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/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.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>
27 #include <linux/nvme-auth.h>
29 #define CREATE_TRACE_POINTS
32 #define NVME_MINORS (1U << MINORBITS)
35 struct nvme_ns_ids ids;
44 unsigned int admin_timeout = 60;
45 module_param(admin_timeout, uint, 0644);
46 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
47 EXPORT_SYMBOL_GPL(admin_timeout);
49 unsigned int nvme_io_timeout = 30;
50 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
51 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
52 EXPORT_SYMBOL_GPL(nvme_io_timeout);
54 static unsigned char shutdown_timeout = 5;
55 module_param(shutdown_timeout, byte, 0644);
56 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
58 static u8 nvme_max_retries = 5;
59 module_param_named(max_retries, nvme_max_retries, byte, 0644);
60 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
62 static unsigned long default_ps_max_latency_us = 100000;
63 module_param(default_ps_max_latency_us, ulong, 0644);
64 MODULE_PARM_DESC(default_ps_max_latency_us,
65 "max power saving latency for new devices; use PM QOS to change per device");
67 static bool force_apst;
68 module_param(force_apst, bool, 0644);
69 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
71 static unsigned long apst_primary_timeout_ms = 100;
72 module_param(apst_primary_timeout_ms, ulong, 0644);
73 MODULE_PARM_DESC(apst_primary_timeout_ms,
74 "primary APST timeout in ms");
76 static unsigned long apst_secondary_timeout_ms = 2000;
77 module_param(apst_secondary_timeout_ms, ulong, 0644);
78 MODULE_PARM_DESC(apst_secondary_timeout_ms,
79 "secondary APST timeout in ms");
81 static unsigned long apst_primary_latency_tol_us = 15000;
82 module_param(apst_primary_latency_tol_us, ulong, 0644);
83 MODULE_PARM_DESC(apst_primary_latency_tol_us,
84 "primary APST latency tolerance in us");
86 static unsigned long apst_secondary_latency_tol_us = 100000;
87 module_param(apst_secondary_latency_tol_us, ulong, 0644);
88 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
89 "secondary APST latency tolerance in us");
92 * nvme_wq - hosts nvme related works that are not reset or delete
93 * nvme_reset_wq - hosts nvme reset works
94 * nvme_delete_wq - hosts nvme delete works
96 * nvme_wq will host works such as scan, aen handling, fw activation,
97 * keep-alive, periodic reconnects etc. nvme_reset_wq
98 * runs reset works which also flush works hosted on nvme_wq for
99 * serialization purposes. nvme_delete_wq host controller deletion
100 * works which flush reset works for serialization.
102 struct workqueue_struct *nvme_wq;
103 EXPORT_SYMBOL_GPL(nvme_wq);
105 struct workqueue_struct *nvme_reset_wq;
106 EXPORT_SYMBOL_GPL(nvme_reset_wq);
108 struct workqueue_struct *nvme_delete_wq;
109 EXPORT_SYMBOL_GPL(nvme_delete_wq);
111 static LIST_HEAD(nvme_subsystems);
112 static DEFINE_MUTEX(nvme_subsystems_lock);
114 static DEFINE_IDA(nvme_instance_ida);
115 static dev_t nvme_ctrl_base_chr_devt;
116 static struct class *nvme_class;
117 static struct class *nvme_subsys_class;
119 static DEFINE_IDA(nvme_ns_chr_minor_ida);
120 static dev_t nvme_ns_chr_devt;
121 static struct class *nvme_ns_chr_class;
123 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
124 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
126 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
127 struct nvme_command *cmd);
129 void nvme_queue_scan(struct nvme_ctrl *ctrl)
132 * Only new queue scan work when admin and IO queues are both alive
134 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
135 queue_work(nvme_wq, &ctrl->scan_work);
139 * Use this function to proceed with scheduling reset_work for a controller
140 * that had previously been set to the resetting state. This is intended for
141 * code paths that can't be interrupted by other reset attempts. A hot removal
142 * may prevent this from succeeding.
144 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
146 if (ctrl->state != NVME_CTRL_RESETTING)
148 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
152 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
154 static void nvme_failfast_work(struct work_struct *work)
156 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
157 struct nvme_ctrl, failfast_work);
159 if (ctrl->state != NVME_CTRL_CONNECTING)
162 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
163 dev_info(ctrl->device, "failfast expired\n");
164 nvme_kick_requeue_lists(ctrl);
167 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
169 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
172 schedule_delayed_work(&ctrl->failfast_work,
173 ctrl->opts->fast_io_fail_tmo * HZ);
176 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
181 cancel_delayed_work_sync(&ctrl->failfast_work);
182 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
186 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
188 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
190 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
194 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
196 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
200 ret = nvme_reset_ctrl(ctrl);
202 flush_work(&ctrl->reset_work);
203 if (ctrl->state != NVME_CTRL_LIVE)
210 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
212 dev_info(ctrl->device,
213 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
215 flush_work(&ctrl->reset_work);
216 nvme_stop_ctrl(ctrl);
217 nvme_remove_namespaces(ctrl);
218 ctrl->ops->delete_ctrl(ctrl);
219 nvme_uninit_ctrl(ctrl);
222 static void nvme_delete_ctrl_work(struct work_struct *work)
224 struct nvme_ctrl *ctrl =
225 container_of(work, struct nvme_ctrl, delete_work);
227 nvme_do_delete_ctrl(ctrl);
230 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
232 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
234 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
238 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
240 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
243 * Keep a reference until nvme_do_delete_ctrl() complete,
244 * since ->delete_ctrl can free the controller.
247 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
248 nvme_do_delete_ctrl(ctrl);
252 static blk_status_t nvme_error_status(u16 status)
254 switch (status & 0x7ff) {
255 case NVME_SC_SUCCESS:
257 case NVME_SC_CAP_EXCEEDED:
258 return BLK_STS_NOSPC;
259 case NVME_SC_LBA_RANGE:
260 case NVME_SC_CMD_INTERRUPTED:
261 case NVME_SC_NS_NOT_READY:
262 return BLK_STS_TARGET;
263 case NVME_SC_BAD_ATTRIBUTES:
264 case NVME_SC_ONCS_NOT_SUPPORTED:
265 case NVME_SC_INVALID_OPCODE:
266 case NVME_SC_INVALID_FIELD:
267 case NVME_SC_INVALID_NS:
268 return BLK_STS_NOTSUPP;
269 case NVME_SC_WRITE_FAULT:
270 case NVME_SC_READ_ERROR:
271 case NVME_SC_UNWRITTEN_BLOCK:
272 case NVME_SC_ACCESS_DENIED:
273 case NVME_SC_READ_ONLY:
274 case NVME_SC_COMPARE_FAILED:
275 return BLK_STS_MEDIUM;
276 case NVME_SC_GUARD_CHECK:
277 case NVME_SC_APPTAG_CHECK:
278 case NVME_SC_REFTAG_CHECK:
279 case NVME_SC_INVALID_PI:
280 return BLK_STS_PROTECTION;
281 case NVME_SC_RESERVATION_CONFLICT:
282 return BLK_STS_NEXUS;
283 case NVME_SC_HOST_PATH_ERROR:
284 return BLK_STS_TRANSPORT;
285 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
286 return BLK_STS_ZONE_ACTIVE_RESOURCE;
287 case NVME_SC_ZONE_TOO_MANY_OPEN:
288 return BLK_STS_ZONE_OPEN_RESOURCE;
290 return BLK_STS_IOERR;
294 static void nvme_retry_req(struct request *req)
296 unsigned long delay = 0;
299 /* The mask and shift result must be <= 3 */
300 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
302 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
304 nvme_req(req)->retries++;
305 blk_mq_requeue_request(req, false);
306 blk_mq_delay_kick_requeue_list(req->q, delay);
309 static void nvme_log_error(struct request *req)
311 struct nvme_ns *ns = req->q->queuedata;
312 struct nvme_request *nr = nvme_req(req);
315 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
316 ns->disk ? ns->disk->disk_name : "?",
317 nvme_get_opcode_str(nr->cmd->common.opcode),
318 nr->cmd->common.opcode,
319 (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
320 (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
321 nvme_get_error_status_str(nr->status),
322 nr->status >> 8 & 7, /* Status Code Type */
323 nr->status & 0xff, /* Status Code */
324 nr->status & NVME_SC_MORE ? "MORE " : "",
325 nr->status & NVME_SC_DNR ? "DNR " : "");
329 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
330 dev_name(nr->ctrl->device),
331 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
332 nr->cmd->common.opcode,
333 nvme_get_error_status_str(nr->status),
334 nr->status >> 8 & 7, /* Status Code Type */
335 nr->status & 0xff, /* Status Code */
336 nr->status & NVME_SC_MORE ? "MORE " : "",
337 nr->status & NVME_SC_DNR ? "DNR " : "");
340 enum nvme_disposition {
347 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
349 if (likely(nvme_req(req)->status == 0))
352 if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
355 if (blk_noretry_request(req) ||
356 (nvme_req(req)->status & NVME_SC_DNR) ||
357 nvme_req(req)->retries >= nvme_max_retries)
360 if (req->cmd_flags & REQ_NVME_MPATH) {
361 if (nvme_is_path_error(nvme_req(req)->status) ||
362 blk_queue_dying(req->q))
365 if (blk_queue_dying(req->q))
372 static inline void nvme_end_req_zoned(struct request *req)
374 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
375 req_op(req) == REQ_OP_ZONE_APPEND)
376 req->__sector = nvme_lba_to_sect(req->q->queuedata,
377 le64_to_cpu(nvme_req(req)->result.u64));
380 static inline void nvme_end_req(struct request *req)
382 blk_status_t status = nvme_error_status(nvme_req(req)->status);
384 if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
386 nvme_end_req_zoned(req);
387 nvme_trace_bio_complete(req);
388 if (req->cmd_flags & REQ_NVME_MPATH)
389 nvme_mpath_end_request(req);
390 blk_mq_end_request(req, status);
393 void nvme_complete_rq(struct request *req)
395 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
397 trace_nvme_complete_rq(req);
398 nvme_cleanup_cmd(req);
401 ctrl->comp_seen = true;
403 switch (nvme_decide_disposition(req)) {
411 nvme_failover_req(req);
414 #ifdef CONFIG_NVME_AUTH
415 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
423 EXPORT_SYMBOL_GPL(nvme_complete_rq);
425 void nvme_complete_batch_req(struct request *req)
427 trace_nvme_complete_rq(req);
428 nvme_cleanup_cmd(req);
429 nvme_end_req_zoned(req);
431 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
434 * Called to unwind from ->queue_rq on a failed command submission so that the
435 * multipathing code gets called to potentially failover to another path.
436 * The caller needs to unwind all transport specific resource allocations and
437 * must return propagate the return value.
439 blk_status_t nvme_host_path_error(struct request *req)
441 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
442 blk_mq_set_request_complete(req);
443 nvme_complete_rq(req);
446 EXPORT_SYMBOL_GPL(nvme_host_path_error);
448 bool nvme_cancel_request(struct request *req, void *data)
450 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
451 "Cancelling I/O %d", req->tag);
453 /* don't abort one completed or idle request */
454 if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
457 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
458 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
459 blk_mq_complete_request(req);
462 EXPORT_SYMBOL_GPL(nvme_cancel_request);
464 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
467 blk_mq_tagset_busy_iter(ctrl->tagset,
468 nvme_cancel_request, ctrl);
469 blk_mq_tagset_wait_completed_request(ctrl->tagset);
472 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
474 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
476 if (ctrl->admin_tagset) {
477 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
478 nvme_cancel_request, ctrl);
479 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
482 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
484 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
485 enum nvme_ctrl_state new_state)
487 enum nvme_ctrl_state old_state;
489 bool changed = false;
491 spin_lock_irqsave(&ctrl->lock, flags);
493 old_state = ctrl->state;
498 case NVME_CTRL_RESETTING:
499 case NVME_CTRL_CONNECTING:
506 case NVME_CTRL_RESETTING:
516 case NVME_CTRL_CONNECTING:
519 case NVME_CTRL_RESETTING:
526 case NVME_CTRL_DELETING:
529 case NVME_CTRL_RESETTING:
530 case NVME_CTRL_CONNECTING:
537 case NVME_CTRL_DELETING_NOIO:
539 case NVME_CTRL_DELETING:
549 case NVME_CTRL_DELETING:
561 ctrl->state = new_state;
562 wake_up_all(&ctrl->state_wq);
565 spin_unlock_irqrestore(&ctrl->lock, flags);
569 if (ctrl->state == NVME_CTRL_LIVE) {
570 if (old_state == NVME_CTRL_CONNECTING)
571 nvme_stop_failfast_work(ctrl);
572 nvme_kick_requeue_lists(ctrl);
573 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
574 old_state == NVME_CTRL_RESETTING) {
575 nvme_start_failfast_work(ctrl);
579 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
582 * Returns true for sink states that can't ever transition back to live.
584 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
586 switch (ctrl->state) {
589 case NVME_CTRL_RESETTING:
590 case NVME_CTRL_CONNECTING:
592 case NVME_CTRL_DELETING:
593 case NVME_CTRL_DELETING_NOIO:
597 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
603 * Waits for the controller state to be resetting, or returns false if it is
604 * not possible to ever transition to that state.
606 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
608 wait_event(ctrl->state_wq,
609 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
610 nvme_state_terminal(ctrl));
611 return ctrl->state == NVME_CTRL_RESETTING;
613 EXPORT_SYMBOL_GPL(nvme_wait_reset);
615 static void nvme_free_ns_head(struct kref *ref)
617 struct nvme_ns_head *head =
618 container_of(ref, struct nvme_ns_head, ref);
620 nvme_mpath_remove_disk(head);
621 ida_free(&head->subsys->ns_ida, head->instance);
622 cleanup_srcu_struct(&head->srcu);
623 nvme_put_subsystem(head->subsys);
627 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
629 return kref_get_unless_zero(&head->ref);
632 void nvme_put_ns_head(struct nvme_ns_head *head)
634 kref_put(&head->ref, nvme_free_ns_head);
637 static void nvme_free_ns(struct kref *kref)
639 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
642 nvme_put_ns_head(ns->head);
643 nvme_put_ctrl(ns->ctrl);
647 static inline bool nvme_get_ns(struct nvme_ns *ns)
649 return kref_get_unless_zero(&ns->kref);
652 void nvme_put_ns(struct nvme_ns *ns)
654 kref_put(&ns->kref, nvme_free_ns);
656 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
658 static inline void nvme_clear_nvme_request(struct request *req)
660 nvme_req(req)->status = 0;
661 nvme_req(req)->retries = 0;
662 nvme_req(req)->flags = 0;
663 req->rq_flags |= RQF_DONTPREP;
666 /* initialize a passthrough request */
667 void nvme_init_request(struct request *req, struct nvme_command *cmd)
669 if (req->q->queuedata)
670 req->timeout = NVME_IO_TIMEOUT;
671 else /* no queuedata implies admin queue */
672 req->timeout = NVME_ADMIN_TIMEOUT;
674 /* passthru commands should let the driver set the SGL flags */
675 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
677 req->cmd_flags |= REQ_FAILFAST_DRIVER;
678 if (req->mq_hctx->type == HCTX_TYPE_POLL)
679 req->cmd_flags |= REQ_POLLED;
680 nvme_clear_nvme_request(req);
681 req->rq_flags |= RQF_QUIET;
682 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
684 EXPORT_SYMBOL_GPL(nvme_init_request);
687 * For something we're not in a state to send to the device the default action
688 * is to busy it and retry it after the controller state is recovered. However,
689 * if the controller is deleting or if anything is marked for failfast or
690 * nvme multipath it is immediately failed.
