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2777 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2778 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2782 if (ctrl->vs >= NVME_VS(1, 2, 1))
2783 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2786 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2787 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2788 "nqn.2014.08.org.nvmexpress:%04x%04x",
2789 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2790 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2791 off += sizeof(id->sn);
2792 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2793 off += sizeof(id->mn);
2794 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2797 static void nvme_release_subsystem(struct device *dev)
2799 struct nvme_subsystem *subsys =
2800 container_of(dev, struct nvme_subsystem, dev);
2802 if (subsys->instance >= 0)
2803 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2807 static void nvme_destroy_subsystem(struct kref *ref)
2809 struct nvme_subsystem *subsys =
2810 container_of(ref, struct nvme_subsystem, ref);
2812 mutex_lock(&nvme_subsystems_lock);
2813 list_del(&subsys->entry);
2814 mutex_unlock(&nvme_subsystems_lock);
2816 ida_destroy(&subsys->ns_ida);
2817 device_del(&subsys->dev);
2818 put_device(&subsys->dev);
2821 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2823 kref_put(&subsys->ref, nvme_destroy_subsystem);
2826 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2828 struct nvme_subsystem *subsys;
2830 lockdep_assert_held(&nvme_subsystems_lock);
2833 * Fail matches for discovery subsystems. This results
2834 * in each discovery controller bound to a unique subsystem.
2835 * This avoids issues with validating controller values
2836 * that can only be true when there is a single unique subsystem.
2837 * There may be multiple and completely independent entities
2838 * that provide discovery controllers.
2840 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2843 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2844 if (strcmp(subsys->subnqn, subsysnqn))
2846 if (!kref_get_unless_zero(&subsys->ref))
2854 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2855 struct device_attribute subsys_attr_##_name = \
2856 __ATTR(_name, _mode, _show, NULL)
2858 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2859 struct device_attribute *attr,
2862 struct nvme_subsystem *subsys =
2863 container_of(dev, struct nvme_subsystem, dev);
2865 return snprintf(buf, PAGE_SIZE, "%s\n", subsys->subnqn);
2867 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2869 #define nvme_subsys_show_str_function(field) \
2870 static ssize_t subsys_##field##_show(struct device *dev, \
2871 struct device_attribute *attr, char *buf) \
2873 struct nvme_subsystem *subsys = \
2874 container_of(dev, struct nvme_subsystem, dev); \
2875 return sprintf(buf, "%.*s\n", \
2876 (int)sizeof(subsys->field), subsys->field); \
2878 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2880 nvme_subsys_show_str_function(model);
2881 nvme_subsys_show_str_function(serial);
2882 nvme_subsys_show_str_function(firmware_rev);
2884 static struct attribute *nvme_subsys_attrs[] = {
2885 &subsys_attr_model.attr,
2886 &subsys_attr_serial.attr,
2887 &subsys_attr_firmware_rev.attr,
2888 &subsys_attr_subsysnqn.attr,
2889 #ifdef CONFIG_NVME_MULTIPATH
2890 &subsys_attr_iopolicy.attr,
2895 static struct attribute_group nvme_subsys_attrs_group = {
2896 .attrs = nvme_subsys_attrs,
2899 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2900 &nvme_subsys_attrs_group,
2904 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2905 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2907 struct nvme_ctrl *tmp;
2909 lockdep_assert_held(&nvme_subsystems_lock);
2911 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2912 if (nvme_state_terminal(tmp))
2915 if (tmp->cntlid == ctrl->cntlid) {
2916 dev_err(ctrl->device,
2917 "Duplicate cntlid %u with %s, rejecting\n",
2918 ctrl->cntlid, dev_name(tmp->device));
2922 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2923 (ctrl->opts && ctrl->opts->discovery_nqn))
2926 dev_err(ctrl->device,
2927 "Subsystem does not support multiple controllers\n");
2934 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2936 struct nvme_subsystem *subsys, *found;
2939 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2943 subsys->instance = -1;
2944 mutex_init(&subsys->lock);
2945 kref_init(&subsys->ref);
2946 INIT_LIST_HEAD(&subsys->ctrls);
2947 INIT_LIST_HEAD(&subsys->nsheads);
2948 nvme_init_subnqn(subsys, ctrl, id);
2949 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2950 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2951 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2952 subsys->vendor_id = le16_to_cpu(id->vid);
2953 subsys->cmic = id->cmic;
2954 subsys->awupf = le16_to_cpu(id->awupf);
2955 #ifdef CONFIG_NVME_MULTIPATH
2956 subsys->iopolicy = NVME_IOPOLICY_NUMA;
2959 subsys->dev.class = nvme_subsys_class;
2960 subsys->dev.release = nvme_release_subsystem;
2961 subsys->dev.groups = nvme_subsys_attrs_groups;
2962 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2963 device_initialize(&subsys->dev);
2965 mutex_lock(&nvme_subsystems_lock);
2966 found = __nvme_find_get_subsystem(subsys->subnqn);
2968 put_device(&subsys->dev);
2971 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2973 goto out_put_subsystem;
2976 ret = device_add(&subsys->dev);
2978 dev_err(ctrl->device,
2979 "failed to register subsystem device.\n");
2980 put_device(&subsys->dev);
2983 ida_init(&subsys->ns_ida);
2984 list_add_tail(&subsys->entry, &nvme_subsystems);
2987 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2988 dev_name(ctrl->device));
2990 dev_err(ctrl->device,
2991 "failed to create sysfs link from subsystem.\n");
2992 goto out_put_subsystem;
2996 subsys->instance = ctrl->instance;
2997 ctrl->subsys = subsys;
2998 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2999 mutex_unlock(&nvme_subsystems_lock);
3003 nvme_put_subsystem(subsys);
3005 mutex_unlock(&nvme_subsystems_lock);
3009 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3010 void *log, size_t size, u64 offset)
3012 struct nvme_command c = { };
3013 u32 dwlen = nvme_bytes_to_numd(size);
3015 c.get_log_page.opcode = nvme_admin_get_log_page;
3016 c.get_log_page.nsid = cpu_to_le32(nsid);
3017 c.get_log_page.lid = log_page;
3018 c.get_log_page.lsp = lsp;
3019 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3020 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3021 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3022 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3023 c.get_log_page.csi = csi;
3025 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3028 static struct nvme_cel *nvme_find_cel(struct nvme_ctrl *ctrl, u8 csi)
3030 struct nvme_cel *cel, *ret = NULL;
3032 spin_lock_irq(&ctrl->lock);
3033 list_for_each_entry(cel, &ctrl->cels, entry) {
3034 if (cel->csi == csi) {
3039 spin_unlock_irq(&ctrl->lock);
3044 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3045 struct nvme_effects_log **log)
3047 struct nvme_cel *cel = nvme_find_cel(ctrl, csi);
3053 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3057 ret = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CMD_EFFECTS, 0, csi,
3058 &cel->log, sizeof(cel->log), 0);
3066 spin_lock_irq(&ctrl->lock);
3067 list_add_tail(&cel->entry, &ctrl->cels);
3068 spin_unlock_irq(&ctrl->lock);
3075 * Initialize the cached copies of the Identify data and various controller
3076 * register in our nvme_ctrl structure. This should be called as soon as
3077 * the admin queue is fully up and running.
