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>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
60 static unsigned long apst_primary_timeout_ms = 100;
61 module_param(apst_primary_timeout_ms, ulong, 0644);
62 MODULE_PARM_DESC(apst_primary_timeout_ms,
63 "primary APST timeout in ms");
65 static unsigned long apst_secondary_timeout_ms = 2000;
66 module_param(apst_secondary_timeout_ms, ulong, 0644);
67 MODULE_PARM_DESC(apst_secondary_timeout_ms,
68 "secondary APST timeout in ms");
70 static unsigned long apst_primary_latency_tol_us = 15000;
71 module_param(apst_primary_latency_tol_us, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_latency_tol_us,
73 "primary APST latency tolerance in us");
75 static unsigned long apst_secondary_latency_tol_us = 100000;
76 module_param(apst_secondary_latency_tol_us, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
78 "secondary APST latency tolerance in us");
81 module_param(streams, bool, 0644);
82 MODULE_PARM_DESC(streams, "turn on support for Streams write directives");
85 * nvme_wq - hosts nvme related works that are not reset or delete
86 * nvme_reset_wq - hosts nvme reset works
87 * nvme_delete_wq - hosts nvme delete works
89 * nvme_wq will host works such as scan, aen handling, fw activation,
90 * keep-alive, periodic reconnects etc. nvme_reset_wq
91 * runs reset works which also flush works hosted on nvme_wq for
92 * serialization purposes. nvme_delete_wq host controller deletion
93 * works which flush reset works for serialization.
95 struct workqueue_struct *nvme_wq;
96 EXPORT_SYMBOL_GPL(nvme_wq);
98 struct workqueue_struct *nvme_reset_wq;
99 EXPORT_SYMBOL_GPL(nvme_reset_wq);
101 struct workqueue_struct *nvme_delete_wq;
102 EXPORT_SYMBOL_GPL(nvme_delete_wq);
104 static LIST_HEAD(nvme_subsystems);
105 static DEFINE_MUTEX(nvme_subsystems_lock);
107 static DEFINE_IDA(nvme_instance_ida);
108 static dev_t nvme_ctrl_base_chr_devt;
109 static struct class *nvme_class;
110 static struct class *nvme_subsys_class;
112 static DEFINE_IDA(nvme_ns_chr_minor_ida);
113 static dev_t nvme_ns_chr_devt;
114 static struct class *nvme_ns_chr_class;
116 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
117 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
119 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
120 struct nvme_command *cmd);
122 void nvme_queue_scan(struct nvme_ctrl *ctrl)
125 * Only new queue scan work when admin and IO queues are both alive
127 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
128 queue_work(nvme_wq, &ctrl->scan_work);
132 * Use this function to proceed with scheduling reset_work for a controller
133 * that had previously been set to the resetting state. This is intended for
134 * code paths that can't be interrupted by other reset attempts. A hot removal
135 * may prevent this from succeeding.
137 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
139 if (ctrl->state != NVME_CTRL_RESETTING)
141 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
145 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
147 static void nvme_failfast_work(struct work_struct *work)
149 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
150 struct nvme_ctrl, failfast_work);
152 if (ctrl->state != NVME_CTRL_CONNECTING)
155 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
156 dev_info(ctrl->device, "failfast expired\n");
157 nvme_kick_requeue_lists(ctrl);
160 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
162 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
165 schedule_delayed_work(&ctrl->failfast_work,
166 ctrl->opts->fast_io_fail_tmo * HZ);
169 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
174 cancel_delayed_work_sync(&ctrl->failfast_work);
175 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
179 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
181 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
183 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
187 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
189 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
193 ret = nvme_reset_ctrl(ctrl);
195 flush_work(&ctrl->reset_work);
196 if (ctrl->state != NVME_CTRL_LIVE)
203 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
205 dev_info(ctrl->device,
206 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
208 flush_work(&ctrl->reset_work);
209 nvme_stop_ctrl(ctrl);
210 nvme_remove_namespaces(ctrl);
211 ctrl->ops->delete_ctrl(ctrl);
212 nvme_uninit_ctrl(ctrl);
215 static void nvme_delete_ctrl_work(struct work_struct *work)
217 struct nvme_ctrl *ctrl =
218 container_of(work, struct nvme_ctrl, delete_work);
220 nvme_do_delete_ctrl(ctrl);
223 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
225 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
227 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
231 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
233 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
236 * Keep a reference until nvme_do_delete_ctrl() complete,
237 * since ->delete_ctrl can free the controller.
240 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
241 nvme_do_delete_ctrl(ctrl);
245 static blk_status_t nvme_error_status(u16 status)
247 switch (status & 0x7ff) {
248 case NVME_SC_SUCCESS:
250 case NVME_SC_CAP_EXCEEDED:
251 return BLK_STS_NOSPC;
252 case NVME_SC_LBA_RANGE:
253 case NVME_SC_CMD_INTERRUPTED:
254 case NVME_SC_NS_NOT_READY:
255 return BLK_STS_TARGET;
256 case NVME_SC_BAD_ATTRIBUTES:
257 case NVME_SC_ONCS_NOT_SUPPORTED:
258 case NVME_SC_INVALID_OPCODE:
259 case NVME_SC_INVALID_FIELD:
260 case NVME_SC_INVALID_NS:
261 return BLK_STS_NOTSUPP;
262 case NVME_SC_WRITE_FAULT:
263 case NVME_SC_READ_ERROR:
264 case NVME_SC_UNWRITTEN_BLOCK:
265 case NVME_SC_ACCESS_DENIED:
266 case NVME_SC_READ_ONLY:
267 case NVME_SC_COMPARE_FAILED:
268 return BLK_STS_MEDIUM;
269 case NVME_SC_GUARD_CHECK:
270 case NVME_SC_APPTAG_CHECK:
271 case NVME_SC_REFTAG_CHECK:
272 case NVME_SC_INVALID_PI:
273 return BLK_STS_PROTECTION;
274 case NVME_SC_RESERVATION_CONFLICT:
275 return BLK_STS_NEXUS;
276 case NVME_SC_HOST_PATH_ERROR:
277 return BLK_STS_TRANSPORT;
278 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
279 return BLK_STS_ZONE_ACTIVE_RESOURCE;
280 case NVME_SC_ZONE_TOO_MANY_OPEN:
281 return BLK_STS_ZONE_OPEN_RESOURCE;
283 return BLK_STS_IOERR;
287 static void nvme_retry_req(struct request *req)
289 unsigned long delay = 0;
292 /* The mask and shift result must be <= 3 */
293 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
295 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
297 nvme_req(req)->retries++;
298 blk_mq_requeue_request(req, false);
299 blk_mq_delay_kick_requeue_list(req->q, delay);
302 enum nvme_disposition {
308 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
310 if (likely(nvme_req(req)->status == 0))
313 if (blk_noretry_request(req) ||
314 (nvme_req(req)->status & NVME_SC_DNR) ||
315 nvme_req(req)->retries >= nvme_max_retries)
318 if (req->cmd_flags & REQ_NVME_MPATH) {
319 if (nvme_is_path_error(nvme_req(req)->status) ||
320 blk_queue_dying(req->q))
323 if (blk_queue_dying(req->q))
330 static inline void nvme_end_req_zoned(struct request *req)
332 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
333 req_op(req) == REQ_OP_ZONE_APPEND)
334 req->__sector = nvme_lba_to_sect(req->q->queuedata,
335 le64_to_cpu(nvme_req(req)->result.u64));
338 static inline void nvme_end_req(struct request *req)
340 blk_status_t status = nvme_error_status(nvme_req(req)->status);
342 nvme_end_req_zoned(req);
343 nvme_trace_bio_complete(req);
344 blk_mq_end_request(req, status);
347 void nvme_complete_rq(struct request *req)
349 trace_nvme_complete_rq(req);
350 nvme_cleanup_cmd(req);
352 if (nvme_req(req)->ctrl->kas)
353 nvme_req(req)->ctrl->comp_seen = true;
355 switch (nvme_decide_disposition(req)) {
363 nvme_failover_req(req);
367 EXPORT_SYMBOL_GPL(nvme_complete_rq);
369 void nvme_complete_batch_req(struct request *req)
371 nvme_cleanup_cmd(req);
372 nvme_end_req_zoned(req);
374 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
377 * Called to unwind from ->queue_rq on a failed command submission so that the
378 * multipathing code gets called to potentially failover to another path.
379 * The caller needs to unwind all transport specific resource allocations and
380 * must return propagate the return value.
382 blk_status_t nvme_host_path_error(struct request *req)
384 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
385 blk_mq_set_request_complete(req);
386 nvme_complete_rq(req);
389 EXPORT_SYMBOL_GPL(nvme_host_path_error);
391 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
393 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
394 "Cancelling I/O %d", req->tag);
396 /* don't abort one completed request */
397 if (blk_mq_request_completed(req))
400 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
401 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
402 blk_mq_complete_request(req);
405 EXPORT_SYMBOL_GPL(nvme_cancel_request);
407 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
410 blk_mq_tagset_busy_iter(ctrl->tagset,
411 nvme_cancel_request, ctrl);
412 blk_mq_tagset_wait_completed_request(ctrl->tagset);
415 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
417 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
419 if (ctrl->admin_tagset) {
420 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
421 nvme_cancel_request, ctrl);
422 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
425 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
427 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
428 enum nvme_ctrl_state new_state)
430 enum nvme_ctrl_state old_state;
432 bool changed = false;
434 spin_lock_irqsave(&ctrl->lock, flags);
436 old_state = ctrl->state;
441 case NVME_CTRL_RESETTING:
442 case NVME_CTRL_CONNECTING:
449 case NVME_CTRL_RESETTING:
459 case NVME_CTRL_CONNECTING:
462 case NVME_CTRL_RESETTING:
469 case NVME_CTRL_DELETING:
472 case NVME_CTRL_RESETTING:
473 case NVME_CTRL_CONNECTING:
480 case NVME_CTRL_DELETING_NOIO:
482 case NVME_CTRL_DELETING:
492 case NVME_CTRL_DELETING:
504 ctrl->state = new_state;
505 wake_up_all(&ctrl->state_wq);
508 spin_unlock_irqrestore(&ctrl->lock, flags);
512 if (ctrl->state == NVME_CTRL_LIVE) {
513 if (old_state == NVME_CTRL_CONNECTING)
514 nvme_stop_failfast_work(ctrl);
515 nvme_kick_requeue_lists(ctrl);
516 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
517 old_state == NVME_CTRL_RESETTING) {
518 nvme_start_failfast_work(ctrl);
522 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
525 * Returns true for sink states that can't ever transition back to live.
527 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
529 switch (ctrl->state) {
532 case NVME_CTRL_RESETTING:
533 case NVME_CTRL_CONNECTING:
535 case NVME_CTRL_DELETING:
536 case NVME_CTRL_DELETING_NOIO:
540 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
546 * Waits for the controller state to be resetting, or returns false if it is
547 * not possible to ever transition to that state.
