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 * Completions of long-running commands should not be able to
402 * defer sending of periodic keep alives, since the controller
403 * may have completed processing such commands a long time ago
404 * (arbitrarily close to command submission time).
405 * req->deadline - req->timeout is the command submission time
409 req->deadline - req->timeout >= ctrl->ka_last_check_time)
410 ctrl->comp_seen = true;
412 switch (nvme_decide_disposition(req)) {
420 nvme_failover_req(req);
423 #ifdef CONFIG_NVME_AUTH
424 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
432 EXPORT_SYMBOL_GPL(nvme_complete_rq);
434 void nvme_complete_batch_req(struct request *req)
436 trace_nvme_complete_rq(req);
437 nvme_cleanup_cmd(req);
438 nvme_end_req_zoned(req);
440 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
443 * Called to unwind from ->queue_rq on a failed command submission so that the
444 * multipathing code gets called to potentially failover to another path.
445 * The caller needs to unwind all transport specific resource allocations and
446 * must return propagate the return value.
448 blk_status_t nvme_host_path_error(struct request *req)
450 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
451 blk_mq_set_request_complete(req);
452 nvme_complete_rq(req);
455 EXPORT_SYMBOL_GPL(nvme_host_path_error);
457 bool nvme_cancel_request(struct request *req, void *data)
459 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
460 "Cancelling I/O %d", req->tag);
462 /* don't abort one completed or idle request */
463 if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
466 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
467 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
468 blk_mq_complete_request(req);
471 EXPORT_SYMBOL_GPL(nvme_cancel_request);
473 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
476 blk_mq_tagset_busy_iter(ctrl->tagset,
477 nvme_cancel_request, ctrl);
478 blk_mq_tagset_wait_completed_request(ctrl->tagset);
481 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
483 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
485 if (ctrl->admin_tagset) {
486 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
487 nvme_cancel_request, ctrl);
488 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
491 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
493 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
494 enum nvme_ctrl_state new_state)
496 enum nvme_ctrl_state old_state;
498 bool changed = false;
500 spin_lock_irqsave(&ctrl->lock, flags);
502 old_state = ctrl->state;
507 case NVME_CTRL_RESETTING:
508 case NVME_CTRL_CONNECTING:
515 case NVME_CTRL_RESETTING:
525 case NVME_CTRL_CONNECTING:
528 case NVME_CTRL_RESETTING:
535 case NVME_CTRL_DELETING:
538 case NVME_CTRL_RESETTING:
539 case NVME_CTRL_CONNECTING:
546 case NVME_CTRL_DELETING_NOIO:
548 case NVME_CTRL_DELETING:
558 case NVME_CTRL_DELETING:
570 ctrl->state = new_state;
571 wake_up_all(&ctrl->state_wq);
574 spin_unlock_irqrestore(&ctrl->lock, flags);
578 if (ctrl->state == NVME_CTRL_LIVE) {
579 if (old_state == NVME_CTRL_CONNECTING)
580 nvme_stop_failfast_work(ctrl);
581 nvme_kick_requeue_lists(ctrl);
582 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
583 old_state == NVME_CTRL_RESETTING) {
584 nvme_start_failfast_work(ctrl);
588 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
591 * Returns true for sink states that can't ever transition back to live.
593 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
595 switch (ctrl->state) {
598 case NVME_CTRL_RESETTING:
599 case NVME_CTRL_CONNECTING:
601 case NVME_CTRL_DELETING:
602 case NVME_CTRL_DELETING_NOIO:
606 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
612 * Waits for the controller state to be resetting, or returns false if it is
613 * not possible to ever transition to that state.
615 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
617 wait_event(ctrl->state_wq,
618 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
619 nvme_state_terminal(ctrl));
620 return ctrl->state == NVME_CTRL_RESETTING;
622 EXPORT_SYMBOL_GPL(nvme_wait_reset);
624 static void nvme_free_ns_head(struct kref *ref)
626 struct nvme_ns_head *head =
627 container_of(ref, struct nvme_ns_head, ref);
629 nvme_mpath_remove_disk(head);
630 ida_free(&head->subsys->ns_ida, head->instance);
631 cleanup_srcu_struct(&head->srcu);
632 nvme_put_subsystem(head->subsys);
636 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
638 return kref_get_unless_zero(&head->ref);
641 void nvme_put_ns_head(struct nvme_ns_head *head)
643 kref_put(&head->ref, nvme_free_ns_head);
646 static void nvme_free_ns(struct kref *kref)
648 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
651 nvme_put_ns_head(ns->head);
652 nvme_put_ctrl(ns->ctrl);
656 static inline bool nvme_get_ns(struct nvme_ns *ns)
658 return kref_get_unless_zero(&ns->kref);
661 void nvme_put_ns(struct nvme_ns *ns)
663 kref_put(&ns->kref, nvme_free_ns);
665 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
667 static inline void nvme_clear_nvme_request(struct request *req)
669 nvme_req(req)->status = 0;
670 nvme_req(req)->retries = 0;
671 nvme_req(req)->flags = 0;
672 req->rq_flags |= RQF_DONTPREP;
675 /* initialize a passthrough request */
676 void nvme_init_request(struct request *req, struct nvme_command *cmd)
678 if (req->q->queuedata)
679 req->timeout = NVME_IO_TIMEOUT;
680 else /* no queuedata implies admin queue */
681 req->timeout = NVME_ADMIN_TIMEOUT;
683 /* passthru commands should let the driver set the SGL flags */
684 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
686 req->cmd_flags |= REQ_FAILFAST_DRIVER;
687 if (req->mq_hctx->type == HCTX_TYPE_POLL)
688 req->cmd_flags |= REQ_POLLED;
689 nvme_clear_nvme_request(req);
690 req->rq_flags |= RQF_QUIET;
691 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
693 EXPORT_SYMBOL_GPL(nvme_init_request);
696 * For something we're not in a state to send to the device the default action
697 * is to busy it and retry it after the controller state is recovered. However,
698 * if the controller is deleting or if anything is marked for failfast or
699 * nvme multipath it is immediately failed.
701 * Note: commands used to initialize the controller will be marked for failfast.
702 * Note: nvme cli/ioctl commands are marked for failfast.
704 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
707 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
708 ctrl->state != NVME_CTRL_DELETING &&
709 ctrl->state != NVME_CTRL_DEAD &&
710 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
711 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
712 return BLK_STS_RESOURCE;
713 return nvme_host_path_error(rq);
715 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
717 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
720 struct nvme_request *req = nvme_req(rq);
723 * currently we have a problem sending passthru commands
724 * on the admin_q if the controller is not LIVE because we can't
725 * make sure that they are going out after the admin connect,
726 * controller enable and/or other commands in the initialization
727 * sequence. until the controller will be LIVE, fail with
728 * BLK_STS_RESOURCE so that they will be rescheduled.
730 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
733 if (ctrl->ops->flags & NVME_F_FABRICS) {
735 * Only allow commands on a live queue, except for the connect
736 * command, which is require to set the queue live in the
737 * appropinquate states.
739 switch (ctrl->state) {
740 case NVME_CTRL_CONNECTING:
741 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
742 (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
743 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
744 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
756 EXPORT_SYMBOL_GPL(__nvme_check_ready);
758 static inline void nvme_setup_flush(struct nvme_ns *ns,
759 struct nvme_command *cmnd)
761 memset(cmnd, 0, sizeof(*cmnd));
762 cmnd->common.opcode = nvme_cmd_flush;
763 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
766 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
767 struct nvme_command *cmnd)
769 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
770 struct nvme_dsm_range *range;
774 * Some devices do not consider the DSM 'Number of Ranges' field when
775 * determining how much data to DMA. Always allocate memory for maximum
776 * number of segments to prevent device reading beyond end of buffer.
778 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
780 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
783 * If we fail allocation our range, fallback to the controller
784 * discard page. If that's also busy, it's safe to return
785 * busy, as we know we can make progress once that's freed.
787 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
788 return BLK_STS_RESOURCE;
790 range = page_address(ns->ctrl->discard_page);
793 if (queue_max_discard_segments(req->q) == 1) {
794 u64 slba = nvme_sect_to_lba(ns, blk_rq_pos(req));
795 u32 nlb = blk_rq_sectors(req) >> (ns->lba_shift - 9);
797 range[0].cattr = cpu_to_le32(0);
798 range[0].nlb = cpu_to_le32(nlb);
799 range[0].slba = cpu_to_le64(slba);
802 __rq_for_each_bio(bio, req) {
803 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
804 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
807 range[n].cattr = cpu_to_le32(0);
808 range[n].nlb = cpu_to_le32(nlb);
809 range[n].slba = cpu_to_le64(slba);
815 if (WARN_ON_ONCE(n != segments)) {
816 if (virt_to_page(range) == ns->ctrl->discard_page)
817 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
820 return BLK_STS_IOERR;
823 memset(cmnd, 0, sizeof(*cmnd));
824 cmnd->dsm.opcode = nvme_cmd_dsm;
825 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
826 cmnd->dsm.nr = cpu_to_le32(segments - 1);
827 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
829 bvec_set_virt(&req->special_vec, range, alloc_size);
830 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
835 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
841 /* both rw and write zeroes share the same reftag format */
842 switch (ns->guard_type) {
843 case NVME_NVM_NS_16B_GUARD:
844 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
846 case NVME_NVM_NS_64B_GUARD:
847 ref48 = ext_pi_ref_tag(req);
848 lower = lower_32_bits(ref48);
849 upper = upper_32_bits(ref48);
851 cmnd->rw.reftag = cpu_to_le32(lower);
852 cmnd->rw.cdw3 = cpu_to_le32(upper);
859 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
860 struct request *req, struct nvme_command *cmnd)
862 memset(cmnd, 0, sizeof(*cmnd));
864 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
865 return nvme_setup_discard(ns, req, cmnd);
867 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
868 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
869 cmnd->write_zeroes.slba =
870 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
871 cmnd->write_zeroes.length =
872 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
874 if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
875 cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
877 if (nvme_ns_has_pi(ns)) {
878 cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
880 switch (ns->pi_type) {
881 case NVME_NS_DPS_PI_TYPE1:
882 case NVME_NS_DPS_PI_TYPE2:
883 nvme_set_ref_tag(ns, cmnd, req);
891 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
892 struct request *req, struct nvme_command *cmnd,
898 if (req->cmd_flags & REQ_FUA)
899 control |= NVME_RW_FUA;
900 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
901 control |= NVME_RW_LR;
903 if (req->cmd_flags & REQ_RAHEAD)
904 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
906 cmnd->rw.opcode = op;
908 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
911 cmnd->rw.metadata = 0;
912 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
913 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
916 cmnd->rw.appmask = 0;
920 * If formated with metadata, the block layer always provides a
921 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
922 * we enable the PRACT bit for protection information or set the
923 * namespace capacity to zero to prevent any I/O.
925 if (!blk_integrity_rq(req)) {
926 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
927 return BLK_STS_NOTSUPP;
928 control |= NVME_RW_PRINFO_PRACT;
931 switch (ns->pi_type) {
932 case NVME_NS_DPS_PI_TYPE3:
933 control |= NVME_RW_PRINFO_PRCHK_GUARD;
935 case NVME_NS_DPS_PI_TYPE1:
936 case NVME_NS_DPS_PI_TYPE2:
937 control |= NVME_RW_PRINFO_PRCHK_GUARD |
938 NVME_RW_PRINFO_PRCHK_REF;
939 if (op == nvme_cmd_zone_append)
940 control |= NVME_RW_APPEND_PIREMAP;
941 nvme_set_ref_tag(ns, cmnd, req);
946 cmnd->rw.control = cpu_to_le16(control);
947 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
951 void nvme_cleanup_cmd(struct request *req)
953 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
954 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
956 if (req->special_vec.bv_page == ctrl->discard_page)
957 clear_bit_unlock(0, &ctrl->discard_page_busy);
959 kfree(bvec_virt(&req->special_vec));
962 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
964 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
966 struct nvme_command *cmd = nvme_req(req)->cmd;
967 blk_status_t ret = BLK_STS_OK;
969 if (!(req->rq_flags & RQF_DONTPREP))
970 nvme_clear_nvme_request(req);
972 switch (req_op(req)) {
975 /* these are setup prior to execution in nvme_init_request() */
978 nvme_setup_flush(ns, cmd);
980 case REQ_OP_ZONE_RESET_ALL:
981 case REQ_OP_ZONE_RESET:
982 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
984 case REQ_OP_ZONE_OPEN:
985 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
987 case REQ_OP_ZONE_CLOSE:
988 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
990 case REQ_OP_ZONE_FINISH:
991 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
993 case REQ_OP_WRITE_ZEROES:
994 ret = nvme_setup_write_zeroes(ns, req, cmd);
997 ret = nvme_setup_discard(ns, req, cmd);
1000 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1003 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1005 case REQ_OP_ZONE_APPEND:
1006 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1010 return BLK_STS_IOERR;
1013 cmd->common.command_id = nvme_cid(req);
1014 trace_nvme_setup_cmd(req, cmd);
1017 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1022 * >0: nvme controller's cqe status response
1023 * <0: kernel error in lieu of controller response
1025 int nvme_execute_rq(struct request *rq, bool at_head)
1027 blk_status_t status;
1029 status = blk_execute_rq(rq, at_head);
1030 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1032 if (nvme_req(rq)->status)
1033 return nvme_req(rq)->status;
1034 return blk_status_to_errno(status);
1036 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1039 * Returns 0 on success. If the result is negative, it's a Linux error code;
1040 * if the result is positive, it's an NVM Express status code
1042 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1043 union nvme_result *result, void *buffer, unsigned bufflen,
1044 int qid, int at_head, blk_mq_req_flags_t flags)
1046 struct request *req;
1049 if (qid == NVME_QID_ANY)
1050 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1052 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1056 return PTR_ERR(req);
1057 nvme_init_request(req, cmd);
1059 if (buffer && bufflen) {
1060 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1065 ret = nvme_execute_rq(req, at_head);
1066 if (result && ret >= 0)
1067 *result = nvme_req(req)->result;
1069 blk_mq_free_request(req);
1072 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1074 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1075 void *buffer, unsigned bufflen)
1077 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1078 NVME_QID_ANY, 0, 0);
1080 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1082 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1087 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1088 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1089 dev_warn_once(ctrl->device,
1090 "IO command:%02x has unusual effects:%08x\n",
1094 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1095 * which would deadlock when done on an I/O command. Note that
1096 * We already warn about an unusual effect above.
