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_DEAD &&
670 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
671 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
672 return BLK_STS_RESOURCE;
673 return nvme_host_path_error(rq);
675 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
677 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
680 struct nvme_request *req = nvme_req(rq);
683 * currently we have a problem sending passthru commands
684 * on the admin_q if the controller is not LIVE because we can't
685 * make sure that they are going out after the admin connect,
686 * controller enable and/or other commands in the initialization
687 * sequence. until the controller will be LIVE, fail with
688 * BLK_STS_RESOURCE so that they will be rescheduled.
690 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
693 if (ctrl->ops->flags & NVME_F_FABRICS) {
695 * Only allow commands on a live queue, except for the connect
696 * command, which is require to set the queue live in the
697 * appropinquate states.
699 switch (ctrl->state) {
700 case NVME_CTRL_CONNECTING:
701 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
702 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
714 EXPORT_SYMBOL_GPL(__nvme_check_ready);
716 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
718 struct nvme_command c = { };
720 c.directive.opcode = nvme_admin_directive_send;
721 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
722 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
723 c.directive.dtype = NVME_DIR_IDENTIFY;
724 c.directive.tdtype = NVME_DIR_STREAMS;
725 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
727 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
730 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
732 return nvme_toggle_streams(ctrl, false);
735 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
737 return nvme_toggle_streams(ctrl, true);
740 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
741 struct streams_directive_params *s, u32 nsid)
743 struct nvme_command c = { };
745 memset(s, 0, sizeof(*s));
747 c.directive.opcode = nvme_admin_directive_recv;
748 c.directive.nsid = cpu_to_le32(nsid);
749 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
750 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
751 c.directive.dtype = NVME_DIR_STREAMS;
753 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
756 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
758 struct streams_directive_params s;
761 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
766 ret = nvme_enable_streams(ctrl);
770 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
772 goto out_disable_stream;
774 ctrl->nssa = le16_to_cpu(s.nssa);
775 if (ctrl->nssa < BLK_MAX_WRITE_HINTS - 1) {
776 dev_info(ctrl->device, "too few streams (%u) available\n",
778 goto out_disable_stream;
781 ctrl->nr_streams = min_t(u16, ctrl->nssa, BLK_MAX_WRITE_HINTS - 1);
782 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
786 nvme_disable_streams(ctrl);
791 * Check if 'req' has a write hint associated with it. If it does, assign
792 * a valid namespace stream to the write.
794 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
795 struct request *req, u16 *control,
798 enum rw_hint streamid = req->write_hint;
800 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
804 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
807 *control |= NVME_RW_DTYPE_STREAMS;
808 *dsmgmt |= streamid << 16;
811 if (streamid < ARRAY_SIZE(req->q->write_hints))
812 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
815 static inline void nvme_setup_flush(struct nvme_ns *ns,
816 struct nvme_command *cmnd)
818 memset(cmnd, 0, sizeof(*cmnd));
819 cmnd->common.opcode = nvme_cmd_flush;
820 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
823 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
824 struct nvme_command *cmnd)
826 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
827 struct nvme_dsm_range *range;
831 * Some devices do not consider the DSM 'Number of Ranges' field when
832 * determining how much data to DMA. Always allocate memory for maximum
833 * number of segments to prevent device reading beyond end of buffer.
835 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
837 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
840 * If we fail allocation our range, fallback to the controller
841 * discard page. If that's also busy, it's safe to return
842 * busy, as we know we can make progress once that's freed.
844 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
845 return BLK_STS_RESOURCE;
847 range = page_address(ns->ctrl->discard_page);
850 __rq_for_each_bio(bio, req) {
851 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
852 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
855 range[n].cattr = cpu_to_le32(0);
856 range[n].nlb = cpu_to_le32(nlb);
857 range[n].slba = cpu_to_le64(slba);
862 if (WARN_ON_ONCE(n != segments)) {
863 if (virt_to_page(range) == ns->ctrl->discard_page)
864 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
867 return BLK_STS_IOERR;
870 memset(cmnd, 0, sizeof(*cmnd));
871 cmnd->dsm.opcode = nvme_cmd_dsm;
872 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
873 cmnd->dsm.nr = cpu_to_le32(segments - 1);
874 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
876 req->special_vec.bv_page = virt_to_page(range);
877 req->special_vec.bv_offset = offset_in_page(range);
878 req->special_vec.bv_len = alloc_size;
879 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
884 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
885 struct request *req, struct nvme_command *cmnd)
887 memset(cmnd, 0, sizeof(*cmnd));
889 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
890 return nvme_setup_discard(ns, req, cmnd);
892 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
893 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
894 cmnd->write_zeroes.slba =
895 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
896 cmnd->write_zeroes.length =
897 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
899 if (nvme_ns_has_pi(ns)) {
900 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
902 switch (ns->pi_type) {
903 case NVME_NS_DPS_PI_TYPE1:
904 case NVME_NS_DPS_PI_TYPE2:
905 cmnd->write_zeroes.reftag =
906 cpu_to_le32(t10_pi_ref_tag(req));
914 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
915 struct request *req, struct nvme_command *cmnd,
918 struct nvme_ctrl *ctrl = ns->ctrl;
922 if (req->cmd_flags & REQ_FUA)
923 control |= NVME_RW_FUA;
924 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
925 control |= NVME_RW_LR;
927 if (req->cmd_flags & REQ_RAHEAD)
928 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
930 cmnd->rw.opcode = op;
932 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
934 cmnd->rw.metadata = 0;
935 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
936 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
939 cmnd->rw.appmask = 0;
941 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
942 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
946 * If formated with metadata, the block layer always provides a
947 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
948 * we enable the PRACT bit for protection information or set the
949 * namespace capacity to zero to prevent any I/O.
951 if (!blk_integrity_rq(req)) {
952 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
953 return BLK_STS_NOTSUPP;
954 control |= NVME_RW_PRINFO_PRACT;
957 switch (ns->pi_type) {
958 case NVME_NS_DPS_PI_TYPE3:
959 control |= NVME_RW_PRINFO_PRCHK_GUARD;
961 case NVME_NS_DPS_PI_TYPE1:
962 case NVME_NS_DPS_PI_TYPE2:
963 control |= NVME_RW_PRINFO_PRCHK_GUARD |
964 NVME_RW_PRINFO_PRCHK_REF;
965 if (op == nvme_cmd_zone_append)
966 control |= NVME_RW_APPEND_PIREMAP;
967 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
972 cmnd->rw.control = cpu_to_le16(control);
973 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
977 void nvme_cleanup_cmd(struct request *req)
979 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
980 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
982 if (req->special_vec.bv_page == ctrl->discard_page)
983 clear_bit_unlock(0, &ctrl->discard_page_busy);
985 kfree(bvec_virt(&req->special_vec));
988 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
990 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
992 struct nvme_command *cmd = nvme_req(req)->cmd;
993 blk_status_t ret = BLK_STS_OK;
995 if (!(req->rq_flags & RQF_DONTPREP))
996 nvme_clear_nvme_request(req);
998 switch (req_op(req)) {
1000 case REQ_OP_DRV_OUT:
1001 /* these are setup prior to execution in nvme_init_request() */
1004 nvme_setup_flush(ns, cmd);
1006 case REQ_OP_ZONE_RESET_ALL:
1007 case REQ_OP_ZONE_RESET:
1008 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1010 case REQ_OP_ZONE_OPEN:
1011 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1013 case REQ_OP_ZONE_CLOSE:
1014 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1016 case REQ_OP_ZONE_FINISH:
1017 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1019 case REQ_OP_WRITE_ZEROES:
1020 ret = nvme_setup_write_zeroes(ns, req, cmd);
1022 case REQ_OP_DISCARD:
1023 ret = nvme_setup_discard(ns, req, cmd);
1026 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1029 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1031 case REQ_OP_ZONE_APPEND:
1032 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1036 return BLK_STS_IOERR;
1039 cmd->common.command_id = nvme_cid(req);
1040 trace_nvme_setup_cmd(req, cmd);
1043 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1048 * >0: nvme controller's cqe status response
1049 * <0: kernel error in lieu of controller response
1051 static int nvme_execute_rq(struct gendisk *disk, struct request *rq,
1054 blk_status_t status;
1056 status = blk_execute_rq(rq, at_head);
1057 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1059 if (nvme_req(rq)->status)
1060 return nvme_req(rq)->status;
1061 return blk_status_to_errno(status);
1065 * Returns 0 on success. If the result is negative, it's a Linux error code;
1066 * if the result is positive, it's an NVM Express status code
1068 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1069 union nvme_result *result, void *buffer, unsigned bufflen,
1070 unsigned timeout, int qid, int at_head,
1071 blk_mq_req_flags_t flags)
1073 struct request *req;
1076 if (qid == NVME_QID_ANY)
1077 req = nvme_alloc_request(q, cmd, flags);
1079 req = nvme_alloc_request_qid(q, cmd, flags, qid);
1081 return PTR_ERR(req);
1084 req->timeout = timeout;
1086 if (buffer && bufflen) {
1087 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1092 ret = nvme_execute_rq(NULL, req, at_head);
1093 if (result && ret >= 0)
1094 *result = nvme_req(req)->result;
1096 blk_mq_free_request(req);
1099 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1101 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1102 void *buffer, unsigned bufflen)
1104 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1105 NVME_QID_ANY, 0, 0);
1107 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1109 static u32 nvme_known_admin_effects(u8 opcode)
1112 case nvme_admin_format_nvm:
1113 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1114 NVME_CMD_EFFECTS_CSE_MASK;
1115 case nvme_admin_sanitize_nvm:
1116 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1123 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1128 if (ns->head->effects)
1129 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1130 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1131 dev_warn_once(ctrl->device,
1132 "IO command:%02x has unhandled effects:%08x\n",
1138 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1139 effects |= nvme_known_admin_effects(opcode);
1143 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1145 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1148 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1151 * For simplicity, IO to all namespaces is quiesced even if the command
1152 * effects say only one namespace is affected.
