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 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_RESV_CONFLICT;
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 if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1139 dev_info(ctrl->device,
1140 "controller capabilities changed, reset may be required to take effect.\n");
1143 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1144 nvme_queue_scan(ctrl);
1145 flush_work(&ctrl->scan_work);
1150 switch (cmd->common.opcode) {
1151 case nvme_admin_set_features:
1152 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1153 case NVME_FEAT_KATO:
1155 * Keep alive commands interval on the host should be
1156 * updated when KATO is modified by Set Features
1160 nvme_update_keep_alive(ctrl, cmd);
1170 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1173 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1175 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1176 * accounting for transport roundtrip times [..].
1178 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1180 unsigned long delay = ctrl->kato * HZ / 2;
1183 * When using Traffic Based Keep Alive, we need to run
1184 * nvme_keep_alive_work at twice the normal frequency, as one
1185 * command completion can postpone sending a keep alive command
1186 * by up to twice the delay between runs.
1188 if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1193 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1195 queue_delayed_work(nvme_wq, &ctrl->ka_work,
1196 nvme_keep_alive_work_period(ctrl));
1199 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1200 blk_status_t status)
1202 struct nvme_ctrl *ctrl = rq->end_io_data;
1203 unsigned long flags;
1204 bool startka = false;
1205 unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1206 unsigned long delay = nvme_keep_alive_work_period(ctrl);
1209 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1210 * at the desired frequency.
1215 dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1216 jiffies_to_msecs(rtt));
1220 blk_mq_free_request(rq);
1223 dev_err(ctrl->device,
1224 "failed nvme_keep_alive_end_io error=%d\n",
1226 return RQ_END_IO_NONE;
1229 ctrl->ka_last_check_time = jiffies;
1230 ctrl->comp_seen = false;
1231 spin_lock_irqsave(&ctrl->lock, flags);
1232 if (ctrl->state == NVME_CTRL_LIVE ||
1233 ctrl->state == NVME_CTRL_CONNECTING)
1235 spin_unlock_irqrestore(&ctrl->lock, flags);
1237 queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1238 return RQ_END_IO_NONE;
1241 static void nvme_keep_alive_work(struct work_struct *work)
1243 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1244 struct nvme_ctrl, ka_work);
1245 bool comp_seen = ctrl->comp_seen;
1248 ctrl->ka_last_check_time = jiffies;
1250 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1251 dev_dbg(ctrl->device,
1252 "reschedule traffic based keep-alive timer\n");
1253 ctrl->comp_seen = false;
1254 nvme_queue_keep_alive_work(ctrl);
1258 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1259 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1261 /* allocation failure, reset the controller */
1262 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1263 nvme_reset_ctrl(ctrl);
1266 nvme_init_request(rq, &ctrl->ka_cmd);
1268 rq->timeout = ctrl->kato * HZ;
1269 rq->end_io = nvme_keep_alive_end_io;
1270 rq->end_io_data = ctrl;
1271 blk_execute_rq_nowait(rq, false);
1274 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1276 if (unlikely(ctrl->kato == 0))
1279 nvme_queue_keep_alive_work(ctrl);
1282 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1284 if (unlikely(ctrl->kato == 0))
1287 cancel_delayed_work_sync(&ctrl->ka_work);
1289 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1291 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1292 struct nvme_command *cmd)
1294 unsigned int new_kato =
1295 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1297 dev_info(ctrl->device,
1298 "keep alive interval updated from %u ms to %u ms\n",
1299 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1301 nvme_stop_keep_alive(ctrl);
1302 ctrl->kato = new_kato;
1303 nvme_start_keep_alive(ctrl);
1307 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1308 * flag, thus sending any new CNS opcodes has a big chance of not working.
1309 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1310 * (but not for any later version).
1312 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1314 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1315 return ctrl->vs < NVME_VS(1, 2, 0);
1316 return ctrl->vs < NVME_VS(1, 1, 0);
1319 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1321 struct nvme_command c = { };
1324 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1325 c.identify.opcode = nvme_admin_identify;
1326 c.identify.cns = NVME_ID_CNS_CTRL;
1328 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1332 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1333 sizeof(struct nvme_id_ctrl));
1339 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1340 struct nvme_ns_id_desc *cur, bool *csi_seen)
1342 const char *warn_str = "ctrl returned bogus length:";
1345 switch (cur->nidt) {
1346 case NVME_NIDT_EUI64:
1347 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1348 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1349 warn_str, cur->nidl);
1352 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1353 return NVME_NIDT_EUI64_LEN;
1354 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1355 return NVME_NIDT_EUI64_LEN;
1356 case NVME_NIDT_NGUID:
1357 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1358 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1359 warn_str, cur->nidl);
1362 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1363 return NVME_NIDT_NGUID_LEN;
1364 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1365 return NVME_NIDT_NGUID_LEN;
1366 case NVME_NIDT_UUID:
1367 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1368 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1369 warn_str, cur->nidl);
1372 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1373 return NVME_NIDT_UUID_LEN;
1374 uuid_copy(&ids->uuid, data + sizeof(*cur));
1375 return NVME_NIDT_UUID_LEN;
1377 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1378 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1379 warn_str, cur->nidl);
1382 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1384 return NVME_NIDT_CSI_LEN;
1386 /* Skip unknown types */
1391 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1392 struct nvme_ns_info *info)
1394 struct nvme_command c = { };
1395 bool csi_seen = false;
1396 int status, pos, len;
1399 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1401 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1404 c.identify.opcode = nvme_admin_identify;
1405 c.identify.nsid = cpu_to_le32(info->nsid);
1406 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1408 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1412 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1413 NVME_IDENTIFY_DATA_SIZE);
1415 dev_warn(ctrl->device,
1416 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1417 info->nsid, status);
1421 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1422 struct nvme_ns_id_desc *cur = data + pos;
1427 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1431 len += sizeof(*cur);
1434 if (nvme_multi_css(ctrl) && !csi_seen) {
1435 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1445 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1446 struct nvme_id_ns **id)
1448 struct nvme_command c = { };
1451 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1452 c.identify.opcode = nvme_admin_identify;
1453 c.identify.nsid = cpu_to_le32(nsid);
1454 c.identify.cns = NVME_ID_CNS_NS;
1456 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1460 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1462 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1468 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1469 struct nvme_ns_info *info)
1471 struct nvme_ns_ids *ids = &info->ids;
1472 struct nvme_id_ns *id;
1475 ret = nvme_identify_ns(ctrl, info->nsid, &id);
1479 if (id->ncap == 0) {
1480 /* namespace not allocated or attached */
1481 info->is_removed = true;
1485 info->anagrpid = id->anagrpid;
1486 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1487 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1488 info->is_ready = true;
1489 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1490 dev_info(ctrl->device,
1491 "Ignoring bogus Namespace Identifiers\n");
1493 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1494 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1495 memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1496 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1497 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1498 memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1504 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1505 struct nvme_ns_info *info)
1507 struct nvme_id_ns_cs_indep *id;
1508 struct nvme_command c = {
1509 .identify.opcode = nvme_admin_identify,
1510 .identify.nsid = cpu_to_le32(info->nsid),
1511 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1515 id = kmalloc(sizeof(*id), GFP_KERNEL);
1519 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1521 info->anagrpid = id->anagrpid;
1522 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1523 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1524 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1530 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1531 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1533 union nvme_result res = { 0 };
1534 struct nvme_command c = { };
1537 c.features.opcode = op;
1538 c.features.fid = cpu_to_le32(fid);
1539 c.features.dword11 = cpu_to_le32(dword11);
1541 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1542 buffer, buflen, NVME_QID_ANY, 0, 0);
1543 if (ret >= 0 && result)
1544 *result = le32_to_cpu(res.u32);
1548 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1549 unsigned int dword11, void *buffer, size_t buflen,
1552 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1555 EXPORT_SYMBOL_GPL(nvme_set_features);
1557 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1558 unsigned int dword11, void *buffer, size_t buflen,
1561 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1564 EXPORT_SYMBOL_GPL(nvme_get_features);
1566 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1568 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1570 int status, nr_io_queues;
1572 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1578 * Degraded controllers might return an error when setting the queue
1579 * count. We still want to be able to bring them online and offer
1580 * access to the admin queue, as that might be only way to fix them up.