549 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
551 wait_event(ctrl->state_wq,
552 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
553 nvme_state_terminal(ctrl));
554 return ctrl->state == NVME_CTRL_RESETTING;
556 EXPORT_SYMBOL_GPL(nvme_wait_reset);
558 static void nvme_free_ns_head(struct kref *ref)
560 struct nvme_ns_head *head =
561 container_of(ref, struct nvme_ns_head, ref);
563 nvme_mpath_remove_disk(head);
564 ida_simple_remove(&head->subsys->ns_ida, head->instance);
565 cleanup_srcu_struct(&head->srcu);
566 nvme_put_subsystem(head->subsys);
570 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
572 return kref_get_unless_zero(&head->ref);
575 void nvme_put_ns_head(struct nvme_ns_head *head)
577 kref_put(&head->ref, nvme_free_ns_head);
580 static void nvme_free_ns(struct kref *kref)
582 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
585 nvme_put_ns_head(ns->head);
586 nvme_put_ctrl(ns->ctrl);
590 static inline bool nvme_get_ns(struct nvme_ns *ns)
592 return kref_get_unless_zero(&ns->kref);
595 void nvme_put_ns(struct nvme_ns *ns)
597 kref_put(&ns->kref, nvme_free_ns);
599 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
601 static inline void nvme_clear_nvme_request(struct request *req)
603 nvme_req(req)->status = 0;
604 nvme_req(req)->retries = 0;
605 nvme_req(req)->flags = 0;
606 req->rq_flags |= RQF_DONTPREP;
609 static inline unsigned int nvme_req_op(struct nvme_command *cmd)
611 return nvme_is_write(cmd) ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN;
614 static inline void nvme_init_request(struct request *req,
615 struct nvme_command *cmd)
617 if (req->q->queuedata)
618 req->timeout = NVME_IO_TIMEOUT;
619 else /* no queuedata implies admin queue */
620 req->timeout = NVME_ADMIN_TIMEOUT;
622 /* passthru commands should let the driver set the SGL flags */
623 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
625 req->cmd_flags |= REQ_FAILFAST_DRIVER;
626 if (req->mq_hctx->type == HCTX_TYPE_POLL)
627 req->cmd_flags |= REQ_POLLED;
628 nvme_clear_nvme_request(req);
629 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
632 struct request *nvme_alloc_request(struct request_queue *q,
633 struct nvme_command *cmd, blk_mq_req_flags_t flags)
637 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
639 nvme_init_request(req, cmd);
642 EXPORT_SYMBOL_GPL(nvme_alloc_request);
644 static struct request *nvme_alloc_request_qid(struct request_queue *q,
645 struct nvme_command *cmd, blk_mq_req_flags_t flags, int qid)
649 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
652 nvme_init_request(req, cmd);
657 * For something we're not in a state to send to the device the default action
658 * is to busy it and retry it after the controller state is recovered. However,
659 * if the controller is deleting or if anything is marked for failfast or
660 * nvme multipath it is immediately failed.
662 * Note: commands used to initialize the controller will be marked for failfast.
663 * Note: nvme cli/ioctl commands are marked for failfast.
665 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
668 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
669 ctrl->state != NVME_CTRL_DELETING &&
670 ctrl->state != NVME_CTRL_DEAD &&
671 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
672 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
673 return BLK_STS_RESOURCE;
674 return nvme_host_path_error(rq);
676 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
678 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
681 struct nvme_request *req = nvme_req(rq);
684 * currently we have a problem sending passthru commands
685 * on the admin_q if the controller is not LIVE because we can't
686 * make sure that they are going out after the admin connect,
687 * controller enable and/or other commands in the initialization
688 * sequence. until the controller will be LIVE, fail with
689 * BLK_STS_RESOURCE so that they will be rescheduled.
691 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
694 if (ctrl->ops->flags & NVME_F_FABRICS) {
696 * Only allow commands on a live queue, except for the connect
697 * command, which is require to set the queue live in the
698 * appropinquate states.
700 switch (ctrl->state) {
701 case NVME_CTRL_CONNECTING:
702 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
703 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
715 EXPORT_SYMBOL_GPL(__nvme_check_ready);
717 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
719 struct nvme_command c = { };
721 c.directive.opcode = nvme_admin_directive_send;
722 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
723 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
724 c.directive.dtype = NVME_DIR_IDENTIFY;
725 c.directive.tdtype = NVME_DIR_STREAMS;
726 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
728 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
731 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
733 return nvme_toggle_streams(ctrl, false);
736 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
738 return nvme_toggle_streams(ctrl, true);
741 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
742 struct streams_directive_params *s, u32 nsid)
744 struct nvme_command c = { };
746 memset(s, 0, sizeof(*s));
748 c.directive.opcode = nvme_admin_directive_recv;
749 c.directive.nsid = cpu_to_le32(nsid);
750 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
751 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
752 c.directive.dtype = NVME_DIR_STREAMS;
754 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
757 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
759 struct streams_directive_params s;
762 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
767 ret = nvme_enable_streams(ctrl);
771 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
773 goto out_disable_stream;
775 ctrl->nssa = le16_to_cpu(s.nssa);
776 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
777 dev_info(ctrl->device, "too few streams (%u) available\n",
779 goto out_disable_stream;
782 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
783 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
787 nvme_disable_streams(ctrl);
792 * Check if 'req' has a write hint associated with it. If it does, assign
793 * a valid namespace stream to the write.
795 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
796 struct request *req, u16 *control,
799 enum rw_hint streamid = req->write_hint;
801 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
805 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
808 *control |= NVME_RW_DTYPE_STREAMS;
809 *dsmgmt |= streamid << 16;
812 if (streamid < ARRAY_SIZE(req->q->write_hints))
813 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
816 static inline void nvme_setup_flush(struct nvme_ns *ns,
817 struct nvme_command *cmnd)
819 memset(cmnd, 0, sizeof(*cmnd));
820 cmnd->common.opcode = nvme_cmd_flush;
821 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
824 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
825 struct nvme_command *cmnd)
827 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
828 struct nvme_dsm_range *range;
832 * Some devices do not consider the DSM 'Number of Ranges' field when
833 * determining how much data to DMA. Always allocate memory for maximum
834 * number of segments to prevent device reading beyond end of buffer.
836 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
838 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
841 * If we fail allocation our range, fallback to the controller
842 * discard page. If that's also busy, it's safe to return
843 * busy, as we know we can make progress once that's freed.
845 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
846 return BLK_STS_RESOURCE;
848 range = page_address(ns->ctrl->discard_page);
851 __rq_for_each_bio(bio, req) {
852 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
853 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
856 range[n].cattr = cpu_to_le32(0);
857 range[n].nlb = cpu_to_le32(nlb);
858 range[n].slba = cpu_to_le64(slba);
863 if (WARN_ON_ONCE(n != segments)) {
864 if (virt_to_page(range) == ns->ctrl->discard_page)
865 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
868 return BLK_STS_IOERR;
871 memset(cmnd, 0, sizeof(*cmnd));
872 cmnd->dsm.opcode = nvme_cmd_dsm;
873 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
874 cmnd->dsm.nr = cpu_to_le32(segments - 1);
875 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
877 req->special_vec.bv_page = virt_to_page(range);
878 req->special_vec.bv_offset = offset_in_page(range);
879 req->special_vec.bv_len = alloc_size;
880 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
885 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
886 struct request *req, struct nvme_command *cmnd)
888 memset(cmnd, 0, sizeof(*cmnd));
890 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
891 return nvme_setup_discard(ns, req, cmnd);
893 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
894 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
895 cmnd->write_zeroes.slba =
896 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
897 cmnd->write_zeroes.length =
898 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
900 if (nvme_ns_has_pi(ns)) {
901 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
903 switch (ns->pi_type) {
904 case NVME_NS_DPS_PI_TYPE1:
905 case NVME_NS_DPS_PI_TYPE2:
906 cmnd->write_zeroes.reftag =
907 cpu_to_le32(t10_pi_ref_tag(req));
915 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
916 struct request *req, struct nvme_command *cmnd,
919 struct nvme_ctrl *ctrl = ns->ctrl;
923 if (req->cmd_flags & REQ_FUA)
924 control |= NVME_RW_FUA;
925 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
926 control |= NVME_RW_LR;
928 if (req->cmd_flags & REQ_RAHEAD)
929 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
931 cmnd->rw.opcode = op;
933 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
935 cmnd->rw.metadata = 0;
936 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
937 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
940 cmnd->rw.appmask = 0;
942 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
943 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
947 * If formated with metadata, the block layer always provides a
948 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
949 * we enable the PRACT bit for protection information or set the
950 * namespace capacity to zero to prevent any I/O.
952 if (!blk_integrity_rq(req)) {
953 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
954 return BLK_STS_NOTSUPP;
955 control |= NVME_RW_PRINFO_PRACT;
958 switch (ns->pi_type) {
959 case NVME_NS_DPS_PI_TYPE3:
960 control |= NVME_RW_PRINFO_PRCHK_GUARD;
962 case NVME_NS_DPS_PI_TYPE1:
963 case NVME_NS_DPS_PI_TYPE2:
964 control |= NVME_RW_PRINFO_PRCHK_GUARD |
965 NVME_RW_PRINFO_PRCHK_REF;
966 if (op == nvme_cmd_zone_append)
967 control |= NVME_RW_APPEND_PIREMAP;
968 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
973 cmnd->rw.control = cpu_to_le16(control);
974 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
978 void nvme_cleanup_cmd(struct request *req)
980 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
981 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
983 if (req->special_vec.bv_page == ctrl->discard_page)
984 clear_bit_unlock(0, &ctrl->discard_page_busy);
986 kfree(bvec_virt(&req->special_vec));
989 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
991 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
993 struct nvme_command *cmd = nvme_req(req)->cmd;
994 blk_status_t ret = BLK_STS_OK;
996 if (!(req->rq_flags & RQF_DONTPREP))
997 nvme_clear_nvme_request(req);
999 switch (req_op(req)) {
1001 case REQ_OP_DRV_OUT:
1002 /* these are setup prior to execution in nvme_init_request() */
1005 nvme_setup_flush(ns, cmd);
1007 case REQ_OP_ZONE_RESET_ALL:
1008 case REQ_OP_ZONE_RESET:
1009 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1011 case REQ_OP_ZONE_OPEN:
1012 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1014 case REQ_OP_ZONE_CLOSE:
1015 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1017 case REQ_OP_ZONE_FINISH:
1018 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1020 case REQ_OP_WRITE_ZEROES:
1021 ret = nvme_setup_write_zeroes(ns, req, cmd);
1023 case REQ_OP_DISCARD:
1024 ret = nvme_setup_discard(ns, req, cmd);
1027 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1030 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1032 case REQ_OP_ZONE_APPEND:
1033 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1037 return BLK_STS_IOERR;
1040 cmd->common.command_id = nvme_cid(req);
1041 trace_nvme_setup_cmd(req, cmd);
1044 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1049 * >0: nvme controller's cqe status response
1050 * <0: kernel error in lieu of controller response
1052 static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1055 blk_status_t status;
1057 status = blk_execute_rq(rq, at_head);
1058 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1060 if (nvme_req(rq)->status)
1061 return nvme_req(rq)->status;
1062 return blk_status_to_errno(status);
1066 * Returns 0 on success. If the result is negative, it's a Linux error code;
1067 * if the result is positive, it's an NVM Express status code
1069 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1070 union nvme_result *result, void *buffer, unsigned bufflen,
1071 unsigned timeout, int qid, int at_head,
1072 blk_mq_req_flags_t flags)
1074 struct request *req;
1077 if (qid == NVME_QID_ANY)
1078 req = nvme_alloc_request(q, cmd, flags);
1080 req = nvme_alloc_request_qid(q, cmd, flags, qid);
1082 return PTR_ERR(req);
1085 req->timeout = timeout;
1087 if (buffer && bufflen) {
1088 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1093 ret = nvme_execute_rq(NULL, req, at_head);
1094 if (result && ret >= 0)
1095 *result = nvme_req(req)->result;
1097 blk_mq_free_request(req);
1100 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1102 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1103 void *buffer, unsigned bufflen)
1105 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1106 NVME_QID_ANY, 0, 0);
1108 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1110 static u32 nvme_known_admin_effects(u8 opcode)
1113 case nvme_admin_format_nvm:
1114 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1115 NVME_CMD_EFFECTS_CSE_MASK;
1116 case nvme_admin_sanitize_nvm:
1117 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1124 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1129 if (ns->head->effects)
1130 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1131 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1132 dev_warn_once(ctrl->device,
1133 "IO command:%02x has unhandled effects:%08x\n",
1139 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1140 effects |= nvme_known_admin_effects(opcode);
1144 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1146 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1149 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1152 * For simplicity, IO to all namespaces is quiesced even if the command
1153 * effects say only one namespace is affected.
1155 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1156 mutex_lock(&ctrl->scan_lock);
1157 mutex_lock(&ctrl->subsys->lock);
1158 nvme_mpath_start_freeze(ctrl->subsys);
1159 nvme_mpath_wait_freeze(ctrl->subsys);
1160 nvme_start_freeze(ctrl);
1161 nvme_wait_freeze(ctrl);
1166 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1167 struct nvme_command *cmd, int status)
1169 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1170 nvme_unfreeze(ctrl);
1171 nvme_mpath_unfreeze(ctrl->subsys);
1172 mutex_unlock(&ctrl->subsys->lock);
1173 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1174 mutex_unlock(&ctrl->scan_lock);
1176 if (effects & NVME_CMD_EFFECTS_CCC)
1177 nvme_init_ctrl_finish(ctrl);
1178 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1179 nvme_queue_scan(ctrl);
1180 flush_work(&ctrl->scan_work);
1183 switch (cmd->common.opcode) {
1184 case nvme_admin_set_features:
1185 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1186 case NVME_FEAT_KATO:
1188 * Keep alive commands interval on the host should be
1189 * updated when KATO is modified by Set Features
1193 nvme_update_keep_alive(ctrl, cmd);
1204 int nvme_execute_passthru_rq(struct request *rq)
1206 struct nvme_command *cmd = nvme_req(rq)->cmd;
1207 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1208 struct nvme_ns *ns = rq->q->queuedata;
1209 struct gendisk *disk = ns ? ns->disk : NULL;
1213 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1214 ret = nvme_execute_rq(disk, rq, false);
1215 if (effects) /* nothing to be done for zero cmd effects */
1216 nvme_passthru_end(ctrl, effects, cmd, ret);
1220 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1223 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1225 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1226 * accounting for transport roundtrip times [..].