1154 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1155 mutex_lock(&ctrl->scan_lock);
1156 mutex_lock(&ctrl->subsys->lock);
1157 nvme_mpath_start_freeze(ctrl->subsys);
1158 nvme_mpath_wait_freeze(ctrl->subsys);
1159 nvme_start_freeze(ctrl);
1160 nvme_wait_freeze(ctrl);
1165 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1166 struct nvme_command *cmd, int status)
1168 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1169 nvme_unfreeze(ctrl);
1170 nvme_mpath_unfreeze(ctrl->subsys);
1171 mutex_unlock(&ctrl->subsys->lock);
1172 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1173 mutex_unlock(&ctrl->scan_lock);
1175 if (effects & NVME_CMD_EFFECTS_CCC)
1176 nvme_init_ctrl_finish(ctrl);
1177 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1178 nvme_queue_scan(ctrl);
1179 flush_work(&ctrl->scan_work);
1182 switch (cmd->common.opcode) {
1183 case nvme_admin_set_features:
1184 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1185 case NVME_FEAT_KATO:
1187 * Keep alive commands interval on the host should be
1188 * updated when KATO is modified by Set Features
1192 nvme_update_keep_alive(ctrl, cmd);
1203 int nvme_execute_passthru_rq(struct request *rq)
1205 struct nvme_command *cmd = nvme_req(rq)->cmd;
1206 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1207 struct nvme_ns *ns = rq->q->queuedata;
1208 struct gendisk *disk = ns ? ns->disk : NULL;
1212 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1213 ret = nvme_execute_rq(disk, rq, false);
1214 if (effects) /* nothing to be done for zero cmd effects */
1215 nvme_passthru_end(ctrl, effects, cmd, ret);
1219 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1222 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1224 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1225 * accounting for transport roundtrip times [..].
1227 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1229 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1232 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1234 struct nvme_ctrl *ctrl = rq->end_io_data;
1235 unsigned long flags;
1236 bool startka = false;
1238 blk_mq_free_request(rq);
1241 dev_err(ctrl->device,
1242 "failed nvme_keep_alive_end_io error=%d\n",
1247 ctrl->comp_seen = false;
1248 spin_lock_irqsave(&ctrl->lock, flags);
1249 if (ctrl->state == NVME_CTRL_LIVE ||
1250 ctrl->state == NVME_CTRL_CONNECTING)
1252 spin_unlock_irqrestore(&ctrl->lock, flags);
1254 nvme_queue_keep_alive_work(ctrl);
1257 static void nvme_keep_alive_work(struct work_struct *work)
1259 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1260 struct nvme_ctrl, ka_work);
1261 bool comp_seen = ctrl->comp_seen;
1264 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1265 dev_dbg(ctrl->device,
1266 "reschedule traffic based keep-alive timer\n");
1267 ctrl->comp_seen = false;
1268 nvme_queue_keep_alive_work(ctrl);
1272 rq = nvme_alloc_request(ctrl->admin_q, &ctrl->ka_cmd,
1273 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1275 /* allocation failure, reset the controller */
1276 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1277 nvme_reset_ctrl(ctrl);
1281 rq->timeout = ctrl->kato * HZ;
1282 rq->end_io_data = ctrl;
1283 blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1286 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1288 if (unlikely(ctrl->kato == 0))
1291 nvme_queue_keep_alive_work(ctrl);
1294 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1296 if (unlikely(ctrl->kato == 0))
1299 cancel_delayed_work_sync(&ctrl->ka_work);
1301 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1303 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1304 struct nvme_command *cmd)
1306 unsigned int new_kato =
1307 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1309 dev_info(ctrl->device,
1310 "keep alive interval updated from %u ms to %u ms\n",
1311 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1313 nvme_stop_keep_alive(ctrl);
1314 ctrl->kato = new_kato;
1315 nvme_start_keep_alive(ctrl);
1319 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1320 * flag, thus sending any new CNS opcodes has a big chance of not working.
1321 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1322 * (but not for any later version).
1324 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1326 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1327 return ctrl->vs < NVME_VS(1, 2, 0);
1328 return ctrl->vs < NVME_VS(1, 1, 0);
1331 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1333 struct nvme_command c = { };
1336 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1337 c.identify.opcode = nvme_admin_identify;
1338 c.identify.cns = NVME_ID_CNS_CTRL;
1340 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1344 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1345 sizeof(struct nvme_id_ctrl));
1351 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1352 struct nvme_ns_id_desc *cur, bool *csi_seen)
1354 const char *warn_str = "ctrl returned bogus length:";
1357 switch (cur->nidt) {
1358 case NVME_NIDT_EUI64:
1359 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1360 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1361 warn_str, cur->nidl);
1364 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1365 return NVME_NIDT_EUI64_LEN;
1366 case NVME_NIDT_NGUID:
1367 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1368 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1369 warn_str, cur->nidl);
1372 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1373 return NVME_NIDT_NGUID_LEN;
1374 case NVME_NIDT_UUID:
1375 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1376 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1377 warn_str, cur->nidl);
1380 uuid_copy(&ids->uuid, data + sizeof(*cur));
1381 return NVME_NIDT_UUID_LEN;
1383 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1384 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1385 warn_str, cur->nidl);
1388 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1390 return NVME_NIDT_CSI_LEN;
1392 /* Skip unknown types */
1397 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1398 struct nvme_ns_ids *ids)
1400 struct nvme_command c = { };
1401 bool csi_seen = false;
1402 int status, pos, len;
1405 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1407 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1410 c.identify.opcode = nvme_admin_identify;
1411 c.identify.nsid = cpu_to_le32(nsid);
1412 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1414 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1418 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1419 NVME_IDENTIFY_DATA_SIZE);
1421 dev_warn(ctrl->device,
1422 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1427 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1428 struct nvme_ns_id_desc *cur = data + pos;
1433 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1437 len += sizeof(*cur);
1440 if (nvme_multi_css(ctrl) && !csi_seen) {
1441 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1451 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1452 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1454 struct nvme_command c = { };
1457 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1458 c.identify.opcode = nvme_admin_identify;
1459 c.identify.nsid = cpu_to_le32(nsid);
1460 c.identify.cns = NVME_ID_CNS_NS;
1462 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1466 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1468 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1472 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1473 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1476 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1477 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1478 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1479 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1480 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1481 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1490 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1491 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1493 union nvme_result res = { 0 };
1494 struct nvme_command c = { };
1497 c.features.opcode = op;
1498 c.features.fid = cpu_to_le32(fid);
1499 c.features.dword11 = cpu_to_le32(dword11);
1501 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1502 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1503 if (ret >= 0 && result)
1504 *result = le32_to_cpu(res.u32);
1508 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1509 unsigned int dword11, void *buffer, size_t buflen,
1512 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1515 EXPORT_SYMBOL_GPL(nvme_set_features);
1517 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1518 unsigned int dword11, void *buffer, size_t buflen,
1521 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1524 EXPORT_SYMBOL_GPL(nvme_get_features);
1526 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1528 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1530 int status, nr_io_queues;
1532 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1538 * Degraded controllers might return an error when setting the queue
1539 * count. We still want to be able to bring them online and offer
1540 * access to the admin queue, as that might be only way to fix them up.
1543 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1546 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1547 *count = min(*count, nr_io_queues);
1552 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1554 #define NVME_AEN_SUPPORTED \
1555 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1556 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1558 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1560 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1563 if (!supported_aens)
1566 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1569 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1572 queue_work(nvme_wq, &ctrl->async_event_work);
1575 static int nvme_ns_open(struct nvme_ns *ns)
1578 /* should never be called due to GENHD_FL_HIDDEN */
1579 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1581 if (!nvme_get_ns(ns))
1583 if (!try_module_get(ns->ctrl->ops->module))
1594 static void nvme_ns_release(struct nvme_ns *ns)
1597 module_put(ns->ctrl->ops->module);
1601 static int nvme_open(struct block_device *bdev, fmode_t mode)
1603 return nvme_ns_open(bdev->bd_disk->private_data);
1606 static void nvme_release(struct gendisk *disk, fmode_t mode)
1608 nvme_ns_release(disk->private_data);
1611 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1613 /* some standard values */
1614 geo->heads = 1 << 6;
1615 geo->sectors = 1 << 5;
1616 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1620 #ifdef CONFIG_BLK_DEV_INTEGRITY
1621 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1622 u32 max_integrity_segments)
1624 struct blk_integrity integrity = { };
1627 case NVME_NS_DPS_PI_TYPE3:
1628 integrity.profile = &t10_pi_type3_crc;
1629 integrity.tag_size = sizeof(u16) + sizeof(u32);
1630 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1632 case NVME_NS_DPS_PI_TYPE1:
1633 case NVME_NS_DPS_PI_TYPE2:
1634 integrity.profile = &t10_pi_type1_crc;
1635 integrity.tag_size = sizeof(u16);
1636 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1639 integrity.profile = NULL;
1642 integrity.tuple_size = ms;
1643 blk_integrity_register(disk, &integrity);
1644 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1647 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1648 u32 max_integrity_segments)
1651 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1653 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1655 struct nvme_ctrl *ctrl = ns->ctrl;
1656 struct request_queue *queue = disk->queue;
1657 u32 size = queue_logical_block_size(queue);
1659 if (ctrl->max_discard_sectors == 0) {
1660 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1664 if (ctrl->nr_streams && ns->sws && ns->sgs)
1665 size *= ns->sws * ns->sgs;
1667 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1668 NVME_DSM_MAX_RANGES);
1670 queue->limits.discard_alignment = 0;
1671 queue->limits.discard_granularity = size;
1673 /* If discard is already enabled, don't reset queue limits */
1674 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1677 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1678 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1680 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1681 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1684 static bool nvme_ns_ids_valid(struct nvme_ns_ids *ids)
1686 return !uuid_is_null(&ids->uuid) ||
1687 memchr_inv(ids->nguid, 0, sizeof(ids->nguid)) ||
1688 memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
1691 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1693 return uuid_equal(&a->uuid, &b->uuid) &&
1694 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1695 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1699 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1700 u32 *phys_bs, u32 *io_opt)
1702 struct streams_directive_params s;
1705 if (!ctrl->nr_streams)
1708 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1712 ns->sws = le32_to_cpu(s.sws);
1713 ns->sgs = le16_to_cpu(s.sgs);
1716 *phys_bs = ns->sws * (1 << ns->lba_shift);
1718 *io_opt = *phys_bs * ns->sgs;
1724 static int nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1726 struct nvme_ctrl *ctrl = ns->ctrl;
1729 * The PI implementation requires the metadata size to be equal to the
1730 * t10 pi tuple size.