3079 int nvme_init_identify(struct nvme_ctrl *ctrl)
3081 struct nvme_id_ctrl *id;
3082 int ret, page_shift;
3084 bool prev_apst_enabled;
3086 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3088 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3091 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12;
3092 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3094 if (ctrl->vs >= NVME_VS(1, 1, 0))
3095 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3097 ret = nvme_identify_ctrl(ctrl, &id);
3099 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3103 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3104 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3109 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3110 ctrl->cntlid = le16_to_cpu(id->cntlid);
3112 if (!ctrl->identified) {
3115 ret = nvme_init_subsystem(ctrl, id);
3120 * Check for quirks. Quirk can depend on firmware version,
3121 * so, in principle, the set of quirks present can change
3122 * across a reset. As a possible future enhancement, we
3123 * could re-scan for quirks every time we reinitialize
3124 * the device, but we'd have to make sure that the driver
3125 * behaves intelligently if the quirks change.
3127 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3128 if (quirk_matches(id, &core_quirks[i]))
3129 ctrl->quirks |= core_quirks[i].quirks;
3133 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3134 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3135 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3138 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3139 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3140 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3142 ctrl->oacs = le16_to_cpu(id->oacs);
3143 ctrl->oncs = le16_to_cpu(id->oncs);
3144 ctrl->mtfa = le16_to_cpu(id->mtfa);
3145 ctrl->oaes = le32_to_cpu(id->oaes);
3146 ctrl->wctemp = le16_to_cpu(id->wctemp);
3147 ctrl->cctemp = le16_to_cpu(id->cctemp);
3149 atomic_set(&ctrl->abort_limit, id->acl + 1);
3150 ctrl->vwc = id->vwc;
3152 max_hw_sectors = 1 << (id->mdts + page_shift - 9);
3154 max_hw_sectors = UINT_MAX;
3155 ctrl->max_hw_sectors =
3156 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3158 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3159 ctrl->sgls = le32_to_cpu(id->sgls);
3160 ctrl->kas = le16_to_cpu(id->kas);
3161 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3162 ctrl->ctratt = le32_to_cpu(id->ctratt);
3166 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3168 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3169 shutdown_timeout, 60);
3171 if (ctrl->shutdown_timeout != shutdown_timeout)
3172 dev_info(ctrl->device,
3173 "Shutdown timeout set to %u seconds\n",
3174 ctrl->shutdown_timeout);
3176 ctrl->shutdown_timeout = shutdown_timeout;
3178 ctrl->npss = id->npss;
3179 ctrl->apsta = id->apsta;
3180 prev_apst_enabled = ctrl->apst_enabled;
3181 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3182 if (force_apst && id->apsta) {
3183 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3184 ctrl->apst_enabled = true;
3186 ctrl->apst_enabled = false;
3189 ctrl->apst_enabled = id->apsta;
3191 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3193 if (ctrl->ops->flags & NVME_F_FABRICS) {
3194 ctrl->icdoff = le16_to_cpu(id->icdoff);
3195 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3196 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3197 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3200 * In fabrics we need to verify the cntlid matches the
3203 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3204 dev_err(ctrl->device,
3205 "Mismatching cntlid: Connect %u vs Identify "
3207 ctrl->cntlid, le16_to_cpu(id->cntlid));
3212 if (!ctrl->opts->discovery_nqn && !ctrl->kas) {
3213 dev_err(ctrl->device,
3214 "keep-alive support is mandatory for fabrics\n");
3219 ctrl->hmpre = le32_to_cpu(id->hmpre);
3220 ctrl->hmmin = le32_to_cpu(id->hmmin);
3221 ctrl->hmminds = le32_to_cpu(id->hmminds);
3222 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3225 ret = nvme_mpath_init(ctrl, id);
3231 if (ctrl->apst_enabled && !prev_apst_enabled)
3232 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3233 else if (!ctrl->apst_enabled && prev_apst_enabled)
3234 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3236 ret = nvme_configure_apst(ctrl);
3240 ret = nvme_configure_timestamp(ctrl);
3244 ret = nvme_configure_directives(ctrl);
3248 ret = nvme_configure_acre(ctrl);
3252 if (!ctrl->identified)
3253 nvme_hwmon_init(ctrl);
3255 ctrl->identified = true;
3263 EXPORT_SYMBOL_GPL(nvme_init_identify);
3265 static int nvme_dev_open(struct inode *inode, struct file *file)
3267 struct nvme_ctrl *ctrl =
3268 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3270 switch (ctrl->state) {
3271 case NVME_CTRL_LIVE:
3274 return -EWOULDBLOCK;
3277 file->private_data = ctrl;
3281 static int nvme_dev_user_cmd(struct nvme_ctrl *ctrl, void __user *argp)
3286 down_read(&ctrl->namespaces_rwsem);
3287 if (list_empty(&ctrl->namespaces)) {
3292 ns = list_first_entry(&ctrl->namespaces, struct nvme_ns, list);
3293 if (ns != list_last_entry(&ctrl->namespaces, struct nvme_ns, list)) {
3294 dev_warn(ctrl->device,
3295 "NVME_IOCTL_IO_CMD not supported when multiple namespaces present!\n");
3300 dev_warn(ctrl->device,
3301 "using deprecated NVME_IOCTL_IO_CMD ioctl on the char device!\n");