1228 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1230 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1233 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1235 struct nvme_ctrl *ctrl = rq->end_io_data;
1236 unsigned long flags;
1237 bool startka = false;
1239 blk_mq_free_request(rq);
1242 dev_err(ctrl->device,
1243 "failed nvme_keep_alive_end_io error=%d\n",
1248 ctrl->comp_seen = false;
1249 spin_lock_irqsave(&ctrl->lock, flags);
1250 if (ctrl->state == NVME_CTRL_LIVE ||
1251 ctrl->state == NVME_CTRL_CONNECTING)
1253 spin_unlock_irqrestore(&ctrl->lock, flags);
1255 nvme_queue_keep_alive_work(ctrl);
1258 static void nvme_keep_alive_work(struct work_struct *work)
1260 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1261 struct nvme_ctrl, ka_work);
1262 bool comp_seen = ctrl->comp_seen;
1265 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1266 dev_dbg(ctrl->device,
1267 "reschedule traffic based keep-alive timer\n");
1268 ctrl->comp_seen = false;
1269 nvme_queue_keep_alive_work(ctrl);
1273 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1274 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1276 /* allocation failure, reset the controller */
1277 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1278 nvme_reset_ctrl(ctrl);
1282 rq->timeout = ctrl->kato * HZ;
1283 rq->end_io_data = ctrl;
1284 blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1287 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1289 if (unlikely(ctrl->kato == 0))
1292 nvme_queue_keep_alive_work(ctrl);
1295 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1297 if (unlikely(ctrl->kato == 0))
1300 cancel_delayed_work_sync(&ctrl->ka_work);
1302 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1304 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1305 struct nvme_command *cmd)
1307 unsigned int new_kato =
1308 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1310 dev_info(ctrl->device,
1311 "keep alive interval updated from %u ms to %u ms\n",
1312 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1314 nvme_stop_keep_alive(ctrl);
1315 ctrl->kato = new_kato;
1316 nvme_start_keep_alive(ctrl);
1320 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1321 * flag, thus sending any new CNS opcodes has a big chance of not working.
1322 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1323 * (but not for any later version).
1325 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1327 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1328 return ctrl->vs < NVME_VS(1, 2, 0);
1329 return ctrl->vs < NVME_VS(1, 1, 0);
1332 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1334 struct nvme_command c = { };
1337 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1338 c.identify.opcode = nvme_admin_identify;
1339 c.identify.cns = NVME_ID_CNS_CTRL;
1341 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1345 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1346 sizeof(struct nvme_id_ctrl));
1352 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1353 struct nvme_ns_id_desc *cur, bool *csi_seen)
1355 const char *warn_str = "ctrl returned bogus length:";
1358 switch (cur->nidt) {
1359 case NVME_NIDT_EUI64:
1360 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1361 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1362 warn_str, cur->nidl);
1365 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1366 return NVME_NIDT_EUI64_LEN;
1367 case NVME_NIDT_NGUID:
1368 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1369 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1370 warn_str, cur->nidl);
1373 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1374 return NVME_NIDT_NGUID_LEN;
1375 case NVME_NIDT_UUID:
1376 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1377 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1378 warn_str, cur->nidl);
1381 uuid_copy(&ids->uuid, data + sizeof(*cur));
1382 return NVME_NIDT_UUID_LEN;
1384 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1385 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1386 warn_str, cur->nidl);
1389 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1391 return NVME_NIDT_CSI_LEN;
1393 /* Skip unknown types */
1398 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1399 struct nvme_ns_ids *ids)
1401 struct nvme_command c = { };
1402 bool csi_seen = false;
1403 int status, pos, len;
1406 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1408 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1411 c.identify.opcode = nvme_admin_identify;
1412 c.identify.nsid = cpu_to_le32(nsid);
1413 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1415 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1419 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1420 NVME_IDENTIFY_DATA_SIZE);
1422 dev_warn(ctrl->device,
1423 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1428 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1429 struct nvme_ns_id_desc *cur = data + pos;
1434 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1438 len += sizeof(*cur);
1441 if (nvme_multi_css(ctrl) && !csi_seen) {
1442 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1452 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1453 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1455 struct nvme_command c = { };
1458 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1459 c.identify.opcode = nvme_admin_identify;
1460 c.identify.nsid = cpu_to_le32(nsid);
1461 c.identify.cns = NVME_ID_CNS_NS;
1463 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1467 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1469 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1473 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1474 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1477 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1478 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1479 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1480 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1481 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1482 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1491 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1492 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1494 union nvme_result res = { 0 };
1495 struct nvme_command c = { };
1498 c.features.opcode = op;
1499 c.features.fid = cpu_to_le32(fid);
1500 c.features.dword11 = cpu_to_le32(dword11);
1502 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1503 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1504 if (ret >= 0 && result)
1505 *result = le32_to_cpu(res.u32);
1509 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1510 unsigned int dword11, void *buffer, size_t buflen,
1513 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1516 EXPORT_SYMBOL_GPL(nvme_set_features);
1518 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1519 unsigned int dword11, void *buffer, size_t buflen,
1522 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1525 EXPORT_SYMBOL_GPL(nvme_get_features);
1527 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1529 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1531 int status, nr_io_queues;
1533 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1539 * Degraded controllers might return an error when setting the queue
1540 * count. We still want to be able to bring them online and offer
1541 * access to the admin queue, as that might be only way to fix them up.
1544 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1547 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1548 *count = min(*count, nr_io_queues);
1553 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1555 #define NVME_AEN_SUPPORTED \
1556 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1557 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1559 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1561 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1564 if (!supported_aens)
1567 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1570 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1573 queue_work(nvme_wq, &ctrl->async_event_work);
1576 static int nvme_ns_open(struct nvme_ns *ns)
1579 /* should never be called due to GENHD_FL_HIDDEN */
1580 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1582 if (!nvme_get_ns(ns))
1584 if (!try_module_get(ns->ctrl->ops->module))
1595 static void nvme_ns_release(struct nvme_ns *ns)
1598 module_put(ns->ctrl->ops->module);
1602 static int nvme_open(struct block_device *bdev, fmode_t mode)
1604 return nvme_ns_open(bdev->bd_disk->private_data);
1607 static void nvme_release(struct gendisk *disk, fmode_t mode)
1609 nvme_ns_release(disk->private_data);
1612 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1614 /* some standard values */
1615 geo->heads = 1 << 6;
1616 geo->sectors = 1 << 5;
1617 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1621 #ifdef CONFIG_BLK_DEV_INTEGRITY
1622 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1623 u32 max_integrity_segments)
1625 struct blk_integrity integrity = { };
1628 case NVME_NS_DPS_PI_TYPE3:
1629 integrity.profile = &t10_pi_type3_crc;
1630 integrity.tag_size = sizeof(u16) + sizeof(u32);
1631 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1633 case NVME_NS_DPS_PI_TYPE1:
1634 case NVME_NS_DPS_PI_TYPE2:
1635 integrity.profile = &t10_pi_type1_crc;
1636 integrity.tag_size = sizeof(u16);
1637 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1640 integrity.profile = NULL;
1643 integrity.tuple_size = ms;
1644 blk_integrity_register(disk, &integrity);
1645 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1648 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1649 u32 max_integrity_segments)
1652 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1654 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1656 struct nvme_ctrl *ctrl = ns->ctrl;
1657 struct request_queue *queue = disk->queue;
1658 u32 size = queue_logical_block_size(queue);
1660 if (ctrl->max_discard_sectors == 0) {
1661 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1665 if (ctrl->nr_streams && ns->sws && ns->sgs)
1666 size *= ns->sws * ns->sgs;
1668 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1669 NVME_DSM_MAX_RANGES);
1671 queue->limits.discard_alignment = 0;
1672 queue->limits.discard_granularity = size;
1674 /* If discard is already enabled, don't reset queue limits */
1675 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1678 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1679 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1681 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1682 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1685 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1687 return !uuid_is_null(&ids->uuid) ||
1688 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1689 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1692 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1694 return uuid_equal(&a->uuid, &b->uuid) &&
1695 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1696 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1700 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1701 u32 *phys_bs, u32 *io_opt)
1703 struct streams_directive_params s;
1706 if (!ctrl->nr_streams)
1709 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1713 ns->sws = le32_to_cpu(s.sws);
1714 ns->sgs = le16_to_cpu(s.sgs);
1717 *phys_bs = ns->sws * (1 << ns->lba_shift);
1719 *io_opt = *phys_bs * ns->sgs;
1725 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1727 struct nvme_ctrl *ctrl = ns->ctrl;
1730 * The PI implementation requires the metadata size to be equal to the
1731 * t10 pi tuple size.
1733 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1734 if (ns->ms == sizeof(struct t10_pi_tuple))
1735 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1739 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1740 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1742 if (ctrl->ops->flags & NVME_F_FABRICS) {
1744 * The NVMe over Fabrics specification only supports metadata as
1745 * part of the extended data LBA. We rely on HCA/HBA support to
1746 * remap the separate metadata buffer from the block layer.
1748 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1751 ns->features |= NVME_NS_EXT_LBAS;
1754 * The current fabrics transport drivers support namespace
1755 * metadata formats only if nvme_ns_has_pi() returns true.
1756 * Suppress support for all other formats so the namespace will
1757 * have a 0 capacity and not be usable through the block stack.
1759 * Note, this check will need to be modified if any drivers
1760 * gain the ability to use other metadata formats.
1762 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1763 ns->features |= NVME_NS_METADATA_SUPPORTED;
1766 * For PCIe controllers, we can't easily remap the separate
1767 * metadata buffer from the block layer and thus require a
1768 * separate metadata buffer for block layer metadata/PI support.
1769 * We allow extended LBAs for the passthrough interface, though.
1771 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1772 ns->features |= NVME_NS_EXT_LBAS;
1774 ns->features |= NVME_NS_METADATA_SUPPORTED;
1780 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1781 struct request_queue *q)
1783 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1785 if (ctrl->max_hw_sectors) {
1787 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1789 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1790 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1791 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1793 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1794 blk_queue_dma_alignment(q, 7);
1795 blk_queue_write_cache(q, vwc, vwc);
1798 static void nvme_update_disk_info(struct gendisk *disk,
1799 struct nvme_ns *ns, struct nvme_id_ns *id)
1801 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1802 unsigned short bs = 1 << ns->lba_shift;
1803 u32 atomic_bs, phys_bs, io_opt = 0;
1806 * The block layer can't support LBA sizes larger than the page size
1807 * yet, so catch this early and don't allow block I/O.
1809 if (ns->lba_shift > PAGE_SHIFT) {
1814 blk_integrity_unregister(disk);
1816 atomic_bs = phys_bs = bs;
1817 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1818 if (id->nabo == 0) {
1820 * Bit 1 indicates whether NAWUPF is defined for this namespace
1821 * and whether it should be used instead of AWUPF. If NAWUPF ==
1822 * 0 then AWUPF must be used instead.
1824 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1825 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1827 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1830 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1831 /* NPWG = Namespace Preferred Write Granularity */
1832 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1833 /* NOWS = Namespace Optimal Write Size */
1834 io_opt = bs * (1 + le16_to_cpu(id->nows));
1837 blk_queue_logical_block_size(disk->queue, bs);
1839 * Linux filesystems assume writing a single physical block is
1840 * an atomic operation. Hence limit the physical block size to the
1841 * value of the Atomic Write Unit Power Fail parameter.