1732 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1733 if (ns->ms == sizeof(struct t10_pi_tuple))
1734 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1738 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1739 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1741 if (ctrl->ops->flags & NVME_F_FABRICS) {
1743 * The NVMe over Fabrics specification only supports metadata as
1744 * part of the extended data LBA. We rely on HCA/HBA support to
1745 * remap the separate metadata buffer from the block layer.
1747 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1749 if (ctrl->max_integrity_segments)
1751 (NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1754 * For PCIe controllers, we can't easily remap the separate
1755 * metadata buffer from the block layer and thus require a
1756 * separate metadata buffer for block layer metadata/PI support.
1757 * We allow extended LBAs for the passthrough interface, though.
1759 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1760 ns->features |= NVME_NS_EXT_LBAS;
1762 ns->features |= NVME_NS_METADATA_SUPPORTED;
1768 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1769 struct request_queue *q)
1771 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1773 if (ctrl->max_hw_sectors) {
1775 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1777 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1778 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1779 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1781 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1782 blk_queue_dma_alignment(q, 7);
1783 blk_queue_write_cache(q, vwc, vwc);
1786 static void nvme_update_disk_info(struct gendisk *disk,
1787 struct nvme_ns *ns, struct nvme_id_ns *id)
1789 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1790 unsigned short bs = 1 << ns->lba_shift;
1791 u32 atomic_bs, phys_bs, io_opt = 0;
1794 * The block layer can't support LBA sizes larger than the page size
1795 * yet, so catch this early and don't allow block I/O.
1797 if (ns->lba_shift > PAGE_SHIFT) {
1802 blk_integrity_unregister(disk);
1804 atomic_bs = phys_bs = bs;
1805 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1806 if (id->nabo == 0) {
1808 * Bit 1 indicates whether NAWUPF is defined for this namespace
1809 * and whether it should be used instead of AWUPF. If NAWUPF ==
1810 * 0 then AWUPF must be used instead.
1812 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1813 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1815 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1818 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1819 /* NPWG = Namespace Preferred Write Granularity */
1820 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1821 /* NOWS = Namespace Optimal Write Size */
1822 io_opt = bs * (1 + le16_to_cpu(id->nows));
1825 blk_queue_logical_block_size(disk->queue, bs);
1827 * Linux filesystems assume writing a single physical block is
1828 * an atomic operation. Hence limit the physical block size to the
1829 * value of the Atomic Write Unit Power Fail parameter.
1831 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1832 blk_queue_io_min(disk->queue, phys_bs);
1833 blk_queue_io_opt(disk->queue, io_opt);
1836 * Register a metadata profile for PI, or the plain non-integrity NVMe
1837 * metadata masquerading as Type 0 if supported, otherwise reject block
1838 * I/O to namespaces with metadata except when the namespace supports
1839 * PI, as it can strip/insert in that case.
1842 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1843 (ns->features & NVME_NS_METADATA_SUPPORTED))
1844 nvme_init_integrity(disk, ns->ms, ns->pi_type,
1845 ns->ctrl->max_integrity_segments);
1846 else if (!nvme_ns_has_pi(ns))
1850 set_capacity_and_notify(disk, capacity);
1852 nvme_config_discard(disk, ns);
1853 blk_queue_max_write_zeroes_sectors(disk->queue,
1854 ns->ctrl->max_zeroes_sectors);
1856 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1857 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1860 static inline bool nvme_first_scan(struct gendisk *disk)
1862 /* nvme_alloc_ns() scans the disk prior to adding it */
1863 return !disk_live(disk);
1866 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1868 struct nvme_ctrl *ctrl = ns->ctrl;
1871 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1872 is_power_of_2(ctrl->max_hw_sectors))
1873 iob = ctrl->max_hw_sectors;
1875 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1880 if (!is_power_of_2(iob)) {
1881 if (nvme_first_scan(ns->disk))
1882 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1883 ns->disk->disk_name, iob);
1887 if (blk_queue_is_zoned(ns->disk->queue)) {
1888 if (nvme_first_scan(ns->disk))
1889 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1890 ns->disk->disk_name);
1894 blk_queue_chunk_sectors(ns->queue, iob);
1897 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1899 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1902 blk_mq_freeze_queue(ns->disk->queue);
1903 ns->lba_shift = id->lbaf[lbaf].ds;
1904 nvme_set_queue_limits(ns->ctrl, ns->queue);
1906 ret = nvme_configure_metadata(ns, id);
1909 nvme_set_chunk_sectors(ns, id);
1910 nvme_update_disk_info(ns->disk, ns, id);
1912 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1913 ret = nvme_update_zone_info(ns, lbaf);
1918 set_bit(NVME_NS_READY, &ns->flags);
1919 blk_mq_unfreeze_queue(ns->disk->queue);
1921 if (blk_queue_is_zoned(ns->queue)) {
1922 ret = nvme_revalidate_zones(ns);
1923 if (ret && !nvme_first_scan(ns->disk))
1927 if (nvme_ns_head_multipath(ns->head)) {
1928 blk_mq_freeze_queue(ns->head->disk->queue);
1929 nvme_update_disk_info(ns->head->disk, ns, id);
1930 nvme_mpath_revalidate_paths(ns);
1931 blk_stack_limits(&ns->head->disk->queue->limits,
1932 &ns->queue->limits, 0);
1933 disk_update_readahead(ns->head->disk);
1934 blk_mq_unfreeze_queue(ns->head->disk->queue);
1939 blk_mq_unfreeze_queue(ns->disk->queue);
1942 * If probing fails due an unsupported feature, hide the block device,
1943 * but still allow other access.
1945 if (ret == -ENODEV) {
1946 ns->disk->flags |= GENHD_FL_HIDDEN;
1952 static char nvme_pr_type(enum pr_type type)
1955 case PR_WRITE_EXCLUSIVE:
1957 case PR_EXCLUSIVE_ACCESS:
1959 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1961 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1963 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1965 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1972 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1973 struct nvme_command *c, u8 data[16])
1975 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1976 int srcu_idx = srcu_read_lock(&head->srcu);
1977 struct nvme_ns *ns = nvme_find_path(head);
1978 int ret = -EWOULDBLOCK;
1981 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1982 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1984 srcu_read_unlock(&head->srcu, srcu_idx);
1988 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
1991 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1992 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
1995 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
1996 u64 key, u64 sa_key, u8 op)
1998 struct nvme_command c = { };
1999 u8 data[16] = { 0, };
2001 put_unaligned_le64(key, &data[0]);
2002 put_unaligned_le64(sa_key, &data[8]);
2004 c.common.opcode = op;
2005 c.common.cdw10 = cpu_to_le32(cdw10);
2007 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2008 bdev->bd_disk->fops == &nvme_ns_head_ops)
2009 return nvme_send_ns_head_pr_command(bdev, &c, data);
2010 return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2013 static int nvme_pr_register(struct block_device *bdev, u64 old,
2014 u64 new, unsigned flags)
2018 if (flags & ~PR_FL_IGNORE_KEY)
2021 cdw10 = old ? 2 : 0;
2022 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2023 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2024 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2027 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2028 enum pr_type type, unsigned flags)
2032 if (flags & ~PR_FL_IGNORE_KEY)
2035 cdw10 = nvme_pr_type(type) << 8;
2036 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2037 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2040 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2041 enum pr_type type, bool abort)
2043 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2045 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2048 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2050 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2052 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2055 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2057 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2059 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2062 const struct pr_ops nvme_pr_ops = {
2063 .pr_register = nvme_pr_register,
2064 .pr_reserve = nvme_pr_reserve,
2065 .pr_release = nvme_pr_release,
2066 .pr_preempt = nvme_pr_preempt,
2067 .pr_clear = nvme_pr_clear,
2070 #ifdef CONFIG_BLK_SED_OPAL
2071 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2074 struct nvme_ctrl *ctrl = data;
2075 struct nvme_command cmd = { };
2078 cmd.common.opcode = nvme_admin_security_send;
2080 cmd.common.opcode = nvme_admin_security_recv;
2081 cmd.common.nsid = 0;
2082 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2083 cmd.common.cdw11 = cpu_to_le32(len);
2085 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2086 NVME_QID_ANY, 1, 0);
2088 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2089 #endif /* CONFIG_BLK_SED_OPAL */
2091 #ifdef CONFIG_BLK_DEV_ZONED
2092 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2093 unsigned int nr_zones, report_zones_cb cb, void *data)
2095 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2099 #define nvme_report_zones NULL
2100 #endif /* CONFIG_BLK_DEV_ZONED */
2102 static const struct block_device_operations nvme_bdev_ops = {
2103 .owner = THIS_MODULE,
2104 .ioctl = nvme_ioctl,
2106 .release = nvme_release,
2107 .getgeo = nvme_getgeo,
2108 .report_zones = nvme_report_zones,
2109 .pr_ops = &nvme_pr_ops,
2112 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2114 unsigned long timeout =
2115 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2116 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2119 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2122 if ((csts & NVME_CSTS_RDY) == bit)
2125 usleep_range(1000, 2000);
2126 if (fatal_signal_pending(current))
2128 if (time_after(jiffies, timeout)) {
2129 dev_err(ctrl->device,
2130 "Device not ready; aborting %s, CSTS=0x%x\n",
2131 enabled ? "initialisation" : "reset", csts);
2140 * If the device has been passed off to us in an enabled state, just clear
2141 * the enabled bit. The spec says we should set the 'shutdown notification
2142 * bits', but doing so may cause the device to complete commands to the
2143 * admin queue ... and we don't know what memory that might be pointing at!