3302 kref_get(&ns->kref);
3303 up_read(&ctrl->namespaces_rwsem);
3305 ret = nvme_user_cmd(ctrl, ns, argp);
3310 up_read(&ctrl->namespaces_rwsem);
3314 static long nvme_dev_ioctl(struct file *file, unsigned int cmd,
3317 struct nvme_ctrl *ctrl = file->private_data;
3318 void __user *argp = (void __user *)arg;
3321 case NVME_IOCTL_ADMIN_CMD:
3322 return nvme_user_cmd(ctrl, NULL, argp);
3323 case NVME_IOCTL_ADMIN64_CMD:
3324 return nvme_user_cmd64(ctrl, NULL, argp);
3325 case NVME_IOCTL_IO_CMD:
3326 return nvme_dev_user_cmd(ctrl, argp);
3327 case NVME_IOCTL_RESET:
3328 dev_warn(ctrl->device, "resetting controller\n");
3329 return nvme_reset_ctrl_sync(ctrl);
3330 case NVME_IOCTL_SUBSYS_RESET:
3331 return nvme_reset_subsystem(ctrl);
3332 case NVME_IOCTL_RESCAN:
3333 nvme_queue_scan(ctrl);
3340 static const struct file_operations nvme_dev_fops = {
3341 .owner = THIS_MODULE,
3342 .open = nvme_dev_open,
3343 .unlocked_ioctl = nvme_dev_ioctl,
3344 .compat_ioctl = compat_ptr_ioctl,
3347 static ssize_t nvme_sysfs_reset(struct device *dev,
3348 struct device_attribute *attr, const char *buf,
3351 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3354 ret = nvme_reset_ctrl_sync(ctrl);
3359 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3361 static ssize_t nvme_sysfs_rescan(struct device *dev,
3362 struct device_attribute *attr, const char *buf,
3365 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3367 nvme_queue_scan(ctrl);
3370 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3372 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3374 struct gendisk *disk = dev_to_disk(dev);
3376 if (disk->fops == &nvme_fops)
3377 return nvme_get_ns_from_dev(dev)->head;
3379 return disk->private_data;
3382 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3385 struct nvme_ns_head *head = dev_to_ns_head(dev);
3386 struct nvme_ns_ids *ids = &head->ids;
3387 struct nvme_subsystem *subsys = head->subsys;
3388 int serial_len = sizeof(subsys->serial);
3389 int model_len = sizeof(subsys->model);
3391 if (!uuid_is_null(&ids->uuid))
3392 return sprintf(buf, "uuid.%pU\n", &ids->uuid);
3394 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3395 return sprintf(buf, "eui.%16phN\n", ids->nguid);
3397 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3398 return sprintf(buf, "eui.%8phN\n", ids->eui64);
3400 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3401 subsys->serial[serial_len - 1] == '\0'))
3403 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3404 subsys->model[model_len - 1] == '\0'))
3407 return sprintf(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3408 serial_len, subsys->serial, model_len, subsys->model,
3411 static DEVICE_ATTR_RO(wwid);
3413 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3416 return sprintf(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3418 static DEVICE_ATTR_RO(nguid);
3420 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3423 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3425 /* For backward compatibility expose the NGUID to userspace if
3426 * we have no UUID set
3428 if (uuid_is_null(&ids->uuid)) {
3429 printk_ratelimited(KERN_WARNING
3430 "No UUID available providing old NGUID\n");
3431 return sprintf(buf, "%pU\n", ids->nguid);
3433 return sprintf(buf, "%pU\n", &ids->uuid);
3435 static DEVICE_ATTR_RO(uuid);
3437 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3440 return sprintf(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3442 static DEVICE_ATTR_RO(eui);
3444 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3447 return sprintf(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3449 static DEVICE_ATTR_RO(nsid);
3451 static struct attribute *nvme_ns_id_attrs[] = {
3452 &dev_attr_wwid.attr,
3453 &dev_attr_uuid.attr,
3454 &dev_attr_nguid.attr,
3456 &dev_attr_nsid.attr,
3457 #ifdef CONFIG_NVME_MULTIPATH
3458 &dev_attr_ana_grpid.attr,
3459 &dev_attr_ana_state.attr,
3464 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3465 struct attribute *a, int n)
3467 struct device *dev = container_of(kobj, struct device, kobj);
3468 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3470 if (a == &dev_attr_uuid.attr) {
3471 if (uuid_is_null(&ids->uuid) &&
3472 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3475 if (a == &dev_attr_nguid.attr) {
3476 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3479 if (a == &dev_attr_eui.attr) {
3480 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3483 #ifdef CONFIG_NVME_MULTIPATH
3484 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3485 if (dev_to_disk(dev)->fops != &nvme_fops) /* per-path attr */
3487 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3494 static const struct attribute_group nvme_ns_id_attr_group = {
3495 .attrs = nvme_ns_id_attrs,
3496 .is_visible = nvme_ns_id_attrs_are_visible,
3499 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3500 &nvme_ns_id_attr_group,
3502 &nvme_nvm_attr_group,
3507 #define nvme_show_str_function(field) \
3508 static ssize_t field##_show(struct device *dev, \
3509 struct device_attribute *attr, char *buf) \
3511 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3512 return sprintf(buf, "%.*s\n", \
3513 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3515 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3517 nvme_show_str_function(model);
3518 nvme_show_str_function(serial);
3519 nvme_show_str_function(firmware_rev);