692 * Note: commands used to initialize the controller will be marked for failfast.
693 * Note: nvme cli/ioctl commands are marked for failfast.
695 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
698 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
699 ctrl->state != NVME_CTRL_DELETING &&
700 ctrl->state != NVME_CTRL_DEAD &&
701 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
702 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
703 return BLK_STS_RESOURCE;
704 return nvme_host_path_error(rq);
706 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
708 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
711 struct nvme_request *req = nvme_req(rq);
714 * currently we have a problem sending passthru commands
715 * on the admin_q if the controller is not LIVE because we can't
716 * make sure that they are going out after the admin connect,
717 * controller enable and/or other commands in the initialization
718 * sequence. until the controller will be LIVE, fail with
719 * BLK_STS_RESOURCE so that they will be rescheduled.
721 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
724 if (ctrl->ops->flags & NVME_F_FABRICS) {
726 * Only allow commands on a live queue, except for the connect
727 * command, which is require to set the queue live in the
728 * appropinquate states.
730 switch (ctrl->state) {
731 case NVME_CTRL_CONNECTING:
732 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
733 (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
734 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
735 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
747 EXPORT_SYMBOL_GPL(__nvme_check_ready);
749 static inline void nvme_setup_flush(struct nvme_ns *ns,
750 struct nvme_command *cmnd)
752 memset(cmnd, 0, sizeof(*cmnd));
753 cmnd->common.opcode = nvme_cmd_flush;
754 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
757 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
758 struct nvme_command *cmnd)
760 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
761 struct nvme_dsm_range *range;
765 * Some devices do not consider the DSM 'Number of Ranges' field when
766 * determining how much data to DMA. Always allocate memory for maximum
767 * number of segments to prevent device reading beyond end of buffer.
769 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
771 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
774 * If we fail allocation our range, fallback to the controller
775 * discard page. If that's also busy, it's safe to return
776 * busy, as we know we can make progress once that's freed.
778 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
779 return BLK_STS_RESOURCE;
781 range = page_address(ns->ctrl->discard_page);
784 if (queue_max_discard_segments(req->q) == 1) {
785 u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
786 u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
788 range[0].cattr = cpu_to_le32(0);
789 range[0].nlb = cpu_to_le32(nlb);
790 range[0].slba = cpu_to_le64(slba);
793 __rq_for_each_bio(bio, req) {
794 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
795 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
798 range[n].cattr = cpu_to_le32(0);
799 range[n].nlb = cpu_to_le32(nlb);
800 range[n].slba = cpu_to_le64(slba);
806 if (WARN_ON_ONCE(n != segments)) {
807 if (virt_to_page(range) == ns->ctrl->discard_page)
808 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
811 return BLK_STS_IOERR;
814 memset(cmnd, 0, sizeof(*cmnd));
815 cmnd->dsm.opcode = nvme_cmd_dsm;
816 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
817 cmnd->dsm.nr = cpu_to_le32(segments - 1);
818 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
820 bvec_set_virt(&req->special_vec, range, alloc_size);
821 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
826 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
832 /* both rw and write zeroes share the same reftag format */
833 switch (ns->guard_type) {
834 case NVME_NVM_NS_16B_GUARD:
835 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
837 case NVME_NVM_NS_64B_GUARD:
838 ref48 = ext_pi_ref_tag(req);
839 lower = lower_32_bits(ref48);
840 upper = upper_32_bits(ref48);
842 cmnd->rw.reftag = cpu_to_le32(lower);
843 cmnd->rw.cdw3 = cpu_to_le32(upper);
850 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
851 struct request *req, struct nvme_command *cmnd)
853 memset(cmnd, 0, sizeof(*cmnd));
855 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
856 return nvme_setup_discard(ns, req, cmnd);
858 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
859 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
860 cmnd->write_zeroes.slba =
861 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
862 cmnd->write_zeroes.length =
863 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
865 if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
866 cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
868 if (nvme_ns_has_pi(ns)) {
869 cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
871 switch (ns->pi_type) {
872 case NVME_NS_DPS_PI_TYPE1:
873 case NVME_NS_DPS_PI_TYPE2:
874 nvme_set_ref_tag(ns, cmnd, req);
882 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
883 struct request *req, struct nvme_command *cmnd,
889 if (req->cmd_flags & REQ_FUA)
890 control |= NVME_RW_FUA;
891 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
892 control |= NVME_RW_LR;
894 if (req->cmd_flags & REQ_RAHEAD)
895 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
897 cmnd->rw.opcode = op;
899 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
902 cmnd->rw.metadata = 0;
903 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
904 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
907 cmnd->rw.appmask = 0;
911 * If formated with metadata, the block layer always provides a
912 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
913 * we enable the PRACT bit for protection information or set the
914 * namespace capacity to zero to prevent any I/O.
916 if (!blk_integrity_rq(req)) {
917 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
918 return BLK_STS_NOTSUPP;
919 control |= NVME_RW_PRINFO_PRACT;
922 switch (ns->pi_type) {
923 case NVME_NS_DPS_PI_TYPE3:
924 control |= NVME_RW_PRINFO_PRCHK_GUARD;
926 case NVME_NS_DPS_PI_TYPE1:
927 case NVME_NS_DPS_PI_TYPE2:
928 control |= NVME_RW_PRINFO_PRCHK_GUARD |
929 NVME_RW_PRINFO_PRCHK_REF;
930 if (op == nvme_cmd_zone_append)
931 control |= NVME_RW_APPEND_PIREMAP;
932 nvme_set_ref_tag(ns, cmnd, req);
937 cmnd->rw.control = cpu_to_le16(control);
938 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
942 void nvme_cleanup_cmd(struct request *req)
944 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
945 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
947 if (req->special_vec.bv_page == ctrl->discard_page)
948 clear_bit_unlock(0, &ctrl->discard_page_busy);
950 kfree(bvec_virt(&req->special_vec));
953 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
955 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
957 struct nvme_command *cmd = nvme_req(req)->cmd;
958 blk_status_t ret = BLK_STS_OK;
960 if (!(req->rq_flags & RQF_DONTPREP))
961 nvme_clear_nvme_request(req);
963 switch (req_op(req)) {
966 /* these are setup prior to execution in nvme_init_request() */
969 nvme_setup_flush(ns, cmd);
971 case REQ_OP_ZONE_RESET_ALL:
972 case REQ_OP_ZONE_RESET:
973 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
975 case REQ_OP_ZONE_OPEN:
976 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
978 case REQ_OP_ZONE_CLOSE:
979 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
981 case REQ_OP_ZONE_FINISH:
982 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
984 case REQ_OP_WRITE_ZEROES:
985 ret = nvme_setup_write_zeroes(ns, req, cmd);
988 ret = nvme_setup_discard(ns, req, cmd);
991 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
994 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
996 case REQ_OP_ZONE_APPEND:
997 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1001 return BLK_STS_IOERR;
1004 cmd->common.command_id = nvme_cid(req);
1005 trace_nvme_setup_cmd(req, cmd);
1008 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1013 * >0: nvme controller's cqe status response
1014 * <0: kernel error in lieu of controller response
1016 int nvme_execute_rq(struct request *rq, bool at_head)
1018 blk_status_t status;
1020 status = blk_execute_rq(rq, at_head);
1021 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1023 if (nvme_req(rq)->status)
1024 return nvme_req(rq)->status;
1025 return blk_status_to_errno(status);
1027 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1030 * Returns 0 on success. If the result is negative, it's a Linux error code;
1031 * if the result is positive, it's an NVM Express status code
1033 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1034 union nvme_result *result, void *buffer, unsigned bufflen,
1035 int qid, int at_head, blk_mq_req_flags_t flags)
1037 struct request *req;
1040 if (qid == NVME_QID_ANY)
1041 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1043 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1047 return PTR_ERR(req);
1048 nvme_init_request(req, cmd);
1050 if (buffer && bufflen) {
1051 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1056 ret = nvme_execute_rq(req, at_head);
1057 if (result && ret >= 0)
1058 *result = nvme_req(req)->result;
1060 blk_mq_free_request(req);
1063 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1065 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1066 void *buffer, unsigned bufflen)
1068 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1069 NVME_QID_ANY, 0, 0);
1071 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1073 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1078 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1079 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1080 dev_warn_once(ctrl->device,
1081 "IO command:%02x has unusual effects:%08x\n",
1085 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1086 * which would deadlock when done on an I/O command. Note that
1087 * We already warn about an unusual effect above.
1089 effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1091 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1096 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1098 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1100 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1103 * For simplicity, IO to all namespaces is quiesced even if the command
1104 * effects say only one namespace is affected.
1106 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1107 mutex_lock(&ctrl->scan_lock);
1108 mutex_lock(&ctrl->subsys->lock);
1109 nvme_mpath_start_freeze(ctrl->subsys);
1110 nvme_mpath_wait_freeze(ctrl->subsys);
1111 nvme_start_freeze(ctrl);
1112 nvme_wait_freeze(ctrl);
1116 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1118 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1119 struct nvme_command *cmd, int status)
1121 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1122 nvme_unfreeze(ctrl);
1123 nvme_mpath_unfreeze(ctrl->subsys);
1124 mutex_unlock(&ctrl->subsys->lock);
1125 mutex_unlock(&ctrl->scan_lock);
1127 if (effects & NVME_CMD_EFFECTS_CCC) {
1128 dev_info(ctrl->device,
1129 "controller capabilities changed, reset may be required to take effect.\n");
1131 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1132 nvme_queue_scan(ctrl);
1133 flush_work(&ctrl->scan_work);
1136 switch (cmd->common.opcode) {
1137 case nvme_admin_set_features:
1138 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1139 case NVME_FEAT_KATO:
1141 * Keep alive commands interval on the host should be
1142 * updated when KATO is modified by Set Features
1146 nvme_update_keep_alive(ctrl, cmd);
1156 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1159 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1161 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1162 * accounting for transport roundtrip times [..].
1164 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1166 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1169 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1170 blk_status_t status)
1172 struct nvme_ctrl *ctrl = rq->end_io_data;
1173 unsigned long flags;
1174 bool startka = false;
1176 blk_mq_free_request(rq);
1179 dev_err(ctrl->device,
1180 "failed nvme_keep_alive_end_io error=%d\n",
1182 return RQ_END_IO_NONE;
1185 ctrl->comp_seen = false;
1186 spin_lock_irqsave(&ctrl->lock, flags);
1187 if (ctrl->state == NVME_CTRL_LIVE ||
1188 ctrl->state == NVME_CTRL_CONNECTING)
1190 spin_unlock_irqrestore(&ctrl->lock, flags);
1192 nvme_queue_keep_alive_work(ctrl);
1193 return RQ_END_IO_NONE;
1196 static void nvme_keep_alive_work(struct work_struct *work)
1198 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1199 struct nvme_ctrl, ka_work);
1200 bool comp_seen = ctrl->comp_seen;
1203 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1204 dev_dbg(ctrl->device,
1205 "reschedule traffic based keep-alive timer\n");
1206 ctrl->comp_seen = false;
1207 nvme_queue_keep_alive_work(ctrl);
1211 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1212 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1214 /* allocation failure, reset the controller */
1215 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1216 nvme_reset_ctrl(ctrl);
1219 nvme_init_request(rq, &ctrl->ka_cmd);
1221 rq->timeout = ctrl->kato * HZ;
1222 rq->end_io = nvme_keep_alive_end_io;
1223 rq->end_io_data = ctrl;
1224 blk_execute_rq_nowait(rq, false);
1227 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1229 if (unlikely(ctrl->kato == 0))
1232 nvme_queue_keep_alive_work(ctrl);
1235 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1237 if (unlikely(ctrl->kato == 0))
1240 cancel_delayed_work_sync(&ctrl->ka_work);
1242 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1244 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1245 struct nvme_command *cmd)
1247 unsigned int new_kato =
1248 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1250 dev_info(ctrl->device,
1251 "keep alive interval updated from %u ms to %u ms\n",
1252 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1254 nvme_stop_keep_alive(ctrl);
1255 ctrl->kato = new_kato;
1256 nvme_start_keep_alive(ctrl);
1260 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1261 * flag, thus sending any new CNS opcodes has a big chance of not working.
1262 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1263 * (but not for any later version).