2145 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2149 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2150 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2152 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2156 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2157 msleep(NVME_QUIRK_DELAY_AMOUNT);
2159 return nvme_wait_ready(ctrl, ctrl->cap, false);
2161 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2163 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2165 unsigned dev_page_min;
2168 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2170 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2173 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2175 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2176 dev_err(ctrl->device,
2177 "Minimum device page size %u too large for host (%u)\n",
2178 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2182 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2183 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2185 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2186 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2187 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2188 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2189 ctrl->ctrl_config |= NVME_CC_ENABLE;
2191 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2194 return nvme_wait_ready(ctrl, ctrl->cap, true);
2196 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2198 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2200 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2204 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2205 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2207 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2211 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2212 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2216 if (fatal_signal_pending(current))
2218 if (time_after(jiffies, timeout)) {
2219 dev_err(ctrl->device,
2220 "Device shutdown incomplete; abort shutdown\n");
2227 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2229 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2234 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2237 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2238 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2241 dev_warn_once(ctrl->device,
2242 "could not set timestamp (%d)\n", ret);
2246 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2248 struct nvme_feat_host_behavior *host;
2251 /* Don't bother enabling the feature if retry delay is not reported */
2255 host = kzalloc(sizeof(*host), GFP_KERNEL);
2259 host->acre = NVME_ENABLE_ACRE;
2260 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2261 host, sizeof(*host), NULL);
2267 * The function checks whether the given total (exlat + enlat) latency of
2268 * a power state allows the latter to be used as an APST transition target.
2269 * It does so by comparing the latency to the primary and secondary latency
2270 * tolerances defined by module params. If there's a match, the corresponding
2271 * timeout value is returned and the matching tolerance index (1 or 2) is
2274 static bool nvme_apst_get_transition_time(u64 total_latency,
2275 u64 *transition_time, unsigned *last_index)
2277 if (total_latency <= apst_primary_latency_tol_us) {
2278 if (*last_index == 1)
2281 *transition_time = apst_primary_timeout_ms;
2284 if (apst_secondary_timeout_ms &&
2285 total_latency <= apst_secondary_latency_tol_us) {
2286 if (*last_index <= 2)
2289 *transition_time = apst_secondary_timeout_ms;
2296 * APST (Autonomous Power State Transition) lets us program a table of power
2297 * state transitions that the controller will perform automatically.
2299 * Depending on module params, one of the two supported techniques will be used:
2301 * - If the parameters provide explicit timeouts and tolerances, they will be
2302 * used to build a table with up to 2 non-operational states to transition to.
2303 * The default parameter values were selected based on the values used by
2304 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2305 * regeneration of the APST table in the event of switching between external
2306 * and battery power, the timeouts and tolerances reflect a compromise
2307 * between values used by Microsoft for AC and battery scenarios.
2308 * - If not, we'll configure the table with a simple heuristic: we are willing
2309 * to spend at most 2% of the time transitioning between power states.
2310 * Therefore, when running in any given state, we will enter the next
2311 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2312 * microseconds, as long as that state's exit latency is under the requested
2315 * We will not autonomously enter any non-operational state for which the total
2316 * latency exceeds ps_max_latency_us.
2318 * Users can set ps_max_latency_us to zero to turn off APST.
2320 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2322 struct nvme_feat_auto_pst *table;
2329 unsigned last_lt_index = UINT_MAX;
2332 * If APST isn't supported or if we haven't been initialized yet,
2333 * then don't do anything.
2338 if (ctrl->npss > 31) {
2339 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2343 table = kzalloc(sizeof(*table), GFP_KERNEL);
2347 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2348 /* Turn off APST. */
2349 dev_dbg(ctrl->device, "APST disabled\n");
2354 * Walk through all states from lowest- to highest-power.
2355 * According to the spec, lower-numbered states use more power. NPSS,
2356 * despite the name, is the index of the lowest-power state, not the
2359 for (state = (int)ctrl->npss; state >= 0; state--) {
2360 u64 total_latency_us, exit_latency_us, transition_ms;
2363 table->entries[state] = target;
2366 * Don't allow transitions to the deepest state if it's quirked
2369 if (state == ctrl->npss &&
2370 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2374 * Is this state a useful non-operational state for higher-power
2375 * states to autonomously transition to?
2377 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2380 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2381 if (exit_latency_us > ctrl->ps_max_latency_us)
2384 total_latency_us = exit_latency_us +
2385 le32_to_cpu(ctrl->psd[state].entry_lat);
2388 * This state is good. It can be used as the APST idle target
2389 * for higher power states.
2391 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2392 if (!nvme_apst_get_transition_time(total_latency_us,
2393 &transition_ms, &last_lt_index))
2396 transition_ms = total_latency_us + 19;
2397 do_div(transition_ms, 20);
2398 if (transition_ms > (1 << 24) - 1)
2399 transition_ms = (1 << 24) - 1;
2402 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2405 if (total_latency_us > max_lat_us)
2406 max_lat_us = total_latency_us;
2410 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2412 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2413 max_ps, max_lat_us, (int)sizeof(*table), table);
2417 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2418 table, sizeof(*table), NULL);
2420 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2425 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2427 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2431 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2432 case PM_QOS_LATENCY_ANY:
2440 if (ctrl->ps_max_latency_us != latency) {
2441 ctrl->ps_max_latency_us = latency;
2442 if (ctrl->state == NVME_CTRL_LIVE)
2443 nvme_configure_apst(ctrl);
2447 struct nvme_core_quirk_entry {
2449 * NVMe model and firmware strings are padded with spaces. For
2450 * simplicity, strings in the quirk table are padded with NULLs
2456 unsigned long quirks;
2459 static const struct nvme_core_quirk_entry core_quirks[] = {
2462 * This Toshiba device seems to die using any APST states. See:
2463 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2466 .mn = "THNSF5256GPUK TOSHIBA",
2467 .quirks = NVME_QUIRK_NO_APST,
2471 * This LiteON CL1-3D*-Q11 firmware version has a race
2472 * condition associated with actions related to suspend to idle
2473 * LiteON has resolved the problem in future firmware
2477 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2481 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2482 * aborts I/O during any load, but more easily reproducible
2483 * with discards (fstrim).
2485 * The device is left in a state where it is also not possible
2486 * to use "nvme set-feature" to disable APST, but booting with
2487 * nvme_core.default_ps_max_latency=0 works.
2490 .mn = "KCD6XVUL6T40",
2491 .quirks = NVME_QUIRK_NO_APST,
2495 /* match is null-terminated but idstr is space-padded. */
2496 static bool string_matches(const char *idstr, const char *match, size_t len)
2503 matchlen = strlen(match);
2504 WARN_ON_ONCE(matchlen > len);
2506 if (memcmp(idstr, match, matchlen))
2509 for (; matchlen < len; matchlen++)
2510 if (idstr[matchlen] != ' ')
2516 static bool quirk_matches(const struct nvme_id_ctrl *id,
2517 const struct nvme_core_quirk_entry *q)
2519 return q->vid == le16_to_cpu(id->vid) &&
2520 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2521 string_matches(id->fr, q->fr, sizeof(id->fr));
2524 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2525 struct nvme_id_ctrl *id)
2530 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2531 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2532 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2533 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2537 if (ctrl->vs >= NVME_VS(1, 2, 1))
2538 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2541 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2542 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2543 "nqn.2014.08.org.nvmexpress:%04x%04x",
2544 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2545 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2546 off += sizeof(id->sn);
2547 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2548 off += sizeof(id->mn);
2549 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2552 static void nvme_release_subsystem(struct device *dev)
2554 struct nvme_subsystem *subsys =
2555 container_of(dev, struct nvme_subsystem, dev);
2557 if (subsys->instance >= 0)
2558 ida_simple_remove(&nvme_instance_ida, subsys->instance);
2562 static void nvme_destroy_subsystem(struct kref *ref)
2564 struct nvme_subsystem *subsys =
2565 container_of(ref, struct nvme_subsystem, ref);
2567 mutex_lock(&nvme_subsystems_lock);
2568 list_del(&subsys->entry);
2569 mutex_unlock(&nvme_subsystems_lock);
2571 ida_destroy(&subsys->ns_ida);
2572 device_del(&subsys->dev);
2573 put_device(&subsys->dev);
2576 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2578 kref_put(&subsys->ref, nvme_destroy_subsystem);
2581 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2583 struct nvme_subsystem *subsys;
2585 lockdep_assert_held(&nvme_subsystems_lock);
2588 * Fail matches for discovery subsystems. This results
2589 * in each discovery controller bound to a unique subsystem.
2590 * This avoids issues with validating controller values
2591 * that can only be true when there is a single unique subsystem.
2592 * There may be multiple and completely independent entities
2593 * that provide discovery controllers.