1843 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1844 blk_queue_io_min(disk->queue, phys_bs);
1845 blk_queue_io_opt(disk->queue, io_opt);
1848 * Register a metadata profile for PI, or the plain non-integrity NVMe
1849 * metadata masquerading as Type 0 if supported, otherwise reject block
1850 * I/O to namespaces with metadata except when the namespace supports
1851 * PI, as it can strip/insert in that case.
1854 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1855 (ns->features & NVME_NS_METADATA_SUPPORTED))
1856 nvme_init_integrity(disk, ns->ms, ns->pi_type,
1857 ns->ctrl->max_integrity_segments);
1858 else if (!nvme_ns_has_pi(ns))
1862 set_capacity_and_notify(disk, capacity);
1864 nvme_config_discard(disk, ns);
1865 blk_queue_max_write_zeroes_sectors(disk->queue,
1866 ns->ctrl->max_zeroes_sectors);
1868 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1869 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1872 static inline bool nvme_first_scan(struct gendisk *disk)
1874 /* nvme_alloc_ns() scans the disk prior to adding it */
1875 return !disk_live(disk);
1878 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1880 struct nvme_ctrl *ctrl = ns->ctrl;
1883 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1884 is_power_of_2(ctrl->max_hw_sectors))
1885 iob = ctrl->max_hw_sectors;
1887 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1892 if (!is_power_of_2(iob)) {
1893 if (nvme_first_scan(ns->disk))
1894 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1895 ns->disk->disk_name, iob);
1899 if (blk_queue_is_zoned(ns->disk->queue)) {
1900 if (nvme_first_scan(ns->disk))
1901 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1902 ns->disk->disk_name);
1906 blk_queue_chunk_sectors(ns->queue, iob);
1909 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1911 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1914 blk_mq_freeze_queue(ns->disk->queue);
1915 ns->lba_shift = id->lbaf[lbaf].ds;
1916 nvme_set_queue_limits(ns->ctrl, ns->queue);
1918 ret = nvme_configure_metadata(ns, id);
1921 nvme_set_chunk_sectors(ns, id);
1922 nvme_update_disk_info(ns->disk, ns, id);
1924 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1925 ret = nvme_update_zone_info(ns, lbaf);
1930 set_bit(NVME_NS_READY, &ns->flags);
1931 blk_mq_unfreeze_queue(ns->disk->queue);
1933 if (blk_queue_is_zoned(ns->queue)) {
1934 ret = nvme_revalidate_zones(ns);
1935 if (ret && !nvme_first_scan(ns->disk))
1939 if (nvme_ns_head_multipath(ns->head)) {
1940 blk_mq_freeze_queue(ns->head->disk->queue);
1941 nvme_update_disk_info(ns->head->disk, ns, id);
1942 nvme_mpath_revalidate_paths(ns);
1943 blk_stack_limits(&ns->head->disk->queue->limits,
1944 &ns->queue->limits, 0);
1945 disk_update_readahead(ns->head->disk);
1946 blk_mq_unfreeze_queue(ns->head->disk->queue);
1951 blk_mq_unfreeze_queue(ns->disk->queue);
1954 * If probing fails due an unsupported feature, hide the block device,
1955 * but still allow other access.
1957 if (ret == -ENODEV) {
1958 ns->disk->flags |= GENHD_FL_HIDDEN;
1964 static char nvme_pr_type(enum pr_type type)
1967 case PR_WRITE_EXCLUSIVE:
1969 case PR_EXCLUSIVE_ACCESS:
1971 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1973 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1975 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1977 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1984 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1985 struct nvme_command *c, u8 data[16])
1987 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1988 int srcu_idx = srcu_read_lock(&head->srcu);
1989 struct nvme_ns *ns = nvme_find_path(head);
1990 int ret = -EWOULDBLOCK;
1993 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1994 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1996 srcu_read_unlock(&head->srcu, srcu_idx);
2000 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2003 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2004 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2007 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2008 u64 key, u64 sa_key, u8 op)
2010 struct nvme_command c = { };
2011 u8 data[16] = { 0, };
2013 put_unaligned_le64(key, &data[0]);
2014 put_unaligned_le64(sa_key, &data[8]);
2016 c.common.opcode = op;
2017 c.common.cdw10 = cpu_to_le32(cdw10);
2019 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2020 bdev->bd_disk->fops == &nvme_ns_head_ops)
2021 return nvme_send_ns_head_pr_command(bdev, &c, data);
2022 return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2025 static int nvme_pr_register(struct block_device *bdev, u64 old,
2026 u64 new, unsigned flags)
2030 if (flags & ~PR_FL_IGNORE_KEY)
2033 cdw10 = old ? 2 : 0;
2034 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2035 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2036 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2039 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2040 enum pr_type type, unsigned flags)
2044 if (flags & ~PR_FL_IGNORE_KEY)
2047 cdw10 = nvme_pr_type(type) << 8;
2048 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2049 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2052 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2053 enum pr_type type, bool abort)
2055 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2057 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2060 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2062 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2064 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2067 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2069 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2071 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2074 const struct pr_ops nvme_pr_ops = {
2075 .pr_register = nvme_pr_register,
2076 .pr_reserve = nvme_pr_reserve,
2077 .pr_release = nvme_pr_release,
2078 .pr_preempt = nvme_pr_preempt,
2079 .pr_clear = nvme_pr_clear,
2082 #ifdef CONFIG_BLK_SED_OPAL
2083 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2086 struct nvme_ctrl *ctrl = data;
2087 struct nvme_command cmd = { };
2090 cmd.common.opcode = nvme_admin_security_send;
2092 cmd.common.opcode = nvme_admin_security_recv;
2093 cmd.common.nsid = 0;
2094 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2095 cmd.common.cdw11 = cpu_to_le32(len);
2097 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2098 NVME_QID_ANY, 1, 0);
2100 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2101 #endif /* CONFIG_BLK_SED_OPAL */
2103 #ifdef CONFIG_BLK_DEV_ZONED
2104 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2105 unsigned int nr_zones, report_zones_cb cb, void *data)
2107 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2111 #define nvme_report_zones NULL
2112 #endif /* CONFIG_BLK_DEV_ZONED */
2114 static const struct block_device_operations nvme_bdev_ops = {
2115 .owner = THIS_MODULE,
2116 .ioctl = nvme_ioctl,
2118 .release = nvme_release,
2119 .getgeo = nvme_getgeo,
2120 .report_zones = nvme_report_zones,
2121 .pr_ops = &nvme_pr_ops,
2124 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2126 unsigned long timeout =
2127 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2128 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2131 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2134 if ((csts & NVME_CSTS_RDY) == bit)
2137 usleep_range(1000, 2000);
2138 if (fatal_signal_pending(current))
2140 if (time_after(jiffies, timeout)) {
2141 dev_err(ctrl->device,
2142 "Device not ready; aborting %s, CSTS=0x%x\n",
2143 enabled ? "initialisation" : "reset", csts);
2152 * If the device has been passed off to us in an enabled state, just clear
2153 * the enabled bit. The spec says we should set the 'shutdown notification
2154 * bits', but doing so may cause the device to complete commands to the
2155 * admin queue ... and we don't know what memory that might be pointing at!
2157 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2161 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2162 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2164 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2168 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2169 msleep(NVME_QUIRK_DELAY_AMOUNT);
2171 return nvme_wait_ready(ctrl, ctrl->cap, false);
2173 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2175 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2177 unsigned dev_page_min;
2180 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2182 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2185 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2187 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2188 dev_err(ctrl->device,
2189 "Minimum device page size %u too large for host (%u)\n",
2190 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2194 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2195 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2197 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2198 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2199 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2200 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2201 ctrl->ctrl_config |= NVME_CC_ENABLE;
2203 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2206 return nvme_wait_ready(ctrl, ctrl->cap, true);
2208 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2210 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2212 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2216 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2217 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2219 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2223 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2224 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2228 if (fatal_signal_pending(current))
2230 if (time_after(jiffies, timeout)) {
2231 dev_err(ctrl->device,
2232 "Device shutdown incomplete; abort shutdown\n");
2239 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2241 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2246 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2249 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2250 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2253 dev_warn_once(ctrl->device,
2254 "could not set timestamp (%d)\n", ret);
2258 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2260 struct nvme_feat_host_behavior *host;
2263 /* Don't bother enabling the feature if retry delay is not reported */
2267 host = kzalloc(sizeof(*host), GFP_KERNEL);
2271 host->acre = NVME_ENABLE_ACRE;
2272 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2273 host, sizeof(*host), NULL);
2279 * The function checks whether the given total (exlat + enlat) latency of
2280 * a power state allows the latter to be used as an APST transition target.
2281 * It does so by comparing the latency to the primary and secondary latency
2282 * tolerances defined by module params. If there's a match, the corresponding
2283 * timeout value is returned and the matching tolerance index (1 or 2) is
2286 static bool nvme_apst_get_transition_time(u64 total_latency,
2287 u64 *transition_time, unsigned *last_index)
2289 if (total_latency <= apst_primary_latency_tol_us) {
2290 if (*last_index == 1)
2293 *transition_time = apst_primary_timeout_ms;
2296 if (apst_secondary_timeout_ms &&
2297 total_latency <= apst_secondary_latency_tol_us) {
2298 if (*last_index <= 2)
2301 *transition_time = apst_secondary_timeout_ms;
2308 * APST (Autonomous Power State Transition) lets us program a table of power
2309 * state transitions that the controller will perform automatically.
2311 * Depending on module params, one of the two supported techniques will be used:
2313 * - If the parameters provide explicit timeouts and tolerances, they will be
2314 * used to build a table with up to 2 non-operational states to transition to.
2315 * The default parameter values were selected based on the values used by
2316 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2317 * regeneration of the APST table in the event of switching between external
2318 * and battery power, the timeouts and tolerances reflect a compromise
2319 * between values used by Microsoft for AC and battery scenarios.
2320 * - If not, we'll configure the table with a simple heuristic: we are willing
2321 * to spend at most 2% of the time transitioning between power states.
2322 * Therefore, when running in any given state, we will enter the next
2323 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2324 * microseconds, as long as that state's exit latency is under the requested
2327 * We will not autonomously enter any non-operational state for which the total
2328 * latency exceeds ps_max_latency_us.
2330 * Users can set ps_max_latency_us to zero to turn off APST.
2332 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2334 struct nvme_feat_auto_pst *table;
2341 unsigned last_lt_index = UINT_MAX;
2344 * If APST isn't supported or if we haven't been initialized yet,
2345 * then don't do anything.
2350 if (ctrl->npss > 31) {
2351 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2355 table = kzalloc(sizeof(*table), GFP_KERNEL);
2359 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2360 /* Turn off APST. */
2361 dev_dbg(ctrl->device, "APST disabled\n");
2366 * Walk through all states from lowest- to highest-power.
2367 * According to the spec, lower-numbered states use more power. NPSS,
2368 * despite the name, is the index of the lowest-power state, not the
2371 for (state = (int)ctrl->npss; state >= 0; state--) {
2372 u64 total_latency_us, exit_latency_us, transition_ms;
2375 table->entries[state] = target;
2378 * Don't allow transitions to the deepest state if it's quirked
2381 if (state == ctrl->npss &&
2382 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2386 * Is this state a useful non-operational state for higher-power
2387 * states to autonomously transition to?
2389 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2392 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2393 if (exit_latency_us > ctrl->ps_max_latency_us)
2396 total_latency_us = exit_latency_us +
2397 le32_to_cpu(ctrl->psd[state].entry_lat);
2400 * This state is good. It can be used as the APST idle target
2401 * for higher power states.
2403 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2404 if (!nvme_apst_get_transition_time(total_latency_us,
2405 &transition_ms, &last_lt_index))
2408 transition_ms = total_latency_us + 19;
2409 do_div(transition_ms, 20);
2410 if (transition_ms > (1 << 24) - 1)
2411 transition_ms = (1 << 24) - 1;
2414 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2417 if (total_latency_us > max_lat_us)
2418 max_lat_us = total_latency_us;
2422 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2424 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2425 max_ps, max_lat_us, (int)sizeof(*table), table);
2429 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2430 table, sizeof(*table), NULL);
2432 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2437 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2439 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2443 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2444 case PM_QOS_LATENCY_ANY:
2452 if (ctrl->ps_max_latency_us != latency) {
2453 ctrl->ps_max_latency_us = latency;
2454 if (ctrl->state == NVME_CTRL_LIVE)
2455 nvme_configure_apst(ctrl);
2459 struct nvme_core_quirk_entry {
2461 * NVMe model and firmware strings are padded with spaces. For
2462 * simplicity, strings in the quirk table are padded with NULLs
2468 unsigned long quirks;
2471 static const struct nvme_core_quirk_entry core_quirks[] = {
2474 * This Toshiba device seems to die using any APST states. See:
2475 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2478 .mn = "THNSF5256GPUK TOSHIBA",
2479 .quirks = NVME_QUIRK_NO_APST,
2483 * This LiteON CL1-3D*-Q11 firmware version has a race
2484 * condition associated with actions related to suspend to idle
2485 * LiteON has resolved the problem in future firmware
2489 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2493 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2494 * aborts I/O during any load, but more easily reproducible
2495 * with discards (fstrim).