3521 #define nvme_show_int_function(field) \
3522 static ssize_t field##_show(struct device *dev, \
3523 struct device_attribute *attr, char *buf) \
3525 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3526 return sprintf(buf, "%d\n", ctrl->field); \
3528 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3530 nvme_show_int_function(cntlid);
3531 nvme_show_int_function(numa_node);
3532 nvme_show_int_function(queue_count);
3533 nvme_show_int_function(sqsize);
3535 static ssize_t nvme_sysfs_delete(struct device *dev,
3536 struct device_attribute *attr, const char *buf,
3539 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3541 /* Can't delete non-created controllers */
3545 if (device_remove_file_self(dev, attr))
3546 nvme_delete_ctrl_sync(ctrl);
3549 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3551 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3552 struct device_attribute *attr,
3555 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3557 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->ops->name);
3559 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3561 static ssize_t nvme_sysfs_show_state(struct device *dev,
3562 struct device_attribute *attr,
3565 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3566 static const char *const state_name[] = {
3567 [NVME_CTRL_NEW] = "new",
3568 [NVME_CTRL_LIVE] = "live",
3569 [NVME_CTRL_RESETTING] = "resetting",
3570 [NVME_CTRL_CONNECTING] = "connecting",
3571 [NVME_CTRL_DELETING] = "deleting",
3572 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3573 [NVME_CTRL_DEAD] = "dead",
3576 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3577 state_name[ctrl->state])
3578 return sprintf(buf, "%s\n", state_name[ctrl->state]);
3580 return sprintf(buf, "unknown state\n");
3583 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3585 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3586 struct device_attribute *attr,
3589 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3591 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->subsys->subnqn);
3593 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3595 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3596 struct device_attribute *attr,
3599 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3601 return snprintf(buf, PAGE_SIZE, "%s\n", ctrl->opts->host->nqn);
3603 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3605 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3606 struct device_attribute *attr,
3609 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3611 return snprintf(buf, PAGE_SIZE, "%pU\n", &ctrl->opts->host->id);
3613 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3615 static ssize_t nvme_sysfs_show_address(struct device *dev,
3616 struct device_attribute *attr,
3619 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3621 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3623 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3625 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3626 struct device_attribute *attr, char *buf)
3628 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3629 struct nvmf_ctrl_options *opts = ctrl->opts;
3631 if (ctrl->opts->max_reconnects == -1)
3632 return sprintf(buf, "off\n");
3633 return sprintf(buf, "%d\n",
3634 opts->max_reconnects * opts->reconnect_delay);
3637 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3638 struct device_attribute *attr, const char *buf, size_t count)
3640 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3641 struct nvmf_ctrl_options *opts = ctrl->opts;
3642 int ctrl_loss_tmo, err;
3644 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3648 else if (ctrl_loss_tmo < 0)
3649 opts->max_reconnects = -1;
3651 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3652 opts->reconnect_delay);
3655 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3656 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3658 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3659 struct device_attribute *attr, char *buf)
3661 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3663 if (ctrl->opts->reconnect_delay == -1)
3664 return sprintf(buf, "off\n");
3665 return sprintf(buf, "%d\n", ctrl->opts->reconnect_delay);
3668 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3669 struct device_attribute *attr, const char *buf, size_t count)
3671 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3675 err = kstrtou32(buf, 10, &v);
3679 ctrl->opts->reconnect_delay = v;
3682 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3683 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3685 static struct attribute *nvme_dev_attrs[] = {
3686 &dev_attr_reset_controller.attr,
3687 &dev_attr_rescan_controller.attr,
3688 &dev_attr_model.attr,
3689 &dev_attr_serial.attr,
3690 &dev_attr_firmware_rev.attr,
3691 &dev_attr_cntlid.attr,
3692 &dev_attr_delete_controller.attr,
3693 &dev_attr_transport.attr,
3694 &dev_attr_subsysnqn.attr,
3695 &dev_attr_address.attr,
3696 &dev_attr_state.attr,
3697 &dev_attr_numa_node.attr,
3698 &dev_attr_queue_count.attr,
3699 &dev_attr_sqsize.attr,
3700 &dev_attr_hostnqn.attr,
3701 &dev_attr_hostid.attr,
3702 &dev_attr_ctrl_loss_tmo.attr,
3703 &dev_attr_reconnect_delay.attr,
3707 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3708 struct attribute *a, int n)
3710 struct device *dev = container_of(kobj, struct device, kobj);