2595 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2598 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2599 if (strcmp(subsys->subnqn, subsysnqn))
2601 if (!kref_get_unless_zero(&subsys->ref))
2609 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2610 struct device_attribute subsys_attr_##_name = \
2611 __ATTR(_name, _mode, _show, NULL)
2613 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2614 struct device_attribute *attr,
2617 struct nvme_subsystem *subsys =
2618 container_of(dev, struct nvme_subsystem, dev);
2620 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2622 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2624 static ssize_t nvme_subsys_show_type(struct device *dev,
2625 struct device_attribute *attr,
2628 struct nvme_subsystem *subsys =
2629 container_of(dev, struct nvme_subsystem, dev);
2631 switch (subsys->subtype) {
2633 return sysfs_emit(buf, "discovery\n");
2635 return sysfs_emit(buf, "nvm\n");
2637 return sysfs_emit(buf, "reserved\n");
2640 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2642 #define nvme_subsys_show_str_function(field) \
2643 static ssize_t subsys_##field##_show(struct device *dev, \
2644 struct device_attribute *attr, char *buf) \
2646 struct nvme_subsystem *subsys = \
2647 container_of(dev, struct nvme_subsystem, dev); \
2648 return sysfs_emit(buf, "%.*s\n", \
2649 (int)sizeof(subsys->field), subsys->field); \
2651 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2653 nvme_subsys_show_str_function(model);
2654 nvme_subsys_show_str_function(serial);
2655 nvme_subsys_show_str_function(firmware_rev);
2657 static struct attribute *nvme_subsys_attrs[] = {
2658 &subsys_attr_model.attr,
2659 &subsys_attr_serial.attr,
2660 &subsys_attr_firmware_rev.attr,
2661 &subsys_attr_subsysnqn.attr,
2662 &subsys_attr_subsystype.attr,
2663 #ifdef CONFIG_NVME_MULTIPATH
2664 &subsys_attr_iopolicy.attr,
2669 static const struct attribute_group nvme_subsys_attrs_group = {
2670 .attrs = nvme_subsys_attrs,
2673 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2674 &nvme_subsys_attrs_group,
2678 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2680 return ctrl->opts && ctrl->opts->discovery_nqn;
2683 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2684 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2686 struct nvme_ctrl *tmp;
2688 lockdep_assert_held(&nvme_subsystems_lock);
2690 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2691 if (nvme_state_terminal(tmp))
2694 if (tmp->cntlid == ctrl->cntlid) {
2695 dev_err(ctrl->device,
2696 "Duplicate cntlid %u with %s, rejecting\n",
2697 ctrl->cntlid, dev_name(tmp->device));
2701 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2702 nvme_discovery_ctrl(ctrl))
2705 dev_err(ctrl->device,
2706 "Subsystem does not support multiple controllers\n");
2713 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2715 struct nvme_subsystem *subsys, *found;
2718 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2722 subsys->instance = -1;
2723 mutex_init(&subsys->lock);
2724 kref_init(&subsys->ref);
2725 INIT_LIST_HEAD(&subsys->ctrls);
2726 INIT_LIST_HEAD(&subsys->nsheads);
2727 nvme_init_subnqn(subsys, ctrl, id);
2728 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2729 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2730 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2731 subsys->vendor_id = le16_to_cpu(id->vid);
2732 subsys->cmic = id->cmic;
2734 /* Versions prior to 1.4 don't necessarily report a valid type */
2735 if (id->cntrltype == NVME_CTRL_DISC ||
2736 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2737 subsys->subtype = NVME_NQN_DISC;
2739 subsys->subtype = NVME_NQN_NVME;
2741 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2742 dev_err(ctrl->device,
2743 "Subsystem %s is not a discovery controller",
2748 subsys->awupf = le16_to_cpu(id->awupf);
2749 nvme_mpath_default_iopolicy(subsys);
2751 subsys->dev.class = nvme_subsys_class;
2752 subsys->dev.release = nvme_release_subsystem;
2753 subsys->dev.groups = nvme_subsys_attrs_groups;
2754 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2755 device_initialize(&subsys->dev);
2757 mutex_lock(&nvme_subsystems_lock);
2758 found = __nvme_find_get_subsystem(subsys->subnqn);
2760 put_device(&subsys->dev);
2763 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2765 goto out_put_subsystem;
2768 ret = device_add(&subsys->dev);
2770 dev_err(ctrl->device,
2771 "failed to register subsystem device.\n");
2772 put_device(&subsys->dev);
2775 ida_init(&subsys->ns_ida);
2776 list_add_tail(&subsys->entry, &nvme_subsystems);
2779 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2780 dev_name(ctrl->device));
2782 dev_err(ctrl->device,
2783 "failed to create sysfs link from subsystem.\n");
2784 goto out_put_subsystem;
2788 subsys->instance = ctrl->instance;
2789 ctrl->subsys = subsys;
2790 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2791 mutex_unlock(&nvme_subsystems_lock);
2795 nvme_put_subsystem(subsys);
2797 mutex_unlock(&nvme_subsystems_lock);
2801 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2802 void *log, size_t size, u64 offset)
2804 struct nvme_command c = { };
2805 u32 dwlen = nvme_bytes_to_numd(size);
2807 c.get_log_page.opcode = nvme_admin_get_log_page;
2808 c.get_log_page.nsid = cpu_to_le32(nsid);
2809 c.get_log_page.lid = log_page;
2810 c.get_log_page.lsp = lsp;
2811 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2812 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2813 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2814 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2815 c.get_log_page.csi = csi;
2817 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2820 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2821 struct nvme_effects_log **log)
2823 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2829 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2833 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2834 cel, sizeof(*cel), 0);
2840 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2846 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2848 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2850 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2855 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2857 struct nvme_command c = { };
2858 struct nvme_id_ctrl_nvm *id;
2861 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2862 ctrl->max_discard_sectors = UINT_MAX;
2863 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2865 ctrl->max_discard_sectors = 0;
2866 ctrl->max_discard_segments = 0;
2870 * Even though NVMe spec explicitly states that MDTS is not applicable
2871 * to the write-zeroes, we are cautious and limit the size to the
2872 * controllers max_hw_sectors value, which is based on the MDTS field
2873 * and possibly other limiting factors.
2875 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2876 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2877 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2879 ctrl->max_zeroes_sectors = 0;
2881 if (nvme_ctrl_limited_cns(ctrl))
2884 id = kzalloc(sizeof(*id), GFP_KERNEL);
2888 c.identify.opcode = nvme_admin_identify;
2889 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2890 c.identify.csi = NVME_CSI_NVM;
2892 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2897 ctrl->max_discard_segments = id->dmrl;
2899 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2901 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2908 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2910 struct nvme_id_ctrl *id;
2912 bool prev_apst_enabled;
2915 ret = nvme_identify_ctrl(ctrl, &id);
2917 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2921 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2922 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2927 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2928 ctrl->cntlid = le16_to_cpu(id->cntlid);
2930 if (!ctrl->identified) {
2933 ret = nvme_init_subsystem(ctrl, id);
2938 * Check for quirks. Quirk can depend on firmware version,
2939 * so, in principle, the set of quirks present can change
2940 * across a reset. As a possible future enhancement, we
2941 * could re-scan for quirks every time we reinitialize
2942 * the device, but we'd have to make sure that the driver
2943 * behaves intelligently if the quirks change.
2945 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2946 if (quirk_matches(id, &core_quirks[i]))
2947 ctrl->quirks |= core_quirks[i].quirks;
2951 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2952 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2953 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2956 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2957 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2958 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2960 ctrl->oacs = le16_to_cpu(id->oacs);
2961 ctrl->oncs = le16_to_cpu(id->oncs);
2962 ctrl->mtfa = le16_to_cpu(id->mtfa);
2963 ctrl->oaes = le32_to_cpu(id->oaes);
2964 ctrl->wctemp = le16_to_cpu(id->wctemp);
2965 ctrl->cctemp = le16_to_cpu(id->cctemp);
2967 atomic_set(&ctrl->abort_limit, id->acl + 1);
2968 ctrl->vwc = id->vwc;
2970 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2972 max_hw_sectors = UINT_MAX;
2973 ctrl->max_hw_sectors =
2974 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2976 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2977 ctrl->sgls = le32_to_cpu(id->sgls);
2978 ctrl->kas = le16_to_cpu(id->kas);
2979 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2980 ctrl->ctratt = le32_to_cpu(id->ctratt);
2984 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
2986 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
2987 shutdown_timeout, 60);
2989 if (ctrl->shutdown_timeout != shutdown_timeout)
2990 dev_info(ctrl->device,
2991 "Shutdown timeout set to %u seconds\n",
2992 ctrl->shutdown_timeout);
2994 ctrl->shutdown_timeout = shutdown_timeout;
2996 ctrl->npss = id->npss;
2997 ctrl->apsta = id->apsta;
2998 prev_apst_enabled = ctrl->apst_enabled;
2999 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3000 if (force_apst && id->apsta) {
3001 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3002 ctrl->apst_enabled = true;
3004 ctrl->apst_enabled = false;
3007 ctrl->apst_enabled = id->apsta;
3009 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3011 if (ctrl->ops->flags & NVME_F_FABRICS) {
3012 ctrl->icdoff = le16_to_cpu(id->icdoff);
3013 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3014 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3015 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3018 * In fabrics we need to verify the cntlid matches the
3021 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3022 dev_err(ctrl->device,
3023 "Mismatching cntlid: Connect %u vs Identify "
3025 ctrl->cntlid, le16_to_cpu(id->cntlid));
3030 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3031 dev_err(ctrl->device,
3032 "keep-alive support is mandatory for fabrics\n");
3037 ctrl->hmpre = le32_to_cpu(id->hmpre);
3038 ctrl->hmmin = le32_to_cpu(id->hmmin);
3039 ctrl->hmminds = le32_to_cpu(id->hmminds);
3040 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3043 ret = nvme_mpath_init_identify(ctrl, id);
3047 if (ctrl->apst_enabled && !prev_apst_enabled)
3048 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3049 else if (!ctrl->apst_enabled && prev_apst_enabled)
3050 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3058 * Initialize the cached copies of the Identify data and various controller
3059 * register in our nvme_ctrl structure. This should be called as soon as
3060 * the admin queue is fully up and running.