1265 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1267 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1268 return ctrl->vs < NVME_VS(1, 2, 0);
1269 return ctrl->vs < NVME_VS(1, 1, 0);
1272 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1274 struct nvme_command c = { };
1277 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1278 c.identify.opcode = nvme_admin_identify;
1279 c.identify.cns = NVME_ID_CNS_CTRL;
1281 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1285 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1286 sizeof(struct nvme_id_ctrl));
1292 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1293 struct nvme_ns_id_desc *cur, bool *csi_seen)
1295 const char *warn_str = "ctrl returned bogus length:";
1298 switch (cur->nidt) {
1299 case NVME_NIDT_EUI64:
1300 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1301 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1302 warn_str, cur->nidl);
1305 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1306 return NVME_NIDT_EUI64_LEN;
1307 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1308 return NVME_NIDT_EUI64_LEN;
1309 case NVME_NIDT_NGUID:
1310 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1311 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1312 warn_str, cur->nidl);
1315 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1316 return NVME_NIDT_NGUID_LEN;
1317 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1318 return NVME_NIDT_NGUID_LEN;
1319 case NVME_NIDT_UUID:
1320 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1321 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1322 warn_str, cur->nidl);
1325 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1326 return NVME_NIDT_UUID_LEN;
1327 uuid_copy(&ids->uuid, data + sizeof(*cur));
1328 return NVME_NIDT_UUID_LEN;
1330 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1331 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1332 warn_str, cur->nidl);
1335 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1337 return NVME_NIDT_CSI_LEN;
1339 /* Skip unknown types */
1344 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1345 struct nvme_ns_info *info)
1347 struct nvme_command c = { };
1348 bool csi_seen = false;
1349 int status, pos, len;
1352 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1354 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1357 c.identify.opcode = nvme_admin_identify;
1358 c.identify.nsid = cpu_to_le32(info->nsid);
1359 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1361 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1365 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1366 NVME_IDENTIFY_DATA_SIZE);
1368 dev_warn(ctrl->device,
1369 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1370 info->nsid, status);
1374 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1375 struct nvme_ns_id_desc *cur = data + pos;
1380 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1384 len += sizeof(*cur);
1387 if (nvme_multi_css(ctrl) && !csi_seen) {
1388 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1398 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1399 struct nvme_id_ns **id)
1401 struct nvme_command c = { };
1404 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1405 c.identify.opcode = nvme_admin_identify;
1406 c.identify.nsid = cpu_to_le32(nsid);
1407 c.identify.cns = NVME_ID_CNS_NS;
1409 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1413 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1415 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1421 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1422 struct nvme_ns_info *info)
1424 struct nvme_ns_ids *ids = &info->ids;
1425 struct nvme_id_ns *id;
1428 ret = nvme_identify_ns(ctrl, info->nsid, &id);
1432 if (id->ncap == 0) {
1433 /* namespace not allocated or attached */
1434 info->is_removed = true;
1438 info->anagrpid = id->anagrpid;
1439 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1440 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1441 info->is_ready = true;
1442 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1443 dev_info(ctrl->device,
1444 "Ignoring bogus Namespace Identifiers\n");
1446 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1447 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1448 memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1449 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1450 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1451 memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1457 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1458 struct nvme_ns_info *info)
1460 struct nvme_id_ns_cs_indep *id;
1461 struct nvme_command c = {
1462 .identify.opcode = nvme_admin_identify,
1463 .identify.nsid = cpu_to_le32(info->nsid),
1464 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1468 id = kmalloc(sizeof(*id), GFP_KERNEL);
1472 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1474 info->anagrpid = id->anagrpid;
1475 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1476 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1477 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1483 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1484 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1486 union nvme_result res = { 0 };
1487 struct nvme_command c = { };
1490 c.features.opcode = op;
1491 c.features.fid = cpu_to_le32(fid);
1492 c.features.dword11 = cpu_to_le32(dword11);
1494 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1495 buffer, buflen, NVME_QID_ANY, 0, 0);
1496 if (ret >= 0 && result)
1497 *result = le32_to_cpu(res.u32);
1501 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1502 unsigned int dword11, void *buffer, size_t buflen,
1505 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1508 EXPORT_SYMBOL_GPL(nvme_set_features);
1510 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1511 unsigned int dword11, void *buffer, size_t buflen,
1514 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1517 EXPORT_SYMBOL_GPL(nvme_get_features);
1519 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1521 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1523 int status, nr_io_queues;
1525 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1531 * Degraded controllers might return an error when setting the queue
1532 * count. We still want to be able to bring them online and offer
1533 * access to the admin queue, as that might be only way to fix them up.
1536 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1539 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1540 *count = min(*count, nr_io_queues);
1545 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1547 #define NVME_AEN_SUPPORTED \
1548 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1549 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1551 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1553 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1556 if (!supported_aens)
1559 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1562 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1565 queue_work(nvme_wq, &ctrl->async_event_work);
1568 static int nvme_ns_open(struct nvme_ns *ns)
1571 /* should never be called due to GENHD_FL_HIDDEN */
1572 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1574 if (!nvme_get_ns(ns))
1576 if (!try_module_get(ns->ctrl->ops->module))
1587 static void nvme_ns_release(struct nvme_ns *ns)
1590 module_put(ns->ctrl->ops->module);
1594 static int nvme_open(struct block_device *bdev, fmode_t mode)
1596 return nvme_ns_open(bdev->bd_disk->private_data);
1599 static void nvme_release(struct gendisk *disk, fmode_t mode)
1601 nvme_ns_release(disk->private_data);
1604 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1606 /* some standard values */
1607 geo->heads = 1 << 6;
1608 geo->sectors = 1 << 5;
1609 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1613 #ifdef CONFIG_BLK_DEV_INTEGRITY
1614 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1615 u32 max_integrity_segments)
1617 struct blk_integrity integrity = { };
1619 switch (ns->pi_type) {
1620 case NVME_NS_DPS_PI_TYPE3:
1621 switch (ns->guard_type) {
1622 case NVME_NVM_NS_16B_GUARD:
1623 integrity.profile = &t10_pi_type3_crc;
1624 integrity.tag_size = sizeof(u16) + sizeof(u32);
1625 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1627 case NVME_NVM_NS_64B_GUARD:
1628 integrity.profile = &ext_pi_type3_crc64;
1629 integrity.tag_size = sizeof(u16) + 6;
1630 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1633 integrity.profile = NULL;
1637 case NVME_NS_DPS_PI_TYPE1:
1638 case NVME_NS_DPS_PI_TYPE2:
1639 switch (ns->guard_type) {
1640 case NVME_NVM_NS_16B_GUARD:
1641 integrity.profile = &t10_pi_type1_crc;
1642 integrity.tag_size = sizeof(u16);
1643 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1645 case NVME_NVM_NS_64B_GUARD:
1646 integrity.profile = &ext_pi_type1_crc64;
1647 integrity.tag_size = sizeof(u16);
1648 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1651 integrity.profile = NULL;
1656 integrity.profile = NULL;
1660 integrity.tuple_size = ns->ms;
1661 blk_integrity_register(disk, &integrity);
1662 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1665 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1666 u32 max_integrity_segments)
1669 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1671 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1673 struct nvme_ctrl *ctrl = ns->ctrl;
1674 struct request_queue *queue = disk->queue;
1675 u32 size = queue_logical_block_size(queue);
1677 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1678 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1680 if (ctrl->max_discard_sectors == 0) {
1681 blk_queue_max_discard_sectors(queue, 0);
1685 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1686 NVME_DSM_MAX_RANGES);
1688 queue->limits.discard_granularity = size;
1690 /* If discard is already enabled, don't reset queue limits */
1691 if (queue->limits.max_discard_sectors)
1694 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1695 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1697 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1698 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1701 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1703 return uuid_equal(&a->uuid, &b->uuid) &&
1704 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1705 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1709 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1711 bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1712 unsigned lbaf = nvme_lbaf_index(id->flbas);
1713 struct nvme_ctrl *ctrl = ns->ctrl;
1714 struct nvme_command c = { };
1715 struct nvme_id_ns_nvm *nvm;
1720 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1721 if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1722 ns->pi_size = sizeof(struct t10_pi_tuple);
1723 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1727 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1731 c.identify.opcode = nvme_admin_identify;
1732 c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1733 c.identify.cns = NVME_ID_CNS_CS_NS;
1734 c.identify.csi = NVME_CSI_NVM;
1736 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1740 elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1742 /* no support for storage tag formats right now */
1743 if (nvme_elbaf_sts(elbaf))
1746 ns->guard_type = nvme_elbaf_guard_type(elbaf);
1747 switch (ns->guard_type) {
1748 case NVME_NVM_NS_64B_GUARD:
1749 ns->pi_size = sizeof(struct crc64_pi_tuple);
1751 case NVME_NVM_NS_16B_GUARD:
1752 ns->pi_size = sizeof(struct t10_pi_tuple);
1761 if (ns->pi_size && (first || ns->ms == ns->pi_size))
1762 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1769 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1771 struct nvme_ctrl *ctrl = ns->ctrl;
1773 if (nvme_init_ms(ns, id))
1776 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1777 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1780 if (ctrl->ops->flags & NVME_F_FABRICS) {
1782 * The NVMe over Fabrics specification only supports metadata as
1783 * part of the extended data LBA. We rely on HCA/HBA support to
1784 * remap the separate metadata buffer from the block layer.
1786 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1789 ns->features |= NVME_NS_EXT_LBAS;
1792 * The current fabrics transport drivers support namespace
1793 * metadata formats only if nvme_ns_has_pi() returns true.
1794 * Suppress support for all other formats so the namespace will
1795 * have a 0 capacity and not be usable through the block stack.
1797 * Note, this check will need to be modified if any drivers
1798 * gain the ability to use other metadata formats.
1800 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1801 ns->features |= NVME_NS_METADATA_SUPPORTED;
1804 * For PCIe controllers, we can't easily remap the separate
1805 * metadata buffer from the block layer and thus require a
1806 * separate metadata buffer for block layer metadata/PI support.
1807 * We allow extended LBAs for the passthrough interface, though.
1809 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1810 ns->features |= NVME_NS_EXT_LBAS;
1812 ns->features |= NVME_NS_METADATA_SUPPORTED;
1816 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1817 struct request_queue *q)
1819 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1821 if (ctrl->max_hw_sectors) {
1823 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1825 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1826 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1827 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1829 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1830 blk_queue_dma_alignment(q, 3);
1831 blk_queue_write_cache(q, vwc, vwc);
1834 static void nvme_update_disk_info(struct gendisk *disk,
1835 struct nvme_ns *ns, struct nvme_id_ns *id)
1837 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1838 unsigned short bs = 1 << ns->lba_shift;
1839 u32 atomic_bs, phys_bs, io_opt = 0;
1842 * The block layer can't support LBA sizes larger than the page size
1843 * yet, so catch this early and don't allow block I/O.
1845 if (ns->lba_shift > PAGE_SHIFT) {
1850 blk_integrity_unregister(disk);
1852 atomic_bs = phys_bs = bs;
1853 if (id->nabo == 0) {
1855 * Bit 1 indicates whether NAWUPF is defined for this namespace
1856 * and whether it should be used instead of AWUPF. If NAWUPF ==
1857 * 0 then AWUPF must be used instead.
1859 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1860 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1862 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1865 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1866 /* NPWG = Namespace Preferred Write Granularity */
1867 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1868 /* NOWS = Namespace Optimal Write Size */
1869 io_opt = bs * (1 + le16_to_cpu(id->nows));
1872 blk_queue_logical_block_size(disk->queue, bs);
1874 * Linux filesystems assume writing a single physical block is
1875 * an atomic operation. Hence limit the physical block size to the
1876 * value of the Atomic Write Unit Power Fail parameter.
1878 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1879 blk_queue_io_min(disk->queue, phys_bs);
1880 blk_queue_io_opt(disk->queue, io_opt);
1883 * Register a metadata profile for PI, or the plain non-integrity NVMe
1884 * metadata masquerading as Type 0 if supported, otherwise reject block
1885 * I/O to namespaces with metadata except when the namespace supports
1886 * PI, as it can strip/insert in that case.
1889 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1890 (ns->features & NVME_NS_METADATA_SUPPORTED))
1891 nvme_init_integrity(disk, ns,
1892 ns->ctrl->max_integrity_segments);
1893 else if (!nvme_ns_has_pi(ns))
1897 set_capacity_and_notify(disk, capacity);
1899 nvme_config_discard(disk, ns);
1900 blk_queue_max_write_zeroes_sectors(disk->queue,
1901 ns->ctrl->max_zeroes_sectors);
1904 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1906 return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1909 static inline bool nvme_first_scan(struct gendisk *disk)
1911 /* nvme_alloc_ns() scans the disk prior to adding it */
1912 return !disk_live(disk);
1915 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1917 struct nvme_ctrl *ctrl = ns->ctrl;
1920 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1921 is_power_of_2(ctrl->max_hw_sectors))
1922 iob = ctrl->max_hw_sectors;
1924 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1929 if (!is_power_of_2(iob)) {
1930 if (nvme_first_scan(ns->disk))
1931 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1932 ns->disk->disk_name, iob);
1936 if (blk_queue_is_zoned(ns->disk->queue)) {
1937 if (nvme_first_scan(ns->disk))
1938 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1939 ns->disk->disk_name);
1943 blk_queue_chunk_sectors(ns->queue, iob);
1946 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1947 struct nvme_ns_info *info)
1949 blk_mq_freeze_queue(ns->disk->queue);
1950 nvme_set_queue_limits(ns->ctrl, ns->queue);
1951 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1952 blk_mq_unfreeze_queue(ns->disk->queue);
1954 if (nvme_ns_head_multipath(ns->head)) {
1955 blk_mq_freeze_queue(ns->head->disk->queue);
1956 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1957 nvme_mpath_revalidate_paths(ns);
1958 blk_stack_limits(&ns->head->disk->queue->limits,
1959 &ns->queue->limits, 0);
1960 ns->head->disk->flags |= GENHD_FL_HIDDEN;
1961 blk_mq_unfreeze_queue(ns->head->disk->queue);
1964 /* Hide the block-interface for these devices */
1965 ns->disk->flags |= GENHD_FL_HIDDEN;
1966 set_bit(NVME_NS_READY, &ns->flags);
1971 static int nvme_update_ns_info_block(struct nvme_ns *ns,
1972 struct nvme_ns_info *info)
1974 struct nvme_id_ns *id;
1978 ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
1982 blk_mq_freeze_queue(ns->disk->queue);
1983 lbaf = nvme_lbaf_index(id->flbas);
1984 ns->lba_shift = id->lbaf[lbaf].ds;
1985 nvme_set_queue_limits(ns->ctrl, ns->queue);
1987 nvme_configure_metadata(ns, id);
1988 nvme_set_chunk_sectors(ns, id);
1989 nvme_update_disk_info(ns->disk, ns, id);
1991 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1992 ret = nvme_update_zone_info(ns, lbaf);
1994 blk_mq_unfreeze_queue(ns->disk->queue);
2000 * Only set the DEAC bit if the device guarantees that reads from
2001 * deallocated data return zeroes. While the DEAC bit does not
2002 * require that, it must be a no-op if reads from deallocated data
2003 * do not return zeroes.
2005 if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2006 ns->features |= NVME_NS_DEAC;
2007 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2008 set_bit(NVME_NS_READY, &ns->flags);
2009 blk_mq_unfreeze_queue(ns->disk->queue);
2011 if (blk_queue_is_zoned(ns->queue)) {
2012 ret = nvme_revalidate_zones(ns);
2013 if (ret && !nvme_first_scan(ns->disk))
2017 if (nvme_ns_head_multipath(ns->head)) {
2018 blk_mq_freeze_queue(ns->head->disk->queue);
2019 nvme_update_disk_info(ns->head->disk, ns, id);
2020 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2021 nvme_mpath_revalidate_paths(ns);
2022 blk_stack_limits(&ns->head->disk->queue->limits,
2023 &ns->queue->limits, 0);
2024 disk_update_readahead(ns->head->disk);
2025 blk_mq_unfreeze_queue(ns->head->disk->queue);
2031 * If probing fails due an unsupported feature, hide the block device,
2032 * but still allow other access.