1098 effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1100 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1105 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1107 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1109 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1112 * For simplicity, IO to all namespaces is quiesced even if the command
1113 * effects say only one namespace is affected.
1115 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1116 mutex_lock(&ctrl->scan_lock);
1117 mutex_lock(&ctrl->subsys->lock);
1118 nvme_mpath_start_freeze(ctrl->subsys);
1119 nvme_mpath_wait_freeze(ctrl->subsys);
1120 nvme_start_freeze(ctrl);
1121 nvme_wait_freeze(ctrl);
1125 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1127 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1128 struct nvme_command *cmd, int status)
1130 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1131 nvme_unfreeze(ctrl);
1132 nvme_mpath_unfreeze(ctrl->subsys);
1133 mutex_unlock(&ctrl->subsys->lock);
1134 mutex_unlock(&ctrl->scan_lock);
1136 if (effects & NVME_CMD_EFFECTS_CCC) {
1137 dev_info(ctrl->device,
1138 "controller capabilities changed, reset may be required to take effect.\n");
1140 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1141 nvme_queue_scan(ctrl);
1142 flush_work(&ctrl->scan_work);
1147 switch (cmd->common.opcode) {
1148 case nvme_admin_set_features:
1149 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1150 case NVME_FEAT_KATO:
1152 * Keep alive commands interval on the host should be
1153 * updated when KATO is modified by Set Features
1157 nvme_update_keep_alive(ctrl, cmd);
1167 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1170 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1172 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1173 * accounting for transport roundtrip times [..].
1175 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1177 unsigned long delay = ctrl->kato * HZ / 2;
1180 * When using Traffic Based Keep Alive, we need to run
1181 * nvme_keep_alive_work at twice the normal frequency, as one
1182 * command completion can postpone sending a keep alive command
1183 * by up to twice the delay between runs.
1185 if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1190 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1192 queue_delayed_work(nvme_wq, &ctrl->ka_work,
1193 nvme_keep_alive_work_period(ctrl));
1196 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1197 blk_status_t status)
1199 struct nvme_ctrl *ctrl = rq->end_io_data;
1200 unsigned long flags;
1201 bool startka = false;
1202 unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1203 unsigned long delay = nvme_keep_alive_work_period(ctrl);
1206 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1207 * at the desired frequency.
1212 dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1213 jiffies_to_msecs(rtt));
1217 blk_mq_free_request(rq);
1220 dev_err(ctrl->device,
1221 "failed nvme_keep_alive_end_io error=%d\n",
1223 return RQ_END_IO_NONE;
1226 ctrl->ka_last_check_time = jiffies;
1227 ctrl->comp_seen = false;
1228 spin_lock_irqsave(&ctrl->lock, flags);
1229 if (ctrl->state == NVME_CTRL_LIVE ||
1230 ctrl->state == NVME_CTRL_CONNECTING)
1232 spin_unlock_irqrestore(&ctrl->lock, flags);
1234 queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1235 return RQ_END_IO_NONE;
1238 static void nvme_keep_alive_work(struct work_struct *work)
1240 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1241 struct nvme_ctrl, ka_work);
1242 bool comp_seen = ctrl->comp_seen;
1245 ctrl->ka_last_check_time = jiffies;
1247 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1248 dev_dbg(ctrl->device,
1249 "reschedule traffic based keep-alive timer\n");
1250 ctrl->comp_seen = false;
1251 nvme_queue_keep_alive_work(ctrl);
1255 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1256 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1258 /* allocation failure, reset the controller */
1259 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1260 nvme_reset_ctrl(ctrl);
1263 nvme_init_request(rq, &ctrl->ka_cmd);
1265 rq->timeout = ctrl->kato * HZ;
1266 rq->end_io = nvme_keep_alive_end_io;
1267 rq->end_io_data = ctrl;
1268 blk_execute_rq_nowait(rq, false);
1271 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1273 if (unlikely(ctrl->kato == 0))
1276 nvme_queue_keep_alive_work(ctrl);
1279 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1281 if (unlikely(ctrl->kato == 0))
1284 cancel_delayed_work_sync(&ctrl->ka_work);
1286 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1288 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1289 struct nvme_command *cmd)
1291 unsigned int new_kato =
1292 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1294 dev_info(ctrl->device,
1295 "keep alive interval updated from %u ms to %u ms\n",
1296 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1298 nvme_stop_keep_alive(ctrl);
1299 ctrl->kato = new_kato;
1300 nvme_start_keep_alive(ctrl);
1304 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1305 * flag, thus sending any new CNS opcodes has a big chance of not working.
1306 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1307 * (but not for any later version).
1309 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1311 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1312 return ctrl->vs < NVME_VS(1, 2, 0);
1313 return ctrl->vs < NVME_VS(1, 1, 0);
1316 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1318 struct nvme_command c = { };
1321 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1322 c.identify.opcode = nvme_admin_identify;
1323 c.identify.cns = NVME_ID_CNS_CTRL;
1325 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1329 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1330 sizeof(struct nvme_id_ctrl));
1336 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1337 struct nvme_ns_id_desc *cur, bool *csi_seen)
1339 const char *warn_str = "ctrl returned bogus length:";
1342 switch (cur->nidt) {
1343 case NVME_NIDT_EUI64:
1344 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1345 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1346 warn_str, cur->nidl);
1349 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1350 return NVME_NIDT_EUI64_LEN;
1351 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1352 return NVME_NIDT_EUI64_LEN;
1353 case NVME_NIDT_NGUID:
1354 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1355 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1356 warn_str, cur->nidl);
1359 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1360 return NVME_NIDT_NGUID_LEN;
1361 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1362 return NVME_NIDT_NGUID_LEN;
1363 case NVME_NIDT_UUID:
1364 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1365 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1366 warn_str, cur->nidl);
1369 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1370 return NVME_NIDT_UUID_LEN;
1371 uuid_copy(&ids->uuid, data + sizeof(*cur));
1372 return NVME_NIDT_UUID_LEN;
1374 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1375 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1376 warn_str, cur->nidl);
1379 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1381 return NVME_NIDT_CSI_LEN;
1383 /* Skip unknown types */
1388 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1389 struct nvme_ns_info *info)
1391 struct nvme_command c = { };
1392 bool csi_seen = false;
1393 int status, pos, len;
1396 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1398 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1401 c.identify.opcode = nvme_admin_identify;
1402 c.identify.nsid = cpu_to_le32(info->nsid);
1403 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1405 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1409 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1410 NVME_IDENTIFY_DATA_SIZE);
1412 dev_warn(ctrl->device,
1413 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1414 info->nsid, status);
1418 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1419 struct nvme_ns_id_desc *cur = data + pos;
1424 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1428 len += sizeof(*cur);
1431 if (nvme_multi_css(ctrl) && !csi_seen) {
1432 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1442 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1443 struct nvme_id_ns **id)
1445 struct nvme_command c = { };
1448 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1449 c.identify.opcode = nvme_admin_identify;
1450 c.identify.nsid = cpu_to_le32(nsid);
1451 c.identify.cns = NVME_ID_CNS_NS;
1453 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1457 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1459 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1465 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1466 struct nvme_ns_info *info)
1468 struct nvme_ns_ids *ids = &info->ids;
1469 struct nvme_id_ns *id;
1472 ret = nvme_identify_ns(ctrl, info->nsid, &id);
1476 if (id->ncap == 0) {
1477 /* namespace not allocated or attached */
1478 info->is_removed = true;
1482 info->anagrpid = id->anagrpid;
1483 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1484 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1485 info->is_ready = true;
1486 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1487 dev_info(ctrl->device,
1488 "Ignoring bogus Namespace Identifiers\n");
1490 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1491 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1492 memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1493 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1494 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1495 memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1501 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1502 struct nvme_ns_info *info)
1504 struct nvme_id_ns_cs_indep *id;
1505 struct nvme_command c = {
1506 .identify.opcode = nvme_admin_identify,
1507 .identify.nsid = cpu_to_le32(info->nsid),
1508 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1512 id = kmalloc(sizeof(*id), GFP_KERNEL);
1516 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1518 info->anagrpid = id->anagrpid;
1519 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1520 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1521 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1527 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1528 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1530 union nvme_result res = { 0 };
1531 struct nvme_command c = { };
1534 c.features.opcode = op;
1535 c.features.fid = cpu_to_le32(fid);
1536 c.features.dword11 = cpu_to_le32(dword11);
1538 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1539 buffer, buflen, NVME_QID_ANY, 0, 0);
1540 if (ret >= 0 && result)
1541 *result = le32_to_cpu(res.u32);
1545 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1546 unsigned int dword11, void *buffer, size_t buflen,
1549 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1552 EXPORT_SYMBOL_GPL(nvme_set_features);
1554 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1555 unsigned int dword11, void *buffer, size_t buflen,
1558 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1561 EXPORT_SYMBOL_GPL(nvme_get_features);
1563 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1565 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1567 int status, nr_io_queues;
1569 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1575 * Degraded controllers might return an error when setting the queue
1576 * count. We still want to be able to bring them online and offer
1577 * access to the admin queue, as that might be only way to fix them up.
1580 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1583 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1584 *count = min(*count, nr_io_queues);
1589 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1591 #define NVME_AEN_SUPPORTED \
1592 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1593 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1595 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1597 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1600 if (!supported_aens)
1603 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1606 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1609 queue_work(nvme_wq, &ctrl->async_event_work);
1612 static int nvme_ns_open(struct nvme_ns *ns)
1615 /* should never be called due to GENHD_FL_HIDDEN */
1616 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1618 if (!nvme_get_ns(ns))
1620 if (!try_module_get(ns->ctrl->ops->module))
1631 static void nvme_ns_release(struct nvme_ns *ns)
1634 module_put(ns->ctrl->ops->module);
1638 static int nvme_open(struct block_device *bdev, fmode_t mode)
1640 return nvme_ns_open(bdev->bd_disk->private_data);
1643 static void nvme_release(struct gendisk *disk, fmode_t mode)
1645 nvme_ns_release(disk->private_data);
1648 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1650 /* some standard values */
1651 geo->heads = 1 << 6;
1652 geo->sectors = 1 << 5;
1653 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1657 #ifdef CONFIG_BLK_DEV_INTEGRITY
1658 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1659 u32 max_integrity_segments)
1661 struct blk_integrity integrity = { };
1663 switch (ns->pi_type) {
1664 case NVME_NS_DPS_PI_TYPE3:
1665 switch (ns->guard_type) {
1666 case NVME_NVM_NS_16B_GUARD:
1667 integrity.profile = &t10_pi_type3_crc;
1668 integrity.tag_size = sizeof(u16) + sizeof(u32);
1669 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1671 case NVME_NVM_NS_64B_GUARD:
1672 integrity.profile = &ext_pi_type3_crc64;
1673 integrity.tag_size = sizeof(u16) + 6;
1674 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1677 integrity.profile = NULL;
1681 case NVME_NS_DPS_PI_TYPE1:
1682 case NVME_NS_DPS_PI_TYPE2:
1683 switch (ns->guard_type) {
1684 case NVME_NVM_NS_16B_GUARD:
1685 integrity.profile = &t10_pi_type1_crc;
1686 integrity.tag_size = sizeof(u16);
1687 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1689 case NVME_NVM_NS_64B_GUARD:
1690 integrity.profile = &ext_pi_type1_crc64;
1691 integrity.tag_size = sizeof(u16);
1692 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1695 integrity.profile = NULL;
1700 integrity.profile = NULL;
1704 integrity.tuple_size = ns->ms;
1705 blk_integrity_register(disk, &integrity);
1706 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1709 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1710 u32 max_integrity_segments)
1713 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1715 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1717 struct nvme_ctrl *ctrl = ns->ctrl;
1718 struct request_queue *queue = disk->queue;
1719 u32 size = queue_logical_block_size(queue);
1721 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1722 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1724 if (ctrl->max_discard_sectors == 0) {
1725 blk_queue_max_discard_sectors(queue, 0);
1729 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1730 NVME_DSM_MAX_RANGES);
1732 queue->limits.discard_granularity = size;
1734 /* If discard is already enabled, don't reset queue limits */
1735 if (queue->limits.max_discard_sectors)
1738 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1739 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1741 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1742 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1745 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1747 return uuid_equal(&a->uuid, &b->uuid) &&
1748 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1749 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1753 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1755 bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1756 unsigned lbaf = nvme_lbaf_index(id->flbas);
1757 struct nvme_ctrl *ctrl = ns->ctrl;
1758 struct nvme_command c = { };
1759 struct nvme_id_ns_nvm *nvm;
1764 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1765 if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1766 ns->pi_size = sizeof(struct t10_pi_tuple);
1767 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1771 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1775 c.identify.opcode = nvme_admin_identify;
1776 c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1777 c.identify.cns = NVME_ID_CNS_CS_NS;
1778 c.identify.csi = NVME_CSI_NVM;
1780 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1784 elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1786 /* no support for storage tag formats right now */
1787 if (nvme_elbaf_sts(elbaf))
1790 ns->guard_type = nvme_elbaf_guard_type(elbaf);
1791 switch (ns->guard_type) {
1792 case NVME_NVM_NS_64B_GUARD:
1793 ns->pi_size = sizeof(struct crc64_pi_tuple);
1795 case NVME_NVM_NS_16B_GUARD:
1796 ns->pi_size = sizeof(struct t10_pi_tuple);
1805 if (ns->pi_size && (first || ns->ms == ns->pi_size))
1806 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1813 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1815 struct nvme_ctrl *ctrl = ns->ctrl;
1817 if (nvme_init_ms(ns, id))
1820 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1821 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1824 if (ctrl->ops->flags & NVME_F_FABRICS) {
1826 * The NVMe over Fabrics specification only supports metadata as
1827 * part of the extended data LBA. We rely on HCA/HBA support to
1828 * remap the separate metadata buffer from the block layer.