1583 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1586 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1587 *count = min(*count, nr_io_queues);
1592 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1594 #define NVME_AEN_SUPPORTED \
1595 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1596 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1598 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1600 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1603 if (!supported_aens)
1606 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1609 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1612 queue_work(nvme_wq, &ctrl->async_event_work);
1615 static int nvme_ns_open(struct nvme_ns *ns)
1618 /* should never be called due to GENHD_FL_HIDDEN */
1619 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1621 if (!nvme_get_ns(ns))
1623 if (!try_module_get(ns->ctrl->ops->module))
1634 static void nvme_ns_release(struct nvme_ns *ns)
1637 module_put(ns->ctrl->ops->module);
1641 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1643 return nvme_ns_open(disk->private_data);
1646 static void nvme_release(struct gendisk *disk)
1648 nvme_ns_release(disk->private_data);
1651 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1653 /* some standard values */
1654 geo->heads = 1 << 6;
1655 geo->sectors = 1 << 5;
1656 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1660 #ifdef CONFIG_BLK_DEV_INTEGRITY
1661 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1662 u32 max_integrity_segments)
1664 struct blk_integrity integrity = { };
1666 switch (ns->pi_type) {
1667 case NVME_NS_DPS_PI_TYPE3:
1668 switch (ns->guard_type) {
1669 case NVME_NVM_NS_16B_GUARD:
1670 integrity.profile = &t10_pi_type3_crc;
1671 integrity.tag_size = sizeof(u16) + sizeof(u32);
1672 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1674 case NVME_NVM_NS_64B_GUARD:
1675 integrity.profile = &ext_pi_type3_crc64;
1676 integrity.tag_size = sizeof(u16) + 6;
1677 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1680 integrity.profile = NULL;
1684 case NVME_NS_DPS_PI_TYPE1:
1685 case NVME_NS_DPS_PI_TYPE2:
1686 switch (ns->guard_type) {
1687 case NVME_NVM_NS_16B_GUARD:
1688 integrity.profile = &t10_pi_type1_crc;
1689 integrity.tag_size = sizeof(u16);
1690 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1692 case NVME_NVM_NS_64B_GUARD:
1693 integrity.profile = &ext_pi_type1_crc64;
1694 integrity.tag_size = sizeof(u16);
1695 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1698 integrity.profile = NULL;
1703 integrity.profile = NULL;
1707 integrity.tuple_size = ns->ms;
1708 blk_integrity_register(disk, &integrity);
1709 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1712 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1713 u32 max_integrity_segments)
1716 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1718 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1720 struct nvme_ctrl *ctrl = ns->ctrl;
1721 struct request_queue *queue = disk->queue;
1722 u32 size = queue_logical_block_size(queue);
1724 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1725 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1727 if (ctrl->max_discard_sectors == 0) {
1728 blk_queue_max_discard_sectors(queue, 0);
1732 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1733 NVME_DSM_MAX_RANGES);
1735 queue->limits.discard_granularity = size;
1737 /* If discard is already enabled, don't reset queue limits */
1738 if (queue->limits.max_discard_sectors)
1741 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1742 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1744 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1745 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1748 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1750 return uuid_equal(&a->uuid, &b->uuid) &&
1751 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1752 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1756 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1758 bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1759 unsigned lbaf = nvme_lbaf_index(id->flbas);
1760 struct nvme_ctrl *ctrl = ns->ctrl;
1761 struct nvme_command c = { };
1762 struct nvme_id_ns_nvm *nvm;
1767 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1768 if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1769 ns->pi_size = sizeof(struct t10_pi_tuple);
1770 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1774 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1778 c.identify.opcode = nvme_admin_identify;
1779 c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1780 c.identify.cns = NVME_ID_CNS_CS_NS;
1781 c.identify.csi = NVME_CSI_NVM;
1783 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1787 elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1789 /* no support for storage tag formats right now */
1790 if (nvme_elbaf_sts(elbaf))
1793 ns->guard_type = nvme_elbaf_guard_type(elbaf);
1794 switch (ns->guard_type) {
1795 case NVME_NVM_NS_64B_GUARD:
1796 ns->pi_size = sizeof(struct crc64_pi_tuple);
1798 case NVME_NVM_NS_16B_GUARD:
1799 ns->pi_size = sizeof(struct t10_pi_tuple);
1808 if (ns->pi_size && (first || ns->ms == ns->pi_size))
1809 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1816 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1818 struct nvme_ctrl *ctrl = ns->ctrl;
1821 ret = nvme_init_ms(ns, id);
1825 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1826 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1829 if (ctrl->ops->flags & NVME_F_FABRICS) {
1831 * The NVMe over Fabrics specification only supports metadata as
1832 * part of the extended data LBA. We rely on HCA/HBA support to
1833 * remap the separate metadata buffer from the block layer.
1835 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1838 ns->features |= NVME_NS_EXT_LBAS;
1841 * The current fabrics transport drivers support namespace
1842 * metadata formats only if nvme_ns_has_pi() returns true.
1843 * Suppress support for all other formats so the namespace will
1844 * have a 0 capacity and not be usable through the block stack.
1846 * Note, this check will need to be modified if any drivers
1847 * gain the ability to use other metadata formats.
1849 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1850 ns->features |= NVME_NS_METADATA_SUPPORTED;
1853 * For PCIe controllers, we can't easily remap the separate
1854 * metadata buffer from the block layer and thus require a
1855 * separate metadata buffer for block layer metadata/PI support.
1856 * We allow extended LBAs for the passthrough interface, though.
1858 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1859 ns->features |= NVME_NS_EXT_LBAS;
1861 ns->features |= NVME_NS_METADATA_SUPPORTED;
1866 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1867 struct request_queue *q)
1869 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1871 if (ctrl->max_hw_sectors) {
1873 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1875 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1876 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1877 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1879 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1880 blk_queue_dma_alignment(q, 3);
1881 blk_queue_write_cache(q, vwc, vwc);
1884 static void nvme_update_disk_info(struct gendisk *disk,
1885 struct nvme_ns *ns, struct nvme_id_ns *id)
1887 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1888 u32 bs = 1U << ns->lba_shift;
1889 u32 atomic_bs, phys_bs, io_opt = 0;
1892 * The block layer can't support LBA sizes larger than the page size
1893 * yet, so catch this early and don't allow block I/O.
1895 if (ns->lba_shift > PAGE_SHIFT) {
1900 blk_integrity_unregister(disk);
1902 atomic_bs = phys_bs = bs;
1903 if (id->nabo == 0) {
1905 * Bit 1 indicates whether NAWUPF is defined for this namespace
1906 * and whether it should be used instead of AWUPF. If NAWUPF ==
1907 * 0 then AWUPF must be used instead.
1909 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1910 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1912 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1915 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1916 /* NPWG = Namespace Preferred Write Granularity */
1917 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1918 /* NOWS = Namespace Optimal Write Size */
1919 io_opt = bs * (1 + le16_to_cpu(id->nows));
1922 blk_queue_logical_block_size(disk->queue, bs);
1924 * Linux filesystems assume writing a single physical block is
1925 * an atomic operation. Hence limit the physical block size to the
1926 * value of the Atomic Write Unit Power Fail parameter.
1928 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1929 blk_queue_io_min(disk->queue, phys_bs);
1930 blk_queue_io_opt(disk->queue, io_opt);
1933 * Register a metadata profile for PI, or the plain non-integrity NVMe
1934 * metadata masquerading as Type 0 if supported, otherwise reject block
1935 * I/O to namespaces with metadata except when the namespace supports
1936 * PI, as it can strip/insert in that case.
1939 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1940 (ns->features & NVME_NS_METADATA_SUPPORTED))
1941 nvme_init_integrity(disk, ns,
1942 ns->ctrl->max_integrity_segments);
1943 else if (!nvme_ns_has_pi(ns))
1947 set_capacity_and_notify(disk, capacity);
1949 nvme_config_discard(disk, ns);
1950 blk_queue_max_write_zeroes_sectors(disk->queue,
1951 ns->ctrl->max_zeroes_sectors);
1954 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1956 return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1959 static inline bool nvme_first_scan(struct gendisk *disk)
1961 /* nvme_alloc_ns() scans the disk prior to adding it */
1962 return !disk_live(disk);
1965 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1967 struct nvme_ctrl *ctrl = ns->ctrl;
1970 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1971 is_power_of_2(ctrl->max_hw_sectors))
1972 iob = ctrl->max_hw_sectors;
1974 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1979 if (!is_power_of_2(iob)) {
1980 if (nvme_first_scan(ns->disk))
1981 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1982 ns->disk->disk_name, iob);
1986 if (blk_queue_is_zoned(ns->disk->queue)) {
1987 if (nvme_first_scan(ns->disk))
1988 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1989 ns->disk->disk_name);
1993 blk_queue_chunk_sectors(ns->queue, iob);
1996 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1997 struct nvme_ns_info *info)
1999 blk_mq_freeze_queue(ns->disk->queue);
2000 nvme_set_queue_limits(ns->ctrl, ns->queue);
2001 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2002 blk_mq_unfreeze_queue(ns->disk->queue);
2004 if (nvme_ns_head_multipath(ns->head)) {
2005 blk_mq_freeze_queue(ns->head->disk->queue);
2006 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2007 nvme_mpath_revalidate_paths(ns);
2008 blk_stack_limits(&ns->head->disk->queue->limits,
2009 &ns->queue->limits, 0);
2010 ns->head->disk->flags |= GENHD_FL_HIDDEN;
2011 blk_mq_unfreeze_queue(ns->head->disk->queue);
2014 /* Hide the block-interface for these devices */
2015 ns->disk->flags |= GENHD_FL_HIDDEN;
2016 set_bit(NVME_NS_READY, &ns->flags);
2021 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2022 struct nvme_ns_info *info)
2024 struct nvme_id_ns *id;
2028 ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2032 if (id->ncap == 0) {
2033 /* namespace not allocated or attached */
2034 info->is_removed = true;
2039 blk_mq_freeze_queue(ns->disk->queue);
2040 lbaf = nvme_lbaf_index(id->flbas);
2041 ns->lba_shift = id->lbaf[lbaf].ds;
2042 nvme_set_queue_limits(ns->ctrl, ns->queue);
2044 ret = nvme_configure_metadata(ns, id);
2046 blk_mq_unfreeze_queue(ns->disk->queue);
2049 nvme_set_chunk_sectors(ns, id);
2050 nvme_update_disk_info(ns->disk, ns, id);
2052 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2053 ret = nvme_update_zone_info(ns, lbaf);
2055 blk_mq_unfreeze_queue(ns->disk->queue);
2061 * Only set the DEAC bit if the device guarantees that reads from
2062 * deallocated data return zeroes. While the DEAC bit does not
2063 * require that, it must be a no-op if reads from deallocated data
2064 * do not return zeroes.
2066 if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2067 ns->features |= NVME_NS_DEAC;
2068 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2069 set_bit(NVME_NS_READY, &ns->flags);
2070 blk_mq_unfreeze_queue(ns->disk->queue);
2072 if (blk_queue_is_zoned(ns->queue)) {
2073 ret = nvme_revalidate_zones(ns);
2074 if (ret && !nvme_first_scan(ns->disk))
2078 if (nvme_ns_head_multipath(ns->head)) {
2079 blk_mq_freeze_queue(ns->head->disk->queue);
2080 nvme_update_disk_info(ns->head->disk, ns, id);
2081 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2082 nvme_mpath_revalidate_paths(ns);
2083 blk_stack_limits(&ns->head->disk->queue->limits,
2084 &ns->queue->limits, 0);
2085 disk_update_readahead(ns->head->disk);
2086 blk_mq_unfreeze_queue(ns->head->disk->queue);
2092 * If probing fails due an unsupported feature, hide the block device,
2093 * but still allow other access.