2034 if (ret == -ENODEV) {
2035 ns->disk->flags |= GENHD_FL_HIDDEN;
2036 set_bit(NVME_NS_READY, &ns->flags);
2043 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2045 switch (info->ids.csi) {
2047 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2048 dev_info(ns->ctrl->device,
2049 "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2051 return nvme_update_ns_info_generic(ns, info);
2053 return nvme_update_ns_info_block(ns, info);
2055 return nvme_update_ns_info_block(ns, info);
2057 dev_info(ns->ctrl->device,
2058 "block device for nsid %u not supported (csi %u)\n",
2059 info->nsid, info->ids.csi);
2060 return nvme_update_ns_info_generic(ns, info);
2064 static char nvme_pr_type(enum pr_type type)
2067 case PR_WRITE_EXCLUSIVE:
2069 case PR_EXCLUSIVE_ACCESS:
2071 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2073 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2075 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2077 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2084 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2085 struct nvme_command *c, u8 data[16])
2087 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2088 int srcu_idx = srcu_read_lock(&head->srcu);
2089 struct nvme_ns *ns = nvme_find_path(head);
2090 int ret = -EWOULDBLOCK;
2093 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2094 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2096 srcu_read_unlock(&head->srcu, srcu_idx);
2100 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2103 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2104 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2107 static int nvme_sc_to_pr_err(int nvme_sc)
2109 if (nvme_is_path_error(nvme_sc))
2110 return PR_STS_PATH_FAILED;
2113 case NVME_SC_SUCCESS:
2114 return PR_STS_SUCCESS;
2115 case NVME_SC_RESERVATION_CONFLICT:
2116 return PR_STS_RESERVATION_CONFLICT;
2117 case NVME_SC_ONCS_NOT_SUPPORTED:
2119 case NVME_SC_BAD_ATTRIBUTES:
2120 case NVME_SC_INVALID_OPCODE:
2121 case NVME_SC_INVALID_FIELD:
2122 case NVME_SC_INVALID_NS:
2125 return PR_STS_IOERR;
2129 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2130 u64 key, u64 sa_key, u8 op)
2132 struct nvme_command c = { };
2133 u8 data[16] = { 0, };
2136 put_unaligned_le64(key, &data[0]);
2137 put_unaligned_le64(sa_key, &data[8]);
2139 c.common.opcode = op;
2140 c.common.cdw10 = cpu_to_le32(cdw10);
2142 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2143 bdev->bd_disk->fops == &nvme_ns_head_ops)
2144 ret = nvme_send_ns_head_pr_command(bdev, &c, data);
2146 ret = nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c,
2151 return nvme_sc_to_pr_err(ret);
2154 static int nvme_pr_register(struct block_device *bdev, u64 old,
2155 u64 new, unsigned flags)
2159 if (flags & ~PR_FL_IGNORE_KEY)
2162 cdw10 = old ? 2 : 0;
2163 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2164 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2165 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2168 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2169 enum pr_type type, unsigned flags)
2173 if (flags & ~PR_FL_IGNORE_KEY)
2176 cdw10 = nvme_pr_type(type) << 8;
2177 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2178 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2181 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2182 enum pr_type type, bool abort)
2184 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2186 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2189 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2191 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2193 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2196 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2198 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2200 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2203 const struct pr_ops nvme_pr_ops = {
2204 .pr_register = nvme_pr_register,
2205 .pr_reserve = nvme_pr_reserve,
2206 .pr_release = nvme_pr_release,
2207 .pr_preempt = nvme_pr_preempt,
2208 .pr_clear = nvme_pr_clear,
2211 #ifdef CONFIG_BLK_SED_OPAL
2212 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2215 struct nvme_ctrl *ctrl = data;
2216 struct nvme_command cmd = { };
2219 cmd.common.opcode = nvme_admin_security_send;
2221 cmd.common.opcode = nvme_admin_security_recv;
2222 cmd.common.nsid = 0;
2223 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2224 cmd.common.cdw11 = cpu_to_le32(len);
2226 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2227 NVME_QID_ANY, 1, 0);
2230 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2232 if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2233 if (!ctrl->opal_dev)
2234 ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2235 else if (was_suspended)
2236 opal_unlock_from_suspend(ctrl->opal_dev);
2238 free_opal_dev(ctrl->opal_dev);
2239 ctrl->opal_dev = NULL;
2243 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2246 #endif /* CONFIG_BLK_SED_OPAL */
2248 #ifdef CONFIG_BLK_DEV_ZONED
2249 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2250 unsigned int nr_zones, report_zones_cb cb, void *data)
2252 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2256 #define nvme_report_zones NULL
2257 #endif /* CONFIG_BLK_DEV_ZONED */
2259 static const struct block_device_operations nvme_bdev_ops = {
2260 .owner = THIS_MODULE,
2261 .ioctl = nvme_ioctl,
2262 .compat_ioctl = blkdev_compat_ptr_ioctl,
2264 .release = nvme_release,
2265 .getgeo = nvme_getgeo,
2266 .report_zones = nvme_report_zones,
2267 .pr_ops = &nvme_pr_ops,
2270 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2271 u32 timeout, const char *op)
2273 unsigned long timeout_jiffies = jiffies + timeout * HZ;
2277 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2280 if ((csts & mask) == val)
2283 usleep_range(1000, 2000);
2284 if (fatal_signal_pending(current))
2286 if (time_after(jiffies, timeout_jiffies)) {
2287 dev_err(ctrl->device,
2288 "Device not ready; aborting %s, CSTS=0x%x\n",
2297 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2301 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2303 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2305 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2307 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2312 return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2313 NVME_CSTS_SHST_CMPLT,
2314 ctrl->shutdown_timeout, "shutdown");
2316 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2317 msleep(NVME_QUIRK_DELAY_AMOUNT);
2318 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2319 (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2321 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2323 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2325 unsigned dev_page_min;
2329 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2331 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2334 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2336 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2337 dev_err(ctrl->device,
2338 "Minimum device page size %u too large for host (%u)\n",
2339 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2343 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2344 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2346 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2348 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2351 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2353 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2358 if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2359 ctrl->ctrl_config |= NVME_CC_CRIME;
2360 timeout = NVME_CRTO_CRIMT(crto);
2362 timeout = NVME_CRTO_CRWMT(crto);
2365 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2368 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2369 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2370 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2371 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2375 /* Flush write to device (required if transport is PCI) */
2376 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2380 ctrl->ctrl_config |= NVME_CC_ENABLE;
2381 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2384 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2385 (timeout + 1) / 2, "initialisation");
2387 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2389 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2394 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2397 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2398 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2401 dev_warn_once(ctrl->device,
2402 "could not set timestamp (%d)\n", ret);
2406 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2408 struct nvme_feat_host_behavior *host;
2409 u8 acre = 0, lbafee = 0;
2412 /* Don't bother enabling the feature if retry delay is not reported */
2414 acre = NVME_ENABLE_ACRE;
2415 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2416 lbafee = NVME_ENABLE_LBAFEE;
2418 if (!acre && !lbafee)
2421 host = kzalloc(sizeof(*host), GFP_KERNEL);
2426 host->lbafee = lbafee;
2427 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2428 host, sizeof(*host), NULL);
2434 * The function checks whether the given total (exlat + enlat) latency of
2435 * a power state allows the latter to be used as an APST transition target.
2436 * It does so by comparing the latency to the primary and secondary latency
2437 * tolerances defined by module params. If there's a match, the corresponding
2438 * timeout value is returned and the matching tolerance index (1 or 2) is
2441 static bool nvme_apst_get_transition_time(u64 total_latency,
2442 u64 *transition_time, unsigned *last_index)
2444 if (total_latency <= apst_primary_latency_tol_us) {
2445 if (*last_index == 1)
2448 *transition_time = apst_primary_timeout_ms;
2451 if (apst_secondary_timeout_ms &&
2452 total_latency <= apst_secondary_latency_tol_us) {
2453 if (*last_index <= 2)
2456 *transition_time = apst_secondary_timeout_ms;
2463 * APST (Autonomous Power State Transition) lets us program a table of power
2464 * state transitions that the controller will perform automatically.
2466 * Depending on module params, one of the two supported techniques will be used:
2468 * - If the parameters provide explicit timeouts and tolerances, they will be
2469 * used to build a table with up to 2 non-operational states to transition to.
2470 * The default parameter values were selected based on the values used by
2471 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2472 * regeneration of the APST table in the event of switching between external
2473 * and battery power, the timeouts and tolerances reflect a compromise
2474 * between values used by Microsoft for AC and battery scenarios.
2475 * - If not, we'll configure the table with a simple heuristic: we are willing
2476 * to spend at most 2% of the time transitioning between power states.
2477 * Therefore, when running in any given state, we will enter the next
2478 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2479 * microseconds, as long as that state's exit latency is under the requested
2482 * We will not autonomously enter any non-operational state for which the total
2483 * latency exceeds ps_max_latency_us.
2485 * Users can set ps_max_latency_us to zero to turn off APST.
2487 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2489 struct nvme_feat_auto_pst *table;
2496 unsigned last_lt_index = UINT_MAX;
2499 * If APST isn't supported or if we haven't been initialized yet,
2500 * then don't do anything.
2505 if (ctrl->npss > 31) {
2506 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2510 table = kzalloc(sizeof(*table), GFP_KERNEL);
2514 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2515 /* Turn off APST. */
2516 dev_dbg(ctrl->device, "APST disabled\n");
2521 * Walk through all states from lowest- to highest-power.
2522 * According to the spec, lower-numbered states use more power. NPSS,
2523 * despite the name, is the index of the lowest-power state, not the
2526 for (state = (int)ctrl->npss; state >= 0; state--) {
2527 u64 total_latency_us, exit_latency_us, transition_ms;
2530 table->entries[state] = target;
2533 * Don't allow transitions to the deepest state if it's quirked
2536 if (state == ctrl->npss &&
2537 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2541 * Is this state a useful non-operational state for higher-power
2542 * states to autonomously transition to?
2544 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2547 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2548 if (exit_latency_us > ctrl->ps_max_latency_us)
2551 total_latency_us = exit_latency_us +
2552 le32_to_cpu(ctrl->psd[state].entry_lat);
2555 * This state is good. It can be used as the APST idle target
2556 * for higher power states.
2558 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2559 if (!nvme_apst_get_transition_time(total_latency_us,
2560 &transition_ms, &last_lt_index))
2563 transition_ms = total_latency_us + 19;
2564 do_div(transition_ms, 20);
2565 if (transition_ms > (1 << 24) - 1)
2566 transition_ms = (1 << 24) - 1;
2569 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2572 if (total_latency_us > max_lat_us)
2573 max_lat_us = total_latency_us;
2577 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2579 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2580 max_ps, max_lat_us, (int)sizeof(*table), table);
2584 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2585 table, sizeof(*table), NULL);
2587 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2592 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2594 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2598 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2599 case PM_QOS_LATENCY_ANY:
2607 if (ctrl->ps_max_latency_us != latency) {
2608 ctrl->ps_max_latency_us = latency;
2609 if (ctrl->state == NVME_CTRL_LIVE)
2610 nvme_configure_apst(ctrl);
2614 struct nvme_core_quirk_entry {
2616 * NVMe model and firmware strings are padded with spaces. For
2617 * simplicity, strings in the quirk table are padded with NULLs
2623 unsigned long quirks;
2626 static const struct nvme_core_quirk_entry core_quirks[] = {
2629 * This Toshiba device seems to die using any APST states. See:
2630 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2633 .mn = "THNSF5256GPUK TOSHIBA",
2634 .quirks = NVME_QUIRK_NO_APST,
2638 * This LiteON CL1-3D*-Q11 firmware version has a race
2639 * condition associated with actions related to suspend to idle
2640 * LiteON has resolved the problem in future firmware
2644 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2648 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2649 * aborts I/O during any load, but more easily reproducible
2650 * with discards (fstrim).
2652 * The device is left in a state where it is also not possible
2653 * to use "nvme set-feature" to disable APST, but booting with
2654 * nvme_core.default_ps_max_latency=0 works.
2657 .mn = "KCD6XVUL6T40",
2658 .quirks = NVME_QUIRK_NO_APST,
2662 * The external Samsung X5 SSD fails initialization without a
2663 * delay before checking if it is ready and has a whole set of
2664 * other problems. To make this even more interesting, it
2665 * shares the PCI ID with internal Samsung 970 Evo Plus that
2666 * does not need or want these quirks.
2669 .mn = "Samsung Portable SSD X5",
2670 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2671 NVME_QUIRK_NO_DEEPEST_PS |
2672 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2676 /* match is null-terminated but idstr is space-padded. */
2677 static bool string_matches(const char *idstr, const char *match, size_t len)
2684 matchlen = strlen(match);
2685 WARN_ON_ONCE(matchlen > len);
2687 if (memcmp(idstr, match, matchlen))
2690 for (; matchlen < len; matchlen++)
2691 if (idstr[matchlen] != ' ')
2697 static bool quirk_matches(const struct nvme_id_ctrl *id,
2698 const struct nvme_core_quirk_entry *q)
2700 return q->vid == le16_to_cpu(id->vid) &&
2701 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2702 string_matches(id->fr, q->fr, sizeof(id->fr));
2705 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2706 struct nvme_id_ctrl *id)
2711 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2712 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2713 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2714 strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2718 if (ctrl->vs >= NVME_VS(1, 2, 1))
2719 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2723 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2724 * Base Specification 2.0. It is slightly different from the format
2725 * specified there due to historic reasons, and we can't change it now.
2727 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2728 "nqn.2014.08.org.nvmexpress:%04x%04x",
2729 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2730 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2731 off += sizeof(id->sn);
2732 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2733 off += sizeof(id->mn);
2734 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2737 static void nvme_release_subsystem(struct device *dev)
2739 struct nvme_subsystem *subsys =
2740 container_of(dev, struct nvme_subsystem, dev);
2742 if (subsys->instance >= 0)
2743 ida_free(&nvme_instance_ida, subsys->instance);
2747 static void nvme_destroy_subsystem(struct kref *ref)
2749 struct nvme_subsystem *subsys =
2750 container_of(ref, struct nvme_subsystem, ref);
2752 mutex_lock(&nvme_subsystems_lock);
2753 list_del(&subsys->entry);
2754 mutex_unlock(&nvme_subsystems_lock);
2756 ida_destroy(&subsys->ns_ida);
2757 device_del(&subsys->dev);
2758 put_device(&subsys->dev);
2761 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2763 kref_put(&subsys->ref, nvme_destroy_subsystem);
2766 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2768 struct nvme_subsystem *subsys;
2770 lockdep_assert_held(&nvme_subsystems_lock);
2773 * Fail matches for discovery subsystems. This results
2774 * in each discovery controller bound to a unique subsystem.
2775 * This avoids issues with validating controller values
2776 * that can only be true when there is a single unique subsystem.
2777 * There may be multiple and completely independent entities
2778 * that provide discovery controllers.