3062 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3066 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3068 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3072 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3074 if (ctrl->vs >= NVME_VS(1, 1, 0))
3075 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3077 ret = nvme_init_identify(ctrl);
3081 ret = nvme_init_non_mdts_limits(ctrl);
3085 ret = nvme_configure_apst(ctrl);
3089 ret = nvme_configure_timestamp(ctrl);
3093 ret = nvme_configure_directives(ctrl);
3097 ret = nvme_configure_acre(ctrl);
3101 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3102 ret = nvme_hwmon_init(ctrl);
3107 ctrl->identified = true;
3111 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3113 static int nvme_dev_open(struct inode *inode, struct file *file)
3115 struct nvme_ctrl *ctrl =
3116 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3118 switch (ctrl->state) {
3119 case NVME_CTRL_LIVE:
3122 return -EWOULDBLOCK;
3125 nvme_get_ctrl(ctrl);
3126 if (!try_module_get(ctrl->ops->module)) {
3127 nvme_put_ctrl(ctrl);
3131 file->private_data = ctrl;
3135 static int nvme_dev_release(struct inode *inode, struct file *file)
3137 struct nvme_ctrl *ctrl =
3138 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3140 module_put(ctrl->ops->module);
3141 nvme_put_ctrl(ctrl);
3145 static const struct file_operations nvme_dev_fops = {
3146 .owner = THIS_MODULE,
3147 .open = nvme_dev_open,
3148 .release = nvme_dev_release,
3149 .unlocked_ioctl = nvme_dev_ioctl,
3150 .compat_ioctl = compat_ptr_ioctl,
3153 static ssize_t nvme_sysfs_reset(struct device *dev,
3154 struct device_attribute *attr, const char *buf,
3157 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3160 ret = nvme_reset_ctrl_sync(ctrl);
3165 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3167 static ssize_t nvme_sysfs_rescan(struct device *dev,
3168 struct device_attribute *attr, const char *buf,
3171 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3173 nvme_queue_scan(ctrl);
3176 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3178 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3180 struct gendisk *disk = dev_to_disk(dev);
3182 if (disk->fops == &nvme_bdev_ops)
3183 return nvme_get_ns_from_dev(dev)->head;
3185 return disk->private_data;
3188 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3191 struct nvme_ns_head *head = dev_to_ns_head(dev);
3192 struct nvme_ns_ids *ids = &head->ids;
3193 struct nvme_subsystem *subsys = head->subsys;
3194 int serial_len = sizeof(subsys->serial);
3195 int model_len = sizeof(subsys->model);
3197 if (!uuid_is_null(&ids->uuid))
3198 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3200 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3201 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3203 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3204 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3206 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3207 subsys->serial[serial_len - 1] == '\0'))
3209 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3210 subsys->model[model_len - 1] == '\0'))
3213 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3214 serial_len, subsys->serial, model_len, subsys->model,
3217 static DEVICE_ATTR_RO(wwid);
3219 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3222 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3224 static DEVICE_ATTR_RO(nguid);
3226 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3229 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3231 /* For backward compatibility expose the NGUID to userspace if
3232 * we have no UUID set
3234 if (uuid_is_null(&ids->uuid)) {
3235 printk_ratelimited(KERN_WARNING
3236 "No UUID available providing old NGUID\n");
3237 return sysfs_emit(buf, "%pU\n", ids->nguid);
3239 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3241 static DEVICE_ATTR_RO(uuid);
3243 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3246 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3248 static DEVICE_ATTR_RO(eui);
3250 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3253 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3255 static DEVICE_ATTR_RO(nsid);
3257 static struct attribute *nvme_ns_id_attrs[] = {
3258 &dev_attr_wwid.attr,
3259 &dev_attr_uuid.attr,
3260 &dev_attr_nguid.attr,
3262 &dev_attr_nsid.attr,
3263 #ifdef CONFIG_NVME_MULTIPATH
3264 &dev_attr_ana_grpid.attr,
3265 &dev_attr_ana_state.attr,
3270 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3271 struct attribute *a, int n)
3273 struct device *dev = container_of(kobj, struct device, kobj);
3274 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3276 if (a == &dev_attr_uuid.attr) {
3277 if (uuid_is_null(&ids->uuid) &&
3278 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3281 if (a == &dev_attr_nguid.attr) {
3282 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3285 if (a == &dev_attr_eui.attr) {
3286 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3289 #ifdef CONFIG_NVME_MULTIPATH
3290 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3291 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3293 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3300 static const struct attribute_group nvme_ns_id_attr_group = {
3301 .attrs = nvme_ns_id_attrs,
3302 .is_visible = nvme_ns_id_attrs_are_visible,
3305 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3306 &nvme_ns_id_attr_group,
3310 #define nvme_show_str_function(field) \
3311 static ssize_t field##_show(struct device *dev, \
3312 struct device_attribute *attr, char *buf) \
3314 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3315 return sysfs_emit(buf, "%.*s\n", \
3316 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3318 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3320 nvme_show_str_function(model);
3321 nvme_show_str_function(serial);
3322 nvme_show_str_function(firmware_rev);
3324 #define nvme_show_int_function(field) \
3325 static ssize_t field##_show(struct device *dev, \
3326 struct device_attribute *attr, char *buf) \
3328 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3329 return sysfs_emit(buf, "%d\n", ctrl->field); \
3331 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3333 nvme_show_int_function(cntlid);
3334 nvme_show_int_function(numa_node);
3335 nvme_show_int_function(queue_count);
3336 nvme_show_int_function(sqsize);
3337 nvme_show_int_function(kato);
3339 static ssize_t nvme_sysfs_delete(struct device *dev,
3340 struct device_attribute *attr, const char *buf,
3343 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3345 if (device_remove_file_self(dev, attr))
3346 nvme_delete_ctrl_sync(ctrl);
3349 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3351 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3352 struct device_attribute *attr,
3355 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3357 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3359 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3361 static ssize_t nvme_sysfs_show_state(struct device *dev,
3362 struct device_attribute *attr,
3365 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3366 static const char *const state_name[] = {
3367 [NVME_CTRL_NEW] = "new",
3368 [NVME_CTRL_LIVE] = "live",
3369 [NVME_CTRL_RESETTING] = "resetting",
3370 [NVME_CTRL_CONNECTING] = "connecting",
3371 [NVME_CTRL_DELETING] = "deleting",
3372 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3373 [NVME_CTRL_DEAD] = "dead",
3376 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3377 state_name[ctrl->state])
3378 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3380 return sysfs_emit(buf, "unknown state\n");
3383 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3385 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3386 struct device_attribute *attr,
3389 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3391 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3393 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3395 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3396 struct device_attribute *attr,
3399 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3401 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3403 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3405 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3406 struct device_attribute *attr,
3409 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3411 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3413 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3415 static ssize_t nvme_sysfs_show_address(struct device *dev,
3416 struct device_attribute *attr,
3419 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3421 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3423 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3425 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3426 struct device_attribute *attr, char *buf)
3428 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3429 struct nvmf_ctrl_options *opts = ctrl->opts;
3431 if (ctrl->opts->max_reconnects == -1)
3432 return sysfs_emit(buf, "off\n");
3433 return sysfs_emit(buf, "%d\n",
3434 opts->max_reconnects * opts->reconnect_delay);
3437 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3438 struct device_attribute *attr, const char *buf, size_t count)
3440 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3441 struct nvmf_ctrl_options *opts = ctrl->opts;
3442 int ctrl_loss_tmo, err;
3444 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3448 if (ctrl_loss_tmo < 0)
3449 opts->max_reconnects = -1;
3451 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3452 opts->reconnect_delay);
3455 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3456 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3458 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3459 struct device_attribute *attr, char *buf)
3461 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3463 if (ctrl->opts->reconnect_delay == -1)
3464 return sysfs_emit(buf, "off\n");
3465 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3468 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3469 struct device_attribute *attr, const char *buf, size_t count)
3471 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3475 err = kstrtou32(buf, 10, &v);
3479 ctrl->opts->reconnect_delay = v;
3482 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3483 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3485 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3486 struct device_attribute *attr, char *buf)
3488 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3490 if (ctrl->opts->fast_io_fail_tmo == -1)
3491 return sysfs_emit(buf, "off\n");