3711 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3713 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3715 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3717 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3719 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3721 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3723 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3729 static struct attribute_group nvme_dev_attrs_group = {
3730 .attrs = nvme_dev_attrs,
3731 .is_visible = nvme_dev_attrs_are_visible,
3734 static const struct attribute_group *nvme_dev_attr_groups[] = {
3735 &nvme_dev_attrs_group,
3739 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3742 struct nvme_ns_head *h;
3744 lockdep_assert_held(&subsys->lock);
3746 list_for_each_entry(h, &subsys->nsheads, entry) {
3747 if (h->ns_id == nsid && kref_get_unless_zero(&h->ref))
3754 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3755 struct nvme_ns_head *new)
3757 struct nvme_ns_head *h;
3759 lockdep_assert_held(&subsys->lock);
3761 list_for_each_entry(h, &subsys->nsheads, entry) {
3762 if (nvme_ns_ids_valid(&new->ids) &&
3763 nvme_ns_ids_equal(&new->ids, &h->ids))
3770 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3771 unsigned nsid, struct nvme_ns_ids *ids)
3773 struct nvme_ns_head *head;
3774 size_t size = sizeof(*head);
3777 #ifdef CONFIG_NVME_MULTIPATH
3778 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3781 head = kzalloc(size, GFP_KERNEL);
3784 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3787 head->instance = ret;
3788 INIT_LIST_HEAD(&head->list);
3789 ret = init_srcu_struct(&head->srcu);
3791 goto out_ida_remove;
3792 head->subsys = ctrl->subsys;
3795 kref_init(&head->ref);
3797 ret = __nvme_check_ids(ctrl->subsys, head);
3799 dev_err(ctrl->device,
3800 "duplicate IDs for nsid %d\n", nsid);
3801 goto out_cleanup_srcu;
3804 if (head->ids.csi) {
3805 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3807 goto out_cleanup_srcu;
3809 head->effects = ctrl->effects;
3811 ret = nvme_mpath_alloc_disk(ctrl, head);
3813 goto out_cleanup_srcu;
3815 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3817 kref_get(&ctrl->subsys->ref);
3821 cleanup_srcu_struct(&head->srcu);
3823 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3828 ret = blk_status_to_errno(nvme_error_status(ret));
3829 return ERR_PTR(ret);
3832 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3833 struct nvme_id_ns *id)
3835 struct nvme_ctrl *ctrl = ns->ctrl;
3836 bool is_shared = id->nmic & NVME_NS_NMIC_SHARED;
3837 struct nvme_ns_head *head = NULL;
3838 struct nvme_ns_ids ids;
3841 ret = nvme_report_ns_ids(ctrl, nsid, id, &ids);
3845 return blk_status_to_errno(nvme_error_status(ret));
3848 mutex_lock(&ctrl->subsys->lock);
3849 head = nvme_find_ns_head(ctrl->subsys, nsid);
3851 head = nvme_alloc_ns_head(ctrl, nsid, &ids);
3853 ret = PTR_ERR(head);
3856 head->shared = is_shared;
3859 if (!is_shared || !head->shared) {
3860 dev_err(ctrl->device,
3861 "Duplicate unshared namespace %d\n", nsid);
3862 goto out_put_ns_head;
3864 if (!nvme_ns_ids_equal(&head->ids, &ids)) {
3865 dev_err(ctrl->device,
3866 "IDs don't match for shared namespace %d\n",
3868 goto out_put_ns_head;
3872 list_add_tail(&ns->siblings, &head->list);
3874 mutex_unlock(&ctrl->subsys->lock);
3878 nvme_put_ns_head(head);
3880 mutex_unlock(&ctrl->subsys->lock);
3884 static int ns_cmp(void *priv, struct list_head *a, struct list_head *b)
3886 struct nvme_ns *nsa = container_of(a, struct nvme_ns, list);
3887 struct nvme_ns *nsb = container_of(b, struct nvme_ns, list);
3889 return nsa->head->ns_id - nsb->head->ns_id;
3892 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3894 struct nvme_ns *ns, *ret = NULL;
3896 down_read(&ctrl->namespaces_rwsem);
3897 list_for_each_entry(ns, &ctrl->namespaces, list) {
3898 if (ns->head->ns_id == nsid) {
3899 if (!kref_get_unless_zero(&ns->kref))
3904 if (ns->head->ns_id > nsid)
3907 up_read(&ctrl->namespaces_rwsem);
3910 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3912 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3915 struct gendisk *disk;
3916 struct nvme_id_ns *id;
3917 char disk_name[DISK_NAME_LEN];
3918 int node = ctrl->numa_node, flags = GENHD_FL_EXT_DEVT, ret;
3920 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3924 ns->queue = blk_mq_init_queue(ctrl->tagset);
3925 if (IS_ERR(ns->queue))
3928 if (ctrl->opts && ctrl->opts->data_digest)
3929 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3931 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3932 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3933 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3935 ns->queue->queuedata = ns;
3938 kref_init(&ns->kref);
3939 ns->lba_shift = 9; /* set to a default value for 512 until disk is validated */
3941 blk_queue_logical_block_size(ns->queue, 1 << ns->lba_shift);
3942 nvme_set_queue_limits(ctrl, ns->queue);
3944 ret = nvme_identify_ns(ctrl, nsid, &id);
3946 goto out_free_queue;
3948 if (id->ncap == 0) /* no namespace (legacy quirk) */
3951 ret = nvme_init_ns_head(ns, nsid, id);
3954 nvme_set_disk_name(disk_name, ns, ctrl, &flags);
3956 disk = alloc_disk_node(0, node);
3960 disk->fops = &nvme_fops;
3961 disk->private_data = ns;
3962 disk->queue = ns->queue;
3963 disk->flags = flags;
3964 memcpy(disk->disk_name, disk_name, DISK_NAME_LEN);
3967 if (__nvme_revalidate_disk(disk, id))