2780 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2783 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2784 if (strcmp(subsys->subnqn, subsysnqn))
2786 if (!kref_get_unless_zero(&subsys->ref))
2794 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2795 struct device_attribute subsys_attr_##_name = \
2796 __ATTR(_name, _mode, _show, NULL)
2798 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2799 struct device_attribute *attr,
2802 struct nvme_subsystem *subsys =
2803 container_of(dev, struct nvme_subsystem, dev);
2805 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2807 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2809 static ssize_t nvme_subsys_show_type(struct device *dev,
2810 struct device_attribute *attr,
2813 struct nvme_subsystem *subsys =
2814 container_of(dev, struct nvme_subsystem, dev);
2816 switch (subsys->subtype) {
2818 return sysfs_emit(buf, "discovery\n");
2820 return sysfs_emit(buf, "nvm\n");
2822 return sysfs_emit(buf, "reserved\n");
2825 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2827 #define nvme_subsys_show_str_function(field) \
2828 static ssize_t subsys_##field##_show(struct device *dev, \
2829 struct device_attribute *attr, char *buf) \
2831 struct nvme_subsystem *subsys = \
2832 container_of(dev, struct nvme_subsystem, dev); \
2833 return sysfs_emit(buf, "%.*s\n", \
2834 (int)sizeof(subsys->field), subsys->field); \
2836 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2838 nvme_subsys_show_str_function(model);
2839 nvme_subsys_show_str_function(serial);
2840 nvme_subsys_show_str_function(firmware_rev);
2842 static struct attribute *nvme_subsys_attrs[] = {
2843 &subsys_attr_model.attr,
2844 &subsys_attr_serial.attr,
2845 &subsys_attr_firmware_rev.attr,
2846 &subsys_attr_subsysnqn.attr,
2847 &subsys_attr_subsystype.attr,
2848 #ifdef CONFIG_NVME_MULTIPATH
2849 &subsys_attr_iopolicy.attr,
2854 static const struct attribute_group nvme_subsys_attrs_group = {
2855 .attrs = nvme_subsys_attrs,
2858 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2859 &nvme_subsys_attrs_group,
2863 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2865 return ctrl->opts && ctrl->opts->discovery_nqn;
2868 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2869 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2871 struct nvme_ctrl *tmp;
2873 lockdep_assert_held(&nvme_subsystems_lock);
2875 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2876 if (nvme_state_terminal(tmp))
2879 if (tmp->cntlid == ctrl->cntlid) {
2880 dev_err(ctrl->device,
2881 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2882 ctrl->cntlid, dev_name(tmp->device),
2887 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2888 nvme_discovery_ctrl(ctrl))
2891 dev_err(ctrl->device,
2892 "Subsystem does not support multiple controllers\n");
2899 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2901 struct nvme_subsystem *subsys, *found;
2904 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2908 subsys->instance = -1;
2909 mutex_init(&subsys->lock);
2910 kref_init(&subsys->ref);
2911 INIT_LIST_HEAD(&subsys->ctrls);
2912 INIT_LIST_HEAD(&subsys->nsheads);
2913 nvme_init_subnqn(subsys, ctrl, id);
2914 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2915 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2916 subsys->vendor_id = le16_to_cpu(id->vid);
2917 subsys->cmic = id->cmic;
2919 /* Versions prior to 1.4 don't necessarily report a valid type */
2920 if (id->cntrltype == NVME_CTRL_DISC ||
2921 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2922 subsys->subtype = NVME_NQN_DISC;
2924 subsys->subtype = NVME_NQN_NVME;
2926 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2927 dev_err(ctrl->device,
2928 "Subsystem %s is not a discovery controller",
2933 subsys->awupf = le16_to_cpu(id->awupf);
2934 nvme_mpath_default_iopolicy(subsys);
2936 subsys->dev.class = nvme_subsys_class;
2937 subsys->dev.release = nvme_release_subsystem;
2938 subsys->dev.groups = nvme_subsys_attrs_groups;
2939 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2940 device_initialize(&subsys->dev);
2942 mutex_lock(&nvme_subsystems_lock);
2943 found = __nvme_find_get_subsystem(subsys->subnqn);
2945 put_device(&subsys->dev);
2948 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2950 goto out_put_subsystem;
2953 ret = device_add(&subsys->dev);
2955 dev_err(ctrl->device,
2956 "failed to register subsystem device.\n");
2957 put_device(&subsys->dev);
2960 ida_init(&subsys->ns_ida);
2961 list_add_tail(&subsys->entry, &nvme_subsystems);
2964 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2965 dev_name(ctrl->device));
2967 dev_err(ctrl->device,
2968 "failed to create sysfs link from subsystem.\n");
2969 goto out_put_subsystem;
2973 subsys->instance = ctrl->instance;
2974 ctrl->subsys = subsys;
2975 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2976 mutex_unlock(&nvme_subsystems_lock);
2980 nvme_put_subsystem(subsys);
2982 mutex_unlock(&nvme_subsystems_lock);
2986 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2987 void *log, size_t size, u64 offset)
2989 struct nvme_command c = { };
2990 u32 dwlen = nvme_bytes_to_numd(size);
2992 c.get_log_page.opcode = nvme_admin_get_log_page;
2993 c.get_log_page.nsid = cpu_to_le32(nsid);
2994 c.get_log_page.lid = log_page;
2995 c.get_log_page.lsp = lsp;
2996 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2997 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2998 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2999 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3000 c.get_log_page.csi = csi;
3002 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3005 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3006 struct nvme_effects_log **log)
3008 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3014 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3018 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3019 cel, sizeof(*cel), 0);
3025 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3031 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3033 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3035 if (check_shl_overflow(1U, units + page_shift - 9, &val))
3040 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3042 struct nvme_command c = { };
3043 struct nvme_id_ctrl_nvm *id;
3046 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3047 ctrl->max_discard_sectors = UINT_MAX;
3048 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3050 ctrl->max_discard_sectors = 0;
3051 ctrl->max_discard_segments = 0;
3055 * Even though NVMe spec explicitly states that MDTS is not applicable
3056 * to the write-zeroes, we are cautious and limit the size to the
3057 * controllers max_hw_sectors value, which is based on the MDTS field
3058 * and possibly other limiting factors.
3060 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3061 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3062 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3064 ctrl->max_zeroes_sectors = 0;
3066 if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3067 nvme_ctrl_limited_cns(ctrl))
3070 id = kzalloc(sizeof(*id), GFP_KERNEL);
3074 c.identify.opcode = nvme_admin_identify;
3075 c.identify.cns = NVME_ID_CNS_CS_CTRL;
3076 c.identify.csi = NVME_CSI_NVM;
3078 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3083 ctrl->max_discard_segments = id->dmrl;
3084 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3086 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3093 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3095 struct nvme_effects_log *log = ctrl->effects;
3097 log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3098 NVME_CMD_EFFECTS_NCC |
3099 NVME_CMD_EFFECTS_CSE_MASK);
3100 log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3101 NVME_CMD_EFFECTS_CSE_MASK);
3104 * The spec says the result of a security receive command depends on
3105 * the previous security send command. As such, many vendors log this
3106 * command as one to submitted only when no other commands to the same
3107 * namespace are outstanding. The intention is to tell the host to
3108 * prevent mixing security send and receive.
3110 * This driver can only enforce such exclusive access against IO
3111 * queues, though. We are not readily able to enforce such a rule for
3112 * two commands to the admin queue, which is the only queue that
3113 * matters for this command.
3115 * Rather than blindly freezing the IO queues for this effect that
3116 * doesn't even apply to IO, mask it off.
3118 log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3120 log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3121 log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3122 log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3125 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3132 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3133 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3138 if (!ctrl->effects) {
3139 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3142 xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3145 nvme_init_known_nvm_effects(ctrl);
3149 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3151 struct nvme_id_ctrl *id;
3153 bool prev_apst_enabled;
3156 ret = nvme_identify_ctrl(ctrl, &id);
3158 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3162 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3163 ctrl->cntlid = le16_to_cpu(id->cntlid);
3165 if (!ctrl->identified) {
3169 * Check for quirks. Quirk can depend on firmware version,
3170 * so, in principle, the set of quirks present can change
3171 * across a reset. As a possible future enhancement, we
3172 * could re-scan for quirks every time we reinitialize
3173 * the device, but we'd have to make sure that the driver
3174 * behaves intelligently if the quirks change.
3176 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3177 if (quirk_matches(id, &core_quirks[i]))
3178 ctrl->quirks |= core_quirks[i].quirks;
3181 ret = nvme_init_subsystem(ctrl, id);
3185 ret = nvme_init_effects(ctrl, id);
3189 memcpy(ctrl->subsys->firmware_rev, id->fr,
3190 sizeof(ctrl->subsys->firmware_rev));
3192 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3193 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3194 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3197 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3198 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3199 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3201 ctrl->oacs = le16_to_cpu(id->oacs);
3202 ctrl->oncs = le16_to_cpu(id->oncs);
3203 ctrl->mtfa = le16_to_cpu(id->mtfa);
3204 ctrl->oaes = le32_to_cpu(id->oaes);
3205 ctrl->wctemp = le16_to_cpu(id->wctemp);
3206 ctrl->cctemp = le16_to_cpu(id->cctemp);
3208 atomic_set(&ctrl->abort_limit, id->acl + 1);
3209 ctrl->vwc = id->vwc;
3211 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3213 max_hw_sectors = UINT_MAX;
3214 ctrl->max_hw_sectors =
3215 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3217 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3218 ctrl->sgls = le32_to_cpu(id->sgls);
3219 ctrl->kas = le16_to_cpu(id->kas);
3220 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3221 ctrl->ctratt = le32_to_cpu(id->ctratt);
3223 ctrl->cntrltype = id->cntrltype;
3224 ctrl->dctype = id->dctype;
3228 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3230 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3231 shutdown_timeout, 60);
3233 if (ctrl->shutdown_timeout != shutdown_timeout)
3234 dev_info(ctrl->device,
3235 "Shutdown timeout set to %u seconds\n",
3236 ctrl->shutdown_timeout);
3238 ctrl->shutdown_timeout = shutdown_timeout;
3240 ctrl->npss = id->npss;
3241 ctrl->apsta = id->apsta;
3242 prev_apst_enabled = ctrl->apst_enabled;
3243 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3244 if (force_apst && id->apsta) {
3245 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3246 ctrl->apst_enabled = true;
3248 ctrl->apst_enabled = false;
3251 ctrl->apst_enabled = id->apsta;
3253 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3255 if (ctrl->ops->flags & NVME_F_FABRICS) {
3256 ctrl->icdoff = le16_to_cpu(id->icdoff);
3257 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3258 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3259 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3262 * In fabrics we need to verify the cntlid matches the
3265 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3266 dev_err(ctrl->device,
3267 "Mismatching cntlid: Connect %u vs Identify "
3269 ctrl->cntlid, le16_to_cpu(id->cntlid));
3274 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3275 dev_err(ctrl->device,
3276 "keep-alive support is mandatory for fabrics\n");
3281 ctrl->hmpre = le32_to_cpu(id->hmpre);
3282 ctrl->hmmin = le32_to_cpu(id->hmmin);
3283 ctrl->hmminds = le32_to_cpu(id->hmminds);
3284 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3287 ret = nvme_mpath_init_identify(ctrl, id);
3291 if (ctrl->apst_enabled && !prev_apst_enabled)
3292 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3293 else if (!ctrl->apst_enabled && prev_apst_enabled)
3294 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3302 * Initialize the cached copies of the Identify data and various controller
3303 * register in our nvme_ctrl structure. This should be called as soon as
3304 * the admin queue is fully up and running.
3306 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3310 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3312 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3316 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3318 if (ctrl->vs >= NVME_VS(1, 1, 0))
3319 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3321 ret = nvme_init_identify(ctrl);
3325 ret = nvme_configure_apst(ctrl);
3329 ret = nvme_configure_timestamp(ctrl);
3333 ret = nvme_configure_host_options(ctrl);
3337 nvme_configure_opal(ctrl, was_suspended);
3339 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3341 * Do not return errors unless we are in a controller reset,
3342 * the controller works perfectly fine without hwmon.