3492 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3495 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3496 struct device_attribute *attr, const char *buf, size_t count)
3498 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3499 struct nvmf_ctrl_options *opts = ctrl->opts;
3500 int fast_io_fail_tmo, err;
3502 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3506 if (fast_io_fail_tmo < 0)
3507 opts->fast_io_fail_tmo = -1;
3509 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3512 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3513 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3515 static struct attribute *nvme_dev_attrs[] = {
3516 &dev_attr_reset_controller.attr,
3517 &dev_attr_rescan_controller.attr,
3518 &dev_attr_model.attr,
3519 &dev_attr_serial.attr,
3520 &dev_attr_firmware_rev.attr,
3521 &dev_attr_cntlid.attr,
3522 &dev_attr_delete_controller.attr,
3523 &dev_attr_transport.attr,
3524 &dev_attr_subsysnqn.attr,
3525 &dev_attr_address.attr,
3526 &dev_attr_state.attr,
3527 &dev_attr_numa_node.attr,
3528 &dev_attr_queue_count.attr,
3529 &dev_attr_sqsize.attr,
3530 &dev_attr_hostnqn.attr,
3531 &dev_attr_hostid.attr,
3532 &dev_attr_ctrl_loss_tmo.attr,
3533 &dev_attr_reconnect_delay.attr,
3534 &dev_attr_fast_io_fail_tmo.attr,
3535 &dev_attr_kato.attr,
3539 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3540 struct attribute *a, int n)
3542 struct device *dev = container_of(kobj, struct device, kobj);
3543 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3545 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3547 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3549 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3551 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3553 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3555 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3557 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3563 static const struct attribute_group nvme_dev_attrs_group = {
3564 .attrs = nvme_dev_attrs,
3565 .is_visible = nvme_dev_attrs_are_visible,
3568 static const struct attribute_group *nvme_dev_attr_groups[] = {
3569 &nvme_dev_attrs_group,
3573 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3576 struct nvme_ns_head *h;
3578 lockdep_assert_held(&subsys->lock);
3580 list_for_each_entry(h, &subsys->nsheads, entry) {
3581 if (h->ns_id != nsid)
3583 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3590 static int __nvme_check_ids(struct nvme_subsystem *subsys,
3591 struct nvme_ns_head *new)
3593 struct nvme_ns_head *h;
3595 lockdep_assert_held(&subsys->lock);
3597 list_for_each_entry(h, &subsys->nsheads, entry) {
3598 if (nvme_ns_ids_valid(&new->ids) &&
3599 nvme_ns_ids_equal(&new->ids, &h->ids))
3606 static void nvme_cdev_rel(struct device *dev)
3608 ida_simple_remove(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3611 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3613 cdev_device_del(cdev, cdev_device);
3614 put_device(cdev_device);
3617 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3618 const struct file_operations *fops, struct module *owner)
3622 minor = ida_simple_get(&nvme_ns_chr_minor_ida, 0, 0, GFP_KERNEL);
3625 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3626 cdev_device->class = nvme_ns_chr_class;
3627 cdev_device->release = nvme_cdev_rel;
3628 device_initialize(cdev_device);
3629 cdev_init(cdev, fops);
3630 cdev->owner = owner;
3631 ret = cdev_device_add(cdev, cdev_device);
3633 put_device(cdev_device);
3638 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3640 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3643 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3645 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3649 static const struct file_operations nvme_ns_chr_fops = {
3650 .owner = THIS_MODULE,
3651 .open = nvme_ns_chr_open,
3652 .release = nvme_ns_chr_release,
3653 .unlocked_ioctl = nvme_ns_chr_ioctl,
3654 .compat_ioctl = compat_ptr_ioctl,
3657 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3661 ns->cdev_device.parent = ns->ctrl->device;
3662 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3663 ns->ctrl->instance, ns->head->instance);
3667 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3668 ns->ctrl->ops->module);
3671 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3672 unsigned nsid, struct nvme_ns_ids *ids)
3674 struct nvme_ns_head *head;
3675 size_t size = sizeof(*head);
3678 #ifdef CONFIG_NVME_MULTIPATH
3679 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3682 head = kzalloc(size, GFP_KERNEL);
3685 ret = ida_simple_get(&ctrl->subsys->ns_ida, 1, 0, GFP_KERNEL);
3688 head->instance = ret;
3689 INIT_LIST_HEAD(&head->list);
3690 ret = init_srcu_struct(&head->srcu);
3692 goto out_ida_remove;
3693 head->subsys = ctrl->subsys;
3696 kref_init(&head->ref);
3698 ret = __nvme_check_ids(ctrl->subsys, head);
3700 dev_err(ctrl->device,
3701 "duplicate IDs for nsid %d\n", nsid);
3702 goto out_cleanup_srcu;
3705 if (head->ids.csi) {
3706 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3708 goto out_cleanup_srcu;
3710 head->effects = ctrl->effects;
3712 ret = nvme_mpath_alloc_disk(ctrl, head);
3714 goto out_cleanup_srcu;
3716 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3718 kref_get(&ctrl->subsys->ref);
3722 cleanup_srcu_struct(&head->srcu);
3724 ida_simple_remove(&ctrl->subsys->ns_ida, head->instance);
3729 ret = blk_status_to_errno(nvme_error_status(ret));
3730 return ERR_PTR(ret);
3733 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3734 struct nvme_ns_ids *ids, bool is_shared)
3736 struct nvme_ctrl *ctrl = ns->ctrl;
3737 struct nvme_ns_head *head = NULL;
3740 mutex_lock(&ctrl->subsys->lock);
3741 head = nvme_find_ns_head(ctrl->subsys, nsid);
3743 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3745 ret = PTR_ERR(head);
3748 head->shared = is_shared;
3751 if (!is_shared || !head->shared) {
3752 dev_err(ctrl->device,
3753 "Duplicate unshared namespace %d\n", nsid);
3754 goto out_put_ns_head;
3756 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3757 dev_err(ctrl->device,
3758 "IDs don't match for shared namespace %d\n",
3760 goto out_put_ns_head;
3764 list_add_tail_rcu(&ns->siblings, &head->list);
3766 mutex_unlock(&ctrl->subsys->lock);
3770 nvme_put_ns_head(head);
3772 mutex_unlock(&ctrl->subsys->lock);
3776 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3778 struct nvme_ns *ns, *ret = NULL;
3780 down_read(&ctrl->namespaces_rwsem);
3781 list_for_each_entry(ns, &ctrl->namespaces, list) {
3782 if (ns->head->ns_id == nsid) {
3783 if (!nvme_get_ns(ns))
3788 if (ns->head->ns_id > nsid)
3791 up_read(&ctrl->namespaces_rwsem);
3794 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3797 * Add the namespace to the controller list while keeping the list ordered.
3799 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3801 struct nvme_ns *tmp;
3803 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3804 if (tmp->head->ns_id < ns->head->ns_id) {
3805 list_add(&ns->list, &tmp->list);
3809 list_add(&ns->list, &ns->ctrl->namespaces);
3812 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3813 struct nvme_ns_ids *ids)
3816 struct gendisk *disk;
3817 struct nvme_id_ns *id;
3818 int node = ctrl->numa_node;
3820 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3823 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3827 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3830 disk->fops = &nvme_bdev_ops;
3831 disk->private_data = ns;
3834 ns->queue = disk->queue;
3836 if (ctrl->opts && ctrl->opts->data_digest)
3837 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3839 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3840 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3841 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3844 kref_init(&ns->kref);
3846 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3847 goto out_cleanup_disk;
3850 * Without the multipath code enabled, multiple controller per
3851 * subsystems are visible as devices and thus we cannot use the
3852 * subsystem instance.
3854 if (!nvme_mpath_set_disk_name(ns, disk->disk_name, &disk->flags))
3855 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3856 ns->head->instance);
3858 if (nvme_update_ns_info(ns, id))
3861 down_write(&ctrl->namespaces_rwsem);
3862 nvme_ns_add_to_ctrl_list(ns);
3863 up_write(&ctrl->namespaces_rwsem);
3864 nvme_get_ctrl(ctrl);
3866 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3867 goto out_cleanup_ns_from_list;
3869 if (!nvme_ns_head_multipath(ns->head))
3870 nvme_add_ns_cdev(ns);
3872 nvme_mpath_add_disk(ns, id);
3873 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3878 out_cleanup_ns_from_list:
3879 nvme_put_ctrl(ctrl);
3880 down_write(&ctrl->namespaces_rwsem);
3881 list_del_init(&ns->list);
3882 up_write(&ctrl->namespaces_rwsem);
3884 mutex_lock(&ctrl->subsys->lock);
3885 list_del_rcu(&ns->siblings);
3886 if (list_empty(&ns->head->list))
3887 list_del_init(&ns->head->entry);
3888 mutex_unlock(&ctrl->subsys->lock);
3889 nvme_put_ns_head(ns->head);
3891 blk_cleanup_disk(disk);
3898 static void nvme_ns_remove(struct nvme_ns *ns)
3900 bool last_path = false;
3902 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3905 clear_bit(NVME_NS_READY, &ns->flags);
3906 set_capacity(ns->disk, 0);
3907 nvme_fault_inject_fini(&ns->fault_inject);
3909 mutex_lock(&ns->ctrl->subsys->lock);
3910 list_del_rcu(&ns->siblings);
3911 if (list_empty(&ns->head->list)) {
3912 list_del_init(&ns->head->entry);
3915 mutex_unlock(&ns->ctrl->subsys->lock);
3917 /* guarantee not available in head->list */
3920 /* wait for concurrent submissions */
3921 if (nvme_mpath_clear_current_path(ns))
3922 synchronize_srcu(&ns->head->srcu);
3924 if (!nvme_ns_head_multipath(ns->head))
3925 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3926 del_gendisk(ns->disk);
3927 blk_cleanup_queue(ns->queue);
3929 down_write(&ns->ctrl->namespaces_rwsem);
3930 list_del_init(&ns->list);
3931 up_write(&ns->ctrl->namespaces_rwsem);
3934 nvme_mpath_shutdown_disk(ns->head);
3938 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3940 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3948 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
3950 struct nvme_id_ns *id;
3951 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3953 if (test_bit(NVME_NS_DEAD, &ns->flags))
3956 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
3960 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3961 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
3962 dev_err(ns->ctrl->device,
3963 "identifiers changed for nsid %d\n", ns->head->ns_id);
3967 ret = nvme_update_ns_info(ns, id);
3973 * Only remove the namespace if we got a fatal error back from the
3974 * device, otherwise ignore the error and just move on.
3976 * TODO: we should probably schedule a delayed retry here.