2095 if (ret == -ENODEV) {
2096 ns->disk->flags |= GENHD_FL_HIDDEN;
2097 set_bit(NVME_NS_READY, &ns->flags);
2106 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2108 switch (info->ids.csi) {
2110 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2111 dev_info(ns->ctrl->device,
2112 "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2114 return nvme_update_ns_info_generic(ns, info);
2116 return nvme_update_ns_info_block(ns, info);
2118 return nvme_update_ns_info_block(ns, info);
2120 dev_info(ns->ctrl->device,
2121 "block device for nsid %u not supported (csi %u)\n",
2122 info->nsid, info->ids.csi);
2123 return nvme_update_ns_info_generic(ns, info);
2127 #ifdef CONFIG_BLK_SED_OPAL
2128 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2131 struct nvme_ctrl *ctrl = data;
2132 struct nvme_command cmd = { };
2135 cmd.common.opcode = nvme_admin_security_send;
2137 cmd.common.opcode = nvme_admin_security_recv;
2138 cmd.common.nsid = 0;
2139 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2140 cmd.common.cdw11 = cpu_to_le32(len);
2142 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2143 NVME_QID_ANY, 1, 0);
2146 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2148 if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2149 if (!ctrl->opal_dev)
2150 ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2151 else if (was_suspended)
2152 opal_unlock_from_suspend(ctrl->opal_dev);
2154 free_opal_dev(ctrl->opal_dev);
2155 ctrl->opal_dev = NULL;
2159 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2162 #endif /* CONFIG_BLK_SED_OPAL */
2164 #ifdef CONFIG_BLK_DEV_ZONED
2165 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2166 unsigned int nr_zones, report_zones_cb cb, void *data)
2168 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2172 #define nvme_report_zones NULL
2173 #endif /* CONFIG_BLK_DEV_ZONED */
2175 const struct block_device_operations nvme_bdev_ops = {
2176 .owner = THIS_MODULE,
2177 .ioctl = nvme_ioctl,
2178 .compat_ioctl = blkdev_compat_ptr_ioctl,
2180 .release = nvme_release,
2181 .getgeo = nvme_getgeo,
2182 .report_zones = nvme_report_zones,
2183 .pr_ops = &nvme_pr_ops,
2186 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2187 u32 timeout, const char *op)
2189 unsigned long timeout_jiffies = jiffies + timeout * HZ;
2193 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2196 if ((csts & mask) == val)
2199 usleep_range(1000, 2000);
2200 if (fatal_signal_pending(current))
2202 if (time_after(jiffies, timeout_jiffies)) {
2203 dev_err(ctrl->device,
2204 "Device not ready; aborting %s, CSTS=0x%x\n",
2213 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2217 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2219 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2221 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2223 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2228 return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2229 NVME_CSTS_SHST_CMPLT,
2230 ctrl->shutdown_timeout, "shutdown");
2232 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2233 msleep(NVME_QUIRK_DELAY_AMOUNT);
2234 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2235 (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2237 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2239 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2241 unsigned dev_page_min;
2245 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2247 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2250 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2252 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2253 dev_err(ctrl->device,
2254 "Minimum device page size %u too large for host (%u)\n",
2255 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2259 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2260 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2262 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2264 if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2265 ctrl->ctrl_config |= NVME_CC_CRIME;
2267 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2268 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2269 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2270 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2274 /* Flush write to device (required if transport is PCI) */
2275 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2279 /* CAP value may change after initial CC write */
2280 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2284 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2285 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2286 u32 crto, ready_timeout;
2288 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2290 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2296 * CRTO should always be greater or equal to CAP.TO, but some
2297 * devices are known to get this wrong. Use the larger of the
2300 if (ctrl->ctrl_config & NVME_CC_CRIME)
2301 ready_timeout = NVME_CRTO_CRIMT(crto);
2303 ready_timeout = NVME_CRTO_CRWMT(crto);
2305 if (ready_timeout < timeout)
2306 dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2309 timeout = ready_timeout;
2312 ctrl->ctrl_config |= NVME_CC_ENABLE;
2313 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2316 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2317 (timeout + 1) / 2, "initialisation");
2319 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2321 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2326 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2329 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2330 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2333 dev_warn_once(ctrl->device,
2334 "could not set timestamp (%d)\n", ret);
2338 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2340 struct nvme_feat_host_behavior *host;
2341 u8 acre = 0, lbafee = 0;
2344 /* Don't bother enabling the feature if retry delay is not reported */
2346 acre = NVME_ENABLE_ACRE;
2347 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2348 lbafee = NVME_ENABLE_LBAFEE;
2350 if (!acre && !lbafee)
2353 host = kzalloc(sizeof(*host), GFP_KERNEL);
2358 host->lbafee = lbafee;
2359 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2360 host, sizeof(*host), NULL);
2366 * The function checks whether the given total (exlat + enlat) latency of
2367 * a power state allows the latter to be used as an APST transition target.
2368 * It does so by comparing the latency to the primary and secondary latency
2369 * tolerances defined by module params. If there's a match, the corresponding
2370 * timeout value is returned and the matching tolerance index (1 or 2) is
2373 static bool nvme_apst_get_transition_time(u64 total_latency,
2374 u64 *transition_time, unsigned *last_index)
2376 if (total_latency <= apst_primary_latency_tol_us) {
2377 if (*last_index == 1)
2380 *transition_time = apst_primary_timeout_ms;
2383 if (apst_secondary_timeout_ms &&
2384 total_latency <= apst_secondary_latency_tol_us) {
2385 if (*last_index <= 2)
2388 *transition_time = apst_secondary_timeout_ms;
2395 * APST (Autonomous Power State Transition) lets us program a table of power
2396 * state transitions that the controller will perform automatically.
2398 * Depending on module params, one of the two supported techniques will be used:
2400 * - If the parameters provide explicit timeouts and tolerances, they will be
2401 * used to build a table with up to 2 non-operational states to transition to.
2402 * The default parameter values were selected based on the values used by
2403 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2404 * regeneration of the APST table in the event of switching between external
2405 * and battery power, the timeouts and tolerances reflect a compromise
2406 * between values used by Microsoft for AC and battery scenarios.
2407 * - If not, we'll configure the table with a simple heuristic: we are willing
2408 * to spend at most 2% of the time transitioning between power states.
2409 * Therefore, when running in any given state, we will enter the next
2410 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2411 * microseconds, as long as that state's exit latency is under the requested
2414 * We will not autonomously enter any non-operational state for which the total
2415 * latency exceeds ps_max_latency_us.
2417 * Users can set ps_max_latency_us to zero to turn off APST.
2419 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2421 struct nvme_feat_auto_pst *table;
2428 unsigned last_lt_index = UINT_MAX;
2431 * If APST isn't supported or if we haven't been initialized yet,
2432 * then don't do anything.
2437 if (ctrl->npss > 31) {
2438 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2442 table = kzalloc(sizeof(*table), GFP_KERNEL);
2446 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2447 /* Turn off APST. */
2448 dev_dbg(ctrl->device, "APST disabled\n");
2453 * Walk through all states from lowest- to highest-power.
2454 * According to the spec, lower-numbered states use more power. NPSS,
2455 * despite the name, is the index of the lowest-power state, not the
2458 for (state = (int)ctrl->npss; state >= 0; state--) {
2459 u64 total_latency_us, exit_latency_us, transition_ms;
2462 table->entries[state] = target;
2465 * Don't allow transitions to the deepest state if it's quirked
2468 if (state == ctrl->npss &&
2469 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2473 * Is this state a useful non-operational state for higher-power
2474 * states to autonomously transition to?
2476 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2479 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2480 if (exit_latency_us > ctrl->ps_max_latency_us)
2483 total_latency_us = exit_latency_us +
2484 le32_to_cpu(ctrl->psd[state].entry_lat);
2487 * This state is good. It can be used as the APST idle target
2488 * for higher power states.
2490 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2491 if (!nvme_apst_get_transition_time(total_latency_us,
2492 &transition_ms, &last_lt_index))
2495 transition_ms = total_latency_us + 19;
2496 do_div(transition_ms, 20);
2497 if (transition_ms > (1 << 24) - 1)
2498 transition_ms = (1 << 24) - 1;
2501 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2504 if (total_latency_us > max_lat_us)
2505 max_lat_us = total_latency_us;
2509 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2511 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2512 max_ps, max_lat_us, (int)sizeof(*table), table);
2516 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2517 table, sizeof(*table), NULL);
2519 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2524 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2526 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2530 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2531 case PM_QOS_LATENCY_ANY:
2539 if (ctrl->ps_max_latency_us != latency) {
2540 ctrl->ps_max_latency_us = latency;
2541 if (ctrl->state == NVME_CTRL_LIVE)
2542 nvme_configure_apst(ctrl);
2546 struct nvme_core_quirk_entry {
2548 * NVMe model and firmware strings are padded with spaces. For
2549 * simplicity, strings in the quirk table are padded with NULLs
2555 unsigned long quirks;
2558 static const struct nvme_core_quirk_entry core_quirks[] = {
2561 * This Toshiba device seems to die using any APST states. See:
2562 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2565 .mn = "THNSF5256GPUK TOSHIBA",
2566 .quirks = NVME_QUIRK_NO_APST,
2570 * This LiteON CL1-3D*-Q11 firmware version has a race
2571 * condition associated with actions related to suspend to idle
2572 * LiteON has resolved the problem in future firmware
2576 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2580 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2581 * aborts I/O during any load, but more easily reproducible
2582 * with discards (fstrim).