3344 ret = nvme_hwmon_init(ctrl);
3349 ctrl->identified = true;
3353 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3355 static int nvme_dev_open(struct inode *inode, struct file *file)
3357 struct nvme_ctrl *ctrl =
3358 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3360 switch (ctrl->state) {
3361 case NVME_CTRL_LIVE:
3364 return -EWOULDBLOCK;
3367 nvme_get_ctrl(ctrl);
3368 if (!try_module_get(ctrl->ops->module)) {
3369 nvme_put_ctrl(ctrl);
3373 file->private_data = ctrl;
3377 static int nvme_dev_release(struct inode *inode, struct file *file)
3379 struct nvme_ctrl *ctrl =
3380 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3382 module_put(ctrl->ops->module);
3383 nvme_put_ctrl(ctrl);
3387 static const struct file_operations nvme_dev_fops = {
3388 .owner = THIS_MODULE,
3389 .open = nvme_dev_open,
3390 .release = nvme_dev_release,
3391 .unlocked_ioctl = nvme_dev_ioctl,
3392 .compat_ioctl = compat_ptr_ioctl,
3393 .uring_cmd = nvme_dev_uring_cmd,
3396 static ssize_t nvme_sysfs_reset(struct device *dev,
3397 struct device_attribute *attr, const char *buf,
3400 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3403 ret = nvme_reset_ctrl_sync(ctrl);
3408 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3410 static ssize_t nvme_sysfs_rescan(struct device *dev,
3411 struct device_attribute *attr, const char *buf,
3414 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3416 nvme_queue_scan(ctrl);
3419 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3421 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3423 struct gendisk *disk = dev_to_disk(dev);
3425 if (disk->fops == &nvme_bdev_ops)
3426 return nvme_get_ns_from_dev(dev)->head;
3428 return disk->private_data;
3431 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3434 struct nvme_ns_head *head = dev_to_ns_head(dev);
3435 struct nvme_ns_ids *ids = &head->ids;
3436 struct nvme_subsystem *subsys = head->subsys;
3437 int serial_len = sizeof(subsys->serial);
3438 int model_len = sizeof(subsys->model);
3440 if (!uuid_is_null(&ids->uuid))
3441 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3443 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3444 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3446 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3447 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3449 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3450 subsys->serial[serial_len - 1] == '\0'))
3452 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3453 subsys->model[model_len - 1] == '\0'))
3456 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3457 serial_len, subsys->serial, model_len, subsys->model,
3460 static DEVICE_ATTR_RO(wwid);
3462 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3465 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3467 static DEVICE_ATTR_RO(nguid);
3469 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3472 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3474 /* For backward compatibility expose the NGUID to userspace if
3475 * we have no UUID set
3477 if (uuid_is_null(&ids->uuid)) {
3478 dev_warn_ratelimited(dev,
3479 "No UUID available providing old NGUID\n");
3480 return sysfs_emit(buf, "%pU\n", ids->nguid);
3482 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3484 static DEVICE_ATTR_RO(uuid);
3486 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3489 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3491 static DEVICE_ATTR_RO(eui);
3493 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3496 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3498 static DEVICE_ATTR_RO(nsid);
3500 static struct attribute *nvme_ns_id_attrs[] = {
3501 &dev_attr_wwid.attr,
3502 &dev_attr_uuid.attr,
3503 &dev_attr_nguid.attr,
3505 &dev_attr_nsid.attr,
3506 #ifdef CONFIG_NVME_MULTIPATH
3507 &dev_attr_ana_grpid.attr,
3508 &dev_attr_ana_state.attr,
3513 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3514 struct attribute *a, int n)
3516 struct device *dev = container_of(kobj, struct device, kobj);
3517 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3519 if (a == &dev_attr_uuid.attr) {
3520 if (uuid_is_null(&ids->uuid) &&
3521 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3524 if (a == &dev_attr_nguid.attr) {
3525 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3528 if (a == &dev_attr_eui.attr) {
3529 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3532 #ifdef CONFIG_NVME_MULTIPATH
3533 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3534 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3536 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3543 static const struct attribute_group nvme_ns_id_attr_group = {
3544 .attrs = nvme_ns_id_attrs,
3545 .is_visible = nvme_ns_id_attrs_are_visible,
3548 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3549 &nvme_ns_id_attr_group,
3553 #define nvme_show_str_function(field) \
3554 static ssize_t field##_show(struct device *dev, \
3555 struct device_attribute *attr, char *buf) \
3557 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3558 return sysfs_emit(buf, "%.*s\n", \
3559 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3561 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3563 nvme_show_str_function(model);
3564 nvme_show_str_function(serial);
3565 nvme_show_str_function(firmware_rev);
3567 #define nvme_show_int_function(field) \
3568 static ssize_t field##_show(struct device *dev, \
3569 struct device_attribute *attr, char *buf) \
3571 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3572 return sysfs_emit(buf, "%d\n", ctrl->field); \
3574 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3576 nvme_show_int_function(cntlid);
3577 nvme_show_int_function(numa_node);
3578 nvme_show_int_function(queue_count);
3579 nvme_show_int_function(sqsize);
3580 nvme_show_int_function(kato);
3582 static ssize_t nvme_sysfs_delete(struct device *dev,
3583 struct device_attribute *attr, const char *buf,
3586 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3588 if (!test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags))
3591 if (device_remove_file_self(dev, attr))
3592 nvme_delete_ctrl_sync(ctrl);
3595 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3597 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3598 struct device_attribute *attr,
3601 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3603 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3605 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3607 static ssize_t nvme_sysfs_show_state(struct device *dev,
3608 struct device_attribute *attr,
3611 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3612 static const char *const state_name[] = {
3613 [NVME_CTRL_NEW] = "new",
3614 [NVME_CTRL_LIVE] = "live",
3615 [NVME_CTRL_RESETTING] = "resetting",
3616 [NVME_CTRL_CONNECTING] = "connecting",
3617 [NVME_CTRL_DELETING] = "deleting",
3618 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3619 [NVME_CTRL_DEAD] = "dead",
3622 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3623 state_name[ctrl->state])
3624 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3626 return sysfs_emit(buf, "unknown state\n");
3629 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3631 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3632 struct device_attribute *attr,
3635 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3637 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3639 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3641 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3642 struct device_attribute *attr,
3645 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3647 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3649 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3651 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3652 struct device_attribute *attr,
3655 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3657 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3659 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3661 static ssize_t nvme_sysfs_show_address(struct device *dev,
3662 struct device_attribute *attr,
3665 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3667 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3669 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3671 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3672 struct device_attribute *attr, char *buf)
3674 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3675 struct nvmf_ctrl_options *opts = ctrl->opts;
3677 if (ctrl->opts->max_reconnects == -1)
3678 return sysfs_emit(buf, "off\n");
3679 return sysfs_emit(buf, "%d\n",
3680 opts->max_reconnects * opts->reconnect_delay);
3683 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3684 struct device_attribute *attr, const char *buf, size_t count)
3686 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3687 struct nvmf_ctrl_options *opts = ctrl->opts;
3688 int ctrl_loss_tmo, err;
3690 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3694 if (ctrl_loss_tmo < 0)
3695 opts->max_reconnects = -1;
3697 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3698 opts->reconnect_delay);
3701 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3702 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3704 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3705 struct device_attribute *attr, char *buf)
3707 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3709 if (ctrl->opts->reconnect_delay == -1)
3710 return sysfs_emit(buf, "off\n");
3711 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3714 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3715 struct device_attribute *attr, const char *buf, size_t count)
3717 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3721 err = kstrtou32(buf, 10, &v);
3725 ctrl->opts->reconnect_delay = v;
3728 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3729 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3731 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3732 struct device_attribute *attr, char *buf)
3734 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3736 if (ctrl->opts->fast_io_fail_tmo == -1)
3737 return sysfs_emit(buf, "off\n");
3738 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3741 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3742 struct device_attribute *attr, const char *buf, size_t count)
3744 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3745 struct nvmf_ctrl_options *opts = ctrl->opts;
3746 int fast_io_fail_tmo, err;
3748 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3752 if (fast_io_fail_tmo < 0)
3753 opts->fast_io_fail_tmo = -1;
3755 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3758 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3759 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3761 static ssize_t cntrltype_show(struct device *dev,
3762 struct device_attribute *attr, char *buf)
3764 static const char * const type[] = {
3765 [NVME_CTRL_IO] = "io\n",
3766 [NVME_CTRL_DISC] = "discovery\n",
3767 [NVME_CTRL_ADMIN] = "admin\n",
3769 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3771 if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3772 return sysfs_emit(buf, "reserved\n");
3774 return sysfs_emit(buf, type[ctrl->cntrltype]);
3776 static DEVICE_ATTR_RO(cntrltype);
3778 static ssize_t dctype_show(struct device *dev,
3779 struct device_attribute *attr, char *buf)
3781 static const char * const type[] = {
3782 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3783 [NVME_DCTYPE_DDC] = "ddc\n",
3784 [NVME_DCTYPE_CDC] = "cdc\n",
3786 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3788 if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3789 return sysfs_emit(buf, "reserved\n");
3791 return sysfs_emit(buf, type[ctrl->dctype]);
3793 static DEVICE_ATTR_RO(dctype);
3795 #ifdef CONFIG_NVME_AUTH
3796 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3797 struct device_attribute *attr, char *buf)
3799 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3800 struct nvmf_ctrl_options *opts = ctrl->opts;
3802 if (!opts->dhchap_secret)
3803 return sysfs_emit(buf, "none\n");
3804 return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3807 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3808 struct device_attribute *attr, const char *buf, size_t count)
3810 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3811 struct nvmf_ctrl_options *opts = ctrl->opts;
3812 char *dhchap_secret;
3814 if (!ctrl->opts->dhchap_secret)
3818 if (memcmp(buf, "DHHC-1:", 7))
3821 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3824 memcpy(dhchap_secret, buf, count);
3825 nvme_auth_stop(ctrl);
3826 if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3827 struct nvme_dhchap_key *key, *host_key;
3830 ret = nvme_auth_generate_key(dhchap_secret, &key);
3833 kfree(opts->dhchap_secret);
3834 opts->dhchap_secret = dhchap_secret;
3835 host_key = ctrl->host_key;
3836 mutex_lock(&ctrl->dhchap_auth_mutex);
3837 ctrl->host_key = key;
3838 mutex_unlock(&ctrl->dhchap_auth_mutex);
3839 nvme_auth_free_key(host_key);
3841 /* Start re-authentication */
3842 dev_info(ctrl->device, "re-authenticating controller\n");
3843 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3847 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3848 nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3850 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3851 struct device_attribute *attr, char *buf)
3853 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3854 struct nvmf_ctrl_options *opts = ctrl->opts;
3856 if (!opts->dhchap_ctrl_secret)
3857 return sysfs_emit(buf, "none\n");
3858 return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3861 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3862 struct device_attribute *attr, const char *buf, size_t count)
3864 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3865 struct nvmf_ctrl_options *opts = ctrl->opts;
3866 char *dhchap_secret;
3868 if (!ctrl->opts->dhchap_ctrl_secret)
3872 if (memcmp(buf, "DHHC-1:", 7))
3875 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3878 memcpy(dhchap_secret, buf, count);
3879 nvme_auth_stop(ctrl);
3880 if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3881 struct nvme_dhchap_key *key, *ctrl_key;
3884 ret = nvme_auth_generate_key(dhchap_secret, &key);
3887 kfree(opts->dhchap_ctrl_secret);
3888 opts->dhchap_ctrl_secret = dhchap_secret;
3889 ctrl_key = ctrl->ctrl_key;
3890 mutex_lock(&ctrl->dhchap_auth_mutex);
3891 ctrl->ctrl_key = key;
3892 mutex_unlock(&ctrl->dhchap_auth_mutex);
3893 nvme_auth_free_key(ctrl_key);
3895 /* Start re-authentication */
3896 dev_info(ctrl->device, "re-authenticating controller\n");
3897 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3901 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3902 nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3905 static struct attribute *nvme_dev_attrs[] = {
3906 &dev_attr_reset_controller.attr,
3907 &dev_attr_rescan_controller.attr,
3908 &dev_attr_model.attr,
3909 &dev_attr_serial.attr,
3910 &dev_attr_firmware_rev.attr,
3911 &dev_attr_cntlid.attr,
3912 &dev_attr_delete_controller.attr,
3913 &dev_attr_transport.attr,
3914 &dev_attr_subsysnqn.attr,
3915 &dev_attr_address.attr,
3916 &dev_attr_state.attr,
3917 &dev_attr_numa_node.attr,
3918 &dev_attr_queue_count.attr,
3919 &dev_attr_sqsize.attr,
3920 &dev_attr_hostnqn.attr,
3921 &dev_attr_hostid.attr,
3922 &dev_attr_ctrl_loss_tmo.attr,
3923 &dev_attr_reconnect_delay.attr,
3924 &dev_attr_fast_io_fail_tmo.attr,
3925 &dev_attr_kato.attr,
3926 &dev_attr_cntrltype.attr,
3927 &dev_attr_dctype.attr,
3928 #ifdef CONFIG_NVME_AUTH
3929 &dev_attr_dhchap_secret.attr,
3930 &dev_attr_dhchap_ctrl_secret.attr,
3935 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3936 struct attribute *a, int n)
3938 struct device *dev = container_of(kobj, struct device, kobj);
3939 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3941 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3943 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3945 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3947 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3949 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3951 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3953 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3955 #ifdef CONFIG_NVME_AUTH
3956 if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3958 if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3965 const struct attribute_group nvme_dev_attrs_group = {
3966 .attrs = nvme_dev_attrs,
3967 .is_visible = nvme_dev_attrs_are_visible,
3969 EXPORT_SYMBOL_GPL(nvme_dev_attrs_group);
3971 static const struct attribute_group *nvme_dev_attr_groups[] = {
3972 &nvme_dev_attrs_group,
3976 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3979 struct nvme_ns_head *h;
3981 lockdep_assert_held(&ctrl->subsys->lock);
3983 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3985 * Private namespaces can share NSIDs under some conditions.
3986 * In that case we can't use the same ns_head for namespaces
3987 * with the same NSID.
3989 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3991 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3998 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3999 struct nvme_ns_ids *ids)
4001 bool has_uuid = !uuid_is_null(&ids->uuid);
4002 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
4003 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
4004 struct nvme_ns_head *h;
4006 lockdep_assert_held(&subsys->lock);
4008 list_for_each_entry(h, &subsys->nsheads, entry) {
4009 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
4012 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
4015 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
4022 static void nvme_cdev_rel(struct device *dev)
4024 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
4027 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
4029 cdev_device_del(cdev, cdev_device);
4030 put_device(cdev_device);
4033 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
4034 const struct file_operations *fops, struct module *owner)
4038 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
4041 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
4042 cdev_device->class = nvme_ns_chr_class;
4043 cdev_device->release = nvme_cdev_rel;
4044 device_initialize(cdev_device);
4045 cdev_init(cdev, fops);
4046 cdev->owner = owner;
4047 ret = cdev_device_add(cdev, cdev_device);
4049 put_device(cdev_device);
4054 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
4056 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
4059 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
4061 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
4065 static const struct file_operations nvme_ns_chr_fops = {
4066 .owner = THIS_MODULE,
4067 .open = nvme_ns_chr_open,
4068 .release = nvme_ns_chr_release,
4069 .unlocked_ioctl = nvme_ns_chr_ioctl,
4070 .compat_ioctl = compat_ptr_ioctl,
4071 .uring_cmd = nvme_ns_chr_uring_cmd,
4072 .uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
4075 static int nvme_add_ns_cdev(struct nvme_ns *ns)
4079 ns->cdev_device.parent = ns->ctrl->device;
4080 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
4081 ns->ctrl->instance, ns->head->instance);
4085 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
4086 ns->ctrl->ops->module);
4089 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
4090 struct nvme_ns_info *info)
4092 struct nvme_ns_head *head;
4093 size_t size = sizeof(*head);
4096 #ifdef CONFIG_NVME_MULTIPATH
4097 size += num_possible_nodes() * sizeof(struct nvme_ns *);
4100 head = kzalloc(size, GFP_KERNEL);
4103 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4106 head->instance = ret;
4107 INIT_LIST_HEAD(&head->list);
4108 ret = init_srcu_struct(&head->srcu);
4110 goto out_ida_remove;
4111 head->subsys = ctrl->subsys;
4112 head->ns_id = info->nsid;
4113 head->ids = info->ids;
4114 head->shared = info->is_shared;
4115 kref_init(&head->ref);
4117 if (head->ids.csi) {
4118 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4120 goto out_cleanup_srcu;
4122 head->effects = ctrl->effects;
4124 ret = nvme_mpath_alloc_disk(ctrl, head);
4126 goto out_cleanup_srcu;
4128 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4130 kref_get(&ctrl->subsys->ref);
4134 cleanup_srcu_struct(&head->srcu);
4136 ida_free(&ctrl->subsys->ns_ida, head->instance);
4141 ret = blk_status_to_errno(nvme_error_status(ret));
4142 return ERR_PTR(ret);
4145 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4146 struct nvme_ns_ids *ids)
4148 struct nvme_subsystem *s;
4152 * Note that this check is racy as we try to avoid holding the global
4153 * lock over the whole ns_head creation. But it is only intended as
4154 * a sanity check anyway.