2497 * The device is left in a state where it is also not possible
2498 * to use "nvme set-feature" to disable APST, but booting with
2499 * nvme_core.default_ps_max_latency=0 works.
2502 .mn = "KCD6XVUL6T40",
2503 .quirks = NVME_QUIRK_NO_APST,
2507 /* match is null-terminated but idstr is space-padded. */
2508 static bool string_matches(const char *idstr, const char *match, size_t len)
2515 matchlen = strlen(match);
2516 WARN_ON_ONCE(matchlen > len);
2518 if (memcmp(idstr, match, matchlen))
2521 for (; matchlen < len; matchlen++)
2522 if (idstr[matchlen] != ' ')
2528 static bool quirk_matches(const struct nvme_id_ctrl *id,
2529 const struct nvme_core_quirk_entry *q)
2531 return q->vid == le16_to_cpu(id->vid) &&
2532 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2533 string_matches(id->fr, q->fr, sizeof(id->fr));
2536 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2537 struct nvme_id_ctrl *id)
2542 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2543 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2544 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2545 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2549 if (ctrl->vs >= NVME_VS(1, 2, 1))
2550 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2553 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2554 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2555 "nqn.2014.08.org.nvmexpress:%04x%04x",
2556 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2557 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2558 off += sizeof(id->sn);
2559 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2560 off += sizeof(id->mn);
2561 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2564 static void nvme_release_subsystem(struct device *dev)
2566 struct nvme_subsystem *subsys =
2567 container_of(dev, struct nvme_subsystem, dev);
2569 if (subsys->instance >= 0)
2570 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2574 static void nvme_destroy_subsystem(struct kref *ref)
2576 struct nvme_subsystem *subsys =
2577 container_of(ref, struct nvme_subsystem, ref);
2579 mutex_lock(&nvme_subsystems_lock);
2580 list_del(&subsys->entry);
2581 mutex_unlock(&nvme_subsystems_lock);
2583 ida_destroy(&subsys->ns_ida);
2584 device_del(&subsys->dev);
2585 put_device(&subsys->dev);
2588 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2590 kref_put(&subsys->ref, nvme_destroy_subsystem);
2593 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2595 struct nvme_subsystem *subsys;
2597 lockdep_assert_held(&nvme_subsystems_lock);
2600 * Fail matches for discovery subsystems. This results
2601 * in each discovery controller bound to a unique subsystem.
2602 * This avoids issues with validating controller values
2603 * that can only be true when there is a single unique subsystem.
2604 * There may be multiple and completely independent entities
2605 * that provide discovery controllers.
2607 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2610 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2611 if (strcmp(subsys->subnqn, subsysnqn))
2613 if (!kref_get_unless_zero(&subsys->ref))
2621 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2622 struct device_attribute subsys_attr_##_name = \
2623 __ATTR(_name, _mode, _show, NULL)
2625 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2626 struct device_attribute *attr,
2629 struct nvme_subsystem *subsys =
2630 container_of(dev, struct nvme_subsystem, dev);
2632 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2634 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2636 static ssize_t nvme_subsys_show_type(struct device *dev,
2637 struct device_attribute *attr,
2640 struct nvme_subsystem *subsys =
2641 container_of(dev, struct nvme_subsystem, dev);
2643 switch (subsys->subtype) {
2645 return sysfs_emit(buf, "discovery\n");
2647 return sysfs_emit(buf, "nvm\n");
2649 return sysfs_emit(buf, "reserved\n");
2652 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2654 #define nvme_subsys_show_str_function(field) \
2655 static ssize_t subsys_##field##_show(struct device *dev, \
2656 struct device_attribute *attr, char *buf) \
2658 struct nvme_subsystem *subsys = \
2659 container_of(dev, struct nvme_subsystem, dev); \
2660 return sysfs_emit(buf, "%.*s\n", \
2661 (int)sizeof(subsys->field), subsys->field); \
2663 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2665 nvme_subsys_show_str_function(model);
2666 nvme_subsys_show_str_function(serial);
2667 nvme_subsys_show_str_function(firmware_rev);
2669 static struct attribute *nvme_subsys_attrs[] = {
2670 &subsys_attr_model.attr,
2671 &subsys_attr_serial.attr,
2672 &subsys_attr_firmware_rev.attr,
2673 &subsys_attr_subsysnqn.attr,
2674 &subsys_attr_subsystype.attr,
2675 #ifdef CONFIG_NVME_MULTIPATH
2676 &subsys_attr_iopolicy.attr,
2681 static const struct attribute_group nvme_subsys_attrs_group = {
2682 .attrs = nvme_subsys_attrs,
2685 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2686 &nvme_subsys_attrs_group,
2690 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2692 return ctrl->opts && ctrl->opts->discovery_nqn;
2695 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2696 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2698 struct nvme_ctrl *tmp;
2700 lockdep_assert_held(&nvme_subsystems_lock);
2702 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2703 if (nvme_state_terminal(tmp))
2706 if (tmp->cntlid == ctrl->cntlid) {
2707 dev_err(ctrl->device,
2708 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2709 ctrl->cntlid, dev_name(tmp->device),
2714 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2715 nvme_discovery_ctrl(ctrl))
2718 dev_err(ctrl->device,
2719 "Subsystem does not support multiple controllers\n");
2726 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2728 struct nvme_subsystem *subsys, *found;
2731 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2735 subsys->instance = -1;
2736 mutex_init(&subsys->lock);
2737 kref_init(&subsys->ref);
2738 INIT_LIST_HEAD(&subsys->ctrls);
2739 INIT_LIST_HEAD(&subsys->nsheads);
2740 nvme_init_subnqn(subsys, ctrl, id);
2741 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2742 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2743 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2744 subsys->vendor_id = le16_to_cpu(id->vid);
2745 subsys->cmic = id->cmic;
2747 /* Versions prior to 1.4 don't necessarily report a valid type */
2748 if (id->cntrltype == NVME_CTRL_DISC ||
2749 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2750 subsys->subtype = NVME_NQN_DISC;
2752 subsys->subtype = NVME_NQN_NVME;
2754 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2755 dev_err(ctrl->device,
2756 "Subsystem %s is not a discovery controller",
2761 subsys->awupf = le16_to_cpu(id->awupf);
2762 nvme_mpath_default_iopolicy(subsys);
2764 subsys->dev.class = nvme_subsys_class;
2765 subsys->dev.release = nvme_release_subsystem;
2766 subsys->dev.groups = nvme_subsys_attrs_groups;
2767 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2768 device_initialize(&subsys->dev);
2770 mutex_lock(&nvme_subsystems_lock);
2771 found = __nvme_find_get_subsystem(subsys->subnqn);
2773 put_device(&subsys->dev);
2776 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2778 goto out_put_subsystem;
2781 ret = device_add(&subsys->dev);
2783 dev_err(ctrl->device,
2784 "failed to register subsystem device.\n");
2785 put_device(&subsys->dev);
2788 ida_init(&subsys->ns_ida);
2789 list_add_tail(&subsys->entry, &nvme_subsystems);
2792 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2793 dev_name(ctrl->device));
2795 dev_err(ctrl->device,
2796 "failed to create sysfs link from subsystem.\n");
2797 goto out_put_subsystem;
2801 subsys->instance = ctrl->instance;
2802 ctrl->subsys = subsys;
2803 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2804 mutex_unlock(&nvme_subsystems_lock);
2808 nvme_put_subsystem(subsys);
2810 mutex_unlock(&nvme_subsystems_lock);
2814 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2815 void *log, size_t size, u64 offset)
2817 struct nvme_command c = { };
2818 u32 dwlen = nvme_bytes_to_numd(size);
2820 c.get_log_page.opcode = nvme_admin_get_log_page;
2821 c.get_log_page.nsid = cpu_to_le32(nsid);
2822 c.get_log_page.lid = log_page;
2823 c.get_log_page.lsp = lsp;
2824 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2825 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2826 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2827 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2828 c.get_log_page.csi = csi;
2830 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2833 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2834 struct nvme_effects_log **log)
2836 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2842 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2846 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2847 cel, sizeof(*cel), 0);
2853 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2859 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2861 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2863 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2868 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2870 struct nvme_command c = { };
2871 struct nvme_id_ctrl_nvm *id;
2874 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2875 ctrl->max_discard_sectors = UINT_MAX;
2876 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2878 ctrl->max_discard_sectors = 0;
2879 ctrl->max_discard_segments = 0;
2883 * Even though NVMe spec explicitly states that MDTS is not applicable
2884 * to the write-zeroes, we are cautious and limit the size to the
2885 * controllers max_hw_sectors value, which is based on the MDTS field
2886 * and possibly other limiting factors.
2888 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2889 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2890 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2892 ctrl->max_zeroes_sectors = 0;
2894 if (nvme_ctrl_limited_cns(ctrl))
2897 id = kzalloc(sizeof(*id), GFP_KERNEL);
2901 c.identify.opcode = nvme_admin_identify;
2902 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2903 c.identify.csi = NVME_CSI_NVM;
2905 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2910 ctrl->max_discard_segments = id->dmrl;
2912 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2914 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2921 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2923 struct nvme_id_ctrl *id;
2925 bool prev_apst_enabled;
2928 ret = nvme_identify_ctrl(ctrl, &id);
2930 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2934 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2935 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2940 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2941 ctrl->cntlid = le16_to_cpu(id->cntlid);
2943 if (!ctrl->identified) {
2946 ret = nvme_init_subsystem(ctrl, id);
2951 * Check for quirks. Quirk can depend on firmware version,
2952 * so, in principle, the set of quirks present can change
2953 * across a reset. As a possible future enhancement, we
2954 * could re-scan for quirks every time we reinitialize
2955 * the device, but we'd have to make sure that the driver
2956 * behaves intelligently if the quirks change.
2958 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2959 if (quirk_matches(id, &core_quirks[i]))
2960 ctrl->quirks |= core_quirks[i].quirks;
2964 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2965 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2966 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2969 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2970 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2971 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2973 ctrl->oacs = le16_to_cpu(id->oacs);
2974 ctrl->oncs = le16_to_cpu(id->oncs);
2975 ctrl->mtfa = le16_to_cpu(id->mtfa);
2976 ctrl->oaes = le32_to_cpu(id->oaes);
2977 ctrl->wctemp = le16_to_cpu(id->wctemp);
2978 ctrl->cctemp = le16_to_cpu(id->cctemp);
2980 atomic_set(&ctrl->abort_limit, id->acl + 1);
2981 ctrl->vwc = id->vwc;
2983 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2985 max_hw_sectors = UINT_MAX;
2986 ctrl->max_hw_sectors =
2987 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2989 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2990 ctrl->sgls = le32_to_cpu(id->sgls);
2991 ctrl->kas = le16_to_cpu(id->kas);
2992 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2993 ctrl->ctratt = le32_to_cpu(id->ctratt);
2997 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2999 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3000 shutdown_timeout, 60);
3002 if (ctrl->shutdown_timeout != shutdown_timeout)
3003 dev_info(ctrl->device,
3004 "Shutdown timeout set to %u seconds\n",
3005 ctrl->shutdown_timeout);
3007 ctrl->shutdown_timeout = shutdown_timeout;
3009 ctrl->npss = id->npss;
3010 ctrl->apsta = id->apsta;
3011 prev_apst_enabled = ctrl->apst_enabled;
3012 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3013 if (force_apst && id->apsta) {
3014 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3015 ctrl->apst_enabled = true;
3017 ctrl->apst_enabled = false;
3020 ctrl->apst_enabled = id->apsta;
3022 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3024 if (ctrl->ops->flags & NVME_F_FABRICS) {
3025 ctrl->icdoff = le16_to_cpu(id->icdoff);
3026 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3027 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3028 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3031 * In fabrics we need to verify the cntlid matches the
3034 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3035 dev_err(ctrl->device,
3036 "Mismatching cntlid: Connect %u vs Identify "
3038 ctrl->cntlid, le16_to_cpu(id->cntlid));
3043 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3044 dev_err(ctrl->device,
3045 "keep-alive support is mandatory for fabrics\n");
3050 ctrl->hmpre = le32_to_cpu(id->hmpre);
3051 ctrl->hmmin = le32_to_cpu(id->hmmin);
3052 ctrl->hmminds = le32_to_cpu(id->hmminds);
3053 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3056 ret = nvme_mpath_init_identify(ctrl, id);
3060 if (ctrl->apst_enabled && !prev_apst_enabled)
3061 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3062 else if (!ctrl->apst_enabled && prev_apst_enabled)
3063 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3071 * Initialize the cached copies of the Identify data and various controller
3072 * register in our nvme_ctrl structure. This should be called as soon as
3073 * the admin queue is fully up and running.