1830 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1833 ns->features |= NVME_NS_EXT_LBAS;
1836 * The current fabrics transport drivers support namespace
1837 * metadata formats only if nvme_ns_has_pi() returns true.
1838 * Suppress support for all other formats so the namespace will
1839 * have a 0 capacity and not be usable through the block stack.
1841 * Note, this check will need to be modified if any drivers
1842 * gain the ability to use other metadata formats.
1844 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1845 ns->features |= NVME_NS_METADATA_SUPPORTED;
1848 * For PCIe controllers, we can't easily remap the separate
1849 * metadata buffer from the block layer and thus require a
1850 * separate metadata buffer for block layer metadata/PI support.
1851 * We allow extended LBAs for the passthrough interface, though.
1853 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1854 ns->features |= NVME_NS_EXT_LBAS;
1856 ns->features |= NVME_NS_METADATA_SUPPORTED;
1860 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1861 struct request_queue *q)
1863 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1865 if (ctrl->max_hw_sectors) {
1867 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1869 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1870 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1871 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1873 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1874 blk_queue_dma_alignment(q, 3);
1875 blk_queue_write_cache(q, vwc, vwc);
1878 static void nvme_update_disk_info(struct gendisk *disk,
1879 struct nvme_ns *ns, struct nvme_id_ns *id)
1881 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1882 unsigned short bs = 1 << ns->lba_shift;
1883 u32 atomic_bs, phys_bs, io_opt = 0;
1886 * The block layer can't support LBA sizes larger than the page size
1887 * yet, so catch this early and don't allow block I/O.
1889 if (ns->lba_shift > PAGE_SHIFT) {
1894 blk_integrity_unregister(disk);
1896 atomic_bs = phys_bs = bs;
1897 if (id->nabo == 0) {
1899 * Bit 1 indicates whether NAWUPF is defined for this namespace
1900 * and whether it should be used instead of AWUPF. If NAWUPF ==
1901 * 0 then AWUPF must be used instead.
1903 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1904 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1906 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1909 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1910 /* NPWG = Namespace Preferred Write Granularity */
1911 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1912 /* NOWS = Namespace Optimal Write Size */
1913 io_opt = bs * (1 + le16_to_cpu(id->nows));
1916 blk_queue_logical_block_size(disk->queue, bs);
1918 * Linux filesystems assume writing a single physical block is
1919 * an atomic operation. Hence limit the physical block size to the
1920 * value of the Atomic Write Unit Power Fail parameter.
1922 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1923 blk_queue_io_min(disk->queue, phys_bs);
1924 blk_queue_io_opt(disk->queue, io_opt);
1927 * Register a metadata profile for PI, or the plain non-integrity NVMe
1928 * metadata masquerading as Type 0 if supported, otherwise reject block
1929 * I/O to namespaces with metadata except when the namespace supports
1930 * PI, as it can strip/insert in that case.
1933 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1934 (ns->features & NVME_NS_METADATA_SUPPORTED))
1935 nvme_init_integrity(disk, ns,
1936 ns->ctrl->max_integrity_segments);
1937 else if (!nvme_ns_has_pi(ns))
1941 set_capacity_and_notify(disk, capacity);
1943 nvme_config_discard(disk, ns);
1944 blk_queue_max_write_zeroes_sectors(disk->queue,
1945 ns->ctrl->max_zeroes_sectors);
1948 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1950 return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1953 static inline bool nvme_first_scan(struct gendisk *disk)
1955 /* nvme_alloc_ns() scans the disk prior to adding it */
1956 return !disk_live(disk);
1959 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1961 struct nvme_ctrl *ctrl = ns->ctrl;
1964 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1965 is_power_of_2(ctrl->max_hw_sectors))
1966 iob = ctrl->max_hw_sectors;
1968 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1973 if (!is_power_of_2(iob)) {
1974 if (nvme_first_scan(ns->disk))
1975 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1976 ns->disk->disk_name, iob);
1980 if (blk_queue_is_zoned(ns->disk->queue)) {
1981 if (nvme_first_scan(ns->disk))
1982 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1983 ns->disk->disk_name);
1987 blk_queue_chunk_sectors(ns->queue, iob);
1990 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1991 struct nvme_ns_info *info)
1993 blk_mq_freeze_queue(ns->disk->queue);
1994 nvme_set_queue_limits(ns->ctrl, ns->queue);
1995 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1996 blk_mq_unfreeze_queue(ns->disk->queue);
1998 if (nvme_ns_head_multipath(ns->head)) {
1999 blk_mq_freeze_queue(ns->head->disk->queue);
2000 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2001 nvme_mpath_revalidate_paths(ns);
2002 blk_stack_limits(&ns->head->disk->queue->limits,
2003 &ns->queue->limits, 0);
2004 ns->head->disk->flags |= GENHD_FL_HIDDEN;
2005 blk_mq_unfreeze_queue(ns->head->disk->queue);
2008 /* Hide the block-interface for these devices */
2009 ns->disk->flags |= GENHD_FL_HIDDEN;
2010 set_bit(NVME_NS_READY, &ns->flags);
2015 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2016 struct nvme_ns_info *info)
2018 struct nvme_id_ns *id;
2022 ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2026 blk_mq_freeze_queue(ns->disk->queue);
2027 lbaf = nvme_lbaf_index(id->flbas);
2028 ns->lba_shift = id->lbaf[lbaf].ds;
2029 nvme_set_queue_limits(ns->ctrl, ns->queue);
2031 nvme_configure_metadata(ns, id);
2032 nvme_set_chunk_sectors(ns, id);
2033 nvme_update_disk_info(ns->disk, ns, id);
2035 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2036 ret = nvme_update_zone_info(ns, lbaf);
2038 blk_mq_unfreeze_queue(ns->disk->queue);
2044 * Only set the DEAC bit if the device guarantees that reads from
2045 * deallocated data return zeroes. While the DEAC bit does not
2046 * require that, it must be a no-op if reads from deallocated data
2047 * do not return zeroes.
2049 if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2050 ns->features |= NVME_NS_DEAC;
2051 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2052 set_bit(NVME_NS_READY, &ns->flags);
2053 blk_mq_unfreeze_queue(ns->disk->queue);
2055 if (blk_queue_is_zoned(ns->queue)) {
2056 ret = nvme_revalidate_zones(ns);
2057 if (ret && !nvme_first_scan(ns->disk))
2061 if (nvme_ns_head_multipath(ns->head)) {
2062 blk_mq_freeze_queue(ns->head->disk->queue);
2063 nvme_update_disk_info(ns->head->disk, ns, id);
2064 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2065 nvme_mpath_revalidate_paths(ns);
2066 blk_stack_limits(&ns->head->disk->queue->limits,
2067 &ns->queue->limits, 0);
2068 disk_update_readahead(ns->head->disk);
2069 blk_mq_unfreeze_queue(ns->head->disk->queue);
2075 * If probing fails due an unsupported feature, hide the block device,
2076 * but still allow other access.
2078 if (ret == -ENODEV) {
2079 ns->disk->flags |= GENHD_FL_HIDDEN;
2080 set_bit(NVME_NS_READY, &ns->flags);
2087 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2089 switch (info->ids.csi) {
2091 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2092 dev_info(ns->ctrl->device,
2093 "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2095 return nvme_update_ns_info_generic(ns, info);
2097 return nvme_update_ns_info_block(ns, info);
2099 return nvme_update_ns_info_block(ns, info);
2101 dev_info(ns->ctrl->device,
2102 "block device for nsid %u not supported (csi %u)\n",
2103 info->nsid, info->ids.csi);
2104 return nvme_update_ns_info_generic(ns, info);
2108 static char nvme_pr_type(enum pr_type type)
2111 case PR_WRITE_EXCLUSIVE:
2113 case PR_EXCLUSIVE_ACCESS:
2115 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2117 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2119 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2121 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2128 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2129 struct nvme_command *c, u8 data[16])
2131 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2132 int srcu_idx = srcu_read_lock(&head->srcu);
2133 struct nvme_ns *ns = nvme_find_path(head);
2134 int ret = -EWOULDBLOCK;
2137 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2138 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2140 srcu_read_unlock(&head->srcu, srcu_idx);
2144 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2147 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2148 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2151 static int nvme_sc_to_pr_err(int nvme_sc)
2153 if (nvme_is_path_error(nvme_sc))
2154 return PR_STS_PATH_FAILED;
2157 case NVME_SC_SUCCESS:
2158 return PR_STS_SUCCESS;
2159 case NVME_SC_RESERVATION_CONFLICT:
2160 return PR_STS_RESERVATION_CONFLICT;
2161 case NVME_SC_ONCS_NOT_SUPPORTED:
2163 case NVME_SC_BAD_ATTRIBUTES:
2164 case NVME_SC_INVALID_OPCODE:
2165 case NVME_SC_INVALID_FIELD:
2166 case NVME_SC_INVALID_NS:
2169 return PR_STS_IOERR;
2173 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2174 u64 key, u64 sa_key, u8 op)
2176 struct nvme_command c = { };
2177 u8 data[16] = { 0, };
2180 put_unaligned_le64(key, &data[0]);
2181 put_unaligned_le64(sa_key, &data[8]);
2183 c.common.opcode = op;
2184 c.common.cdw10 = cpu_to_le32(cdw10);
2186 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2187 bdev->bd_disk->fops == &nvme_ns_head_ops)
2188 ret = nvme_send_ns_head_pr_command(bdev, &c, data);
2190 ret = nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c,
2195 return nvme_sc_to_pr_err(ret);
2198 static int nvme_pr_register(struct block_device *bdev, u64 old,
2199 u64 new, unsigned flags)
2203 if (flags & ~PR_FL_IGNORE_KEY)
2206 cdw10 = old ? 2 : 0;
2207 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2208 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2209 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2212 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2213 enum pr_type type, unsigned flags)
2217 if (flags & ~PR_FL_IGNORE_KEY)
2220 cdw10 = nvme_pr_type(type) << 8;
2221 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2222 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2225 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2226 enum pr_type type, bool abort)
2228 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2230 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2233 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2235 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2237 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2240 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2242 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2244 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2247 const struct pr_ops nvme_pr_ops = {
2248 .pr_register = nvme_pr_register,
2249 .pr_reserve = nvme_pr_reserve,
2250 .pr_release = nvme_pr_release,
2251 .pr_preempt = nvme_pr_preempt,
2252 .pr_clear = nvme_pr_clear,
2255 #ifdef CONFIG_BLK_SED_OPAL
2256 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2259 struct nvme_ctrl *ctrl = data;
2260 struct nvme_command cmd = { };
2263 cmd.common.opcode = nvme_admin_security_send;
2265 cmd.common.opcode = nvme_admin_security_recv;
2266 cmd.common.nsid = 0;
2267 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2268 cmd.common.cdw11 = cpu_to_le32(len);
2270 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2271 NVME_QID_ANY, 1, 0);
2274 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2276 if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2277 if (!ctrl->opal_dev)
2278 ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2279 else if (was_suspended)
2280 opal_unlock_from_suspend(ctrl->opal_dev);
2282 free_opal_dev(ctrl->opal_dev);
2283 ctrl->opal_dev = NULL;
2287 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2290 #endif /* CONFIG_BLK_SED_OPAL */
2292 #ifdef CONFIG_BLK_DEV_ZONED
2293 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2294 unsigned int nr_zones, report_zones_cb cb, void *data)
2296 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2300 #define nvme_report_zones NULL
2301 #endif /* CONFIG_BLK_DEV_ZONED */
2303 static const struct block_device_operations nvme_bdev_ops = {
2304 .owner = THIS_MODULE,
2305 .ioctl = nvme_ioctl,
2306 .compat_ioctl = blkdev_compat_ptr_ioctl,
2308 .release = nvme_release,
2309 .getgeo = nvme_getgeo,
2310 .report_zones = nvme_report_zones,
2311 .pr_ops = &nvme_pr_ops,
2314 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2315 u32 timeout, const char *op)
2317 unsigned long timeout_jiffies = jiffies + timeout * HZ;
2321 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2324 if ((csts & mask) == val)
2327 usleep_range(1000, 2000);
2328 if (fatal_signal_pending(current))
2330 if (time_after(jiffies, timeout_jiffies)) {
2331 dev_err(ctrl->device,
2332 "Device not ready; aborting %s, CSTS=0x%x\n",
2341 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2345 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2347 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2349 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2351 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2356 return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2357 NVME_CSTS_SHST_CMPLT,
2358 ctrl->shutdown_timeout, "shutdown");
2360 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2361 msleep(NVME_QUIRK_DELAY_AMOUNT);
2362 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2363 (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2365 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2367 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2369 unsigned dev_page_min;
2373 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2375 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2378 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2380 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2381 dev_err(ctrl->device,
2382 "Minimum device page size %u too large for host (%u)\n",
2383 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2387 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2388 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2390 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2392 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2395 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2397 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2402 if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2403 ctrl->ctrl_config |= NVME_CC_CRIME;
2404 timeout = NVME_CRTO_CRIMT(crto);
2406 timeout = NVME_CRTO_CRWMT(crto);
2409 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2412 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2413 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2414 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2415 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2419 /* Flush write to device (required if transport is PCI) */
2420 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2424 ctrl->ctrl_config |= NVME_CC_ENABLE;
2425 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2428 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2429 (timeout + 1) / 2, "initialisation");
2431 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2433 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2438 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2441 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2442 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2445 dev_warn_once(ctrl->device,
2446 "could not set timestamp (%d)\n", ret);
2450 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2452 struct nvme_feat_host_behavior *host;
2453 u8 acre = 0, lbafee = 0;
2456 /* Don't bother enabling the feature if retry delay is not reported */
2458 acre = NVME_ENABLE_ACRE;
2459 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2460 lbafee = NVME_ENABLE_LBAFEE;
2462 if (!acre && !lbafee)
2465 host = kzalloc(sizeof(*host), GFP_KERNEL);
2470 host->lbafee = lbafee;
2471 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2472 host, sizeof(*host), NULL);
2478 * The function checks whether the given total (exlat + enlat) latency of
2479 * a power state allows the latter to be used as an APST transition target.