4156 mutex_lock(&nvme_subsystems_lock);
4157 list_for_each_entry(s, &nvme_subsystems, entry) {
4160 mutex_lock(&s->lock);
4161 ret = nvme_subsys_check_duplicate_ids(s, ids);
4162 mutex_unlock(&s->lock);
4166 mutex_unlock(&nvme_subsystems_lock);
4171 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4173 struct nvme_ctrl *ctrl = ns->ctrl;
4174 struct nvme_ns_head *head = NULL;
4177 ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4179 dev_err(ctrl->device,
4180 "globally duplicate IDs for nsid %d\n", info->nsid);
4181 nvme_print_device_info(ctrl);
4185 mutex_lock(&ctrl->subsys->lock);
4186 head = nvme_find_ns_head(ctrl, info->nsid);
4188 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4190 dev_err(ctrl->device,
4191 "duplicate IDs in subsystem for nsid %d\n",
4195 head = nvme_alloc_ns_head(ctrl, info);
4197 ret = PTR_ERR(head);
4202 if (!info->is_shared || !head->shared) {
4203 dev_err(ctrl->device,
4204 "Duplicate unshared namespace %d\n",
4206 goto out_put_ns_head;
4208 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4209 dev_err(ctrl->device,
4210 "IDs don't match for shared namespace %d\n",
4212 goto out_put_ns_head;
4215 if (!multipath && !list_empty(&head->list)) {
4216 dev_warn(ctrl->device,
4217 "Found shared namespace %d, but multipathing not supported.\n",
4219 dev_warn_once(ctrl->device,
4220 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4224 list_add_tail_rcu(&ns->siblings, &head->list);
4226 mutex_unlock(&ctrl->subsys->lock);
4230 nvme_put_ns_head(head);
4232 mutex_unlock(&ctrl->subsys->lock);
4236 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4238 struct nvme_ns *ns, *ret = NULL;
4240 down_read(&ctrl->namespaces_rwsem);
4241 list_for_each_entry(ns, &ctrl->namespaces, list) {
4242 if (ns->head->ns_id == nsid) {
4243 if (!nvme_get_ns(ns))
4248 if (ns->head->ns_id > nsid)
4251 up_read(&ctrl->namespaces_rwsem);
4254 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4257 * Add the namespace to the controller list while keeping the list ordered.
4259 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4261 struct nvme_ns *tmp;
4263 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4264 if (tmp->head->ns_id < ns->head->ns_id) {
4265 list_add(&ns->list, &tmp->list);
4269 list_add(&ns->list, &ns->ctrl->namespaces);
4272 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4275 struct gendisk *disk;
4276 int node = ctrl->numa_node;
4278 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4282 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4285 disk->fops = &nvme_bdev_ops;
4286 disk->private_data = ns;
4289 ns->queue = disk->queue;
4291 if (ctrl->opts && ctrl->opts->data_digest)
4292 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4294 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4295 if (ctrl->ops->supports_pci_p2pdma &&
4296 ctrl->ops->supports_pci_p2pdma(ctrl))
4297 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4300 kref_init(&ns->kref);
4302 if (nvme_init_ns_head(ns, info))
4303 goto out_cleanup_disk;
4306 * If multipathing is enabled, the device name for all disks and not
4307 * just those that represent shared namespaces needs to be based on the
4308 * subsystem instance. Using the controller instance for private
4309 * namespaces could lead to naming collisions between shared and private
4310 * namespaces if they don't use a common numbering scheme.
4312 * If multipathing is not enabled, disk names must use the controller
4313 * instance as shared namespaces will show up as multiple block
4316 if (ns->head->disk) {
4317 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4318 ctrl->instance, ns->head->instance);
4319 disk->flags |= GENHD_FL_HIDDEN;
4320 } else if (multipath) {
4321 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4322 ns->head->instance);
4324 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4325 ns->head->instance);
4328 if (nvme_update_ns_info(ns, info))
4331 down_write(&ctrl->namespaces_rwsem);
4332 nvme_ns_add_to_ctrl_list(ns);
4333 up_write(&ctrl->namespaces_rwsem);
4334 nvme_get_ctrl(ctrl);
4336 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4337 goto out_cleanup_ns_from_list;
4339 if (!nvme_ns_head_multipath(ns->head))
4340 nvme_add_ns_cdev(ns);
4342 nvme_mpath_add_disk(ns, info->anagrpid);
4343 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4347 out_cleanup_ns_from_list:
4348 nvme_put_ctrl(ctrl);
4349 down_write(&ctrl->namespaces_rwsem);
4350 list_del_init(&ns->list);
4351 up_write(&ctrl->namespaces_rwsem);
4353 mutex_lock(&ctrl->subsys->lock);
4354 list_del_rcu(&ns->siblings);
4355 if (list_empty(&ns->head->list))
4356 list_del_init(&ns->head->entry);
4357 mutex_unlock(&ctrl->subsys->lock);
4358 nvme_put_ns_head(ns->head);
4365 static void nvme_ns_remove(struct nvme_ns *ns)
4367 bool last_path = false;
4369 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4372 clear_bit(NVME_NS_READY, &ns->flags);
4373 set_capacity(ns->disk, 0);
4374 nvme_fault_inject_fini(&ns->fault_inject);
4377 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4378 * this ns going back into current_path.
4380 synchronize_srcu(&ns->head->srcu);
4382 /* wait for concurrent submissions */
4383 if (nvme_mpath_clear_current_path(ns))
4384 synchronize_srcu(&ns->head->srcu);
4386 mutex_lock(&ns->ctrl->subsys->lock);
4387 list_del_rcu(&ns->siblings);
4388 if (list_empty(&ns->head->list)) {
4389 list_del_init(&ns->head->entry);
4392 mutex_unlock(&ns->ctrl->subsys->lock);
4394 /* guarantee not available in head->list */
4395 synchronize_srcu(&ns->head->srcu);
4397 if (!nvme_ns_head_multipath(ns->head))
4398 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4399 del_gendisk(ns->disk);
4401 down_write(&ns->ctrl->namespaces_rwsem);
4402 list_del_init(&ns->list);
4403 up_write(&ns->ctrl->namespaces_rwsem);
4406 nvme_mpath_shutdown_disk(ns->head);
4410 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4412 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4420 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4422 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4424 if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4425 dev_err(ns->ctrl->device,
4426 "identifiers changed for nsid %d\n", ns->head->ns_id);
4430 ret = nvme_update_ns_info(ns, info);
4433 * Only remove the namespace if we got a fatal error back from the
4434 * device, otherwise ignore the error and just move on.
4436 * TODO: we should probably schedule a delayed retry here.
4438 if (ret > 0 && (ret & NVME_SC_DNR))
4442 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4444 struct nvme_ns_info info = { .nsid = nsid };
4448 if (nvme_identify_ns_descs(ctrl, &info))
4451 if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4452 dev_warn(ctrl->device,
4453 "command set not reported for nsid: %d\n", nsid);
4458 * If available try to use the Command Set Idependent Identify Namespace
4459 * data structure to find all the generic information that is needed to
4460 * set up a namespace. If not fall back to the legacy version.
4462 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4463 (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
4464 ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
4466 ret = nvme_ns_info_from_identify(ctrl, &info);
4468 if (info.is_removed)
4469 nvme_ns_remove_by_nsid(ctrl, nsid);
4472 * Ignore the namespace if it is not ready. We will get an AEN once it
4473 * becomes ready and restart the scan.
4475 if (ret || !info.is_ready)
4478 ns = nvme_find_get_ns(ctrl, nsid);
4480 nvme_validate_ns(ns, &info);
4483 nvme_alloc_ns(ctrl, &info);
4487 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4490 struct nvme_ns *ns, *next;
4493 down_write(&ctrl->namespaces_rwsem);
4494 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4495 if (ns->head->ns_id > nsid)
4496 list_move_tail(&ns->list, &rm_list);
4498 up_write(&ctrl->namespaces_rwsem);
4500 list_for_each_entry_safe(ns, next, &rm_list, list)
4505 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4507 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4512 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4517 struct nvme_command cmd = {
4518 .identify.opcode = nvme_admin_identify,
4519 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4520 .identify.nsid = cpu_to_le32(prev),
4523 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4524 NVME_IDENTIFY_DATA_SIZE);
4526 dev_warn(ctrl->device,
4527 "Identify NS List failed (status=0x%x)\n", ret);
4531 for (i = 0; i < nr_entries; i++) {
4532 u32 nsid = le32_to_cpu(ns_list[i]);
4534 if (!nsid) /* end of the list? */
4536 nvme_scan_ns(ctrl, nsid);
4537 while (++prev < nsid)
4538 nvme_ns_remove_by_nsid(ctrl, prev);
4542 nvme_remove_invalid_namespaces(ctrl, prev);
4548 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4550 struct nvme_id_ctrl *id;
4553 if (nvme_identify_ctrl(ctrl, &id))
4555 nn = le32_to_cpu(id->nn);
4558 for (i = 1; i <= nn; i++)
4559 nvme_scan_ns(ctrl, i);
4561 nvme_remove_invalid_namespaces(ctrl, nn);
4564 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4566 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4570 log = kzalloc(log_size, GFP_KERNEL);
4575 * We need to read the log to clear the AEN, but we don't want to rely
4576 * on it for the changed namespace information as userspace could have
4577 * raced with us in reading the log page, which could cause us to miss
4580 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4581 NVME_CSI_NVM, log, log_size, 0);
4583 dev_warn(ctrl->device,
4584 "reading changed ns log failed: %d\n", error);
4589 static void nvme_scan_work(struct work_struct *work)
4591 struct nvme_ctrl *ctrl =
4592 container_of(work, struct nvme_ctrl, scan_work);
4595 /* No tagset on a live ctrl means IO queues could not created */
4596 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4600 * Identify controller limits can change at controller reset due to
4601 * new firmware download, even though it is not common we cannot ignore
4602 * such scenario. Controller's non-mdts limits are reported in the unit
4603 * of logical blocks that is dependent on the format of attached
4604 * namespace. Hence re-read the limits at the time of ns allocation.
4606 ret = nvme_init_non_mdts_limits(ctrl);
4608 dev_warn(ctrl->device,
4609 "reading non-mdts-limits failed: %d\n", ret);
4613 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4614 dev_info(ctrl->device, "rescanning namespaces.\n");
4615 nvme_clear_changed_ns_log(ctrl);
4618 mutex_lock(&ctrl->scan_lock);
4619 if (nvme_ctrl_limited_cns(ctrl)) {
4620 nvme_scan_ns_sequential(ctrl);
4623 * Fall back to sequential scan if DNR is set to handle broken
4624 * devices which should support Identify NS List (as per the VS
4625 * they report) but don't actually support it.
4627 ret = nvme_scan_ns_list(ctrl);
4628 if (ret > 0 && ret & NVME_SC_DNR)
4629 nvme_scan_ns_sequential(ctrl);
4631 mutex_unlock(&ctrl->scan_lock);
4635 * This function iterates the namespace list unlocked to allow recovery from
4636 * controller failure. It is up to the caller to ensure the namespace list is
4637 * not modified by scan work while this function is executing.
4639 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4641 struct nvme_ns *ns, *next;
4645 * make sure to requeue I/O to all namespaces as these
4646 * might result from the scan itself and must complete
4647 * for the scan_work to make progress
4649 nvme_mpath_clear_ctrl_paths(ctrl);
4651 /* prevent racing with ns scanning */
4652 flush_work(&ctrl->scan_work);
4655 * The dead states indicates the controller was not gracefully
4656 * disconnected. In that case, we won't be able to flush any data while
4657 * removing the namespaces' disks; fail all the queues now to avoid
4658 * potentially having to clean up the failed sync later.
4660 if (ctrl->state == NVME_CTRL_DEAD) {
4661 nvme_mark_namespaces_dead(ctrl);
4662 nvme_unquiesce_io_queues(ctrl);
4665 /* this is a no-op when called from the controller reset handler */
4666 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4668 down_write(&ctrl->namespaces_rwsem);
4669 list_splice_init(&ctrl->namespaces, &ns_list);
4670 up_write(&ctrl->namespaces_rwsem);
4672 list_for_each_entry_safe(ns, next, &ns_list, list)
4675 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4677 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4679 const struct nvme_ctrl *ctrl =
4680 container_of(dev, struct nvme_ctrl, ctrl_device);
4681 struct nvmf_ctrl_options *opts = ctrl->opts;
4684 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4689 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4693 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4694 opts->trsvcid ?: "none");
4698 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4699 opts->host_traddr ?: "none");
4703 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4704 opts->host_iface ?: "none");
4709 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4711 char *envp[2] = { envdata, NULL };
4713 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4716 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4718 char *envp[2] = { NULL, NULL };
4719 u32 aen_result = ctrl->aen_result;
4721 ctrl->aen_result = 0;
4725 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4728 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4732 static void nvme_async_event_work(struct work_struct *work)
4734 struct nvme_ctrl *ctrl =
4735 container_of(work, struct nvme_ctrl, async_event_work);
4737 nvme_aen_uevent(ctrl);
4740 * The transport drivers must guarantee AER submission here is safe by
4741 * flushing ctrl async_event_work after changing the controller state
4742 * from LIVE and before freeing the admin queue.
4744 if (ctrl->state == NVME_CTRL_LIVE)
4745 ctrl->ops->submit_async_event(ctrl);
4748 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4753 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4759 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4762 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4764 struct nvme_fw_slot_info_log *log;
4766 log = kmalloc(sizeof(*log), GFP_KERNEL);
4770 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4771 log, sizeof(*log), 0))
4772 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4776 static void nvme_fw_act_work(struct work_struct *work)
4778 struct nvme_ctrl *ctrl = container_of(work,
4779 struct nvme_ctrl, fw_act_work);
4780 unsigned long fw_act_timeout;
4783 fw_act_timeout = jiffies +
4784 msecs_to_jiffies(ctrl->mtfa * 100);
4786 fw_act_timeout = jiffies +
4787 msecs_to_jiffies(admin_timeout * 1000);
4789 nvme_quiesce_io_queues(ctrl);
4790 while (nvme_ctrl_pp_status(ctrl)) {
4791 if (time_after(jiffies, fw_act_timeout)) {
4792 dev_warn(ctrl->device,
4793 "Fw activation timeout, reset controller\n");
4794 nvme_try_sched_reset(ctrl);
4800 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4803 nvme_unquiesce_io_queues(ctrl);
4804 /* read FW slot information to clear the AER */
4805 nvme_get_fw_slot_info(ctrl);
4807 queue_work(nvme_wq, &ctrl->async_event_work);
4810 static u32 nvme_aer_type(u32 result)
4812 return result & 0x7;
4815 static u32 nvme_aer_subtype(u32 result)
4817 return (result & 0xff00) >> 8;
4820 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4822 u32 aer_notice_type = nvme_aer_subtype(result);
4823 bool requeue = true;
4825 switch (aer_notice_type) {
4826 case NVME_AER_NOTICE_NS_CHANGED:
4827 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4828 nvme_queue_scan(ctrl);
4830 case NVME_AER_NOTICE_FW_ACT_STARTING:
4832 * We are (ab)using the RESETTING state to prevent subsequent
4833 * recovery actions from interfering with the controller's
4834 * firmware activation.