3970 if ((ctrl->quirks & NVME_QUIRK_LIGHTNVM) && id->vs[0] == 0x1) {
3971 ret = nvme_nvm_register(ns, disk_name, node);
3973 dev_warn(ctrl->device, "LightNVM init failure\n");
3978 down_write(&ctrl->namespaces_rwsem);
3979 list_add_tail(&ns->list, &ctrl->namespaces);
3980 up_write(&ctrl->namespaces_rwsem);
3982 nvme_get_ctrl(ctrl);
3984 device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups);
3986 nvme_mpath_add_disk(ns, id);
3987 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3992 /* prevent double queue cleanup */
3993 ns->disk->queue = NULL;
3996 mutex_lock(&ctrl->subsys->lock);
3997 list_del_rcu(&ns->siblings);
3998 if (list_empty(&ns->head->list))
3999 list_del_init(&ns->head->entry);
4000 mutex_unlock(&ctrl->subsys->lock);
4001 nvme_put_ns_head(ns->head);
4005 blk_cleanup_queue(ns->queue);
4010 static void nvme_ns_remove(struct nvme_ns *ns)
4012 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4015 nvme_fault_inject_fini(&ns->fault_inject);
4017 mutex_lock(&ns->ctrl->subsys->lock);
4018 list_del_rcu(&ns->siblings);
4019 if (list_empty(&ns->head->list))
4020 list_del_init(&ns->head->entry);
4021 mutex_unlock(&ns->ctrl->subsys->lock);
4023 synchronize_rcu(); /* guarantee not available in head->list */
4024 nvme_mpath_clear_current_path(ns);
4025 synchronize_srcu(&ns->head->srcu); /* wait for concurrent submissions */
4027 if (ns->disk->flags & GENHD_FL_UP) {
4028 del_gendisk(ns->disk);
4029 blk_cleanup_queue(ns->queue);
4030 if (blk_get_integrity(ns->disk))
4031 blk_integrity_unregister(ns->disk);
4034 down_write(&ns->ctrl->namespaces_rwsem);
4035 list_del_init(&ns->list);
4036 up_write(&ns->ctrl->namespaces_rwsem);
4038 nvme_mpath_check_last_path(ns);
4042 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4044 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4052 static void nvme_validate_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4057 ns = nvme_find_get_ns(ctrl, nsid);
4059 nvme_alloc_ns(ctrl, nsid);
4063 ret = nvme_revalidate_disk(ns->disk);
4064 revalidate_disk_size(ns->disk, ret == 0);
4070 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4073 struct nvme_ns *ns, *next;
4076 down_write(&ctrl->namespaces_rwsem);
4077 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4078 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4079 list_move_tail(&ns->list, &rm_list);
4081 up_write(&ctrl->namespaces_rwsem);
4083 list_for_each_entry_safe(ns, next, &rm_list, list)
4088 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4090 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4095 if (nvme_ctrl_limited_cns(ctrl))
4098 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4103 ret = nvme_identify_ns_list(ctrl, prev, ns_list);
4107 for (i = 0; i < nr_entries; i++) {
4108 u32 nsid = le32_to_cpu(ns_list[i]);
4110 if (!nsid) /* end of the list? */
4112 nvme_validate_ns(ctrl, nsid);
4113 while (++prev < nsid)
4114 nvme_ns_remove_by_nsid(ctrl, prev);
4118 nvme_remove_invalid_namespaces(ctrl, prev);
4124 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4126 struct nvme_id_ctrl *id;
4129 if (nvme_identify_ctrl(ctrl, &id))
4131 nn = le32_to_cpu(id->nn);
4134 for (i = 1; i <= nn; i++)
4135 nvme_validate_ns(ctrl, i);
4137 nvme_remove_invalid_namespaces(ctrl, nn);
4140 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4142 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4146 log = kzalloc(log_size, GFP_KERNEL);
4151 * We need to read the log to clear the AEN, but we don't want to rely
4152 * on it for the changed namespace information as userspace could have
4153 * raced with us in reading the log page, which could cause us to miss
4156 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4157 NVME_CSI_NVM, log, log_size, 0);
4159 dev_warn(ctrl->device,
4160 "reading changed ns log failed: %d\n", error);
4165 static void nvme_scan_work(struct work_struct *work)
4167 struct nvme_ctrl *ctrl =
4168 container_of(work, struct nvme_ctrl, scan_work);
4170 /* No tagset on a live ctrl means IO queues could not created */
4171 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4174 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4175 dev_info(ctrl->device, "rescanning namespaces.\n");
4176 nvme_clear_changed_ns_log(ctrl);
4179 mutex_lock(&ctrl->scan_lock);
4180 if (nvme_scan_ns_list(ctrl) != 0)
4181 nvme_scan_ns_sequential(ctrl);
4182 mutex_unlock(&ctrl->scan_lock);
4184 down_write(&ctrl->namespaces_rwsem);
4185 list_sort(NULL, &ctrl->namespaces, ns_cmp);
4186 up_write(&ctrl->namespaces_rwsem);
4190 * This function iterates the namespace list unlocked to allow recovery from
4191 * controller failure. It is up to the caller to ensure the namespace list is
4192 * not modified by scan work while this function is executing.
4194 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4196 struct nvme_ns *ns, *next;
4200 * make sure to requeue I/O to all namespaces as these
4201 * might result from the scan itself and must complete
4202 * for the scan_work to make progress
4204 nvme_mpath_clear_ctrl_paths(ctrl);
4206 /* prevent racing with ns scanning */
4207 flush_work(&ctrl->scan_work);
4210 * The dead states indicates the controller was not gracefully
4211 * disconnected. In that case, we won't be able to flush any data while
4212 * removing the namespaces' disks; fail all the queues now to avoid
4213 * potentially having to clean up the failed sync later.