2480 * It does so by comparing the latency to the primary and secondary latency
2481 * tolerances defined by module params. If there's a match, the corresponding
2482 * timeout value is returned and the matching tolerance index (1 or 2) is
2485 static bool nvme_apst_get_transition_time(u64 total_latency,
2486 u64 *transition_time, unsigned *last_index)
2488 if (total_latency <= apst_primary_latency_tol_us) {
2489 if (*last_index == 1)
2492 *transition_time = apst_primary_timeout_ms;
2495 if (apst_secondary_timeout_ms &&
2496 total_latency <= apst_secondary_latency_tol_us) {
2497 if (*last_index <= 2)
2500 *transition_time = apst_secondary_timeout_ms;
2507 * APST (Autonomous Power State Transition) lets us program a table of power
2508 * state transitions that the controller will perform automatically.
2510 * Depending on module params, one of the two supported techniques will be used:
2512 * - If the parameters provide explicit timeouts and tolerances, they will be
2513 * used to build a table with up to 2 non-operational states to transition to.
2514 * The default parameter values were selected based on the values used by
2515 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2516 * regeneration of the APST table in the event of switching between external
2517 * and battery power, the timeouts and tolerances reflect a compromise
2518 * between values used by Microsoft for AC and battery scenarios.
2519 * - If not, we'll configure the table with a simple heuristic: we are willing
2520 * to spend at most 2% of the time transitioning between power states.
2521 * Therefore, when running in any given state, we will enter the next
2522 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2523 * microseconds, as long as that state's exit latency is under the requested
2526 * We will not autonomously enter any non-operational state for which the total
2527 * latency exceeds ps_max_latency_us.
2529 * Users can set ps_max_latency_us to zero to turn off APST.
2531 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2533 struct nvme_feat_auto_pst *table;
2540 unsigned last_lt_index = UINT_MAX;
2543 * If APST isn't supported or if we haven't been initialized yet,
2544 * then don't do anything.
2549 if (ctrl->npss > 31) {
2550 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2554 table = kzalloc(sizeof(*table), GFP_KERNEL);
2558 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2559 /* Turn off APST. */
2560 dev_dbg(ctrl->device, "APST disabled\n");
2565 * Walk through all states from lowest- to highest-power.
2566 * According to the spec, lower-numbered states use more power. NPSS,
2567 * despite the name, is the index of the lowest-power state, not the
2570 for (state = (int)ctrl->npss; state >= 0; state--) {
2571 u64 total_latency_us, exit_latency_us, transition_ms;
2574 table->entries[state] = target;
2577 * Don't allow transitions to the deepest state if it's quirked
2580 if (state == ctrl->npss &&
2581 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2585 * Is this state a useful non-operational state for higher-power
2586 * states to autonomously transition to?
2588 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2591 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2592 if (exit_latency_us > ctrl->ps_max_latency_us)
2595 total_latency_us = exit_latency_us +
2596 le32_to_cpu(ctrl->psd[state].entry_lat);
2599 * This state is good. It can be used as the APST idle target
2600 * for higher power states.
2602 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2603 if (!nvme_apst_get_transition_time(total_latency_us,
2604 &transition_ms, &last_lt_index))
2607 transition_ms = total_latency_us + 19;
2608 do_div(transition_ms, 20);
2609 if (transition_ms > (1 << 24) - 1)
2610 transition_ms = (1 << 24) - 1;
2613 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2616 if (total_latency_us > max_lat_us)
2617 max_lat_us = total_latency_us;
2621 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2623 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2624 max_ps, max_lat_us, (int)sizeof(*table), table);
2628 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2629 table, sizeof(*table), NULL);
2631 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2636 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2638 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2642 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2643 case PM_QOS_LATENCY_ANY:
2651 if (ctrl->ps_max_latency_us != latency) {
2652 ctrl->ps_max_latency_us = latency;
2653 if (ctrl->state == NVME_CTRL_LIVE)
2654 nvme_configure_apst(ctrl);
2658 struct nvme_core_quirk_entry {
2660 * NVMe model and firmware strings are padded with spaces. For
2661 * simplicity, strings in the quirk table are padded with NULLs
2667 unsigned long quirks;
2670 static const struct nvme_core_quirk_entry core_quirks[] = {
2673 * This Toshiba device seems to die using any APST states. See:
2674 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2677 .mn = "THNSF5256GPUK TOSHIBA",
2678 .quirks = NVME_QUIRK_NO_APST,
2682 * This LiteON CL1-3D*-Q11 firmware version has a race
2683 * condition associated with actions related to suspend to idle
2684 * LiteON has resolved the problem in future firmware
2688 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2692 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2693 * aborts I/O during any load, but more easily reproducible
2694 * with discards (fstrim).
2696 * The device is left in a state where it is also not possible
2697 * to use "nvme set-feature" to disable APST, but booting with
2698 * nvme_core.default_ps_max_latency=0 works.
2701 .mn = "KCD6XVUL6T40",
2702 .quirks = NVME_QUIRK_NO_APST,
2706 * The external Samsung X5 SSD fails initialization without a
2707 * delay before checking if it is ready and has a whole set of
2708 * other problems. To make this even more interesting, it
2709 * shares the PCI ID with internal Samsung 970 Evo Plus that
2710 * does not need or want these quirks.
2713 .mn = "Samsung Portable SSD X5",
2714 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2715 NVME_QUIRK_NO_DEEPEST_PS |
2716 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2720 /* match is null-terminated but idstr is space-padded. */
2721 static bool string_matches(const char *idstr, const char *match, size_t len)
2728 matchlen = strlen(match);
2729 WARN_ON_ONCE(matchlen > len);
2731 if (memcmp(idstr, match, matchlen))
2734 for (; matchlen < len; matchlen++)
2735 if (idstr[matchlen] != ' ')
2741 static bool quirk_matches(const struct nvme_id_ctrl *id,
2742 const struct nvme_core_quirk_entry *q)
2744 return q->vid == le16_to_cpu(id->vid) &&
2745 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2746 string_matches(id->fr, q->fr, sizeof(id->fr));
2749 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2750 struct nvme_id_ctrl *id)
2755 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2756 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2757 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2758 strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2762 if (ctrl->vs >= NVME_VS(1, 2, 1))
2763 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2767 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2768 * Base Specification 2.0. It is slightly different from the format
2769 * specified there due to historic reasons, and we can't change it now.
2771 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2772 "nqn.2014.08.org.nvmexpress:%04x%04x",
2773 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2774 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2775 off += sizeof(id->sn);
2776 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2777 off += sizeof(id->mn);
2778 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2781 static void nvme_release_subsystem(struct device *dev)
2783 struct nvme_subsystem *subsys =
2784 container_of(dev, struct nvme_subsystem, dev);
2786 if (subsys->instance >= 0)
2787 ida_free(&nvme_instance_ida, subsys->instance);
2791 static void nvme_destroy_subsystem(struct kref *ref)
2793 struct nvme_subsystem *subsys =
2794 container_of(ref, struct nvme_subsystem, ref);
2796 mutex_lock(&nvme_subsystems_lock);
2797 list_del(&subsys->entry);
2798 mutex_unlock(&nvme_subsystems_lock);
2800 ida_destroy(&subsys->ns_ida);
2801 device_del(&subsys->dev);
2802 put_device(&subsys->dev);
2805 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2807 kref_put(&subsys->ref, nvme_destroy_subsystem);
2810 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2812 struct nvme_subsystem *subsys;
2814 lockdep_assert_held(&nvme_subsystems_lock);
2817 * Fail matches for discovery subsystems. This results
2818 * in each discovery controller bound to a unique subsystem.
2819 * This avoids issues with validating controller values
2820 * that can only be true when there is a single unique subsystem.
2821 * There may be multiple and completely independent entities
2822 * that provide discovery controllers.
2824 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2827 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2828 if (strcmp(subsys->subnqn, subsysnqn))
2830 if (!kref_get_unless_zero(&subsys->ref))
2838 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2839 struct device_attribute subsys_attr_##_name = \
2840 __ATTR(_name, _mode, _show, NULL)
2842 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2843 struct device_attribute *attr,
2846 struct nvme_subsystem *subsys =
2847 container_of(dev, struct nvme_subsystem, dev);
2849 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2851 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2853 static ssize_t nvme_subsys_show_type(struct device *dev,
2854 struct device_attribute *attr,
2857 struct nvme_subsystem *subsys =
2858 container_of(dev, struct nvme_subsystem, dev);
2860 switch (subsys->subtype) {
2862 return sysfs_emit(buf, "discovery\n");
2864 return sysfs_emit(buf, "nvm\n");
2866 return sysfs_emit(buf, "reserved\n");
2869 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2871 #define nvme_subsys_show_str_function(field) \
2872 static ssize_t subsys_##field##_show(struct device *dev, \
2873 struct device_attribute *attr, char *buf) \
2875 struct nvme_subsystem *subsys = \
2876 container_of(dev, struct nvme_subsystem, dev); \
2877 return sysfs_emit(buf, "%.*s\n", \
2878 (int)sizeof(subsys->field), subsys->field); \
2880 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2882 nvme_subsys_show_str_function(model);
2883 nvme_subsys_show_str_function(serial);
2884 nvme_subsys_show_str_function(firmware_rev);
2886 static struct attribute *nvme_subsys_attrs[] = {
2887 &subsys_attr_model.attr,
2888 &subsys_attr_serial.attr,
2889 &subsys_attr_firmware_rev.attr,
2890 &subsys_attr_subsysnqn.attr,
2891 &subsys_attr_subsystype.attr,
2892 #ifdef CONFIG_NVME_MULTIPATH
2893 &subsys_attr_iopolicy.attr,
2898 static const struct attribute_group nvme_subsys_attrs_group = {
2899 .attrs = nvme_subsys_attrs,
2902 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2903 &nvme_subsys_attrs_group,
2907 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2909 return ctrl->opts && ctrl->opts->discovery_nqn;
2912 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2913 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2915 struct nvme_ctrl *tmp;
2917 lockdep_assert_held(&nvme_subsystems_lock);
2919 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2920 if (nvme_state_terminal(tmp))
2923 if (tmp->cntlid == ctrl->cntlid) {
2924 dev_err(ctrl->device,
2925 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2926 ctrl->cntlid, dev_name(tmp->device),
2931 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2932 nvme_discovery_ctrl(ctrl))
2935 dev_err(ctrl->device,
2936 "Subsystem does not support multiple controllers\n");
2943 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2945 struct nvme_subsystem *subsys, *found;
2948 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2952 subsys->instance = -1;
2953 mutex_init(&subsys->lock);
2954 kref_init(&subsys->ref);
2955 INIT_LIST_HEAD(&subsys->ctrls);
2956 INIT_LIST_HEAD(&subsys->nsheads);
2957 nvme_init_subnqn(subsys, ctrl, id);
2958 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2959 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2960 subsys->vendor_id = le16_to_cpu(id->vid);
2961 subsys->cmic = id->cmic;
2963 /* Versions prior to 1.4 don't necessarily report a valid type */
2964 if (id->cntrltype == NVME_CTRL_DISC ||
2965 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2966 subsys->subtype = NVME_NQN_DISC;
2968 subsys->subtype = NVME_NQN_NVME;
2970 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2971 dev_err(ctrl->device,
2972 "Subsystem %s is not a discovery controller",
2977 subsys->awupf = le16_to_cpu(id->awupf);
2978 nvme_mpath_default_iopolicy(subsys);
2980 subsys->dev.class = nvme_subsys_class;
2981 subsys->dev.release = nvme_release_subsystem;
2982 subsys->dev.groups = nvme_subsys_attrs_groups;
2983 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2984 device_initialize(&subsys->dev);
2986 mutex_lock(&nvme_subsystems_lock);
2987 found = __nvme_find_get_subsystem(subsys->subnqn);
2989 put_device(&subsys->dev);
2992 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2994 goto out_put_subsystem;
2997 ret = device_add(&subsys->dev);
2999 dev_err(ctrl->device,
3000 "failed to register subsystem device.\n");
3001 put_device(&subsys->dev);
3004 ida_init(&subsys->ns_ida);
3005 list_add_tail(&subsys->entry, &nvme_subsystems);
3008 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
3009 dev_name(ctrl->device));
3011 dev_err(ctrl->device,
3012 "failed to create sysfs link from subsystem.\n");
3013 goto out_put_subsystem;
3017 subsys->instance = ctrl->instance;
3018 ctrl->subsys = subsys;
3019 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
3020 mutex_unlock(&nvme_subsystems_lock);
3024 nvme_put_subsystem(subsys);
3026 mutex_unlock(&nvme_subsystems_lock);
3030 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3031 void *log, size_t size, u64 offset)
3033 struct nvme_command c = { };
3034 u32 dwlen = nvme_bytes_to_numd(size);
3036 c.get_log_page.opcode = nvme_admin_get_log_page;
3037 c.get_log_page.nsid = cpu_to_le32(nsid);
3038 c.get_log_page.lid = log_page;
3039 c.get_log_page.lsp = lsp;
3040 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3041 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3042 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3043 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3044 c.get_log_page.csi = csi;
3046 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3049 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3050 struct nvme_effects_log **log)
3052 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3058 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3062 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3063 cel, sizeof(*cel), 0);
3069 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3075 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3077 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3079 if (check_shl_overflow(1U, units + page_shift - 9, &val))
3084 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3086 struct nvme_command c = { };
3087 struct nvme_id_ctrl_nvm *id;
3090 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3091 ctrl->max_discard_sectors = UINT_MAX;
3092 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3094 ctrl->max_discard_sectors = 0;
3095 ctrl->max_discard_segments = 0;
3099 * Even though NVMe spec explicitly states that MDTS is not applicable
3100 * to the write-zeroes, we are cautious and limit the size to the
3101 * controllers max_hw_sectors value, which is based on the MDTS field
3102 * and possibly other limiting factors.