2584 * The device is left in a state where it is also not possible
2585 * to use "nvme set-feature" to disable APST, but booting with
2586 * nvme_core.default_ps_max_latency=0 works.
2589 .mn = "KCD6XVUL6T40",
2590 .quirks = NVME_QUIRK_NO_APST,
2594 * The external Samsung X5 SSD fails initialization without a
2595 * delay before checking if it is ready and has a whole set of
2596 * other problems. To make this even more interesting, it
2597 * shares the PCI ID with internal Samsung 970 Evo Plus that
2598 * does not need or want these quirks.
2601 .mn = "Samsung Portable SSD X5",
2602 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2603 NVME_QUIRK_NO_DEEPEST_PS |
2604 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2608 /* match is null-terminated but idstr is space-padded. */
2609 static bool string_matches(const char *idstr, const char *match, size_t len)
2616 matchlen = strlen(match);
2617 WARN_ON_ONCE(matchlen > len);
2619 if (memcmp(idstr, match, matchlen))
2622 for (; matchlen < len; matchlen++)
2623 if (idstr[matchlen] != ' ')
2629 static bool quirk_matches(const struct nvme_id_ctrl *id,
2630 const struct nvme_core_quirk_entry *q)
2632 return q->vid == le16_to_cpu(id->vid) &&
2633 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2634 string_matches(id->fr, q->fr, sizeof(id->fr));
2637 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2638 struct nvme_id_ctrl *id)
2643 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2644 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2645 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2646 strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2650 if (ctrl->vs >= NVME_VS(1, 2, 1))
2651 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2655 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2656 * Base Specification 2.0. It is slightly different from the format
2657 * specified there due to historic reasons, and we can't change it now.
2659 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2660 "nqn.2014.08.org.nvmexpress:%04x%04x",
2661 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2662 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2663 off += sizeof(id->sn);
2664 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2665 off += sizeof(id->mn);
2666 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2669 static void nvme_release_subsystem(struct device *dev)
2671 struct nvme_subsystem *subsys =
2672 container_of(dev, struct nvme_subsystem, dev);
2674 if (subsys->instance >= 0)
2675 ida_free(&nvme_instance_ida, subsys->instance);
2679 static void nvme_destroy_subsystem(struct kref *ref)
2681 struct nvme_subsystem *subsys =
2682 container_of(ref, struct nvme_subsystem, ref);
2684 mutex_lock(&nvme_subsystems_lock);
2685 list_del(&subsys->entry);
2686 mutex_unlock(&nvme_subsystems_lock);
2688 ida_destroy(&subsys->ns_ida);
2689 device_del(&subsys->dev);
2690 put_device(&subsys->dev);
2693 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2695 kref_put(&subsys->ref, nvme_destroy_subsystem);
2698 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2700 struct nvme_subsystem *subsys;
2702 lockdep_assert_held(&nvme_subsystems_lock);
2705 * Fail matches for discovery subsystems. This results
2706 * in each discovery controller bound to a unique subsystem.
2707 * This avoids issues with validating controller values
2708 * that can only be true when there is a single unique subsystem.
2709 * There may be multiple and completely independent entities
2710 * that provide discovery controllers.
2712 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2715 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2716 if (strcmp(subsys->subnqn, subsysnqn))
2718 if (!kref_get_unless_zero(&subsys->ref))
2726 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2728 return ctrl->opts && ctrl->opts->discovery_nqn;
2731 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2732 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2734 struct nvme_ctrl *tmp;
2736 lockdep_assert_held(&nvme_subsystems_lock);
2738 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2739 if (nvme_state_terminal(tmp))
2742 if (tmp->cntlid == ctrl->cntlid) {
2743 dev_err(ctrl->device,
2744 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2745 ctrl->cntlid, dev_name(tmp->device),
2750 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2751 nvme_discovery_ctrl(ctrl))
2754 dev_err(ctrl->device,
2755 "Subsystem does not support multiple controllers\n");
2762 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2764 struct nvme_subsystem *subsys, *found;
2767 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2771 subsys->instance = -1;
2772 mutex_init(&subsys->lock);
2773 kref_init(&subsys->ref);
2774 INIT_LIST_HEAD(&subsys->ctrls);
2775 INIT_LIST_HEAD(&subsys->nsheads);
2776 nvme_init_subnqn(subsys, ctrl, id);
2777 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2778 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2779 subsys->vendor_id = le16_to_cpu(id->vid);
2780 subsys->cmic = id->cmic;
2782 /* Versions prior to 1.4 don't necessarily report a valid type */
2783 if (id->cntrltype == NVME_CTRL_DISC ||
2784 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2785 subsys->subtype = NVME_NQN_DISC;
2787 subsys->subtype = NVME_NQN_NVME;
2789 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2790 dev_err(ctrl->device,
2791 "Subsystem %s is not a discovery controller",
2796 subsys->awupf = le16_to_cpu(id->awupf);
2797 nvme_mpath_default_iopolicy(subsys);
2799 subsys->dev.class = nvme_subsys_class;
2800 subsys->dev.release = nvme_release_subsystem;
2801 subsys->dev.groups = nvme_subsys_attrs_groups;
2802 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2803 device_initialize(&subsys->dev);
2805 mutex_lock(&nvme_subsystems_lock);
2806 found = __nvme_find_get_subsystem(subsys->subnqn);
2808 put_device(&subsys->dev);
2811 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2813 goto out_put_subsystem;
2816 ret = device_add(&subsys->dev);
2818 dev_err(ctrl->device,
2819 "failed to register subsystem device.\n");
2820 put_device(&subsys->dev);
2823 ida_init(&subsys->ns_ida);
2824 list_add_tail(&subsys->entry, &nvme_subsystems);
2827 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2828 dev_name(ctrl->device));
2830 dev_err(ctrl->device,
2831 "failed to create sysfs link from subsystem.\n");
2832 goto out_put_subsystem;
2836 subsys->instance = ctrl->instance;
2837 ctrl->subsys = subsys;
2838 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2839 mutex_unlock(&nvme_subsystems_lock);
2843 nvme_put_subsystem(subsys);
2845 mutex_unlock(&nvme_subsystems_lock);
2849 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2850 void *log, size_t size, u64 offset)
2852 struct nvme_command c = { };
2853 u32 dwlen = nvme_bytes_to_numd(size);
2855 c.get_log_page.opcode = nvme_admin_get_log_page;
2856 c.get_log_page.nsid = cpu_to_le32(nsid);
2857 c.get_log_page.lid = log_page;
2858 c.get_log_page.lsp = lsp;
2859 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2860 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2861 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2862 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2863 c.get_log_page.csi = csi;
2865 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2868 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2869 struct nvme_effects_log **log)
2871 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2877 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2881 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2882 cel, sizeof(*cel), 0);
2888 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2894 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2896 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2898 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2903 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2905 struct nvme_command c = { };
2906 struct nvme_id_ctrl_nvm *id;
2909 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2910 ctrl->max_discard_sectors = UINT_MAX;
2911 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2913 ctrl->max_discard_sectors = 0;
2914 ctrl->max_discard_segments = 0;
2918 * Even though NVMe spec explicitly states that MDTS is not applicable
2919 * to the write-zeroes, we are cautious and limit the size to the
2920 * controllers max_hw_sectors value, which is based on the MDTS field
2921 * and possibly other limiting factors.
2923 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2924 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2925 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2927 ctrl->max_zeroes_sectors = 0;
2929 if (ctrl->subsys->subtype != NVME_NQN_NVME ||
2930 nvme_ctrl_limited_cns(ctrl) ||
2931 test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
2934 id = kzalloc(sizeof(*id), GFP_KERNEL);
2938 c.identify.opcode = nvme_admin_identify;
2939 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2940 c.identify.csi = NVME_CSI_NVM;
2942 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2947 ctrl->max_discard_segments = id->dmrl;
2948 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2950 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2954 set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
2959 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
2961 struct nvme_effects_log *log = ctrl->effects;
2963 log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2964 NVME_CMD_EFFECTS_NCC |
2965 NVME_CMD_EFFECTS_CSE_MASK);
2966 log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
2967 NVME_CMD_EFFECTS_CSE_MASK);
2970 * The spec says the result of a security receive command depends on
2971 * the previous security send command. As such, many vendors log this
2972 * command as one to submitted only when no other commands to the same
2973 * namespace are outstanding. The intention is to tell the host to
2974 * prevent mixing security send and receive.
2976 * This driver can only enforce such exclusive access against IO
2977 * queues, though. We are not readily able to enforce such a rule for
2978 * two commands to the admin queue, which is the only queue that
2979 * matters for this command.
2981 * Rather than blindly freezing the IO queues for this effect that
2982 * doesn't even apply to IO, mask it off.
2984 log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
2986 log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2987 log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2988 log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
2991 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2998 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2999 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3004 if (!ctrl->effects) {
3005 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3008 xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3011 nvme_init_known_nvm_effects(ctrl);
3015 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3017 struct nvme_id_ctrl *id;
3019 bool prev_apst_enabled;
3022 ret = nvme_identify_ctrl(ctrl, &id);
3024 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3028 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3029 ctrl->cntlid = le16_to_cpu(id->cntlid);
3031 if (!ctrl->identified) {
3035 * Check for quirks. Quirk can depend on firmware version,
3036 * so, in principle, the set of quirks present can change
3037 * across a reset. As a possible future enhancement, we
3038 * could re-scan for quirks every time we reinitialize
3039 * the device, but we'd have to make sure that the driver
3040 * behaves intelligently if the quirks change.