3075 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3079 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3081 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3085 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3087 if (ctrl->vs >= NVME_VS(1, 1, 0))
3088 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3090 ret = nvme_init_identify(ctrl);
3094 ret = nvme_init_non_mdts_limits(ctrl);
3098 ret = nvme_configure_apst(ctrl);
3102 ret = nvme_configure_timestamp(ctrl);
3106 ret = nvme_configure_directives(ctrl);
3110 ret = nvme_configure_acre(ctrl);
3114 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3115 ret = nvme_hwmon_init(ctrl);
3120 ctrl->identified = true;
3124 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3126 static int nvme_dev_open(struct inode *inode, struct file *file)
3128 struct nvme_ctrl *ctrl =
3129 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3131 switch (ctrl->state) {
3132 case NVME_CTRL_LIVE:
3135 return -EWOULDBLOCK;
3138 nvme_get_ctrl(ctrl);
3139 if (!try_module_get(ctrl->ops->module)) {
3140 nvme_put_ctrl(ctrl);
3144 file->private_data = ctrl;
3148 static int nvme_dev_release(struct inode *inode, struct file *file)
3150 struct nvme_ctrl *ctrl =
3151 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3153 module_put(ctrl->ops->module);
3154 nvme_put_ctrl(ctrl);
3158 static const struct file_operations nvme_dev_fops = {
3159 .owner = THIS_MODULE,
3160 .open = nvme_dev_open,
3161 .release = nvme_dev_release,
3162 .unlocked_ioctl = nvme_dev_ioctl,
3163 .compat_ioctl = compat_ptr_ioctl,
3166 static ssize_t nvme_sysfs_reset(struct device *dev,
3167 struct device_attribute *attr, const char *buf,
3170 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3173 ret = nvme_reset_ctrl_sync(ctrl);
3178 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3180 static ssize_t nvme_sysfs_rescan(struct device *dev,
3181 struct device_attribute *attr, const char *buf,
3184 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3186 nvme_queue_scan(ctrl);
3189 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3191 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3193 struct gendisk *disk = dev_to_disk(dev);
3195 if (disk->fops == &nvme_bdev_ops)
3196 return nvme_get_ns_from_dev(dev)->head;
3198 return disk->private_data;
3201 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3204 struct nvme_ns_head *head = dev_to_ns_head(dev);
3205 struct nvme_ns_ids *ids = &head->ids;
3206 struct nvme_subsystem *subsys = head->subsys;
3207 int serial_len = sizeof(subsys->serial);
3208 int model_len = sizeof(subsys->model);
3210 if (!uuid_is_null(&ids->uuid))
3211 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3213 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3214 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3216 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3217 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3219 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3220 subsys->serial[serial_len - 1] == '\0'))
3222 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3223 subsys->model[model_len - 1] == '\0'))
3226 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3227 serial_len, subsys->serial, model_len, subsys->model,
3230 static DEVICE_ATTR_RO(wwid);
3232 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3235 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3237 static DEVICE_ATTR_RO(nguid);
3239 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3242 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3244 /* For backward compatibility expose the NGUID to userspace if
3245 * we have no UUID set
3247 if (uuid_is_null(&ids->uuid)) {
3248 printk_ratelimited(KERN_WARNING
3249 "No UUID available providing old NGUID\n");
3250 return sysfs_emit(buf, "%pU\n", ids->nguid);
3252 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3254 static DEVICE_ATTR_RO(uuid);
3256 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3259 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3261 static DEVICE_ATTR_RO(eui);
3263 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3266 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3268 static DEVICE_ATTR_RO(nsid);
3270 static struct attribute *nvme_ns_id_attrs[] = {
3271 &dev_attr_wwid.attr,
3272 &dev_attr_uuid.attr,
3273 &dev_attr_nguid.attr,
3275 &dev_attr_nsid.attr,
3276 #ifdef CONFIG_NVME_MULTIPATH
3277 &dev_attr_ana_grpid.attr,
3278 &dev_attr_ana_state.attr,
3283 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3284 struct attribute *a, int n)
3286 struct device *dev = container_of(kobj, struct device, kobj);
3287 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3289 if (a == &dev_attr_uuid.attr) {
3290 if (uuid_is_null(&ids->uuid) &&
3291 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3294 if (a == &dev_attr_nguid.attr) {
3295 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3298 if (a == &dev_attr_eui.attr) {
3299 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3302 #ifdef CONFIG_NVME_MULTIPATH
3303 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3304 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3306 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3313 static const struct attribute_group nvme_ns_id_attr_group = {
3314 .attrs = nvme_ns_id_attrs,
3315 .is_visible = nvme_ns_id_attrs_are_visible,
3318 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3319 &nvme_ns_id_attr_group,
3323 #define nvme_show_str_function(field) \
3324 static ssize_t field##_show(struct device *dev, \
3325 struct device_attribute *attr, char *buf) \
3327 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3328 return sysfs_emit(buf, "%.*s\n", \
3329 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3331 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3333 nvme_show_str_function(model);
3334 nvme_show_str_function(serial);
3335 nvme_show_str_function(firmware_rev);
3337 #define nvme_show_int_function(field) \
3338 static ssize_t field##_show(struct device *dev, \
3339 struct device_attribute *attr, char *buf) \
3341 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3342 return sysfs_emit(buf, "%d\n", ctrl->field); \
3344 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3346 nvme_show_int_function(cntlid);
3347 nvme_show_int_function(numa_node);
3348 nvme_show_int_function(queue_count);
3349 nvme_show_int_function(sqsize);
3350 nvme_show_int_function(kato);
3352 static ssize_t nvme_sysfs_delete(struct device *dev,
3353 struct device_attribute *attr, const char *buf,
3356 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3358 if (device_remove_file_self(dev, attr))
3359 nvme_delete_ctrl_sync(ctrl);
3362 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3364 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3365 struct device_attribute *attr,
3368 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3370 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3372 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3374 static ssize_t nvme_sysfs_show_state(struct device *dev,
3375 struct device_attribute *attr,
3378 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3379 static const char *const state_name[] = {
3380 [NVME_CTRL_NEW] = "new",
3381 [NVME_CTRL_LIVE] = "live",
3382 [NVME_CTRL_RESETTING] = "resetting",
3383 [NVME_CTRL_CONNECTING] = "connecting",
3384 [NVME_CTRL_DELETING] = "deleting",
3385 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3386 [NVME_CTRL_DEAD] = "dead",
3389 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3390 state_name[ctrl->state])
3391 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3393 return sysfs_emit(buf, "unknown state\n");
3396 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3398 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3399 struct device_attribute *attr,
3402 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3404 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3406 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3408 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3409 struct device_attribute *attr,
3412 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3414 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3416 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3418 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3419 struct device_attribute *attr,
3422 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3424 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3426 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3428 static ssize_t nvme_sysfs_show_address(struct device *dev,
3429 struct device_attribute *attr,
3432 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3434 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3436 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3438 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3439 struct device_attribute *attr, char *buf)
3441 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3442 struct nvmf_ctrl_options *opts = ctrl->opts;
3444 if (ctrl->opts->max_reconnects == -1)
3445 return sysfs_emit(buf, "off\n");
3446 return sysfs_emit(buf, "%d\n",
3447 opts->max_reconnects * opts->reconnect_delay);
3450 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3451 struct device_attribute *attr, const char *buf, size_t count)
3453 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3454 struct nvmf_ctrl_options *opts = ctrl->opts;
3455 int ctrl_loss_tmo, err;
3457 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3461 if (ctrl_loss_tmo < 0)
3462 opts->max_reconnects = -1;
3464 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3465 opts->reconnect_delay);
3468 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3469 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3471 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3472 struct device_attribute *attr, char *buf)
3474 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3476 if (ctrl->opts->reconnect_delay == -1)
3477 return sysfs_emit(buf, "off\n");
3478 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3481 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3482 struct device_attribute *attr, const char *buf, size_t count)
3484 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3488 err = kstrtou32(buf, 10, &v);
3492 ctrl->opts->reconnect_delay = v;
3495 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3496 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3498 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3499 struct device_attribute *attr, char *buf)
3501 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3503 if (ctrl->opts->fast_io_fail_tmo == -1)
3504 return sysfs_emit(buf, "off\n");
3505 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3508 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3509 struct device_attribute *attr, const char *buf, size_t count)
3511 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3512 struct nvmf_ctrl_options *opts = ctrl->opts;
3513 int fast_io_fail_tmo, err;
3515 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3519 if (fast_io_fail_tmo < 0)
3520 opts->fast_io_fail_tmo = -1;
3522 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3525 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3526 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3528 static struct attribute *nvme_dev_attrs[] = {
3529 &dev_attr_reset_controller.attr,
3530 &dev_attr_rescan_controller.attr,
3531 &dev_attr_model.attr,
3532 &dev_attr_serial.attr,
3533 &dev_attr_firmware_rev.attr,
3534 &dev_attr_cntlid.attr,
3535 &dev_attr_delete_controller.attr,
3536 &dev_attr_transport.attr,
3537 &dev_attr_subsysnqn.attr,
3538 &dev_attr_address.attr,
3539 &dev_attr_state.attr,
3540 &dev_attr_numa_node.attr,
3541 &dev_attr_queue_count.attr,
3542 &dev_attr_sqsize.attr,
3543 &dev_attr_hostnqn.attr,
3544 &dev_attr_hostid.attr,
3545 &dev_attr_ctrl_loss_tmo.attr,
3546 &dev_attr_reconnect_delay.attr,
3547 &dev_attr_fast_io_fail_tmo.attr,
3548 &dev_attr_kato.attr,
3552 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3553 struct attribute *a, int n)
3555 struct device *dev = container_of(kobj, struct device, kobj);
3556 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3558 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3560 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3562 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3564 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3566 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3568 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3570 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3576 static const struct attribute_group nvme_dev_attrs_group = {
3577 .attrs = nvme_dev_attrs,
3578 .is_visible = nvme_dev_attrs_are_visible,
3581 static const struct attribute_group *nvme_dev_attr_groups[] = {
3582 &nvme_dev_attrs_group,
3586 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3589 struct nvme_ns_head *h;
3591 lockdep_assert_held(&subsys->lock);
3593 list_for_each_entry(h, &subsys->nsheads, entry) {
3594 if (h->ns_id != nsid)
3596 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3603 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3604 struct nvme_ns_head *new)
3606 struct nvme_ns_head *h;
3608 lockdep_assert_held(&subsys->lock);
3610 list_for_each_entry(h, &subsys->nsheads, entry) {
3611 if (nvme_ns_ids_valid(&new->ids) &&
3612 nvme_ns_ids_equal(&new->ids, &h->ids))
3619 static void nvme_cdev_rel(struct device *dev)
3621 ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3624 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3626 cdev_device_del(cdev, cdev_device);
3627 put_device(cdev_device);
3630 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3631 const struct file_operations *fops, struct module *owner)
3635 minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3638 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3639 cdev_device->class = nvme_ns_chr_class;
3640 cdev_device->release = nvme_cdev_rel;
3641 device_initialize(cdev_device);
3642 cdev_init(cdev, fops);
3643 cdev->owner = owner;
3644 ret = cdev_device_add(cdev, cdev_device);
3646 put_device(cdev_device);
3651 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3653 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3656 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3658 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3662 static const struct file_operations nvme_ns_chr_fops = {
3663 .owner = THIS_MODULE,
3664 .open = nvme_ns_chr_open,
3665 .release = nvme_ns_chr_release,
3666 .unlocked_ioctl = nvme_ns_chr_ioctl,
3667 .compat_ioctl = compat_ptr_ioctl,
3670 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3674 ns->cdev_device.parent = ns->ctrl->device;
3675 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3676 ns->ctrl->instance, ns->head->instance);
3680 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3681 ns->ctrl->ops->module);
3684 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3685 unsigned nsid, struct nvme_ns_ids *ids)
3687 struct nvme_ns_head *head;
3688 size_t size = sizeof(*head);
3691 #ifdef CONFIG_NVME_MULTIPATH
3692 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3695 head = kzalloc(size, GFP_KERNEL);
3698 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3701 head->instance = ret;
3702 INIT_LIST_HEAD(&head->list);
3703 ret = init_srcu_struct(&head->srcu);
3705 goto out_ida_remove;
3706 head->subsys = ctrl->subsys;
3709 kref_init(&head->ref);
3711 ret = __nvme_check_ids(ctrl->subsys, head);
3713 dev_err(ctrl->device,
3714 "duplicate IDs for nsid %d\n", nsid);
3715 goto out_cleanup_srcu;
3718 if (head->ids.csi) {
3719 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3721 goto out_cleanup_srcu;
3723 head->effects = ctrl->effects;
3725 ret = nvme_mpath_alloc_disk(ctrl, head);
3727 goto out_cleanup_srcu;
3729 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3731 kref_get(&ctrl->subsys->ref);
3735 cleanup_srcu_struct(&head->srcu);
3737 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3742 ret = blk_status_to_errno(nvme_error_status(ret));
3743 return ERR_PTR(ret);
3746 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3747 struct nvme_ns_ids *ids, bool is_shared)
3749 struct nvme_ctrl *ctrl = ns->ctrl;
3750 struct nvme_ns_head *head = NULL;
3753 mutex_lock(&ctrl->subsys->lock);
3754 head = nvme_find_ns_head(ctrl->subsys, nsid);
3756 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3758 ret = PTR_ERR(head);
3761 head->shared = is_shared;
3764 if (!is_shared || !head->shared) {
3765 dev_err(ctrl->device,
3766 "Duplicate unshared namespace %d\n", nsid);
3767 goto out_put_ns_head;
3769 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3770 dev_err(ctrl->device,
3771 "IDs don't match for shared namespace %d\n",
3773 goto out_put_ns_head;
3777 list_add_tail_rcu(&ns->siblings, &head->list);
3779 mutex_unlock(&ctrl->subsys->lock);
3783 nvme_put_ns_head(head);
3785 mutex_unlock(&ctrl->subsys->lock);
3789 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3791 struct nvme_ns *ns, *ret = NULL;
3793 down_read(&ctrl->namespaces_rwsem);
3794 list_for_each_entry(ns, &ctrl->namespaces, list) {
3795 if (ns->head->ns_id == nsid) {
3796 if (!nvme_get_ns(ns))
3801 if (ns->head->ns_id > nsid)
3804 up_read(&ctrl->namespaces_rwsem);
3807 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3810 * Add the namespace to the controller list while keeping the list ordered.