3104 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3105 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3106 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3108 ctrl->max_zeroes_sectors = 0;
3110 if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3111 nvme_ctrl_limited_cns(ctrl))
3114 id = kzalloc(sizeof(*id), GFP_KERNEL);
3118 c.identify.opcode = nvme_admin_identify;
3119 c.identify.cns = NVME_ID_CNS_CS_CTRL;
3120 c.identify.csi = NVME_CSI_NVM;
3122 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3127 ctrl->max_discard_segments = id->dmrl;
3128 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3130 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3137 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3139 struct nvme_effects_log *log = ctrl->effects;
3141 log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3142 NVME_CMD_EFFECTS_NCC |
3143 NVME_CMD_EFFECTS_CSE_MASK);
3144 log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3145 NVME_CMD_EFFECTS_CSE_MASK);
3148 * The spec says the result of a security receive command depends on
3149 * the previous security send command. As such, many vendors log this
3150 * command as one to submitted only when no other commands to the same
3151 * namespace are outstanding. The intention is to tell the host to
3152 * prevent mixing security send and receive.
3154 * This driver can only enforce such exclusive access against IO
3155 * queues, though. We are not readily able to enforce such a rule for
3156 * two commands to the admin queue, which is the only queue that
3157 * matters for this command.
3159 * Rather than blindly freezing the IO queues for this effect that
3160 * doesn't even apply to IO, mask it off.
3162 log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3164 log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3165 log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3166 log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3169 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3176 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3177 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3182 if (!ctrl->effects) {
3183 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3186 xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3189 nvme_init_known_nvm_effects(ctrl);
3193 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3195 struct nvme_id_ctrl *id;
3197 bool prev_apst_enabled;
3200 ret = nvme_identify_ctrl(ctrl, &id);
3202 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3206 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3207 ctrl->cntlid = le16_to_cpu(id->cntlid);
3209 if (!ctrl->identified) {
3213 * Check for quirks. Quirk can depend on firmware version,
3214 * so, in principle, the set of quirks present can change
3215 * across a reset. As a possible future enhancement, we
3216 * could re-scan for quirks every time we reinitialize
3217 * the device, but we'd have to make sure that the driver
3218 * behaves intelligently if the quirks change.
3220 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3221 if (quirk_matches(id, &core_quirks[i]))
3222 ctrl->quirks |= core_quirks[i].quirks;
3225 ret = nvme_init_subsystem(ctrl, id);
3229 ret = nvme_init_effects(ctrl, id);
3233 memcpy(ctrl->subsys->firmware_rev, id->fr,
3234 sizeof(ctrl->subsys->firmware_rev));
3236 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3237 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3238 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3241 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3242 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3243 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3245 ctrl->oacs = le16_to_cpu(id->oacs);
3246 ctrl->oncs = le16_to_cpu(id->oncs);
3247 ctrl->mtfa = le16_to_cpu(id->mtfa);
3248 ctrl->oaes = le32_to_cpu(id->oaes);
3249 ctrl->wctemp = le16_to_cpu(id->wctemp);
3250 ctrl->cctemp = le16_to_cpu(id->cctemp);
3252 atomic_set(&ctrl->abort_limit, id->acl + 1);
3253 ctrl->vwc = id->vwc;
3255 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3257 max_hw_sectors = UINT_MAX;
3258 ctrl->max_hw_sectors =
3259 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3261 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3262 ctrl->sgls = le32_to_cpu(id->sgls);
3263 ctrl->kas = le16_to_cpu(id->kas);
3264 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3265 ctrl->ctratt = le32_to_cpu(id->ctratt);
3267 ctrl->cntrltype = id->cntrltype;
3268 ctrl->dctype = id->dctype;
3272 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3274 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3275 shutdown_timeout, 60);
3277 if (ctrl->shutdown_timeout != shutdown_timeout)
3278 dev_info(ctrl->device,
3279 "Shutdown timeout set to %u seconds\n",
3280 ctrl->shutdown_timeout);
3282 ctrl->shutdown_timeout = shutdown_timeout;
3284 ctrl->npss = id->npss;
3285 ctrl->apsta = id->apsta;
3286 prev_apst_enabled = ctrl->apst_enabled;
3287 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3288 if (force_apst && id->apsta) {
3289 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3290 ctrl->apst_enabled = true;
3292 ctrl->apst_enabled = false;
3295 ctrl->apst_enabled = id->apsta;
3297 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3299 if (ctrl->ops->flags & NVME_F_FABRICS) {
3300 ctrl->icdoff = le16_to_cpu(id->icdoff);
3301 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3302 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3303 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3306 * In fabrics we need to verify the cntlid matches the
3309 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3310 dev_err(ctrl->device,
3311 "Mismatching cntlid: Connect %u vs Identify "
3313 ctrl->cntlid, le16_to_cpu(id->cntlid));
3318 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3319 dev_err(ctrl->device,
3320 "keep-alive support is mandatory for fabrics\n");
3325 ctrl->hmpre = le32_to_cpu(id->hmpre);
3326 ctrl->hmmin = le32_to_cpu(id->hmmin);
3327 ctrl->hmminds = le32_to_cpu(id->hmminds);
3328 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3331 ret = nvme_mpath_init_identify(ctrl, id);
3335 if (ctrl->apst_enabled && !prev_apst_enabled)
3336 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3337 else if (!ctrl->apst_enabled && prev_apst_enabled)
3338 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3346 * Initialize the cached copies of the Identify data and various controller
3347 * register in our nvme_ctrl structure. This should be called as soon as
3348 * the admin queue is fully up and running.
3350 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3354 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3356 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3360 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3362 if (ctrl->vs >= NVME_VS(1, 1, 0))
3363 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3365 ret = nvme_init_identify(ctrl);
3369 ret = nvme_configure_apst(ctrl);
3373 ret = nvme_configure_timestamp(ctrl);
3377 ret = nvme_configure_host_options(ctrl);
3381 nvme_configure_opal(ctrl, was_suspended);
3383 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3385 * Do not return errors unless we are in a controller reset,
3386 * the controller works perfectly fine without hwmon.
3388 ret = nvme_hwmon_init(ctrl);
3393 ctrl->identified = true;
3397 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3399 static int nvme_dev_open(struct inode *inode, struct file *file)
3401 struct nvme_ctrl *ctrl =
3402 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3404 switch (ctrl->state) {
3405 case NVME_CTRL_LIVE:
3408 return -EWOULDBLOCK;
3411 nvme_get_ctrl(ctrl);
3412 if (!try_module_get(ctrl->ops->module)) {
3413 nvme_put_ctrl(ctrl);
3417 file->private_data = ctrl;
3421 static int nvme_dev_release(struct inode *inode, struct file *file)
3423 struct nvme_ctrl *ctrl =
3424 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3426 module_put(ctrl->ops->module);
3427 nvme_put_ctrl(ctrl);
3431 static const struct file_operations nvme_dev_fops = {
3432 .owner = THIS_MODULE,
3433 .open = nvme_dev_open,
3434 .release = nvme_dev_release,
3435 .unlocked_ioctl = nvme_dev_ioctl,
3436 .compat_ioctl = compat_ptr_ioctl,
3437 .uring_cmd = nvme_dev_uring_cmd,
3440 static ssize_t nvme_sysfs_reset(struct device *dev,
3441 struct device_attribute *attr, const char *buf,
3444 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3447 ret = nvme_reset_ctrl_sync(ctrl);
3452 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3454 static ssize_t nvme_sysfs_rescan(struct device *dev,
3455 struct device_attribute *attr, const char *buf,
3458 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3460 nvme_queue_scan(ctrl);
3463 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3465 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3467 struct gendisk *disk = dev_to_disk(dev);
3469 if (disk->fops == &nvme_bdev_ops)
3470 return nvme_get_ns_from_dev(dev)->head;
3472 return disk->private_data;
3475 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3478 struct nvme_ns_head *head = dev_to_ns_head(dev);
3479 struct nvme_ns_ids *ids = &head->ids;
3480 struct nvme_subsystem *subsys = head->subsys;
3481 int serial_len = sizeof(subsys->serial);
3482 int model_len = sizeof(subsys->model);
3484 if (!uuid_is_null(&ids->uuid))
3485 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3487 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3488 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3490 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3491 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3493 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3494 subsys->serial[serial_len - 1] == '\0'))
3496 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3497 subsys->model[model_len - 1] == '\0'))
3500 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3501 serial_len, subsys->serial, model_len, subsys->model,
3504 static DEVICE_ATTR_RO(wwid);
3506 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3509 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3511 static DEVICE_ATTR_RO(nguid);
3513 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3516 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3518 /* For backward compatibility expose the NGUID to userspace if
3519 * we have no UUID set
3521 if (uuid_is_null(&ids->uuid)) {
3522 dev_warn_ratelimited(dev,
3523 "No UUID available providing old NGUID\n");
3524 return sysfs_emit(buf, "%pU\n", ids->nguid);
3526 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3528 static DEVICE_ATTR_RO(uuid);
3530 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3533 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3535 static DEVICE_ATTR_RO(eui);
3537 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3540 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3542 static DEVICE_ATTR_RO(nsid);
3544 static struct attribute *nvme_ns_id_attrs[] = {
3545 &dev_attr_wwid.attr,
3546 &dev_attr_uuid.attr,
3547 &dev_attr_nguid.attr,
3549 &dev_attr_nsid.attr,
3550 #ifdef CONFIG_NVME_MULTIPATH
3551 &dev_attr_ana_grpid.attr,
3552 &dev_attr_ana_state.attr,
3557 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3558 struct attribute *a, int n)
3560 struct device *dev = container_of(kobj, struct device, kobj);
3561 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3563 if (a == &dev_attr_uuid.attr) {
3564 if (uuid_is_null(&ids->uuid) &&
3565 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3568 if (a == &dev_attr_nguid.attr) {
3569 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3572 if (a == &dev_attr_eui.attr) {
3573 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3576 #ifdef CONFIG_NVME_MULTIPATH
3577 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3578 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3580 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3587 static const struct attribute_group nvme_ns_id_attr_group = {
3588 .attrs = nvme_ns_id_attrs,
3589 .is_visible = nvme_ns_id_attrs_are_visible,
3592 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3593 &nvme_ns_id_attr_group,
3597 #define nvme_show_str_function(field) \
3598 static ssize_t field##_show(struct device *dev, \
3599 struct device_attribute *attr, char *buf) \
3601 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3602 return sysfs_emit(buf, "%.*s\n", \
3603 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3605 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3607 nvme_show_str_function(model);
3608 nvme_show_str_function(serial);
3609 nvme_show_str_function(firmware_rev);
3611 #define nvme_show_int_function(field) \
3612 static ssize_t field##_show(struct device *dev, \
3613 struct device_attribute *attr, char *buf) \
3615 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3616 return sysfs_emit(buf, "%d\n", ctrl->field); \
3618 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3620 nvme_show_int_function(cntlid);
3621 nvme_show_int_function(numa_node);
3622 nvme_show_int_function(queue_count);
3623 nvme_show_int_function(sqsize);
3624 nvme_show_int_function(kato);
3626 static ssize_t nvme_sysfs_delete(struct device *dev,
3627 struct device_attribute *attr, const char *buf,
3630 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3632 if (!test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags))
3635 if (device_remove_file_self(dev, attr))
3636 nvme_delete_ctrl_sync(ctrl);
3639 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3641 static ssize_t nvme_sysfs_show_transport(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->ops->name);
3649 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3651 static ssize_t nvme_sysfs_show_state(struct device *dev,
3652 struct device_attribute *attr,
3655 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3656 static const char *const state_name[] = {
3657 [NVME_CTRL_NEW] = "new",
3658 [NVME_CTRL_LIVE] = "live",
3659 [NVME_CTRL_RESETTING] = "resetting",
3660 [NVME_CTRL_CONNECTING] = "connecting",
3661 [NVME_CTRL_DELETING] = "deleting",
3662 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3663 [NVME_CTRL_DEAD] = "dead",
3666 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3667 state_name[ctrl->state])
3668 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3670 return sysfs_emit(buf, "unknown state\n");
3673 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3675 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3676 struct device_attribute *attr,
3679 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3681 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3683 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3685 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3686 struct device_attribute *attr,
3689 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3691 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3693 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3695 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3696 struct device_attribute *attr,
3699 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3701 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3703 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3705 static ssize_t nvme_sysfs_show_address(struct device *dev,
3706 struct device_attribute *attr,
3709 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3711 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3713 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3715 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3716 struct device_attribute *attr, char *buf)
3718 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3719 struct nvmf_ctrl_options *opts = ctrl->opts;