3042 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3043 if (quirk_matches(id, &core_quirks[i]))
3044 ctrl->quirks |= core_quirks[i].quirks;
3047 ret = nvme_init_subsystem(ctrl, id);
3051 ret = nvme_init_effects(ctrl, id);
3055 memcpy(ctrl->subsys->firmware_rev, id->fr,
3056 sizeof(ctrl->subsys->firmware_rev));
3058 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3059 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3060 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3063 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3064 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3065 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3067 ctrl->oacs = le16_to_cpu(id->oacs);
3068 ctrl->oncs = le16_to_cpu(id->oncs);
3069 ctrl->mtfa = le16_to_cpu(id->mtfa);
3070 ctrl->oaes = le32_to_cpu(id->oaes);
3071 ctrl->wctemp = le16_to_cpu(id->wctemp);
3072 ctrl->cctemp = le16_to_cpu(id->cctemp);
3074 atomic_set(&ctrl->abort_limit, id->acl + 1);
3075 ctrl->vwc = id->vwc;
3077 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3079 max_hw_sectors = UINT_MAX;
3080 ctrl->max_hw_sectors =
3081 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3083 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3084 ctrl->sgls = le32_to_cpu(id->sgls);
3085 ctrl->kas = le16_to_cpu(id->kas);
3086 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3087 ctrl->ctratt = le32_to_cpu(id->ctratt);
3089 ctrl->cntrltype = id->cntrltype;
3090 ctrl->dctype = id->dctype;
3094 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3096 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3097 shutdown_timeout, 60);
3099 if (ctrl->shutdown_timeout != shutdown_timeout)
3100 dev_info(ctrl->device,
3101 "Shutdown timeout set to %u seconds\n",
3102 ctrl->shutdown_timeout);
3104 ctrl->shutdown_timeout = shutdown_timeout;
3106 ctrl->npss = id->npss;
3107 ctrl->apsta = id->apsta;
3108 prev_apst_enabled = ctrl->apst_enabled;
3109 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3110 if (force_apst && id->apsta) {
3111 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3112 ctrl->apst_enabled = true;
3114 ctrl->apst_enabled = false;
3117 ctrl->apst_enabled = id->apsta;
3119 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3121 if (ctrl->ops->flags & NVME_F_FABRICS) {
3122 ctrl->icdoff = le16_to_cpu(id->icdoff);
3123 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3124 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3125 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3128 * In fabrics we need to verify the cntlid matches the
3131 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3132 dev_err(ctrl->device,
3133 "Mismatching cntlid: Connect %u vs Identify "
3135 ctrl->cntlid, le16_to_cpu(id->cntlid));
3140 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3141 dev_err(ctrl->device,
3142 "keep-alive support is mandatory for fabrics\n");
3147 ctrl->hmpre = le32_to_cpu(id->hmpre);
3148 ctrl->hmmin = le32_to_cpu(id->hmmin);
3149 ctrl->hmminds = le32_to_cpu(id->hmminds);
3150 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3153 ret = nvme_mpath_init_identify(ctrl, id);
3157 if (ctrl->apst_enabled && !prev_apst_enabled)
3158 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3159 else if (!ctrl->apst_enabled && prev_apst_enabled)
3160 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3168 * Initialize the cached copies of the Identify data and various controller
3169 * register in our nvme_ctrl structure. This should be called as soon as
3170 * the admin queue is fully up and running.
3172 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3176 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3178 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3182 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3184 if (ctrl->vs >= NVME_VS(1, 1, 0))
3185 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3187 ret = nvme_init_identify(ctrl);
3191 ret = nvme_configure_apst(ctrl);
3195 ret = nvme_configure_timestamp(ctrl);
3199 ret = nvme_configure_host_options(ctrl);
3203 nvme_configure_opal(ctrl, was_suspended);
3205 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3207 * Do not return errors unless we are in a controller reset,
3208 * the controller works perfectly fine without hwmon.
3210 ret = nvme_hwmon_init(ctrl);
3215 clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3216 ctrl->identified = true;
3220 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3222 static int nvme_dev_open(struct inode *inode, struct file *file)
3224 struct nvme_ctrl *ctrl =
3225 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3227 switch (ctrl->state) {
3228 case NVME_CTRL_LIVE:
3231 return -EWOULDBLOCK;
3234 nvme_get_ctrl(ctrl);
3235 if (!try_module_get(ctrl->ops->module)) {
3236 nvme_put_ctrl(ctrl);
3240 file->private_data = ctrl;
3244 static int nvme_dev_release(struct inode *inode, struct file *file)
3246 struct nvme_ctrl *ctrl =
3247 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3249 module_put(ctrl->ops->module);
3250 nvme_put_ctrl(ctrl);
3254 static const struct file_operations nvme_dev_fops = {
3255 .owner = THIS_MODULE,
3256 .open = nvme_dev_open,
3257 .release = nvme_dev_release,
3258 .unlocked_ioctl = nvme_dev_ioctl,
3259 .compat_ioctl = compat_ptr_ioctl,
3260 .uring_cmd = nvme_dev_uring_cmd,
3263 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3266 struct nvme_ns_head *h;
3268 lockdep_assert_held(&ctrl->subsys->lock);
3270 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3272 * Private namespaces can share NSIDs under some conditions.
3273 * In that case we can't use the same ns_head for namespaces
3274 * with the same NSID.
3276 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3278 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3285 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3286 struct nvme_ns_ids *ids)
3288 bool has_uuid = !uuid_is_null(&ids->uuid);
3289 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3290 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3291 struct nvme_ns_head *h;
3293 lockdep_assert_held(&subsys->lock);
3295 list_for_each_entry(h, &subsys->nsheads, entry) {
3296 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3299 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3302 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3309 static void nvme_cdev_rel(struct device *dev)
3311 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3314 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3316 cdev_device_del(cdev, cdev_device);
3317 put_device(cdev_device);
3320 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3321 const struct file_operations *fops, struct module *owner)
3325 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3328 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3329 cdev_device->class = nvme_ns_chr_class;
3330 cdev_device->release = nvme_cdev_rel;
3331 device_initialize(cdev_device);
3332 cdev_init(cdev, fops);
3333 cdev->owner = owner;
3334 ret = cdev_device_add(cdev, cdev_device);
3336 put_device(cdev_device);
3341 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3343 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3346 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3348 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3352 static const struct file_operations nvme_ns_chr_fops = {
3353 .owner = THIS_MODULE,
3354 .open = nvme_ns_chr_open,
3355 .release = nvme_ns_chr_release,
3356 .unlocked_ioctl = nvme_ns_chr_ioctl,
3357 .compat_ioctl = compat_ptr_ioctl,
3358 .uring_cmd = nvme_ns_chr_uring_cmd,
3359 .uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3362 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3366 ns->cdev_device.parent = ns->ctrl->device;
3367 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3368 ns->ctrl->instance, ns->head->instance);
3372 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3373 ns->ctrl->ops->module);
3376 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3377 struct nvme_ns_info *info)
3379 struct nvme_ns_head *head;
3380 size_t size = sizeof(*head);
3383 #ifdef CONFIG_NVME_MULTIPATH
3384 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3387 head = kzalloc(size, GFP_KERNEL);
3390 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3393 head->instance = ret;
3394 INIT_LIST_HEAD(&head->list);
3395 ret = init_srcu_struct(&head->srcu);
3397 goto out_ida_remove;
3398 head->subsys = ctrl->subsys;
3399 head->ns_id = info->nsid;
3400 head->ids = info->ids;
3401 head->shared = info->is_shared;
3402 kref_init(&head->ref);
3404 if (head->ids.csi) {
3405 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3407 goto out_cleanup_srcu;
3409 head->effects = ctrl->effects;
3411 ret = nvme_mpath_alloc_disk(ctrl, head);
3413 goto out_cleanup_srcu;
3415 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3417 kref_get(&ctrl->subsys->ref);
3421 cleanup_srcu_struct(&head->srcu);
3423 ida_free(&ctrl->subsys->ns_ida, head->instance);
3428 ret = blk_status_to_errno(nvme_error_status(ret));
3429 return ERR_PTR(ret);
3432 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3433 struct nvme_ns_ids *ids)
3435 struct nvme_subsystem *s;
3439 * Note that this check is racy as we try to avoid holding the global
3440 * lock over the whole ns_head creation. But it is only intended as
3441 * a sanity check anyway.
3443 mutex_lock(&nvme_subsystems_lock);
3444 list_for_each_entry(s, &nvme_subsystems, entry) {
3447 mutex_lock(&s->lock);
3448 ret = nvme_subsys_check_duplicate_ids(s, ids);
3449 mutex_unlock(&s->lock);
3453 mutex_unlock(&nvme_subsystems_lock);
3458 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3460 struct nvme_ctrl *ctrl = ns->ctrl;
3461 struct nvme_ns_head *head = NULL;
3464 ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3467 * We've found two different namespaces on two different
3468 * subsystems that report the same ID. This is pretty nasty
3469 * for anything that actually requires unique device
3470 * identification. In the kernel we need this for multipathing,
3471 * and in user space the /dev/disk/by-id/ links rely on it.
3473 * If the device also claims to be multi-path capable back off
3474 * here now and refuse the probe the second device as this is a
3475 * recipe for data corruption. If not this is probably a
3476 * cheap consumer device if on the PCIe bus, so let the user
3477 * proceed and use the shiny toy, but warn that with changing
3478 * probing order (which due to our async probing could just be
3479 * device taking longer to startup) the other device could show
3482 nvme_print_device_info(ctrl);
3483 if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3484 ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3486 dev_err(ctrl->device,
3487 "ignoring nsid %d because of duplicate IDs\n",
3492 dev_err(ctrl->device,
3493 "clearing duplicate IDs for nsid %d\n", info->nsid);
3494 dev_err(ctrl->device,
3495 "use of /dev/disk/by-id/ may cause data corruption\n");
3496 memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3497 memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3498 memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3499 ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3502 mutex_lock(&ctrl->subsys->lock);
3503 head = nvme_find_ns_head(ctrl, info->nsid);
3505 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3507 dev_err(ctrl->device,
3508 "duplicate IDs in subsystem for nsid %d\n",
3512 head = nvme_alloc_ns_head(ctrl, info);
3514 ret = PTR_ERR(head);
3519 if (!info->is_shared || !head->shared) {
3520 dev_err(ctrl->device,
3521 "Duplicate unshared namespace %d\n",
3523 goto out_put_ns_head;
3525 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3526 dev_err(ctrl->device,
3527 "IDs don't match for shared namespace %d\n",
3529 goto out_put_ns_head;
3533 dev_warn(ctrl->device,
3534 "Found shared namespace %d, but multipathing not supported.\n",
3536 dev_warn_once(ctrl->device,
3537 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3541 list_add_tail_rcu(&ns->siblings, &head->list);
3543 mutex_unlock(&ctrl->subsys->lock);
3547 nvme_put_ns_head(head);
3549 mutex_unlock(&ctrl->subsys->lock);
3553 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3555 struct nvme_ns *ns, *ret = NULL;
3557 down_read(&ctrl->namespaces_rwsem);
3558 list_for_each_entry(ns, &ctrl->namespaces, list) {
3559 if (ns->head->ns_id == nsid) {
3560 if (!nvme_get_ns(ns))
3565 if (ns->head->ns_id > nsid)
3568 up_read(&ctrl->namespaces_rwsem);
3571 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3574 * Add the namespace to the controller list while keeping the list ordered.