3812 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3814 struct nvme_ns *tmp;
3816 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3817 if (tmp->head->ns_id < ns->head->ns_id) {
3818 list_add(&ns->list, &tmp->list);
3822 list_add(&ns->list, &ns->ctrl->namespaces);
3825 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3826 struct nvme_ns_ids *ids)
3829 struct gendisk *disk;
3830 struct nvme_id_ns *id;
3831 int node = ctrl->numa_node;
3833 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3836 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3840 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3843 disk->fops = &nvme_bdev_ops;
3844 disk->private_data = ns;
3847 ns->queue = disk->queue;
3849 if (ctrl->opts && ctrl->opts->data_digest)
3850 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3852 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3853 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3854 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3857 kref_init(&ns->kref);
3859 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3860 goto out_cleanup_disk;
3863 * Without the multipath code enabled, multiple controller per
3864 * subsystems are visible as devices and thus we cannot use the
3865 * subsystem instance.
3867 if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3868 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3869 ns->head->instance);
3871 if (nvme_update_ns_info(ns, id))
3874 down_write(&ctrl->namespaces_rwsem);
3875 nvme_ns_add_to_ctrl_list(ns);
3876 up_write(&ctrl->namespaces_rwsem);
3877 nvme_get_ctrl(ctrl);
3879 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3880 goto out_cleanup_ns_from_list;
3882 if (!nvme_ns_head_multipath(ns->head))
3883 nvme_add_ns_cdev(ns);
3885 nvme_mpath_add_disk(ns, id);
3886 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3891 out_cleanup_ns_from_list:
3892 nvme_put_ctrl(ctrl);
3893 down_write(&ctrl->namespaces_rwsem);
3894 list_del_init(&ns->list);
3895 up_write(&ctrl->namespaces_rwsem);
3897 mutex_lock(&ctrl->subsys->lock);
3898 list_del_rcu(&ns->siblings);
3899 if (list_empty(&ns->head->list))
3900 list_del_init(&ns->head->entry);
3901 mutex_unlock(&ctrl->subsys->lock);
3902 nvme_put_ns_head(ns->head);
3904 blk_cleanup_disk(disk);
3911 static void nvme_ns_remove(struct nvme_ns *ns)
3913 bool last_path = false;
3915 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3918 clear_bit(NVME_NS_READY, &ns->flags);
3919 set_capacity(ns->disk, 0);
3920 nvme_fault_inject_fini(&ns->fault_inject);
3922 mutex_lock(&ns->ctrl->subsys->lock);
3923 list_del_rcu(&ns->siblings);
3924 if (list_empty(&ns->head->list)) {
3925 list_del_init(&ns->head->entry);
3928 mutex_unlock(&ns->ctrl->subsys->lock);
3930 /* guarantee not available in head->list */
3933 /* wait for concurrent submissions */
3934 if (nvme_mpath_clear_current_path(ns))
3935 synchronize_srcu(&ns->head->srcu);
3937 if (!nvme_ns_head_multipath(ns->head))
3938 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3939 del_gendisk(ns->disk);
3940 blk_cleanup_queue(ns->queue);
3942 down_write(&ns->ctrl->namespaces_rwsem);
3943 list_del_init(&ns->list);
3944 up_write(&ns->ctrl->namespaces_rwsem);
3947 nvme_mpath_shutdown_disk(ns->head);
3951 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3953 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3961 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3963 struct nvme_id_ns *id;
3964 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3966 if (test_bit(NVME_NS_DEAD, &ns->flags))
3969 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3973 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3974 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3975 dev_err(ns->ctrl->device,
3976 "identifiers changed for nsid %d\n", ns->head->ns_id);
3980 ret = nvme_update_ns_info(ns, id);
3986 * Only remove the namespace if we got a fatal error back from the
3987 * device, otherwise ignore the error and just move on.
3989 * TODO: we should probably schedule a delayed retry here.
3991 if (ret > 0 && (ret & NVME_SC_DNR))
3995 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3997 struct nvme_ns_ids ids = { };
4000 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4003 ns = nvme_find_get_ns(ctrl, nsid);
4005 nvme_validate_ns(ns, &ids);
4012 nvme_alloc_ns(ctrl, nsid, &ids);
4015 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4016 dev_warn(ctrl->device,
4017 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4021 if (!nvme_multi_css(ctrl)) {
4022 dev_warn(ctrl->device,
4023 "command set not reported for nsid: %d\n",
4027 nvme_alloc_ns(ctrl, nsid, &ids);
4030 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4036 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4039 struct nvme_ns *ns, *next;
4042 down_write(&ctrl->namespaces_rwsem);
4043 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4044 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4045 list_move_tail(&ns->list, &rm_list);
4047 up_write(&ctrl->namespaces_rwsem);
4049 list_for_each_entry_safe(ns, next, &rm_list, list)
4054 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4056 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4061 if (nvme_ctrl_limited_cns(ctrl))
4064 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4069 struct nvme_command cmd = {
4070 .identify.opcode = nvme_admin_identify,
4071 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4072 .identify.nsid = cpu_to_le32(prev),
4075 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4076 NVME_IDENTIFY_DATA_SIZE);
4078 dev_warn(ctrl->device,
4079 "Identify NS List failed (status=0x%x)\n", ret);
4083 for (i = 0; i < nr_entries; i++) {
4084 u32 nsid = le32_to_cpu(ns_list[i]);
4086 if (!nsid) /* end of the list? */
4088 nvme_validate_or_alloc_ns(ctrl, nsid);
4089 while (++prev < nsid)
4090 nvme_ns_remove_by_nsid(ctrl, prev);
4094 nvme_remove_invalid_namespaces(ctrl, prev);
4100 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4102 struct nvme_id_ctrl *id;
4105 if (nvme_identify_ctrl(ctrl, &id))
4107 nn = le32_to_cpu(id->nn);
4110 for (i = 1; i <= nn; i++)
4111 nvme_validate_or_alloc_ns(ctrl, i);
4113 nvme_remove_invalid_namespaces(ctrl, nn);
4116 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4118 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4122 log = kzalloc(log_size, GFP_KERNEL);
4127 * We need to read the log to clear the AEN, but we don't want to rely
4128 * on it for the changed namespace information as userspace could have
4129 * raced with us in reading the log page, which could cause us to miss
4132 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4133 NVME_CSI_NVM, log, log_size, 0);
4135 dev_warn(ctrl->device,
4136 "reading changed ns log failed: %d\n", error);
4141 static void nvme_scan_work(struct work_struct *work)
4143 struct nvme_ctrl *ctrl =
4144 container_of(work, struct nvme_ctrl, scan_work);
4146 /* No tagset on a live ctrl means IO queues could not created */
4147 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4150 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4151 dev_info(ctrl->device, "rescanning namespaces.\n");
4152 nvme_clear_changed_ns_log(ctrl);
4155 mutex_lock(&ctrl->scan_lock);
4156 if (nvme_scan_ns_list(ctrl) != 0)
4157 nvme_scan_ns_sequential(ctrl);
4158 mutex_unlock(&ctrl->scan_lock);
4162 * This function iterates the namespace list unlocked to allow recovery from
4163 * controller failure. It is up to the caller to ensure the namespace list is
4164 * not modified by scan work while this function is executing.
4166 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4168 struct nvme_ns *ns, *next;
4172 * make sure to requeue I/O to all namespaces as these
4173 * might result from the scan itself and must complete
4174 * for the scan_work to make progress
4176 nvme_mpath_clear_ctrl_paths(ctrl);
4178 /* prevent racing with ns scanning */
4179 flush_work(&ctrl->scan_work);
4182 * The dead states indicates the controller was not gracefully
4183 * disconnected. In that case, we won't be able to flush any data while
4184 * removing the namespaces' disks; fail all the queues now to avoid
4185 * potentially having to clean up the failed sync later.