3721 if (ctrl->opts->max_reconnects == -1)
3722 return sysfs_emit(buf, "off\n");
3723 return sysfs_emit(buf, "%d\n",
3724 opts->max_reconnects * opts->reconnect_delay);
3727 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3728 struct device_attribute *attr, const char *buf, size_t count)
3730 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3731 struct nvmf_ctrl_options *opts = ctrl->opts;
3732 int ctrl_loss_tmo, err;
3734 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3738 if (ctrl_loss_tmo < 0)
3739 opts->max_reconnects = -1;
3741 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3742 opts->reconnect_delay);
3745 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3746 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3748 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3749 struct device_attribute *attr, char *buf)
3751 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3753 if (ctrl->opts->reconnect_delay == -1)
3754 return sysfs_emit(buf, "off\n");
3755 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3758 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3759 struct device_attribute *attr, const char *buf, size_t count)
3761 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3765 err = kstrtou32(buf, 10, &v);
3769 ctrl->opts->reconnect_delay = v;
3772 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3773 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3775 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3776 struct device_attribute *attr, char *buf)
3778 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3780 if (ctrl->opts->fast_io_fail_tmo == -1)
3781 return sysfs_emit(buf, "off\n");
3782 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3785 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3786 struct device_attribute *attr, const char *buf, size_t count)
3788 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3789 struct nvmf_ctrl_options *opts = ctrl->opts;
3790 int fast_io_fail_tmo, err;
3792 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3796 if (fast_io_fail_tmo < 0)
3797 opts->fast_io_fail_tmo = -1;
3799 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3802 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3803 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3805 static ssize_t cntrltype_show(struct device *dev,
3806 struct device_attribute *attr, char *buf)
3808 static const char * const type[] = {
3809 [NVME_CTRL_IO] = "io\n",
3810 [NVME_CTRL_DISC] = "discovery\n",
3811 [NVME_CTRL_ADMIN] = "admin\n",
3813 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3815 if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3816 return sysfs_emit(buf, "reserved\n");
3818 return sysfs_emit(buf, type[ctrl->cntrltype]);
3820 static DEVICE_ATTR_RO(cntrltype);
3822 static ssize_t dctype_show(struct device *dev,
3823 struct device_attribute *attr, char *buf)
3825 static const char * const type[] = {
3826 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3827 [NVME_DCTYPE_DDC] = "ddc\n",
3828 [NVME_DCTYPE_CDC] = "cdc\n",
3830 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3832 if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3833 return sysfs_emit(buf, "reserved\n");
3835 return sysfs_emit(buf, type[ctrl->dctype]);
3837 static DEVICE_ATTR_RO(dctype);
3839 #ifdef CONFIG_NVME_AUTH
3840 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3841 struct device_attribute *attr, char *buf)
3843 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3844 struct nvmf_ctrl_options *opts = ctrl->opts;
3846 if (!opts->dhchap_secret)
3847 return sysfs_emit(buf, "none\n");
3848 return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3851 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3852 struct device_attribute *attr, const char *buf, size_t count)
3854 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3855 struct nvmf_ctrl_options *opts = ctrl->opts;
3856 char *dhchap_secret;
3858 if (!ctrl->opts->dhchap_secret)
3862 if (memcmp(buf, "DHHC-1:", 7))
3865 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3868 memcpy(dhchap_secret, buf, count);
3869 nvme_auth_stop(ctrl);
3870 if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3871 struct nvme_dhchap_key *key, *host_key;
3874 ret = nvme_auth_generate_key(dhchap_secret, &key);
3877 kfree(opts->dhchap_secret);
3878 opts->dhchap_secret = dhchap_secret;
3879 host_key = ctrl->host_key;
3880 mutex_lock(&ctrl->dhchap_auth_mutex);
3881 ctrl->host_key = key;
3882 mutex_unlock(&ctrl->dhchap_auth_mutex);
3883 nvme_auth_free_key(host_key);
3885 /* Start re-authentication */
3886 dev_info(ctrl->device, "re-authenticating controller\n");
3887 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3891 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3892 nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3894 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3895 struct device_attribute *attr, char *buf)
3897 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3898 struct nvmf_ctrl_options *opts = ctrl->opts;
3900 if (!opts->dhchap_ctrl_secret)
3901 return sysfs_emit(buf, "none\n");
3902 return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3905 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3906 struct device_attribute *attr, const char *buf, size_t count)
3908 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3909 struct nvmf_ctrl_options *opts = ctrl->opts;
3910 char *dhchap_secret;
3912 if (!ctrl->opts->dhchap_ctrl_secret)
3916 if (memcmp(buf, "DHHC-1:", 7))
3919 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3922 memcpy(dhchap_secret, buf, count);
3923 nvme_auth_stop(ctrl);
3924 if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3925 struct nvme_dhchap_key *key, *ctrl_key;
3928 ret = nvme_auth_generate_key(dhchap_secret, &key);
3931 kfree(opts->dhchap_ctrl_secret);
3932 opts->dhchap_ctrl_secret = dhchap_secret;
3933 ctrl_key = ctrl->ctrl_key;
3934 mutex_lock(&ctrl->dhchap_auth_mutex);
3935 ctrl->ctrl_key = key;
3936 mutex_unlock(&ctrl->dhchap_auth_mutex);
3937 nvme_auth_free_key(ctrl_key);
3939 /* Start re-authentication */
3940 dev_info(ctrl->device, "re-authenticating controller\n");
3941 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3945 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3946 nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3949 static struct attribute *nvme_dev_attrs[] = {
3950 &dev_attr_reset_controller.attr,
3951 &dev_attr_rescan_controller.attr,
3952 &dev_attr_model.attr,
3953 &dev_attr_serial.attr,
3954 &dev_attr_firmware_rev.attr,
3955 &dev_attr_cntlid.attr,
3956 &dev_attr_delete_controller.attr,
3957 &dev_attr_transport.attr,
3958 &dev_attr_subsysnqn.attr,
3959 &dev_attr_address.attr,
3960 &dev_attr_state.attr,
3961 &dev_attr_numa_node.attr,
3962 &dev_attr_queue_count.attr,
3963 &dev_attr_sqsize.attr,
3964 &dev_attr_hostnqn.attr,
3965 &dev_attr_hostid.attr,
3966 &dev_attr_ctrl_loss_tmo.attr,
3967 &dev_attr_reconnect_delay.attr,
3968 &dev_attr_fast_io_fail_tmo.attr,
3969 &dev_attr_kato.attr,
3970 &dev_attr_cntrltype.attr,
3971 &dev_attr_dctype.attr,
3972 #ifdef CONFIG_NVME_AUTH
3973 &dev_attr_dhchap_secret.attr,
3974 &dev_attr_dhchap_ctrl_secret.attr,
3979 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3980 struct attribute *a, int n)
3982 struct device *dev = container_of(kobj, struct device, kobj);
3983 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3985 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3987 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3989 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3991 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3993 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3995 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3997 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3999 #ifdef CONFIG_NVME_AUTH
4000 if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
4002 if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
4009 const struct attribute_group nvme_dev_attrs_group = {
4010 .attrs = nvme_dev_attrs,
4011 .is_visible = nvme_dev_attrs_are_visible,
4013 EXPORT_SYMBOL_GPL(nvme_dev_attrs_group);
4015 static const struct attribute_group *nvme_dev_attr_groups[] = {
4016 &nvme_dev_attrs_group,
4020 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
4023 struct nvme_ns_head *h;
4025 lockdep_assert_held(&ctrl->subsys->lock);
4027 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
4029 * Private namespaces can share NSIDs under some conditions.
4030 * In that case we can't use the same ns_head for namespaces
4031 * with the same NSID.
4033 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
4035 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
4042 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
4043 struct nvme_ns_ids *ids)
4045 bool has_uuid = !uuid_is_null(&ids->uuid);
4046 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
4047 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
4048 struct nvme_ns_head *h;
4050 lockdep_assert_held(&subsys->lock);
4052 list_for_each_entry(h, &subsys->nsheads, entry) {
4053 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
4056 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
4059 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
4066 static void nvme_cdev_rel(struct device *dev)
4068 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
4071 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
4073 cdev_device_del(cdev, cdev_device);
4074 put_device(cdev_device);
4077 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
4078 const struct file_operations *fops, struct module *owner)
4082 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
4085 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
4086 cdev_device->class = nvme_ns_chr_class;
4087 cdev_device->release = nvme_cdev_rel;
4088 device_initialize(cdev_device);
4089 cdev_init(cdev, fops);
4090 cdev->owner = owner;
4091 ret = cdev_device_add(cdev, cdev_device);
4093 put_device(cdev_device);
4098 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
4100 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
4103 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
4105 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
4109 static const struct file_operations nvme_ns_chr_fops = {
4110 .owner = THIS_MODULE,
4111 .open = nvme_ns_chr_open,
4112 .release = nvme_ns_chr_release,
4113 .unlocked_ioctl = nvme_ns_chr_ioctl,
4114 .compat_ioctl = compat_ptr_ioctl,
4115 .uring_cmd = nvme_ns_chr_uring_cmd,
4116 .uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
4119 static int nvme_add_ns_cdev(struct nvme_ns *ns)
4123 ns->cdev_device.parent = ns->ctrl->device;
4124 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
4125 ns->ctrl->instance, ns->head->instance);
4129 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
4130 ns->ctrl->ops->module);
4133 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
4134 struct nvme_ns_info *info)
4136 struct nvme_ns_head *head;
4137 size_t size = sizeof(*head);
4140 #ifdef CONFIG_NVME_MULTIPATH
4141 size += num_possible_nodes() * sizeof(struct nvme_ns *);
4144 head = kzalloc(size, GFP_KERNEL);
4147 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4150 head->instance = ret;
4151 INIT_LIST_HEAD(&head->list);
4152 ret = init_srcu_struct(&head->srcu);
4154 goto out_ida_remove;
4155 head->subsys = ctrl->subsys;
4156 head->ns_id = info->nsid;
4157 head->ids = info->ids;
4158 head->shared = info->is_shared;
4159 kref_init(&head->ref);
4161 if (head->ids.csi) {
4162 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4164 goto out_cleanup_srcu;
4166 head->effects = ctrl->effects;
4168 ret = nvme_mpath_alloc_disk(ctrl, head);
4170 goto out_cleanup_srcu;
4172 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4174 kref_get(&ctrl->subsys->ref);
4178 cleanup_srcu_struct(&head->srcu);
4180 ida_free(&ctrl->subsys->ns_ida, head->instance);
4185 ret = blk_status_to_errno(nvme_error_status(ret));
4186 return ERR_PTR(ret);
4189 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4190 struct nvme_ns_ids *ids)
4192 struct nvme_subsystem *s;
4196 * Note that this check is racy as we try to avoid holding the global
4197 * lock over the whole ns_head creation. But it is only intended as
4198 * a sanity check anyway.
4200 mutex_lock(&nvme_subsystems_lock);
4201 list_for_each_entry(s, &nvme_subsystems, entry) {
4204 mutex_lock(&s->lock);
4205 ret = nvme_subsys_check_duplicate_ids(s, ids);
4206 mutex_unlock(&s->lock);
4210 mutex_unlock(&nvme_subsystems_lock);
4215 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4217 struct nvme_ctrl *ctrl = ns->ctrl;
4218 struct nvme_ns_head *head = NULL;
4221 ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4223 dev_err(ctrl->device,
4224 "globally duplicate IDs for nsid %d\n", info->nsid);
4225 nvme_print_device_info(ctrl);
4229 mutex_lock(&ctrl->subsys->lock);
4230 head = nvme_find_ns_head(ctrl, info->nsid);
4232 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4234 dev_err(ctrl->device,
4235 "duplicate IDs in subsystem for nsid %d\n",
4239 head = nvme_alloc_ns_head(ctrl, info);
4241 ret = PTR_ERR(head);
4246 if (!info->is_shared || !head->shared) {
4247 dev_err(ctrl->device,
4248 "Duplicate unshared namespace %d\n",
4250 goto out_put_ns_head;
4252 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4253 dev_err(ctrl->device,
4254 "IDs don't match for shared namespace %d\n",
4256 goto out_put_ns_head;
4259 if (!multipath && !list_empty(&head->list)) {
4260 dev_warn(ctrl->device,
4261 "Found shared namespace %d, but multipathing not supported.\n",
4263 dev_warn_once(ctrl->device,
4264 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4268 list_add_tail_rcu(&ns->siblings, &head->list);
4270 mutex_unlock(&ctrl->subsys->lock);
4274 nvme_put_ns_head(head);
4276 mutex_unlock(&ctrl->subsys->lock);
4280 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4282 struct nvme_ns *ns, *ret = NULL;
4284 down_read(&ctrl->namespaces_rwsem);
4285 list_for_each_entry(ns, &ctrl->namespaces, list) {
4286 if (ns->head->ns_id == nsid) {
4287 if (!nvme_get_ns(ns))
4292 if (ns->head->ns_id > nsid)
4295 up_read(&ctrl->namespaces_rwsem);
4298 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4301 * Add the namespace to the controller list while keeping the list ordered.
4303 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4305 struct nvme_ns *tmp;
4307 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4308 if (tmp->head->ns_id < ns->head->ns_id) {
4309 list_add(&ns->list, &tmp->list);
4313 list_add(&ns->list, &ns->ctrl->namespaces);
4316 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4319 struct gendisk *disk;
4320 int node = ctrl->numa_node;
4322 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4326 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4329 disk->fops = &nvme_bdev_ops;
4330 disk->private_data = ns;
4333 ns->queue = disk->queue;
4335 if (ctrl->opts && ctrl->opts->data_digest)
4336 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4338 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4339 if (ctrl->ops->supports_pci_p2pdma &&
4340 ctrl->ops->supports_pci_p2pdma(ctrl))
4341 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4344 kref_init(&ns->kref);
4346 if (nvme_init_ns_head(ns, info))
4347 goto out_cleanup_disk;
4350 * If multipathing is enabled, the device name for all disks and not
4351 * just those that represent shared namespaces needs to be based on the
4352 * subsystem instance. Using the controller instance for private
4353 * namespaces could lead to naming collisions between shared and private
4354 * namespaces if they don't use a common numbering scheme.