3576 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3578 struct nvme_ns *tmp;
3580 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3581 if (tmp->head->ns_id < ns->head->ns_id) {
3582 list_add(&ns->list, &tmp->list);
3586 list_add(&ns->list, &ns->ctrl->namespaces);
3589 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3592 struct gendisk *disk;
3593 int node = ctrl->numa_node;
3595 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3599 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3602 disk->fops = &nvme_bdev_ops;
3603 disk->private_data = ns;
3606 ns->queue = disk->queue;
3608 if (ctrl->opts && ctrl->opts->data_digest)
3609 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3611 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3612 if (ctrl->ops->supports_pci_p2pdma &&
3613 ctrl->ops->supports_pci_p2pdma(ctrl))
3614 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3617 kref_init(&ns->kref);
3619 if (nvme_init_ns_head(ns, info))
3620 goto out_cleanup_disk;
3623 * If multipathing is enabled, the device name for all disks and not
3624 * just those that represent shared namespaces needs to be based on the
3625 * subsystem instance. Using the controller instance for private
3626 * namespaces could lead to naming collisions between shared and private
3627 * namespaces if they don't use a common numbering scheme.
3629 * If multipathing is not enabled, disk names must use the controller
3630 * instance as shared namespaces will show up as multiple block
3633 if (nvme_ns_head_multipath(ns->head)) {
3634 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3635 ctrl->instance, ns->head->instance);
3636 disk->flags |= GENHD_FL_HIDDEN;
3637 } else if (multipath) {
3638 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3639 ns->head->instance);
3641 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3642 ns->head->instance);
3645 if (nvme_update_ns_info(ns, info))
3648 down_write(&ctrl->namespaces_rwsem);
3649 nvme_ns_add_to_ctrl_list(ns);
3650 up_write(&ctrl->namespaces_rwsem);
3651 nvme_get_ctrl(ctrl);
3653 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3654 goto out_cleanup_ns_from_list;
3656 if (!nvme_ns_head_multipath(ns->head))
3657 nvme_add_ns_cdev(ns);
3659 nvme_mpath_add_disk(ns, info->anagrpid);
3660 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3664 out_cleanup_ns_from_list:
3665 nvme_put_ctrl(ctrl);
3666 down_write(&ctrl->namespaces_rwsem);
3667 list_del_init(&ns->list);
3668 up_write(&ctrl->namespaces_rwsem);
3670 mutex_lock(&ctrl->subsys->lock);
3671 list_del_rcu(&ns->siblings);
3672 if (list_empty(&ns->head->list))
3673 list_del_init(&ns->head->entry);
3674 mutex_unlock(&ctrl->subsys->lock);
3675 nvme_put_ns_head(ns->head);
3682 static void nvme_ns_remove(struct nvme_ns *ns)
3684 bool last_path = false;
3686 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3689 clear_bit(NVME_NS_READY, &ns->flags);
3690 set_capacity(ns->disk, 0);
3691 nvme_fault_inject_fini(&ns->fault_inject);
3694 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3695 * this ns going back into current_path.
3697 synchronize_srcu(&ns->head->srcu);
3699 /* wait for concurrent submissions */
3700 if (nvme_mpath_clear_current_path(ns))
3701 synchronize_srcu(&ns->head->srcu);
3703 mutex_lock(&ns->ctrl->subsys->lock);
3704 list_del_rcu(&ns->siblings);
3705 if (list_empty(&ns->head->list)) {
3706 list_del_init(&ns->head->entry);
3709 mutex_unlock(&ns->ctrl->subsys->lock);
3711 /* guarantee not available in head->list */
3712 synchronize_srcu(&ns->head->srcu);
3714 if (!nvme_ns_head_multipath(ns->head))
3715 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3716 del_gendisk(ns->disk);
3718 down_write(&ns->ctrl->namespaces_rwsem);
3719 list_del_init(&ns->list);
3720 up_write(&ns->ctrl->namespaces_rwsem);
3723 nvme_mpath_shutdown_disk(ns->head);
3727 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3729 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3737 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3739 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3741 if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3742 dev_err(ns->ctrl->device,
3743 "identifiers changed for nsid %d\n", ns->head->ns_id);
3747 ret = nvme_update_ns_info(ns, info);
3750 * Only remove the namespace if we got a fatal error back from the
3751 * device, otherwise ignore the error and just move on.
3753 * TODO: we should probably schedule a delayed retry here.
3755 if (ret > 0 && (ret & NVME_SC_DNR))
3759 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3761 struct nvme_ns_info info = { .nsid = nsid };
3765 if (nvme_identify_ns_descs(ctrl, &info))
3768 if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3769 dev_warn(ctrl->device,
3770 "command set not reported for nsid: %d\n", nsid);
3775 * If available try to use the Command Set Idependent Identify Namespace
3776 * data structure to find all the generic information that is needed to
3777 * set up a namespace. If not fall back to the legacy version.
3779 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3780 (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3781 ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3783 ret = nvme_ns_info_from_identify(ctrl, &info);
3785 if (info.is_removed)
3786 nvme_ns_remove_by_nsid(ctrl, nsid);
3789 * Ignore the namespace if it is not ready. We will get an AEN once it
3790 * becomes ready and restart the scan.
3792 if (ret || !info.is_ready)
3795 ns = nvme_find_get_ns(ctrl, nsid);
3797 nvme_validate_ns(ns, &info);
3800 nvme_alloc_ns(ctrl, &info);
3804 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3807 struct nvme_ns *ns, *next;
3810 down_write(&ctrl->namespaces_rwsem);
3811 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3812 if (ns->head->ns_id > nsid)
3813 list_move_tail(&ns->list, &rm_list);
3815 up_write(&ctrl->namespaces_rwsem);
3817 list_for_each_entry_safe(ns, next, &rm_list, list)
3822 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3824 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3829 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3834 struct nvme_command cmd = {
3835 .identify.opcode = nvme_admin_identify,
3836 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
3837 .identify.nsid = cpu_to_le32(prev),
3840 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3841 NVME_IDENTIFY_DATA_SIZE);
3843 dev_warn(ctrl->device,
3844 "Identify NS List failed (status=0x%x)\n", ret);
3848 for (i = 0; i < nr_entries; i++) {
3849 u32 nsid = le32_to_cpu(ns_list[i]);
3851 if (!nsid) /* end of the list? */
3853 nvme_scan_ns(ctrl, nsid);
3854 while (++prev < nsid)
3855 nvme_ns_remove_by_nsid(ctrl, prev);
3859 nvme_remove_invalid_namespaces(ctrl, prev);
3865 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
3867 struct nvme_id_ctrl *id;
3870 if (nvme_identify_ctrl(ctrl, &id))
3872 nn = le32_to_cpu(id->nn);
3875 for (i = 1; i <= nn; i++)
3876 nvme_scan_ns(ctrl, i);
3878 nvme_remove_invalid_namespaces(ctrl, nn);
3881 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
3883 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
3887 log = kzalloc(log_size, GFP_KERNEL);
3892 * We need to read the log to clear the AEN, but we don't want to rely
3893 * on it for the changed namespace information as userspace could have
3894 * raced with us in reading the log page, which could cause us to miss
3897 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
3898 NVME_CSI_NVM, log, log_size, 0);
3900 dev_warn(ctrl->device,
3901 "reading changed ns log failed: %d\n", error);
3906 static void nvme_scan_work(struct work_struct *work)
3908 struct nvme_ctrl *ctrl =
3909 container_of(work, struct nvme_ctrl, scan_work);
3912 /* No tagset on a live ctrl means IO queues could not created */
3913 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
3917 * Identify controller limits can change at controller reset due to
3918 * new firmware download, even though it is not common we cannot ignore
3919 * such scenario. Controller's non-mdts limits are reported in the unit
3920 * of logical blocks that is dependent on the format of attached
3921 * namespace. Hence re-read the limits at the time of ns allocation.
3923 ret = nvme_init_non_mdts_limits(ctrl);
3925 dev_warn(ctrl->device,
3926 "reading non-mdts-limits failed: %d\n", ret);
3930 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
3931 dev_info(ctrl->device, "rescanning namespaces.\n");
3932 nvme_clear_changed_ns_log(ctrl);
3935 mutex_lock(&ctrl->scan_lock);
3936 if (nvme_ctrl_limited_cns(ctrl)) {
3937 nvme_scan_ns_sequential(ctrl);
3940 * Fall back to sequential scan if DNR is set to handle broken
3941 * devices which should support Identify NS List (as per the VS
3942 * they report) but don't actually support it.
3944 ret = nvme_scan_ns_list(ctrl);
3945 if (ret > 0 && ret & NVME_SC_DNR)
3946 nvme_scan_ns_sequential(ctrl);
3948 mutex_unlock(&ctrl->scan_lock);
3952 * This function iterates the namespace list unlocked to allow recovery from
3953 * controller failure. It is up to the caller to ensure the namespace list is
3954 * not modified by scan work while this function is executing.