4215 if (ctrl->state == NVME_CTRL_DEAD)
4216 nvme_kill_queues(ctrl);
4218 /* this is a no-op when called from the controller reset handler */
4219 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4221 down_write(&ctrl->namespaces_rwsem);
4222 list_splice_init(&ctrl->namespaces, &ns_list);
4223 up_write(&ctrl->namespaces_rwsem);
4225 list_for_each_entry_safe(ns, next, &ns_list, list)
4228 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4230 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4232 struct nvme_ctrl *ctrl =
4233 container_of(dev, struct nvme_ctrl, ctrl_device);
4234 struct nvmf_ctrl_options *opts = ctrl->opts;
4237 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4242 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4246 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4247 opts->trsvcid ?: "none");
4251 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4252 opts->host_traddr ?: "none");
4257 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4259 char *envp[2] = { NULL, NULL };
4260 u32 aen_result = ctrl->aen_result;
4262 ctrl->aen_result = 0;
4266 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4269 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4273 static void nvme_async_event_work(struct work_struct *work)
4275 struct nvme_ctrl *ctrl =
4276 container_of(work, struct nvme_ctrl, async_event_work);
4278 nvme_aen_uevent(ctrl);
4279 ctrl->ops->submit_async_event(ctrl);
4282 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4287 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4293 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4296 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4298 struct nvme_fw_slot_info_log *log;
4300 log = kmalloc(sizeof(*log), GFP_KERNEL);
4304 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4305 log, sizeof(*log), 0))
4306 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4310 static void nvme_fw_act_work(struct work_struct *work)
4312 struct nvme_ctrl *ctrl = container_of(work,
4313 struct nvme_ctrl, fw_act_work);
4314 unsigned long fw_act_timeout;
4317 fw_act_timeout = jiffies +
4318 msecs_to_jiffies(ctrl->mtfa * 100);
4320 fw_act_timeout = jiffies +
4321 msecs_to_jiffies(admin_timeout * 1000);
4323 nvme_stop_queues(ctrl);
4324 while (nvme_ctrl_pp_status(ctrl)) {
4325 if (time_after(jiffies, fw_act_timeout)) {
4326 dev_warn(ctrl->device,
4327 "Fw activation timeout, reset controller\n");
4328 nvme_try_sched_reset(ctrl);
4334 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4337 nvme_start_queues(ctrl);
4338 /* read FW slot information to clear the AER */
4339 nvme_get_fw_slot_info(ctrl);
4342 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4344 u32 aer_notice_type = (result & 0xff00) >> 8;
4346 trace_nvme_async_event(ctrl, aer_notice_type);
4348 switch (aer_notice_type) {
4349 case NVME_AER_NOTICE_NS_CHANGED:
4350 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4351 nvme_queue_scan(ctrl);
4353 case NVME_AER_NOTICE_FW_ACT_STARTING:
4355 * We are (ab)using the RESETTING state to prevent subsequent
4356 * recovery actions from interfering with the controller's
4357 * firmware activation.
4359 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4360 queue_work(nvme_wq, &ctrl->fw_act_work);
4362 #ifdef CONFIG_NVME_MULTIPATH
4363 case NVME_AER_NOTICE_ANA:
4364 if (!ctrl->ana_log_buf)
4366 queue_work(nvme_wq, &ctrl->ana_work);
4369 case NVME_AER_NOTICE_DISC_CHANGED:
4370 ctrl->aen_result = result;
4373 dev_warn(ctrl->device, "async event result %08x\n", result);
4377 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4378 volatile union nvme_result *res)
4380 u32 result = le32_to_cpu(res->u32);
4381 u32 aer_type = result & 0x07;
4383 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4387 case NVME_AER_NOTICE:
4388 nvme_handle_aen_notice(ctrl, result);
4390 case NVME_AER_ERROR:
4391 case NVME_AER_SMART:
4394 trace_nvme_async_event(ctrl, aer_type);
4395 ctrl->aen_result = result;
4400 queue_work(nvme_wq, &ctrl->async_event_work);
4402 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4404 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4406 nvme_mpath_stop(ctrl);
4407 nvme_stop_keep_alive(ctrl);
4408 flush_work(&ctrl->async_event_work);
4409 cancel_work_sync(&ctrl->fw_act_work);
4411 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4413 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4415 nvme_start_keep_alive(ctrl);
4417 nvme_enable_aen(ctrl);
4419 if (ctrl->queue_count > 1) {
4420 nvme_queue_scan(ctrl);
4421 nvme_start_queues(ctrl);
4423 ctrl->created = true;
4425 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4427 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4429 nvme_fault_inject_fini(&ctrl->fault_inject);
4430 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4431 cdev_device_del(&ctrl->cdev, ctrl->device);
4432 nvme_put_ctrl(ctrl);
4434 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4436 static void nvme_free_ctrl(struct device *dev)
4438 struct nvme_ctrl *ctrl =
4439 container_of(dev, struct nvme_ctrl, ctrl_device);
4440 struct nvme_subsystem *subsys = ctrl->subsys;
4441 struct nvme_cel *cel, *next;
4443 if (!subsys || ctrl->instance != subsys->instance)
4444 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4446 list_for_each_entry_safe(cel, next, &ctrl->cels, entry) {
4447 list_del(&cel->entry);
4451 nvme_mpath_uninit(ctrl);
4452 __free_page(ctrl->discard_page);
4455 mutex_lock(&nvme_subsystems_lock);
4456 list_del(&ctrl->subsys_entry);
4457 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4458 mutex_unlock(&nvme_subsystems_lock);
4461 ctrl->ops->free_ctrl(ctrl);
4464 nvme_put_subsystem(subsys);
4468 * Initialize a NVMe controller structures. This needs to be called during
4469 * earliest initialization so that we have the initialized structured around
4472 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4473 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4477 ctrl->state = NVME_CTRL_NEW;
4478 spin_lock_init(&ctrl->lock);
4479 mutex_init(&ctrl->scan_lock);
4480 INIT_LIST_HEAD(&ctrl->namespaces);
4481 INIT_LIST_HEAD(&ctrl->cels);
4482 init_rwsem(&ctrl->namespaces_rwsem);
4485 ctrl->quirks = quirks;
4486 ctrl->numa_node = NUMA_NO_NODE;
4487 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4488 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4489 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4490 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4491 init_waitqueue_head(&ctrl->state_wq);
4493 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4494 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4495 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4497 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4499 ctrl->discard_page = alloc_page(GFP_KERNEL);
4500 if (!ctrl->discard_page) {
4505 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4508 ctrl->instance = ret;
4510 device_initialize(&ctrl->ctrl_device);
4511 ctrl->device = &ctrl->ctrl_device;
4512 ctrl->device->devt = MKDEV(MAJOR(nvme_chr_devt), ctrl->instance);
4513 ctrl->device->class = nvme_class;
4514 ctrl->device->parent = ctrl->dev;
4515 ctrl->device->groups = nvme_dev_attr_groups;
4516 ctrl->device->release = nvme_free_ctrl;
4517 dev_set_drvdata(ctrl->device, ctrl);
4518 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4520 goto out_release_instance;
4522 nvme_get_ctrl(ctrl);
4523 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4524 ctrl->cdev.owner = ops->module;
4525 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4530 * Initialize latency tolerance controls. The sysfs files won't
4531 * be visible to userspace unless the device actually supports APST.