4356 * If multipathing is not enabled, disk names must use the controller
4357 * instance as shared namespaces will show up as multiple block
4360 if (ns->head->disk) {
4361 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4362 ctrl->instance, ns->head->instance);
4363 disk->flags |= GENHD_FL_HIDDEN;
4364 } else if (multipath) {
4365 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4366 ns->head->instance);
4368 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4369 ns->head->instance);
4372 if (nvme_update_ns_info(ns, info))
4375 down_write(&ctrl->namespaces_rwsem);
4376 nvme_ns_add_to_ctrl_list(ns);
4377 up_write(&ctrl->namespaces_rwsem);
4378 nvme_get_ctrl(ctrl);
4380 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4381 goto out_cleanup_ns_from_list;
4383 if (!nvme_ns_head_multipath(ns->head))
4384 nvme_add_ns_cdev(ns);
4386 nvme_mpath_add_disk(ns, info->anagrpid);
4387 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4391 out_cleanup_ns_from_list:
4392 nvme_put_ctrl(ctrl);
4393 down_write(&ctrl->namespaces_rwsem);
4394 list_del_init(&ns->list);
4395 up_write(&ctrl->namespaces_rwsem);
4397 mutex_lock(&ctrl->subsys->lock);
4398 list_del_rcu(&ns->siblings);
4399 if (list_empty(&ns->head->list))
4400 list_del_init(&ns->head->entry);
4401 mutex_unlock(&ctrl->subsys->lock);
4402 nvme_put_ns_head(ns->head);
4409 static void nvme_ns_remove(struct nvme_ns *ns)
4411 bool last_path = false;
4413 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4416 clear_bit(NVME_NS_READY, &ns->flags);
4417 set_capacity(ns->disk, 0);
4418 nvme_fault_inject_fini(&ns->fault_inject);
4421 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4422 * this ns going back into current_path.
4424 synchronize_srcu(&ns->head->srcu);
4426 /* wait for concurrent submissions */
4427 if (nvme_mpath_clear_current_path(ns))
4428 synchronize_srcu(&ns->head->srcu);
4430 mutex_lock(&ns->ctrl->subsys->lock);
4431 list_del_rcu(&ns->siblings);
4432 if (list_empty(&ns->head->list)) {
4433 list_del_init(&ns->head->entry);
4436 mutex_unlock(&ns->ctrl->subsys->lock);
4438 /* guarantee not available in head->list */
4439 synchronize_srcu(&ns->head->srcu);
4441 if (!nvme_ns_head_multipath(ns->head))
4442 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4443 del_gendisk(ns->disk);
4445 down_write(&ns->ctrl->namespaces_rwsem);
4446 list_del_init(&ns->list);
4447 up_write(&ns->ctrl->namespaces_rwsem);
4450 nvme_mpath_shutdown_disk(ns->head);
4454 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4456 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4464 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4466 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4468 if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4469 dev_err(ns->ctrl->device,
4470 "identifiers changed for nsid %d\n", ns->head->ns_id);
4474 ret = nvme_update_ns_info(ns, info);
4477 * Only remove the namespace if we got a fatal error back from the
4478 * device, otherwise ignore the error and just move on.
4480 * TODO: we should probably schedule a delayed retry here.
4482 if (ret > 0 && (ret & NVME_SC_DNR))
4486 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4488 struct nvme_ns_info info = { .nsid = nsid };
4492 if (nvme_identify_ns_descs(ctrl, &info))
4495 if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4496 dev_warn(ctrl->device,
4497 "command set not reported for nsid: %d\n", nsid);
4502 * If available try to use the Command Set Idependent Identify Namespace
4503 * data structure to find all the generic information that is needed to
4504 * set up a namespace. If not fall back to the legacy version.
4506 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4507 (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
4508 ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
4510 ret = nvme_ns_info_from_identify(ctrl, &info);
4512 if (info.is_removed)
4513 nvme_ns_remove_by_nsid(ctrl, nsid);
4516 * Ignore the namespace if it is not ready. We will get an AEN once it
4517 * becomes ready and restart the scan.
4519 if (ret || !info.is_ready)
4522 ns = nvme_find_get_ns(ctrl, nsid);
4524 nvme_validate_ns(ns, &info);
4527 nvme_alloc_ns(ctrl, &info);
4531 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4534 struct nvme_ns *ns, *next;
4537 down_write(&ctrl->namespaces_rwsem);
4538 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4539 if (ns->head->ns_id > nsid)
4540 list_move_tail(&ns->list, &rm_list);
4542 up_write(&ctrl->namespaces_rwsem);
4544 list_for_each_entry_safe(ns, next, &rm_list, list)
4549 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4551 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4556 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4561 struct nvme_command cmd = {
4562 .identify.opcode = nvme_admin_identify,
4563 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4564 .identify.nsid = cpu_to_le32(prev),
4567 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4568 NVME_IDENTIFY_DATA_SIZE);
4570 dev_warn(ctrl->device,
4571 "Identify NS List failed (status=0x%x)\n", ret);
4575 for (i = 0; i < nr_entries; i++) {
4576 u32 nsid = le32_to_cpu(ns_list[i]);
4578 if (!nsid) /* end of the list? */
4580 nvme_scan_ns(ctrl, nsid);
4581 while (++prev < nsid)
4582 nvme_ns_remove_by_nsid(ctrl, prev);
4586 nvme_remove_invalid_namespaces(ctrl, prev);
4592 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4594 struct nvme_id_ctrl *id;
4597 if (nvme_identify_ctrl(ctrl, &id))
4599 nn = le32_to_cpu(id->nn);
4602 for (i = 1; i <= nn; i++)
4603 nvme_scan_ns(ctrl, i);
4605 nvme_remove_invalid_namespaces(ctrl, nn);
4608 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4610 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4614 log = kzalloc(log_size, GFP_KERNEL);
4619 * We need to read the log to clear the AEN, but we don't want to rely
4620 * on it for the changed namespace information as userspace could have
4621 * raced with us in reading the log page, which could cause us to miss
4624 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4625 NVME_CSI_NVM, log, log_size, 0);
4627 dev_warn(ctrl->device,
4628 "reading changed ns log failed: %d\n", error);
4633 static void nvme_scan_work(struct work_struct *work)
4635 struct nvme_ctrl *ctrl =
4636 container_of(work, struct nvme_ctrl, scan_work);
4639 /* No tagset on a live ctrl means IO queues could not created */
4640 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4644 * Identify controller limits can change at controller reset due to
4645 * new firmware download, even though it is not common we cannot ignore
4646 * such scenario. Controller's non-mdts limits are reported in the unit
4647 * of logical blocks that is dependent on the format of attached
4648 * namespace. Hence re-read the limits at the time of ns allocation.
4650 ret = nvme_init_non_mdts_limits(ctrl);
4652 dev_warn(ctrl->device,
4653 "reading non-mdts-limits failed: %d\n", ret);
4657 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4658 dev_info(ctrl->device, "rescanning namespaces.\n");
4659 nvme_clear_changed_ns_log(ctrl);
4662 mutex_lock(&ctrl->scan_lock);
4663 if (nvme_ctrl_limited_cns(ctrl)) {
4664 nvme_scan_ns_sequential(ctrl);
4667 * Fall back to sequential scan if DNR is set to handle broken
4668 * devices which should support Identify NS List (as per the VS
4669 * they report) but don't actually support it.
4671 ret = nvme_scan_ns_list(ctrl);
4672 if (ret > 0 && ret & NVME_SC_DNR)
4673 nvme_scan_ns_sequential(ctrl);
4675 mutex_unlock(&ctrl->scan_lock);
4679 * This function iterates the namespace list unlocked to allow recovery from
4680 * controller failure. It is up to the caller to ensure the namespace list is
4681 * not modified by scan work while this function is executing.
4683 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4685 struct nvme_ns *ns, *next;
4689 * make sure to requeue I/O to all namespaces as these
4690 * might result from the scan itself and must complete
4691 * for the scan_work to make progress
4693 nvme_mpath_clear_ctrl_paths(ctrl);
4695 /* prevent racing with ns scanning */
4696 flush_work(&ctrl->scan_work);
4699 * The dead states indicates the controller was not gracefully
4700 * disconnected. In that case, we won't be able to flush any data while
4701 * removing the namespaces' disks; fail all the queues now to avoid
4702 * potentially having to clean up the failed sync later.
4704 if (ctrl->state == NVME_CTRL_DEAD) {
4705 nvme_mark_namespaces_dead(ctrl);
4706 nvme_unquiesce_io_queues(ctrl);
4709 /* this is a no-op when called from the controller reset handler */
4710 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4712 down_write(&ctrl->namespaces_rwsem);
4713 list_splice_init(&ctrl->namespaces, &ns_list);
4714 up_write(&ctrl->namespaces_rwsem);
4716 list_for_each_entry_safe(ns, next, &ns_list, list)
4719 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4721 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4723 const struct nvme_ctrl *ctrl =
4724 container_of(dev, struct nvme_ctrl, ctrl_device);
4725 struct nvmf_ctrl_options *opts = ctrl->opts;
4728 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4733 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4737 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4738 opts->trsvcid ?: "none");
4742 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4743 opts->host_traddr ?: "none");
4747 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4748 opts->host_iface ?: "none");
4753 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4755 char *envp[2] = { envdata, NULL };
4757 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4760 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4762 char *envp[2] = { NULL, NULL };
4763 u32 aen_result = ctrl->aen_result;
4765 ctrl->aen_result = 0;
4769 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4772 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4776 static void nvme_async_event_work(struct work_struct *work)
4778 struct nvme_ctrl *ctrl =
4779 container_of(work, struct nvme_ctrl, async_event_work);
4781 nvme_aen_uevent(ctrl);
4784 * The transport drivers must guarantee AER submission here is safe by
4785 * flushing ctrl async_event_work after changing the controller state
4786 * from LIVE and before freeing the admin queue.
4788 if (ctrl->state == NVME_CTRL_LIVE)
4789 ctrl->ops->submit_async_event(ctrl);
4792 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4797 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4803 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4806 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4808 struct nvme_fw_slot_info_log *log;
4810 log = kmalloc(sizeof(*log), GFP_KERNEL);
4814 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4815 log, sizeof(*log), 0))
4816 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4820 static void nvme_fw_act_work(struct work_struct *work)
4822 struct nvme_ctrl *ctrl = container_of(work,
4823 struct nvme_ctrl, fw_act_work);
4824 unsigned long fw_act_timeout;
4827 fw_act_timeout = jiffies +
4828 msecs_to_jiffies(ctrl->mtfa * 100);
4830 fw_act_timeout = jiffies +
4831 msecs_to_jiffies(admin_timeout * 1000);
4833 nvme_quiesce_io_queues(ctrl);
4834 while (nvme_ctrl_pp_status(ctrl)) {
4835 if (time_after(jiffies, fw_act_timeout)) {
4836 dev_warn(ctrl->device,
4837 "Fw activation timeout, reset controller\n");
4838 nvme_try_sched_reset(ctrl);
4844 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4847 nvme_unquiesce_io_queues(ctrl);
4848 /* read FW slot information to clear the AER */
4849 nvme_get_fw_slot_info(ctrl);
4851 queue_work(nvme_wq, &ctrl->async_event_work);
4854 static u32 nvme_aer_type(u32 result)
4856 return result & 0x7;
4859 static u32 nvme_aer_subtype(u32 result)
4861 return (result & 0xff00) >> 8;
4864 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4866 u32 aer_notice_type = nvme_aer_subtype(result);
4867 bool requeue = true;
4869 switch (aer_notice_type) {
4870 case NVME_AER_NOTICE_NS_CHANGED:
4871 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4872 nvme_queue_scan(ctrl);
4874 case NVME_AER_NOTICE_FW_ACT_STARTING:
4876 * We are (ab)using the RESETTING state to prevent subsequent
4877 * recovery actions from interfering with the controller's
4878 * firmware activation.
4880 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4881 nvme_auth_stop(ctrl);
4883 queue_work(nvme_wq, &ctrl->fw_act_work);
4886 #ifdef CONFIG_NVME_MULTIPATH
4887 case NVME_AER_NOTICE_ANA:
4888 if (!ctrl->ana_log_buf)
4890 queue_work(nvme_wq, &ctrl->ana_work);
4893 case NVME_AER_NOTICE_DISC_CHANGED:
4894 ctrl->aen_result = result;
4897 dev_warn(ctrl->device, "async event result %08x\n", result);
4902 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4904 dev_warn(ctrl->device, "resetting controller due to AER\n");
4905 nvme_reset_ctrl(ctrl);
4908 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4909 volatile union nvme_result *res)
4911 u32 result = le32_to_cpu(res->u32);
4912 u32 aer_type = nvme_aer_type(result);
4913 u32 aer_subtype = nvme_aer_subtype(result);
4914 bool requeue = true;
4916 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4919 trace_nvme_async_event(ctrl, result);
4921 case NVME_AER_NOTICE:
4922 requeue = nvme_handle_aen_notice(ctrl, result);
4924 case NVME_AER_ERROR:
4926 * For a persistent internal error, don't run async_event_work
4927 * to submit a new AER. The controller reset will do it.