3956 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
3958 struct nvme_ns *ns, *next;
3962 * make sure to requeue I/O to all namespaces as these
3963 * might result from the scan itself and must complete
3964 * for the scan_work to make progress
3966 nvme_mpath_clear_ctrl_paths(ctrl);
3969 * Unquiesce io queues so any pending IO won't hang, especially
3970 * those submitted from scan work
3972 nvme_unquiesce_io_queues(ctrl);
3974 /* prevent racing with ns scanning */
3975 flush_work(&ctrl->scan_work);
3978 * The dead states indicates the controller was not gracefully
3979 * disconnected. In that case, we won't be able to flush any data while
3980 * removing the namespaces' disks; fail all the queues now to avoid
3981 * potentially having to clean up the failed sync later.
3983 if (ctrl->state == NVME_CTRL_DEAD)
3984 nvme_mark_namespaces_dead(ctrl);
3986 /* this is a no-op when called from the controller reset handler */
3987 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
3989 down_write(&ctrl->namespaces_rwsem);
3990 list_splice_init(&ctrl->namespaces, &ns_list);
3991 up_write(&ctrl->namespaces_rwsem);
3993 list_for_each_entry_safe(ns, next, &ns_list, list)
3996 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
3998 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4000 const struct nvme_ctrl *ctrl =
4001 container_of(dev, struct nvme_ctrl, ctrl_device);
4002 struct nvmf_ctrl_options *opts = ctrl->opts;
4005 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4010 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4014 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4015 opts->trsvcid ?: "none");
4019 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4020 opts->host_traddr ?: "none");
4024 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4025 opts->host_iface ?: "none");
4030 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4032 char *envp[2] = { envdata, NULL };
4034 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4037 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4039 char *envp[2] = { NULL, NULL };
4040 u32 aen_result = ctrl->aen_result;
4042 ctrl->aen_result = 0;
4046 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4049 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4053 static void nvme_async_event_work(struct work_struct *work)
4055 struct nvme_ctrl *ctrl =
4056 container_of(work, struct nvme_ctrl, async_event_work);
4058 nvme_aen_uevent(ctrl);
4061 * The transport drivers must guarantee AER submission here is safe by
4062 * flushing ctrl async_event_work after changing the controller state
4063 * from LIVE and before freeing the admin queue.
4065 if (ctrl->state == NVME_CTRL_LIVE)
4066 ctrl->ops->submit_async_event(ctrl);
4069 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4074 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4080 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4083 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4085 struct nvme_fw_slot_info_log *log;
4087 log = kmalloc(sizeof(*log), GFP_KERNEL);
4091 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4092 log, sizeof(*log), 0))
4093 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4097 static void nvme_fw_act_work(struct work_struct *work)
4099 struct nvme_ctrl *ctrl = container_of(work,
4100 struct nvme_ctrl, fw_act_work);
4101 unsigned long fw_act_timeout;
4104 fw_act_timeout = jiffies +
4105 msecs_to_jiffies(ctrl->mtfa * 100);
4107 fw_act_timeout = jiffies +
4108 msecs_to_jiffies(admin_timeout * 1000);
4110 nvme_quiesce_io_queues(ctrl);
4111 while (nvme_ctrl_pp_status(ctrl)) {
4112 if (time_after(jiffies, fw_act_timeout)) {
4113 dev_warn(ctrl->device,
4114 "Fw activation timeout, reset controller\n");
4115 nvme_try_sched_reset(ctrl);
4121 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4124 nvme_unquiesce_io_queues(ctrl);
4125 /* read FW slot information to clear the AER */
4126 nvme_get_fw_slot_info(ctrl);
4128 queue_work(nvme_wq, &ctrl->async_event_work);
4131 static u32 nvme_aer_type(u32 result)
4133 return result & 0x7;
4136 static u32 nvme_aer_subtype(u32 result)
4138 return (result & 0xff00) >> 8;
4141 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4143 u32 aer_notice_type = nvme_aer_subtype(result);
4144 bool requeue = true;
4146 switch (aer_notice_type) {
4147 case NVME_AER_NOTICE_NS_CHANGED:
4148 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4149 nvme_queue_scan(ctrl);
4151 case NVME_AER_NOTICE_FW_ACT_STARTING:
4153 * We are (ab)using the RESETTING state to prevent subsequent
4154 * recovery actions from interfering with the controller's
4155 * firmware activation.
4157 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4158 nvme_auth_stop(ctrl);
4160 queue_work(nvme_wq, &ctrl->fw_act_work);
4163 #ifdef CONFIG_NVME_MULTIPATH
4164 case NVME_AER_NOTICE_ANA:
4165 if (!ctrl->ana_log_buf)
4167 queue_work(nvme_wq, &ctrl->ana_work);
4170 case NVME_AER_NOTICE_DISC_CHANGED:
4171 ctrl->aen_result = result;
4174 dev_warn(ctrl->device, "async event result %08x\n", result);
4179 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4181 dev_warn(ctrl->device, "resetting controller due to AER\n");
4182 nvme_reset_ctrl(ctrl);
4185 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4186 volatile union nvme_result *res)
4188 u32 result = le32_to_cpu(res->u32);
4189 u32 aer_type = nvme_aer_type(result);
4190 u32 aer_subtype = nvme_aer_subtype(result);
4191 bool requeue = true;
4193 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4196 trace_nvme_async_event(ctrl, result);
4198 case NVME_AER_NOTICE:
4199 requeue = nvme_handle_aen_notice(ctrl, result);
4201 case NVME_AER_ERROR:
4203 * For a persistent internal error, don't run async_event_work
4204 * to submit a new AER. The controller reset will do it.
4206 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4207 nvme_handle_aer_persistent_error(ctrl);
4211 case NVME_AER_SMART:
4214 ctrl->aen_result = result;
4221 queue_work(nvme_wq, &ctrl->async_event_work);
4223 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4225 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4226 const struct blk_mq_ops *ops, unsigned int cmd_size)
4230 memset(set, 0, sizeof(*set));
4232 set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4233 if (ctrl->ops->flags & NVME_F_FABRICS)
4234 set->reserved_tags = NVMF_RESERVED_TAGS;
4235 set->numa_node = ctrl->numa_node;
4236 set->flags = BLK_MQ_F_NO_SCHED;
4237 if (ctrl->ops->flags & NVME_F_BLOCKING)
4238 set->flags |= BLK_MQ_F_BLOCKING;
4239 set->cmd_size = cmd_size;
4240 set->driver_data = ctrl;
4241 set->nr_hw_queues = 1;
4242 set->timeout = NVME_ADMIN_TIMEOUT;
4243 ret = blk_mq_alloc_tag_set(set);
4247 ctrl->admin_q = blk_mq_init_queue(set);
4248 if (IS_ERR(ctrl->admin_q)) {
4249 ret = PTR_ERR(ctrl->admin_q);
4250 goto out_free_tagset;
4253 if (ctrl->ops->flags & NVME_F_FABRICS) {
4254 ctrl->fabrics_q = blk_mq_init_queue(set);
4255 if (IS_ERR(ctrl->fabrics_q)) {
4256 ret = PTR_ERR(ctrl->fabrics_q);
4257 goto out_cleanup_admin_q;
4261 ctrl->admin_tagset = set;
4264 out_cleanup_admin_q:
4265 blk_mq_destroy_queue(ctrl->admin_q);
4266 blk_put_queue(ctrl->admin_q);
4268 blk_mq_free_tag_set(set);
4269 ctrl->admin_q = NULL;
4270 ctrl->fabrics_q = NULL;
4273 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4275 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4277 blk_mq_destroy_queue(ctrl->admin_q);
4278 blk_put_queue(ctrl->admin_q);
4279 if (ctrl->ops->flags & NVME_F_FABRICS) {
4280 blk_mq_destroy_queue(ctrl->fabrics_q);
4281 blk_put_queue(ctrl->fabrics_q);
4283 blk_mq_free_tag_set(ctrl->admin_tagset);
4285 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4287 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4288 const struct blk_mq_ops *ops, unsigned int nr_maps,
4289 unsigned int cmd_size)
4293 memset(set, 0, sizeof(*set));
4295 set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4297 * Some Apple controllers requires tags to be unique across admin and
4298 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4300 if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4301 set->reserved_tags = NVME_AQ_DEPTH;
4302 else if (ctrl->ops->flags & NVME_F_FABRICS)
4303 set->reserved_tags = NVMF_RESERVED_TAGS;
4304 set->numa_node = ctrl->numa_node;
4305 set->flags = BLK_MQ_F_SHOULD_MERGE;
4306 if (ctrl->ops->flags & NVME_F_BLOCKING)
4307 set->flags |= BLK_MQ_F_BLOCKING;
4308 set->cmd_size = cmd_size,
4309 set->driver_data = ctrl;
4310 set->nr_hw_queues = ctrl->queue_count - 1;
4311 set->timeout = NVME_IO_TIMEOUT;
4312 set->nr_maps = nr_maps;
4313 ret = blk_mq_alloc_tag_set(set);
4317 if (ctrl->ops->flags & NVME_F_FABRICS) {
4318 ctrl->connect_q = blk_mq_init_queue(set);
4319 if (IS_ERR(ctrl->connect_q)) {
4320 ret = PTR_ERR(ctrl->connect_q);
4321 goto out_free_tag_set;
4323 blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4331 blk_mq_free_tag_set(set);
4332 ctrl->connect_q = NULL;
4335 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4337 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4339 if (ctrl->ops->flags & NVME_F_FABRICS) {
4340 blk_mq_destroy_queue(ctrl->connect_q);
4341 blk_put_queue(ctrl->connect_q);
4343 blk_mq_free_tag_set(ctrl->tagset);
4345 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4347 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4349 nvme_mpath_stop(ctrl);
4350 nvme_auth_stop(ctrl);
4351 nvme_stop_keep_alive(ctrl);
4352 nvme_stop_failfast_work(ctrl);
4353 flush_work(&ctrl->async_event_work);
4354 cancel_work_sync(&ctrl->fw_act_work);
4355 if (ctrl->ops->stop_ctrl)
4356 ctrl->ops->stop_ctrl(ctrl);
4358 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4360 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4362 nvme_start_keep_alive(ctrl);
4364 nvme_enable_aen(ctrl);
4367 * persistent discovery controllers need to send indication to userspace
4368 * to re-read the discovery log page to learn about possible changes
4369 * that were missed. We identify persistent discovery controllers by
4370 * checking that they started once before, hence are reconnecting back.