4836 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4837 nvme_auth_stop(ctrl);
4839 queue_work(nvme_wq, &ctrl->fw_act_work);
4842 #ifdef CONFIG_NVME_MULTIPATH
4843 case NVME_AER_NOTICE_ANA:
4844 if (!ctrl->ana_log_buf)
4846 queue_work(nvme_wq, &ctrl->ana_work);
4849 case NVME_AER_NOTICE_DISC_CHANGED:
4850 ctrl->aen_result = result;
4853 dev_warn(ctrl->device, "async event result %08x\n", result);
4858 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4860 dev_warn(ctrl->device, "resetting controller due to AER\n");
4861 nvme_reset_ctrl(ctrl);
4864 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4865 volatile union nvme_result *res)
4867 u32 result = le32_to_cpu(res->u32);
4868 u32 aer_type = nvme_aer_type(result);
4869 u32 aer_subtype = nvme_aer_subtype(result);
4870 bool requeue = true;
4872 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4875 trace_nvme_async_event(ctrl, result);
4877 case NVME_AER_NOTICE:
4878 requeue = nvme_handle_aen_notice(ctrl, result);
4880 case NVME_AER_ERROR:
4882 * For a persistent internal error, don't run async_event_work
4883 * to submit a new AER. The controller reset will do it.
4885 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4886 nvme_handle_aer_persistent_error(ctrl);
4890 case NVME_AER_SMART:
4893 ctrl->aen_result = result;
4900 queue_work(nvme_wq, &ctrl->async_event_work);
4902 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4904 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4905 const struct blk_mq_ops *ops, unsigned int cmd_size)
4909 memset(set, 0, sizeof(*set));
4911 set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4912 if (ctrl->ops->flags & NVME_F_FABRICS)
4913 set->reserved_tags = NVMF_RESERVED_TAGS;
4914 set->numa_node = ctrl->numa_node;
4915 set->flags = BLK_MQ_F_NO_SCHED;
4916 if (ctrl->ops->flags & NVME_F_BLOCKING)
4917 set->flags |= BLK_MQ_F_BLOCKING;
4918 set->cmd_size = cmd_size;
4919 set->driver_data = ctrl;
4920 set->nr_hw_queues = 1;
4921 set->timeout = NVME_ADMIN_TIMEOUT;
4922 ret = blk_mq_alloc_tag_set(set);
4926 ctrl->admin_q = blk_mq_init_queue(set);
4927 if (IS_ERR(ctrl->admin_q)) {
4928 ret = PTR_ERR(ctrl->admin_q);
4929 goto out_free_tagset;
4932 if (ctrl->ops->flags & NVME_F_FABRICS) {
4933 ctrl->fabrics_q = blk_mq_init_queue(set);
4934 if (IS_ERR(ctrl->fabrics_q)) {
4935 ret = PTR_ERR(ctrl->fabrics_q);
4936 goto out_cleanup_admin_q;
4940 ctrl->admin_tagset = set;
4943 out_cleanup_admin_q:
4944 blk_mq_destroy_queue(ctrl->admin_q);
4945 blk_put_queue(ctrl->admin_q);
4947 blk_mq_free_tag_set(set);
4948 ctrl->admin_q = NULL;
4949 ctrl->fabrics_q = NULL;
4952 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4954 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4956 blk_mq_destroy_queue(ctrl->admin_q);
4957 blk_put_queue(ctrl->admin_q);
4958 if (ctrl->ops->flags & NVME_F_FABRICS) {
4959 blk_mq_destroy_queue(ctrl->fabrics_q);
4960 blk_put_queue(ctrl->fabrics_q);
4962 blk_mq_free_tag_set(ctrl->admin_tagset);
4964 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4966 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4967 const struct blk_mq_ops *ops, unsigned int nr_maps,
4968 unsigned int cmd_size)
4972 memset(set, 0, sizeof(*set));
4974 set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4976 * Some Apple controllers requires tags to be unique across admin and
4977 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4979 if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4980 set->reserved_tags = NVME_AQ_DEPTH;
4981 else if (ctrl->ops->flags & NVME_F_FABRICS)
4982 set->reserved_tags = NVMF_RESERVED_TAGS;
4983 set->numa_node = ctrl->numa_node;
4984 set->flags = BLK_MQ_F_SHOULD_MERGE;
4985 if (ctrl->ops->flags & NVME_F_BLOCKING)
4986 set->flags |= BLK_MQ_F_BLOCKING;
4987 set->cmd_size = cmd_size,
4988 set->driver_data = ctrl;
4989 set->nr_hw_queues = ctrl->queue_count - 1;
4990 set->timeout = NVME_IO_TIMEOUT;
4991 set->nr_maps = nr_maps;
4992 ret = blk_mq_alloc_tag_set(set);
4996 if (ctrl->ops->flags & NVME_F_FABRICS) {
4997 ctrl->connect_q = blk_mq_init_queue(set);
4998 if (IS_ERR(ctrl->connect_q)) {
4999 ret = PTR_ERR(ctrl->connect_q);
5000 goto out_free_tag_set;
5002 blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
5010 blk_mq_free_tag_set(set);
5011 ctrl->connect_q = NULL;
5014 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
5016 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
5018 if (ctrl->ops->flags & NVME_F_FABRICS) {
5019 blk_mq_destroy_queue(ctrl->connect_q);
5020 blk_put_queue(ctrl->connect_q);
5022 blk_mq_free_tag_set(ctrl->tagset);
5024 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
5026 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
5028 nvme_mpath_stop(ctrl);
5029 nvme_auth_stop(ctrl);
5030 nvme_stop_keep_alive(ctrl);
5031 nvme_stop_failfast_work(ctrl);
5032 flush_work(&ctrl->async_event_work);
5033 cancel_work_sync(&ctrl->fw_act_work);
5034 if (ctrl->ops->stop_ctrl)
5035 ctrl->ops->stop_ctrl(ctrl);
5037 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
5039 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
5041 nvme_start_keep_alive(ctrl);
5043 nvme_enable_aen(ctrl);
5046 * persistent discovery controllers need to send indication to userspace
5047 * to re-read the discovery log page to learn about possible changes
5048 * that were missed. We identify persistent discovery controllers by
5049 * checking that they started once before, hence are reconnecting back.
5051 if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
5052 nvme_discovery_ctrl(ctrl))
5053 nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
5055 if (ctrl->queue_count > 1) {
5056 nvme_queue_scan(ctrl);
5057 nvme_unquiesce_io_queues(ctrl);
5058 nvme_mpath_update(ctrl);
5061 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
5062 set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
5064 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
5066 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
5068 nvme_hwmon_exit(ctrl);
5069 nvme_fault_inject_fini(&ctrl->fault_inject);
5070 dev_pm_qos_hide_latency_tolerance(ctrl->device);
5071 cdev_device_del(&ctrl->cdev, ctrl->device);
5072 nvme_put_ctrl(ctrl);
5074 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
5076 static void nvme_free_cels(struct nvme_ctrl *ctrl)
5078 struct nvme_effects_log *cel;
5081 xa_for_each(&ctrl->cels, i, cel) {
5082 xa_erase(&ctrl->cels, i);
5086 xa_destroy(&ctrl->cels);
5089 static void nvme_free_ctrl(struct device *dev)
5091 struct nvme_ctrl *ctrl =
5092 container_of(dev, struct nvme_ctrl, ctrl_device);
5093 struct nvme_subsystem *subsys = ctrl->subsys;
5095 if (!subsys || ctrl->instance != subsys->instance)
5096 ida_free(&nvme_instance_ida, ctrl->instance);
5098 nvme_free_cels(ctrl);
5099 nvme_mpath_uninit(ctrl);
5100 nvme_auth_stop(ctrl);
5101 nvme_auth_free(ctrl);
5102 __free_page(ctrl->discard_page);
5103 free_opal_dev(ctrl->opal_dev);
5106 mutex_lock(&nvme_subsystems_lock);
5107 list_del(&ctrl->subsys_entry);
5108 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
5109 mutex_unlock(&nvme_subsystems_lock);
5112 ctrl->ops->free_ctrl(ctrl);
5115 nvme_put_subsystem(subsys);
5119 * Initialize a NVMe controller structures. This needs to be called during
5120 * earliest initialization so that we have the initialized structured around
5123 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5124 const struct nvme_ctrl_ops *ops, unsigned long quirks)
5128 ctrl->state = NVME_CTRL_NEW;
5129 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5130 spin_lock_init(&ctrl->lock);
5131 mutex_init(&ctrl->scan_lock);
5132 INIT_LIST_HEAD(&ctrl->namespaces);
5133 xa_init(&ctrl->cels);
5134 init_rwsem(&ctrl->namespaces_rwsem);
5137 ctrl->quirks = quirks;
5138 ctrl->numa_node = NUMA_NO_NODE;
5139 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5140 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5141 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5142 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5143 init_waitqueue_head(&ctrl->state_wq);
5145 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5146 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5147 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5148 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5150 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5152 ctrl->discard_page = alloc_page(GFP_KERNEL);
5153 if (!ctrl->discard_page) {
5158 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5161 ctrl->instance = ret;
5163 device_initialize(&ctrl->ctrl_device);
5164 ctrl->device = &ctrl->ctrl_device;
5165 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5167 ctrl->device->class = nvme_class;
5168 ctrl->device->parent = ctrl->dev;
5169 if (ops->dev_attr_groups)
5170 ctrl->device->groups = ops->dev_attr_groups;
5172 ctrl->device->groups = nvme_dev_attr_groups;
5173 ctrl->device->release = nvme_free_ctrl;
5174 dev_set_drvdata(ctrl->device, ctrl);
5175 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5177 goto out_release_instance;
5179 nvme_get_ctrl(ctrl);
5180 cdev_init(&ctrl->cdev, &nvme_dev_fops);
5181 ctrl->cdev.owner = ops->module;
5182 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5187 * Initialize latency tolerance controls. The sysfs files won't
5188 * be visible to userspace unless the device actually supports APST.
5190 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5191 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5192 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5194 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5195 nvme_mpath_init_ctrl(ctrl);
5196 ret = nvme_auth_init_ctrl(ctrl);
5202 cdev_device_del(&ctrl->cdev, ctrl->device);
5204 nvme_put_ctrl(ctrl);
5205 kfree_const(ctrl->device->kobj.name);
5206 out_release_instance:
5207 ida_free(&nvme_instance_ida, ctrl->instance);
5209 if (ctrl->discard_page)
5210 __free_page(ctrl->discard_page);
5213 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5215 /* let I/O to all namespaces fail in preparation for surprise removal */
5216 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
5220 down_read(&ctrl->namespaces_rwsem);
5221 list_for_each_entry(ns, &ctrl->namespaces, list)
5222 blk_mark_disk_dead(ns->disk);
5223 up_read(&ctrl->namespaces_rwsem);
5225 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
5227 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5231 down_read(&ctrl->namespaces_rwsem);
5232 list_for_each_entry(ns, &ctrl->namespaces, list)
5233 blk_mq_unfreeze_queue(ns->queue);
5234 up_read(&ctrl->namespaces_rwsem);
5236 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5238 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5242 down_read(&ctrl->namespaces_rwsem);
5243 list_for_each_entry(ns, &ctrl->namespaces, list) {
5244 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5248 up_read(&ctrl->namespaces_rwsem);
5251 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5253 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5257 down_read(&ctrl->namespaces_rwsem);
5258 list_for_each_entry(ns, &ctrl->namespaces, list)
5259 blk_mq_freeze_queue_wait(ns->queue);
5260 up_read(&ctrl->namespaces_rwsem);
5262 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5264 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5268 down_read(&ctrl->namespaces_rwsem);
5269 list_for_each_entry(ns, &ctrl->namespaces, list)
5270 blk_freeze_queue_start(ns->queue);
5271 up_read(&ctrl->namespaces_rwsem);
5273 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5275 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
5279 if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5280 blk_mq_quiesce_tagset(ctrl->tagset);
5282 blk_mq_wait_quiesce_done(ctrl->tagset);
5284 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
5286 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
5290 if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5291 blk_mq_unquiesce_tagset(ctrl->tagset);
5293 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
5295 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
5297 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5298 blk_mq_quiesce_queue(ctrl->admin_q);
5300 blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
5302 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
5304 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
5306 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5307 blk_mq_unquiesce_queue(ctrl->admin_q);
5309 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
5311 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5315 down_read(&ctrl->namespaces_rwsem);
5316 list_for_each_entry(ns, &ctrl->namespaces, list)
5317 blk_sync_queue(ns->queue);
5318 up_read(&ctrl->namespaces_rwsem);
5320 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5322 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5324 nvme_sync_io_queues(ctrl);
5326 blk_sync_queue(ctrl->admin_q);
5328 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5330 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5332 if (file->f_op != &nvme_dev_fops)
5334 return file->private_data;
5336 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5339 * Check we didn't inadvertently grow the command structure sizes:
5341 static inline void _nvme_check_size(void)
5343 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5344 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5345 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5346 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5347 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5348 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5349 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5350 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5351 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5352 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5353 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5354 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5355 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5356 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5357 NVME_IDENTIFY_DATA_SIZE);
5358 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5359 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5360 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5361 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5362 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5363 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5364 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5365 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5366 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5370 static int __init nvme_core_init(void)
5372 int result = -ENOMEM;
5376 nvme_wq = alloc_workqueue("nvme-wq",
5377 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5381 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5382 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5386 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5387 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5388 if (!nvme_delete_wq)
5389 goto destroy_reset_wq;
5391 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5392 NVME_MINORS, "nvme");
5394 goto destroy_delete_wq;
5396 nvme_class = class_create("nvme");
5397 if (IS_ERR(nvme_class)) {
5398 result = PTR_ERR(nvme_class);
5399 goto unregister_chrdev;
5401 nvme_class->dev_uevent = nvme_class_uevent;
5403 nvme_subsys_class = class_create("nvme-subsystem");
5404 if (IS_ERR(nvme_subsys_class)) {
5405 result = PTR_ERR(nvme_subsys_class);
5409 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5412 goto destroy_subsys_class;
5414 nvme_ns_chr_class = class_create("nvme-generic");
5415 if (IS_ERR(nvme_ns_chr_class)) {
5416 result = PTR_ERR(nvme_ns_chr_class);
5417 goto unregister_generic_ns;
5420 result = nvme_init_auth();
5422 goto destroy_ns_chr;
5426 class_destroy(nvme_ns_chr_class);
5427 unregister_generic_ns:
5428 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5429 destroy_subsys_class:
5430 class_destroy(nvme_subsys_class);
5432 class_destroy(nvme_class);
5434 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5436 destroy_workqueue(nvme_delete_wq);
5438 destroy_workqueue(nvme_reset_wq);
5440 destroy_workqueue(nvme_wq);
5445 static void __exit nvme_core_exit(void)
5448 class_destroy(nvme_ns_chr_class);
5449 class_destroy(nvme_subsys_class);
5450 class_destroy(nvme_class);
5451 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5452 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5453 destroy_workqueue(nvme_delete_wq);
5454 destroy_workqueue(nvme_reset_wq);
5455 destroy_workqueue(nvme_wq);
5456 ida_destroy(&nvme_ns_chr_minor_ida);
5457 ida_destroy(&nvme_instance_ida);
5460 MODULE_LICENSE("GPL");
5461 MODULE_VERSION("1.0");
5462 module_init(nvme_core_init);
5463 module_exit(nvme_core_exit);