4929 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4930 nvme_handle_aer_persistent_error(ctrl);
4934 case NVME_AER_SMART:
4937 ctrl->aen_result = result;
4944 queue_work(nvme_wq, &ctrl->async_event_work);
4946 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4948 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4949 const struct blk_mq_ops *ops, unsigned int cmd_size)
4953 memset(set, 0, sizeof(*set));
4955 set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4956 if (ctrl->ops->flags & NVME_F_FABRICS)
4957 set->reserved_tags = NVMF_RESERVED_TAGS;
4958 set->numa_node = ctrl->numa_node;
4959 set->flags = BLK_MQ_F_NO_SCHED;
4960 if (ctrl->ops->flags & NVME_F_BLOCKING)
4961 set->flags |= BLK_MQ_F_BLOCKING;
4962 set->cmd_size = cmd_size;
4963 set->driver_data = ctrl;
4964 set->nr_hw_queues = 1;
4965 set->timeout = NVME_ADMIN_TIMEOUT;
4966 ret = blk_mq_alloc_tag_set(set);
4970 ctrl->admin_q = blk_mq_init_queue(set);
4971 if (IS_ERR(ctrl->admin_q)) {
4972 ret = PTR_ERR(ctrl->admin_q);
4973 goto out_free_tagset;
4976 if (ctrl->ops->flags & NVME_F_FABRICS) {
4977 ctrl->fabrics_q = blk_mq_init_queue(set);
4978 if (IS_ERR(ctrl->fabrics_q)) {
4979 ret = PTR_ERR(ctrl->fabrics_q);
4980 goto out_cleanup_admin_q;
4984 ctrl->admin_tagset = set;
4987 out_cleanup_admin_q:
4988 blk_mq_destroy_queue(ctrl->admin_q);
4989 blk_put_queue(ctrl->admin_q);
4991 blk_mq_free_tag_set(set);
4992 ctrl->admin_q = NULL;
4993 ctrl->fabrics_q = NULL;
4996 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4998 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
5000 blk_mq_destroy_queue(ctrl->admin_q);
5001 blk_put_queue(ctrl->admin_q);
5002 if (ctrl->ops->flags & NVME_F_FABRICS) {
5003 blk_mq_destroy_queue(ctrl->fabrics_q);
5004 blk_put_queue(ctrl->fabrics_q);
5006 blk_mq_free_tag_set(ctrl->admin_tagset);
5008 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
5010 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
5011 const struct blk_mq_ops *ops, unsigned int nr_maps,
5012 unsigned int cmd_size)
5016 memset(set, 0, sizeof(*set));
5018 set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
5020 * Some Apple controllers requires tags to be unique across admin and
5021 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
5023 if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
5024 set->reserved_tags = NVME_AQ_DEPTH;
5025 else if (ctrl->ops->flags & NVME_F_FABRICS)
5026 set->reserved_tags = NVMF_RESERVED_TAGS;
5027 set->numa_node = ctrl->numa_node;
5028 set->flags = BLK_MQ_F_SHOULD_MERGE;
5029 if (ctrl->ops->flags & NVME_F_BLOCKING)
5030 set->flags |= BLK_MQ_F_BLOCKING;
5031 set->cmd_size = cmd_size,
5032 set->driver_data = ctrl;
5033 set->nr_hw_queues = ctrl->queue_count - 1;
5034 set->timeout = NVME_IO_TIMEOUT;
5035 set->nr_maps = nr_maps;
5036 ret = blk_mq_alloc_tag_set(set);
5040 if (ctrl->ops->flags & NVME_F_FABRICS) {
5041 ctrl->connect_q = blk_mq_init_queue(set);
5042 if (IS_ERR(ctrl->connect_q)) {
5043 ret = PTR_ERR(ctrl->connect_q);
5044 goto out_free_tag_set;
5046 blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
5054 blk_mq_free_tag_set(set);
5055 ctrl->connect_q = NULL;
5058 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
5060 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
5062 if (ctrl->ops->flags & NVME_F_FABRICS) {
5063 blk_mq_destroy_queue(ctrl->connect_q);
5064 blk_put_queue(ctrl->connect_q);
5066 blk_mq_free_tag_set(ctrl->tagset);
5068 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
5070 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
5072 nvme_mpath_stop(ctrl);
5073 nvme_auth_stop(ctrl);
5074 nvme_stop_keep_alive(ctrl);
5075 nvme_stop_failfast_work(ctrl);
5076 flush_work(&ctrl->async_event_work);
5077 cancel_work_sync(&ctrl->fw_act_work);
5078 if (ctrl->ops->stop_ctrl)
5079 ctrl->ops->stop_ctrl(ctrl);
5081 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
5083 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
5085 nvme_start_keep_alive(ctrl);
5087 nvme_enable_aen(ctrl);
5090 * persistent discovery controllers need to send indication to userspace
5091 * to re-read the discovery log page to learn about possible changes
5092 * that were missed. We identify persistent discovery controllers by
5093 * checking that they started once before, hence are reconnecting back.
5095 if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
5096 nvme_discovery_ctrl(ctrl))
5097 nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
5099 if (ctrl->queue_count > 1) {
5100 nvme_queue_scan(ctrl);
5101 nvme_unquiesce_io_queues(ctrl);
5102 nvme_mpath_update(ctrl);
5105 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
5106 set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
5108 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
5110 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
5112 nvme_hwmon_exit(ctrl);
5113 nvme_fault_inject_fini(&ctrl->fault_inject);
5114 dev_pm_qos_hide_latency_tolerance(ctrl->device);
5115 cdev_device_del(&ctrl->cdev, ctrl->device);
5116 nvme_put_ctrl(ctrl);
5118 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
5120 static void nvme_free_cels(struct nvme_ctrl *ctrl)
5122 struct nvme_effects_log *cel;
5125 xa_for_each(&ctrl->cels, i, cel) {
5126 xa_erase(&ctrl->cels, i);
5130 xa_destroy(&ctrl->cels);
5133 static void nvme_free_ctrl(struct device *dev)
5135 struct nvme_ctrl *ctrl =
5136 container_of(dev, struct nvme_ctrl, ctrl_device);
5137 struct nvme_subsystem *subsys = ctrl->subsys;
5139 if (!subsys || ctrl->instance != subsys->instance)
5140 ida_free(&nvme_instance_ida, ctrl->instance);
5142 nvme_free_cels(ctrl);
5143 nvme_mpath_uninit(ctrl);
5144 nvme_auth_stop(ctrl);
5145 nvme_auth_free(ctrl);
5146 __free_page(ctrl->discard_page);
5147 free_opal_dev(ctrl->opal_dev);
5150 mutex_lock(&nvme_subsystems_lock);
5151 list_del(&ctrl->subsys_entry);
5152 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
5153 mutex_unlock(&nvme_subsystems_lock);
5156 ctrl->ops->free_ctrl(ctrl);
5159 nvme_put_subsystem(subsys);
5163 * Initialize a NVMe controller structures. This needs to be called during
5164 * earliest initialization so that we have the initialized structured around
5167 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5168 const struct nvme_ctrl_ops *ops, unsigned long quirks)
5172 ctrl->state = NVME_CTRL_NEW;
5173 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5174 spin_lock_init(&ctrl->lock);
5175 mutex_init(&ctrl->scan_lock);
5176 INIT_LIST_HEAD(&ctrl->namespaces);
5177 xa_init(&ctrl->cels);
5178 init_rwsem(&ctrl->namespaces_rwsem);
5181 ctrl->quirks = quirks;
5182 ctrl->numa_node = NUMA_NO_NODE;
5183 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5184 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5185 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5186 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5187 init_waitqueue_head(&ctrl->state_wq);
5189 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5190 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5191 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5192 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5194 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5196 ctrl->discard_page = alloc_page(GFP_KERNEL);
5197 if (!ctrl->discard_page) {
5202 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5205 ctrl->instance = ret;
5207 device_initialize(&ctrl->ctrl_device);
5208 ctrl->device = &ctrl->ctrl_device;
5209 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5211 ctrl->device->class = nvme_class;
5212 ctrl->device->parent = ctrl->dev;
5213 if (ops->dev_attr_groups)
5214 ctrl->device->groups = ops->dev_attr_groups;
5216 ctrl->device->groups = nvme_dev_attr_groups;
5217 ctrl->device->release = nvme_free_ctrl;
5218 dev_set_drvdata(ctrl->device, ctrl);
5219 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5221 goto out_release_instance;
5223 nvme_get_ctrl(ctrl);
5224 cdev_init(&ctrl->cdev, &nvme_dev_fops);
5225 ctrl->cdev.owner = ops->module;
5226 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5231 * Initialize latency tolerance controls. The sysfs files won't
5232 * be visible to userspace unless the device actually supports APST.
5234 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5235 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5236 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5238 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5239 nvme_mpath_init_ctrl(ctrl);
5240 ret = nvme_auth_init_ctrl(ctrl);
5246 cdev_device_del(&ctrl->cdev, ctrl->device);
5248 nvme_put_ctrl(ctrl);
5249 kfree_const(ctrl->device->kobj.name);
5250 out_release_instance:
5251 ida_free(&nvme_instance_ida, ctrl->instance);
5253 if (ctrl->discard_page)
5254 __free_page(ctrl->discard_page);
5257 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5259 /* let I/O to all namespaces fail in preparation for surprise removal */
5260 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
5264 down_read(&ctrl->namespaces_rwsem);
5265 list_for_each_entry(ns, &ctrl->namespaces, list)
5266 blk_mark_disk_dead(ns->disk);
5267 up_read(&ctrl->namespaces_rwsem);
5269 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
5271 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5275 down_read(&ctrl->namespaces_rwsem);
5276 list_for_each_entry(ns, &ctrl->namespaces, list)
5277 blk_mq_unfreeze_queue(ns->queue);
5278 up_read(&ctrl->namespaces_rwsem);
5280 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5282 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5286 down_read(&ctrl->namespaces_rwsem);
5287 list_for_each_entry(ns, &ctrl->namespaces, list) {
5288 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5292 up_read(&ctrl->namespaces_rwsem);
5295 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5297 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5301 down_read(&ctrl->namespaces_rwsem);
5302 list_for_each_entry(ns, &ctrl->namespaces, list)
5303 blk_mq_freeze_queue_wait(ns->queue);
5304 up_read(&ctrl->namespaces_rwsem);
5306 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5308 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5312 down_read(&ctrl->namespaces_rwsem);
5313 list_for_each_entry(ns, &ctrl->namespaces, list)
5314 blk_freeze_queue_start(ns->queue);
5315 up_read(&ctrl->namespaces_rwsem);
5317 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5319 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
5323 if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5324 blk_mq_quiesce_tagset(ctrl->tagset);
5326 blk_mq_wait_quiesce_done(ctrl->tagset);
5328 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
5330 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
5334 if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5335 blk_mq_unquiesce_tagset(ctrl->tagset);
5337 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
5339 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
5341 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5342 blk_mq_quiesce_queue(ctrl->admin_q);
5344 blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
5346 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
5348 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
5350 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5351 blk_mq_unquiesce_queue(ctrl->admin_q);
5353 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
5355 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5359 down_read(&ctrl->namespaces_rwsem);
5360 list_for_each_entry(ns, &ctrl->namespaces, list)
5361 blk_sync_queue(ns->queue);
5362 up_read(&ctrl->namespaces_rwsem);
5364 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5366 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5368 nvme_sync_io_queues(ctrl);
5370 blk_sync_queue(ctrl->admin_q);
5372 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5374 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5376 if (file->f_op != &nvme_dev_fops)
5378 return file->private_data;
5380 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5383 * Check we didn't inadvertently grow the command structure sizes:
5385 static inline void _nvme_check_size(void)
5387 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5388 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5389 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5390 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5391 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5392 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5393 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5394 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5395 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5396 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5397 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5398 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5399 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5400 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5401 NVME_IDENTIFY_DATA_SIZE);
5402 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5403 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5404 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5405 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5406 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5407 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5408 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5409 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5410 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5414 static int __init nvme_core_init(void)
5416 int result = -ENOMEM;
5420 nvme_wq = alloc_workqueue("nvme-wq",
5421 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5425 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5426 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5430 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5431 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5432 if (!nvme_delete_wq)
5433 goto destroy_reset_wq;
5435 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5436 NVME_MINORS, "nvme");
5438 goto destroy_delete_wq;
5440 nvme_class = class_create("nvme");
5441 if (IS_ERR(nvme_class)) {
5442 result = PTR_ERR(nvme_class);
5443 goto unregister_chrdev;
5445 nvme_class->dev_uevent = nvme_class_uevent;
5447 nvme_subsys_class = class_create("nvme-subsystem");
5448 if (IS_ERR(nvme_subsys_class)) {
5449 result = PTR_ERR(nvme_subsys_class);
5453 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5456 goto destroy_subsys_class;
5458 nvme_ns_chr_class = class_create("nvme-generic");
5459 if (IS_ERR(nvme_ns_chr_class)) {
5460 result = PTR_ERR(nvme_ns_chr_class);
5461 goto unregister_generic_ns;
5464 result = nvme_init_auth();
5466 goto destroy_ns_chr;
5470 class_destroy(nvme_ns_chr_class);
5471 unregister_generic_ns:
5472 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5473 destroy_subsys_class:
5474 class_destroy(nvme_subsys_class);
5476 class_destroy(nvme_class);
5478 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5480 destroy_workqueue(nvme_delete_wq);
5482 destroy_workqueue(nvme_reset_wq);
5484 destroy_workqueue(nvme_wq);
5489 static void __exit nvme_core_exit(void)
5492 class_destroy(nvme_ns_chr_class);
5493 class_destroy(nvme_subsys_class);
5494 class_destroy(nvme_class);
5495 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5496 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5497 destroy_workqueue(nvme_delete_wq);
5498 destroy_workqueue(nvme_reset_wq);
5499 destroy_workqueue(nvme_wq);
5500 ida_destroy(&nvme_ns_chr_minor_ida);
5501 ida_destroy(&nvme_instance_ida);
5504 MODULE_LICENSE("GPL");
5505 MODULE_VERSION("1.0");
5506 module_init(nvme_core_init);
5507 module_exit(nvme_core_exit);