4533 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4534 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4535 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4537 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4541 nvme_put_ctrl(ctrl);
4542 kfree_const(ctrl->device->kobj.name);
4543 out_release_instance:
4544 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4546 if (ctrl->discard_page)
4547 __free_page(ctrl->discard_page);
4550 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4553 * nvme_kill_queues(): Ends all namespace queues
4554 * @ctrl: the dead controller that needs to end
4556 * Call this function when the driver determines it is unable to get the
4557 * controller in a state capable of servicing IO.
4559 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4563 down_read(&ctrl->namespaces_rwsem);
4565 /* Forcibly unquiesce queues to avoid blocking dispatch */
4566 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4567 blk_mq_unquiesce_queue(ctrl->admin_q);
4569 list_for_each_entry(ns, &ctrl->namespaces, list)
4570 nvme_set_queue_dying(ns);
4572 up_read(&ctrl->namespaces_rwsem);
4574 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4576 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4580 down_read(&ctrl->namespaces_rwsem);
4581 list_for_each_entry(ns, &ctrl->namespaces, list)
4582 blk_mq_unfreeze_queue(ns->queue);
4583 up_read(&ctrl->namespaces_rwsem);
4585 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4587 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4591 down_read(&ctrl->namespaces_rwsem);
4592 list_for_each_entry(ns, &ctrl->namespaces, list) {
4593 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4597 up_read(&ctrl->namespaces_rwsem);
4600 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4602 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4606 down_read(&ctrl->namespaces_rwsem);
4607 list_for_each_entry(ns, &ctrl->namespaces, list)
4608 blk_mq_freeze_queue_wait(ns->queue);
4609 up_read(&ctrl->namespaces_rwsem);
4611 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4613 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4617 down_read(&ctrl->namespaces_rwsem);
4618 list_for_each_entry(ns, &ctrl->namespaces, list)
4619 blk_freeze_queue_start(ns->queue);
4620 up_read(&ctrl->namespaces_rwsem);
4622 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4624 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4628 down_read(&ctrl->namespaces_rwsem);
4629 list_for_each_entry(ns, &ctrl->namespaces, list)
4630 blk_mq_quiesce_queue(ns->queue);
4631 up_read(&ctrl->namespaces_rwsem);
4633 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4635 void nvme_start_queues(struct nvme_ctrl *ctrl)
4639 down_read(&ctrl->namespaces_rwsem);
4640 list_for_each_entry(ns, &ctrl->namespaces, list)
4641 blk_mq_unquiesce_queue(ns->queue);
4642 up_read(&ctrl->namespaces_rwsem);
4644 EXPORT_SYMBOL_GPL(nvme_start_queues);
4647 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4651 down_read(&ctrl->namespaces_rwsem);
4652 list_for_each_entry(ns, &ctrl->namespaces, list)
4653 blk_sync_queue(ns->queue);
4654 up_read(&ctrl->namespaces_rwsem);
4657 blk_sync_queue(ctrl->admin_q);
4659 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4661 struct nvme_ctrl *nvme_ctrl_get_by_path(const char *path)
4663 struct nvme_ctrl *ctrl;
4666 f = filp_open(path, O_RDWR, 0);
4670 if (f->f_op != &nvme_dev_fops) {
4671 ctrl = ERR_PTR(-EINVAL);
4675 ctrl = f->private_data;
4676 nvme_get_ctrl(ctrl);
4679 filp_close(f, NULL);
4682 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_get_by_path, NVME_TARGET_PASSTHRU);
4685 * Check we didn't inadvertently grow the command structure sizes:
4687 static inline void _nvme_check_size(void)
4689 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4690 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4691 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4692 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4693 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4694 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4695 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4696 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4697 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4698 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4699 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4700 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4701 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4702 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4703 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4704 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4705 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4706 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4707 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4711 static int __init nvme_core_init(void)
4713 int result = -ENOMEM;
4717 nvme_wq = alloc_workqueue("nvme-wq",
4718 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4722 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4723 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4727 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4728 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4729 if (!nvme_delete_wq)
4730 goto destroy_reset_wq;
4732 result = alloc_chrdev_region(&nvme_chr_devt, 0, NVME_MINORS, "nvme");
4734 goto destroy_delete_wq;
4736 nvme_class = class_create(THIS_MODULE, "nvme");
4737 if (IS_ERR(nvme_class)) {
4738 result = PTR_ERR(nvme_class);
4739 goto unregister_chrdev;
4741 nvme_class->dev_uevent = nvme_class_uevent;
4743 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4744 if (IS_ERR(nvme_subsys_class)) {
4745 result = PTR_ERR(nvme_subsys_class);
4751 class_destroy(nvme_class);
4753 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4755 destroy_workqueue(nvme_delete_wq);
4757 destroy_workqueue(nvme_reset_wq);
4759 destroy_workqueue(nvme_wq);
4764 static void __exit nvme_core_exit(void)
4766 class_destroy(nvme_subsys_class);
4767 class_destroy(nvme_class);
4768 unregister_chrdev_region(nvme_chr_devt, NVME_MINORS);
4769 destroy_workqueue(nvme_delete_wq);
4770 destroy_workqueue(nvme_reset_wq);
4771 destroy_workqueue(nvme_wq);
4772 ida_destroy(&nvme_instance_ida);
4775 MODULE_LICENSE("GPL");
4776 MODULE_VERSION("1.0");
4777 module_init(nvme_core_init);
4778 module_exit(nvme_core_exit);