4372 if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4373 nvme_discovery_ctrl(ctrl))
4374 nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4376 if (ctrl->queue_count > 1) {
4377 nvme_queue_scan(ctrl);
4378 nvme_unquiesce_io_queues(ctrl);
4379 nvme_mpath_update(ctrl);
4382 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4383 set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4385 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4387 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4389 nvme_hwmon_exit(ctrl);
4390 nvme_fault_inject_fini(&ctrl->fault_inject);
4391 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4392 cdev_device_del(&ctrl->cdev, ctrl->device);
4393 nvme_put_ctrl(ctrl);
4395 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4397 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4399 struct nvme_effects_log *cel;
4402 xa_for_each(&ctrl->cels, i, cel) {
4403 xa_erase(&ctrl->cels, i);
4407 xa_destroy(&ctrl->cels);
4410 static void nvme_free_ctrl(struct device *dev)
4412 struct nvme_ctrl *ctrl =
4413 container_of(dev, struct nvme_ctrl, ctrl_device);
4414 struct nvme_subsystem *subsys = ctrl->subsys;
4416 if (!subsys || ctrl->instance != subsys->instance)
4417 ida_free(&nvme_instance_ida, ctrl->instance);
4419 nvme_free_cels(ctrl);
4420 nvme_mpath_uninit(ctrl);
4421 nvme_auth_stop(ctrl);
4422 nvme_auth_free(ctrl);
4423 __free_page(ctrl->discard_page);
4424 free_opal_dev(ctrl->opal_dev);
4427 mutex_lock(&nvme_subsystems_lock);
4428 list_del(&ctrl->subsys_entry);
4429 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4430 mutex_unlock(&nvme_subsystems_lock);
4433 ctrl->ops->free_ctrl(ctrl);
4436 nvme_put_subsystem(subsys);
4440 * Initialize a NVMe controller structures. This needs to be called during
4441 * earliest initialization so that we have the initialized structured around
4444 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4445 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4449 ctrl->state = NVME_CTRL_NEW;
4450 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4451 spin_lock_init(&ctrl->lock);
4452 mutex_init(&ctrl->scan_lock);
4453 INIT_LIST_HEAD(&ctrl->namespaces);
4454 xa_init(&ctrl->cels);
4455 init_rwsem(&ctrl->namespaces_rwsem);
4458 ctrl->quirks = quirks;
4459 ctrl->numa_node = NUMA_NO_NODE;
4460 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4461 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4462 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4463 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4464 init_waitqueue_head(&ctrl->state_wq);
4466 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4467 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4468 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4469 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4471 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4473 ctrl->discard_page = alloc_page(GFP_KERNEL);
4474 if (!ctrl->discard_page) {
4479 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4482 ctrl->instance = ret;
4484 device_initialize(&ctrl->ctrl_device);
4485 ctrl->device = &ctrl->ctrl_device;
4486 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4488 ctrl->device->class = nvme_class;
4489 ctrl->device->parent = ctrl->dev;
4490 if (ops->dev_attr_groups)
4491 ctrl->device->groups = ops->dev_attr_groups;
4493 ctrl->device->groups = nvme_dev_attr_groups;
4494 ctrl->device->release = nvme_free_ctrl;
4495 dev_set_drvdata(ctrl->device, ctrl);
4496 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4498 goto out_release_instance;
4500 nvme_get_ctrl(ctrl);
4501 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4502 ctrl->cdev.owner = ops->module;
4503 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4508 * Initialize latency tolerance controls. The sysfs files won't
4509 * be visible to userspace unless the device actually supports APST.
4511 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4512 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4513 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4515 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4516 nvme_mpath_init_ctrl(ctrl);
4517 ret = nvme_auth_init_ctrl(ctrl);
4523 nvme_fault_inject_fini(&ctrl->fault_inject);
4524 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4525 cdev_device_del(&ctrl->cdev, ctrl->device);
4527 nvme_put_ctrl(ctrl);
4528 kfree_const(ctrl->device->kobj.name);
4529 out_release_instance:
4530 ida_free(&nvme_instance_ida, ctrl->instance);
4532 if (ctrl->discard_page)
4533 __free_page(ctrl->discard_page);
4536 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4538 /* let I/O to all namespaces fail in preparation for surprise removal */
4539 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4543 down_read(&ctrl->namespaces_rwsem);
4544 list_for_each_entry(ns, &ctrl->namespaces, list)
4545 blk_mark_disk_dead(ns->disk);
4546 up_read(&ctrl->namespaces_rwsem);
4548 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4550 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4554 down_read(&ctrl->namespaces_rwsem);
4555 list_for_each_entry(ns, &ctrl->namespaces, list)
4556 blk_mq_unfreeze_queue(ns->queue);
4557 up_read(&ctrl->namespaces_rwsem);
4559 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4561 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4565 down_read(&ctrl->namespaces_rwsem);
4566 list_for_each_entry(ns, &ctrl->namespaces, list) {
4567 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4571 up_read(&ctrl->namespaces_rwsem);
4574 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4576 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4580 down_read(&ctrl->namespaces_rwsem);
4581 list_for_each_entry(ns, &ctrl->namespaces, list)
4582 blk_mq_freeze_queue_wait(ns->queue);
4583 up_read(&ctrl->namespaces_rwsem);
4585 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4587 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4591 down_read(&ctrl->namespaces_rwsem);
4592 list_for_each_entry(ns, &ctrl->namespaces, list)
4593 blk_freeze_queue_start(ns->queue);
4594 up_read(&ctrl->namespaces_rwsem);
4596 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4598 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4602 if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4603 blk_mq_quiesce_tagset(ctrl->tagset);
4605 blk_mq_wait_quiesce_done(ctrl->tagset);
4607 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4609 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4613 if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4614 blk_mq_unquiesce_tagset(ctrl->tagset);
4616 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4618 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4620 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4621 blk_mq_quiesce_queue(ctrl->admin_q);
4623 blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4625 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4627 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4629 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4630 blk_mq_unquiesce_queue(ctrl->admin_q);
4632 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4634 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4638 down_read(&ctrl->namespaces_rwsem);
4639 list_for_each_entry(ns, &ctrl->namespaces, list)
4640 blk_sync_queue(ns->queue);
4641 up_read(&ctrl->namespaces_rwsem);
4643 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4645 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4647 nvme_sync_io_queues(ctrl);
4649 blk_sync_queue(ctrl->admin_q);
4651 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4653 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4655 if (file->f_op != &nvme_dev_fops)
4657 return file->private_data;
4659 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4662 * Check we didn't inadvertently grow the command structure sizes:
4664 static inline void _nvme_check_size(void)
4666 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4667 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4668 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4669 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4670 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4671 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4672 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4673 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4674 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4675 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4676 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4677 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4678 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4679 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4680 NVME_IDENTIFY_DATA_SIZE);
4681 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4682 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4683 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4684 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4685 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4686 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4687 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4688 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4689 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4693 static int __init nvme_core_init(void)
4695 int result = -ENOMEM;
4699 nvme_wq = alloc_workqueue("nvme-wq",
4700 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4704 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4705 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4709 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4710 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4711 if (!nvme_delete_wq)
4712 goto destroy_reset_wq;
4714 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4715 NVME_MINORS, "nvme");
4717 goto destroy_delete_wq;
4719 nvme_class = class_create("nvme");
4720 if (IS_ERR(nvme_class)) {
4721 result = PTR_ERR(nvme_class);
4722 goto unregister_chrdev;
4724 nvme_class->dev_uevent = nvme_class_uevent;
4726 nvme_subsys_class = class_create("nvme-subsystem");
4727 if (IS_ERR(nvme_subsys_class)) {
4728 result = PTR_ERR(nvme_subsys_class);
4732 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4735 goto destroy_subsys_class;
4737 nvme_ns_chr_class = class_create("nvme-generic");
4738 if (IS_ERR(nvme_ns_chr_class)) {
4739 result = PTR_ERR(nvme_ns_chr_class);
4740 goto unregister_generic_ns;
4743 result = nvme_init_auth();
4745 goto destroy_ns_chr;
4749 class_destroy(nvme_ns_chr_class);
4750 unregister_generic_ns:
4751 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4752 destroy_subsys_class:
4753 class_destroy(nvme_subsys_class);
4755 class_destroy(nvme_class);
4757 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4759 destroy_workqueue(nvme_delete_wq);
4761 destroy_workqueue(nvme_reset_wq);
4763 destroy_workqueue(nvme_wq);
4768 static void __exit nvme_core_exit(void)
4771 class_destroy(nvme_ns_chr_class);
4772 class_destroy(nvme_subsys_class);
4773 class_destroy(nvme_class);
4774 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4775 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4776 destroy_workqueue(nvme_delete_wq);
4777 destroy_workqueue(nvme_reset_wq);
4778 destroy_workqueue(nvme_wq);
4779 ida_destroy(&nvme_ns_chr_minor_ida);
4780 ida_destroy(&nvme_instance_ida);
4783 MODULE_LICENSE("GPL");
4784 MODULE_VERSION("1.0");
4785 module_init(nvme_core_init);
4786 module_exit(nvme_core_exit);