4187 if (ctrl->state == NVME_CTRL_DEAD)
4188 nvme_kill_queues(ctrl);
4190 /* this is a no-op when called from the controller reset handler */
4191 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4193 down_write(&ctrl->namespaces_rwsem);
4194 list_splice_init(&ctrl->namespaces, &ns_list);
4195 up_write(&ctrl->namespaces_rwsem);
4197 list_for_each_entry_safe(ns, next, &ns_list, list)
4200 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4202 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4204 struct nvme_ctrl *ctrl =
4205 container_of(dev, struct nvme_ctrl, ctrl_device);
4206 struct nvmf_ctrl_options *opts = ctrl->opts;
4209 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4214 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4218 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4219 opts->trsvcid ?: "none");
4223 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4224 opts->host_traddr ?: "none");
4228 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4229 opts->host_iface ?: "none");
4234 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4236 char *envp[2] = { NULL, NULL };
4237 u32 aen_result = ctrl->aen_result;
4239 ctrl->aen_result = 0;
4243 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4246 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4250 static void nvme_async_event_work(struct work_struct *work)
4252 struct nvme_ctrl *ctrl =
4253 container_of(work, struct nvme_ctrl, async_event_work);
4255 nvme_aen_uevent(ctrl);
4256 ctrl->ops->submit_async_event(ctrl);
4259 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4264 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4270 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4273 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4275 struct nvme_fw_slot_info_log *log;
4277 log = kmalloc(sizeof(*log), GFP_KERNEL);
4281 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4282 log, sizeof(*log), 0))
4283 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4287 static void nvme_fw_act_work(struct work_struct *work)
4289 struct nvme_ctrl *ctrl = container_of(work,
4290 struct nvme_ctrl, fw_act_work);
4291 unsigned long fw_act_timeout;
4294 fw_act_timeout = jiffies +
4295 msecs_to_jiffies(ctrl->mtfa * 100);
4297 fw_act_timeout = jiffies +
4298 msecs_to_jiffies(admin_timeout * 1000);
4300 nvme_stop_queues(ctrl);
4301 while (nvme_ctrl_pp_status(ctrl)) {
4302 if (time_after(jiffies, fw_act_timeout)) {
4303 dev_warn(ctrl->device,
4304 "Fw activation timeout, reset controller\n");
4305 nvme_try_sched_reset(ctrl);
4311 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4314 nvme_start_queues(ctrl);
4315 /* read FW slot information to clear the AER */
4316 nvme_get_fw_slot_info(ctrl);
4319 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4321 u32 aer_notice_type = (result & 0xff00) >> 8;
4323 trace_nvme_async_event(ctrl, aer_notice_type);
4325 switch (aer_notice_type) {
4326 case NVME_AER_NOTICE_NS_CHANGED:
4327 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4328 nvme_queue_scan(ctrl);
4330 case NVME_AER_NOTICE_FW_ACT_STARTING:
4332 * We are (ab)using the RESETTING state to prevent subsequent
4333 * recovery actions from interfering with the controller's
4334 * firmware activation.
4336 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4337 queue_work(nvme_wq, &ctrl->fw_act_work);
4339 #ifdef CONFIG_NVME_MULTIPATH
4340 case NVME_AER_NOTICE_ANA:
4341 if (!ctrl->ana_log_buf)
4343 queue_work(nvme_wq, &ctrl->ana_work);
4346 case NVME_AER_NOTICE_DISC_CHANGED:
4347 ctrl->aen_result = result;
4350 dev_warn(ctrl->device, "async event result %08x\n", result);
4354 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4355 volatile union nvme_result *res)
4357 u32 result = le32_to_cpu(res->u32);
4358 u32 aer_type = result & 0x07;
4360 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4364 case NVME_AER_NOTICE:
4365 nvme_handle_aen_notice(ctrl, result);
4367 case NVME_AER_ERROR:
4368 case NVME_AER_SMART:
4371 trace_nvme_async_event(ctrl, aer_type);
4372 ctrl->aen_result = result;
4377 queue_work(nvme_wq, &ctrl->async_event_work);
4379 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4381 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4383 nvme_mpath_stop(ctrl);
4384 nvme_stop_keep_alive(ctrl);
4385 nvme_stop_failfast_work(ctrl);
4386 flush_work(&ctrl->async_event_work);
4387 cancel_work_sync(&ctrl->fw_act_work);
4389 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4391 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4393 nvme_start_keep_alive(ctrl);
4395 nvme_enable_aen(ctrl);
4397 if (ctrl->queue_count > 1) {
4398 nvme_queue_scan(ctrl);
4399 nvme_start_queues(ctrl);
4402 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4404 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4406 nvme_hwmon_exit(ctrl);
4407 nvme_fault_inject_fini(&ctrl->fault_inject);
4408 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4409 cdev_device_del(&ctrl->cdev, ctrl->device);
4410 nvme_put_ctrl(ctrl);
4412 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4414 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4416 struct nvme_effects_log *cel;
4419 xa_for_each(&ctrl->cels, i, cel) {
4420 xa_erase(&ctrl->cels, i);
4424 xa_destroy(&ctrl->cels);
4427 static void nvme_free_ctrl(struct device *dev)
4429 struct nvme_ctrl *ctrl =
4430 container_of(dev, struct nvme_ctrl, ctrl_device);
4431 struct nvme_subsystem *subsys = ctrl->subsys;
4433 if (!subsys || ctrl->instance != subsys->instance)
4434 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4436 nvme_free_cels(ctrl);
4437 nvme_mpath_uninit(ctrl);
4438 __free_page(ctrl->discard_page);
4441 mutex_lock(&nvme_subsystems_lock);
4442 list_del(&ctrl->subsys_entry);
4443 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4444 mutex_unlock(&nvme_subsystems_lock);
4447 ctrl->ops->free_ctrl(ctrl);
4450 nvme_put_subsystem(subsys);
4454 * Initialize a NVMe controller structures. This needs to be called during
4455 * earliest initialization so that we have the initialized structured around
4458 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4459 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4463 ctrl->state = NVME_CTRL_NEW;
4464 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4465 spin_lock_init(&ctrl->lock);
4466 mutex_init(&ctrl->scan_lock);
4467 INIT_LIST_HEAD(&ctrl->namespaces);
4468 xa_init(&ctrl->cels);
4469 init_rwsem(&ctrl->namespaces_rwsem);
4472 ctrl->quirks = quirks;
4473 ctrl->numa_node = NUMA_NO_NODE;
4474 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4475 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4476 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4477 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4478 init_waitqueue_head(&ctrl->state_wq);
4480 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4481 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4482 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4483 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4485 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4487 ctrl->discard_page = alloc_page(GFP_KERNEL);
4488 if (!ctrl->discard_page) {
4493 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4496 ctrl->instance = ret;
4498 device_initialize(&ctrl->ctrl_device);
4499 ctrl->device = &ctrl->ctrl_device;
4500 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4502 ctrl->device->class = nvme_class;
4503 ctrl->device->parent = ctrl->dev;
4504 ctrl->device->groups = nvme_dev_attr_groups;
4505 ctrl->device->release = nvme_free_ctrl;
4506 dev_set_drvdata(ctrl->device, ctrl);
4507 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4509 goto out_release_instance;
4511 nvme_get_ctrl(ctrl);
4512 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4513 ctrl->cdev.owner = ops->module;
4514 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4519 * Initialize latency tolerance controls. The sysfs files won't
4520 * be visible to userspace unless the device actually supports APST.
4522 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4523 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4524 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4526 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4527 nvme_mpath_init_ctrl(ctrl);
4531 nvme_put_ctrl(ctrl);
4532 kfree_const(ctrl->device->kobj.name);
4533 out_release_instance:
4534 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4536 if (ctrl->discard_page)
4537 __free_page(ctrl->discard_page);
4540 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4542 static void nvme_start_ns_queue(struct nvme_ns *ns)
4544 if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4545 blk_mq_unquiesce_queue(ns->queue);
4548 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4550 if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4551 blk_mq_quiesce_queue(ns->queue);
4553 blk_mq_wait_quiesce_done(ns->queue);
4557 * Prepare a queue for teardown.
4559 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4560 * the capacity to 0 after that to avoid blocking dispatchers that may be
4561 * holding bd_butex. This will end buffered writers dirtying pages that can't
4564 static void nvme_set_queue_dying(struct nvme_ns *ns)
4566 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4569 blk_set_queue_dying(ns->queue);
4570 nvme_start_ns_queue(ns);
4572 set_capacity_and_notify(ns->disk, 0);
4576 * nvme_kill_queues(): Ends all namespace queues
4577 * @ctrl: the dead controller that needs to end
4579 * Call this function when the driver determines it is unable to get the
4580 * controller in a state capable of servicing IO.
4582 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4586 down_read(&ctrl->namespaces_rwsem);
4588 /* Forcibly unquiesce queues to avoid blocking dispatch */
4589 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4590 nvme_start_admin_queue(ctrl);
4592 list_for_each_entry(ns, &ctrl->namespaces, list)
4593 nvme_set_queue_dying(ns);
4595 up_read(&ctrl->namespaces_rwsem);
4597 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4599 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4603 down_read(&ctrl->namespaces_rwsem);
4604 list_for_each_entry(ns, &ctrl->namespaces, list)
4605 blk_mq_unfreeze_queue(ns->queue);
4606 up_read(&ctrl->namespaces_rwsem);
4608 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4610 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4614 down_read(&ctrl->namespaces_rwsem);
4615 list_for_each_entry(ns, &ctrl->namespaces, list) {
4616 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4620 up_read(&ctrl->namespaces_rwsem);
4623 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4625 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4629 down_read(&ctrl->namespaces_rwsem);
4630 list_for_each_entry(ns, &ctrl->namespaces, list)
4631 blk_mq_freeze_queue_wait(ns->queue);
4632 up_read(&ctrl->namespaces_rwsem);
4634 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4636 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4640 down_read(&ctrl->namespaces_rwsem);
4641 list_for_each_entry(ns, &ctrl->namespaces, list)
4642 blk_freeze_queue_start(ns->queue);
4643 up_read(&ctrl->namespaces_rwsem);
4645 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4647 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4651 down_read(&ctrl->namespaces_rwsem);
4652 list_for_each_entry(ns, &ctrl->namespaces, list)
4653 nvme_stop_ns_queue(ns);
4654 up_read(&ctrl->namespaces_rwsem);
4656 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4658 void nvme_start_queues(struct nvme_ctrl *ctrl)
4662 down_read(&ctrl->namespaces_rwsem);
4663 list_for_each_entry(ns, &ctrl->namespaces, list)
4664 nvme_start_ns_queue(ns);
4665 up_read(&ctrl->namespaces_rwsem);
4667 EXPORT_SYMBOL_GPL(nvme_start_queues);
4669 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4671 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4672 blk_mq_quiesce_queue(ctrl->admin_q);
4674 blk_mq_wait_quiesce_done(ctrl->admin_q);
4676 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4678 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4680 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4681 blk_mq_unquiesce_queue(ctrl->admin_q);
4683 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4685 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4689 down_read(&ctrl->namespaces_rwsem);
4690 list_for_each_entry(ns, &ctrl->namespaces, list)
4691 blk_sync_queue(ns->queue);
4692 up_read(&ctrl->namespaces_rwsem);
4694 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4696 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4698 nvme_sync_io_queues(ctrl);
4700 blk_sync_queue(ctrl->admin_q);
4702 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4704 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4706 if (file->f_op != &nvme_dev_fops)
4708 return file->private_data;
4710 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4713 * Check we didn't inadvertently grow the command structure sizes:
4715 static inline void _nvme_check_size(void)
4717 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4718 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4719 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4720 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4721 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4722 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4723 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4724 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4725 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4726 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4727 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4728 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4729 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4730 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4731 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4732 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4733 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4734 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4735 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4736 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4740 static int __init nvme_core_init(void)
4742 int result = -ENOMEM;
4746 nvme_wq = alloc_workqueue("nvme-wq",
4747 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4751 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4752 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4756 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4757 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4758 if (!nvme_delete_wq)
4759 goto destroy_reset_wq;
4761 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4762 NVME_MINORS, "nvme");
4764 goto destroy_delete_wq;
4766 nvme_class = class_create(THIS_MODULE, "nvme");
4767 if (IS_ERR(nvme_class)) {
4768 result = PTR_ERR(nvme_class);
4769 goto unregister_chrdev;
4771 nvme_class->dev_uevent = nvme_class_uevent;
4773 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4774 if (IS_ERR(nvme_subsys_class)) {
4775 result = PTR_ERR(nvme_subsys_class);
4779 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4782 goto destroy_subsys_class;
4784 nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4785 if (IS_ERR(nvme_ns_chr_class)) {
4786 result = PTR_ERR(nvme_ns_chr_class);
4787 goto unregister_generic_ns;
4792 unregister_generic_ns:
4793 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4794 destroy_subsys_class:
4795 class_destroy(nvme_subsys_class);
4797 class_destroy(nvme_class);
4799 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4801 destroy_workqueue(nvme_delete_wq);
4803 destroy_workqueue(nvme_reset_wq);
4805 destroy_workqueue(nvme_wq);
4810 static void __exit nvme_core_exit(void)
4812 class_destroy(nvme_ns_chr_class);
4813 class_destroy(nvme_subsys_class);
4814 class_destroy(nvme_class);
4815 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4816 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4817 destroy_workqueue(nvme_delete_wq);
4818 destroy_workqueue(nvme_reset_wq);
4819 destroy_workqueue(nvme_wq);
4820 ida_destroy(&nvme_ns_chr_minor_ida);
4821 ida_destroy(&nvme_instance_ida);
4824 MODULE_LICENSE("GPL");
4825 MODULE_VERSION("1.0");
4826 module_init(nvme_core_init);
4827 module_exit(nvme_core_exit);