3978 if (ret > 0 && (ret & NVME_SC_DNR))
3982 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3984 struct nvme_ns_ids ids = { };
3987 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
3990 ns = nvme_find_get_ns(ctrl, nsid);
3992 nvme_validate_ns(ns, &ids);
3999 nvme_alloc_ns(ctrl, nsid, &ids);
4002 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4003 dev_warn(ctrl->device,
4004 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4008 if (!nvme_multi_css(ctrl)) {
4009 dev_warn(ctrl->device,
4010 "command set not reported for nsid: %d\n",
4014 nvme_alloc_ns(ctrl, nsid, &ids);
4017 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4023 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4026 struct nvme_ns *ns, *next;
4029 down_write(&ctrl->namespaces_rwsem);
4030 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4031 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4032 list_move_tail(&ns->list, &rm_list);
4034 up_write(&ctrl->namespaces_rwsem);
4036 list_for_each_entry_safe(ns, next, &rm_list, list)
4041 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4043 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4048 if (nvme_ctrl_limited_cns(ctrl))
4051 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4056 struct nvme_command cmd = {
4057 .identify.opcode = nvme_admin_identify,
4058 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4059 .identify.nsid = cpu_to_le32(prev),
4062 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4063 NVME_IDENTIFY_DATA_SIZE);
4065 dev_warn(ctrl->device,
4066 "Identify NS List failed (status=0x%x)\n", ret);
4070 for (i = 0; i < nr_entries; i++) {
4071 u32 nsid = le32_to_cpu(ns_list[i]);
4073 if (!nsid) /* end of the list? */
4075 nvme_validate_or_alloc_ns(ctrl, nsid);
4076 while (++prev < nsid)
4077 nvme_ns_remove_by_nsid(ctrl, prev);
4081 nvme_remove_invalid_namespaces(ctrl, prev);
4087 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4089 struct nvme_id_ctrl *id;
4092 if (nvme_identify_ctrl(ctrl, &id))
4094 nn = le32_to_cpu(id->nn);
4097 for (i = 1; i <= nn; i++)
4098 nvme_validate_or_alloc_ns(ctrl, i);
4100 nvme_remove_invalid_namespaces(ctrl, nn);
4103 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4105 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4109 log = kzalloc(log_size, GFP_KERNEL);
4114 * We need to read the log to clear the AEN, but we don't want to rely
4115 * on it for the changed namespace information as userspace could have
4116 * raced with us in reading the log page, which could cause us to miss
4119 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4120 NVME_CSI_NVM, log, log_size, 0);
4122 dev_warn(ctrl->device,
4123 "reading changed ns log failed: %d\n", error);
4128 static void nvme_scan_work(struct work_struct *work)
4130 struct nvme_ctrl *ctrl =
4131 container_of(work, struct nvme_ctrl, scan_work);
4133 /* No tagset on a live ctrl means IO queues could not created */
4134 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4137 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4138 dev_info(ctrl->device, "rescanning namespaces.\n");
4139 nvme_clear_changed_ns_log(ctrl);
4142 mutex_lock(&ctrl->scan_lock);
4143 if (nvme_scan_ns_list(ctrl) != 0)
4144 nvme_scan_ns_sequential(ctrl);
4145 mutex_unlock(&ctrl->scan_lock);
4149 * This function iterates the namespace list unlocked to allow recovery from
4150 * controller failure. It is up to the caller to ensure the namespace list is
4151 * not modified by scan work while this function is executing.
4153 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4155 struct nvme_ns *ns, *next;
4159 * make sure to requeue I/O to all namespaces as these
4160 * might result from the scan itself and must complete
4161 * for the scan_work to make progress
4163 nvme_mpath_clear_ctrl_paths(ctrl);
4165 /* prevent racing with ns scanning */
4166 flush_work(&ctrl->scan_work);
4169 * The dead states indicates the controller was not gracefully
4170 * disconnected. In that case, we won't be able to flush any data while
4171 * removing the namespaces' disks; fail all the queues now to avoid
4172 * potentially having to clean up the failed sync later.
4174 if (ctrl->state == NVME_CTRL_DEAD)
4175 nvme_kill_queues(ctrl);
4177 /* this is a no-op when called from the controller reset handler */
4178 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4180 down_write(&ctrl->namespaces_rwsem);
4181 list_splice_init(&ctrl->namespaces, &ns_list);
4182 up_write(&ctrl->namespaces_rwsem);
4184 list_for_each_entry_safe(ns, next, &ns_list, list)
4187 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4189 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4191 struct nvme_ctrl *ctrl =
4192 container_of(dev, struct nvme_ctrl, ctrl_device);
4193 struct nvmf_ctrl_options *opts = ctrl->opts;
4196 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4201 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4205 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4206 opts->trsvcid ?: "none");
4210 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4211 opts->host_traddr ?: "none");
4215 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4216 opts->host_iface ?: "none");
4221 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4223 char *envp[2] = { NULL, NULL };
4224 u32 aen_result = ctrl->aen_result;
4226 ctrl->aen_result = 0;
4230 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4233 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4237 static void nvme_async_event_work(struct work_struct *work)
4239 struct nvme_ctrl *ctrl =
4240 container_of(work, struct nvme_ctrl, async_event_work);
4242 nvme_aen_uevent(ctrl);
4243 ctrl->ops->submit_async_event(ctrl);
4246 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4251 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4257 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4260 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4262 struct nvme_fw_slot_info_log *log;
4264 log = kmalloc(sizeof(*log), GFP_KERNEL);
4268 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4269 log, sizeof(*log), 0))
4270 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4274 static void nvme_fw_act_work(struct work_struct *work)
4276 struct nvme_ctrl *ctrl = container_of(work,
4277 struct nvme_ctrl, fw_act_work);
4278 unsigned long fw_act_timeout;
4281 fw_act_timeout = jiffies +
4282 msecs_to_jiffies(ctrl->mtfa * 100);
4284 fw_act_timeout = jiffies +
4285 msecs_to_jiffies(admin_timeout * 1000);
4287 nvme_stop_queues(ctrl);
4288 while (nvme_ctrl_pp_status(ctrl)) {
4289 if (time_after(jiffies, fw_act_timeout)) {
4290 dev_warn(ctrl->device,
4291 "Fw activation timeout, reset controller\n");
4292 nvme_try_sched_reset(ctrl);
4298 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4301 nvme_start_queues(ctrl);
4302 /* read FW slot information to clear the AER */
4303 nvme_get_fw_slot_info(ctrl);
4306 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4308 u32 aer_notice_type = (result & 0xff00) >> 8;
4310 trace_nvme_async_event(ctrl, aer_notice_type);
4312 switch (aer_notice_type) {
4313 case NVME_AER_NOTICE_NS_CHANGED:
4314 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4315 nvme_queue_scan(ctrl);
4317 case NVME_AER_NOTICE_FW_ACT_STARTING:
4319 * We are (ab)using the RESETTING state to prevent subsequent
4320 * recovery actions from interfering with the controller's
4321 * firmware activation.
4323 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4324 queue_work(nvme_wq, &ctrl->fw_act_work);
4326 #ifdef CONFIG_NVME_MULTIPATH
4327 case NVME_AER_NOTICE_ANA:
4328 if (!ctrl->ana_log_buf)
4330 queue_work(nvme_wq, &ctrl->ana_work);
4333 case NVME_AER_NOTICE_DISC_CHANGED:
4334 ctrl->aen_result = result;
4337 dev_warn(ctrl->device, "async event result %08x\n", result);
4341 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4342 volatile union nvme_result *res)
4344 u32 result = le32_to_cpu(res->u32);
4345 u32 aer_type = result & 0x07;
4347 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4351 case NVME_AER_NOTICE:
4352 nvme_handle_aen_notice(ctrl, result);
4354 case NVME_AER_ERROR:
4355 case NVME_AER_SMART:
4358 trace_nvme_async_event(ctrl, aer_type);
4359 ctrl->aen_result = result;
4364 queue_work(nvme_wq, &ctrl->async_event_work);
4366 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4368 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4370 nvme_mpath_stop(ctrl);
4371 nvme_stop_keep_alive(ctrl);
4372 nvme_stop_failfast_work(ctrl);
4373 flush_work(&ctrl->async_event_work);
4374 cancel_work_sync(&ctrl->fw_act_work);
4376 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4378 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4380 nvme_start_keep_alive(ctrl);
4382 nvme_enable_aen(ctrl);
4384 if (ctrl->queue_count > 1) {
4385 nvme_queue_scan(ctrl);
4386 nvme_start_queues(ctrl);
4389 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4391 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4393 nvme_hwmon_exit(ctrl);
4394 nvme_fault_inject_fini(&ctrl->fault_inject);
4395 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4396 cdev_device_del(&ctrl->cdev, ctrl->device);
4397 nvme_put_ctrl(ctrl);
4399 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4401 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4403 struct nvme_effects_log *cel;
4406 xa_for_each(&ctrl->cels, i, cel) {
4407 xa_erase(&ctrl->cels, i);
4411 xa_destroy(&ctrl->cels);
4414 static void nvme_free_ctrl(struct device *dev)
4416 struct nvme_ctrl *ctrl =
4417 container_of(dev, struct nvme_ctrl, ctrl_device);
4418 struct nvme_subsystem *subsys = ctrl->subsys;
4420 if (!subsys || ctrl->instance != subsys->instance)
4421 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4423 nvme_free_cels(ctrl);
4424 nvme_mpath_uninit(ctrl);
4425 __free_page(ctrl->discard_page);
4428 mutex_lock(&nvme_subsystems_lock);
4429 list_del(&ctrl->subsys_entry);
4430 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4431 mutex_unlock(&nvme_subsystems_lock);
4434 ctrl->ops->free_ctrl(ctrl);
4437 nvme_put_subsystem(subsys);
4441 * Initialize a NVMe controller structures. This needs to be called during
4442 * earliest initialization so that we have the initialized structured around
4445 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4446 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4450 ctrl->state = NVME_CTRL_NEW;
4451 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4452 spin_lock_init(&ctrl->lock);
4453 mutex_init(&ctrl->scan_lock);
4454 INIT_LIST_HEAD(&ctrl->namespaces);
4455 xa_init(&ctrl->cels);
4456 init_rwsem(&ctrl->namespaces_rwsem);
4459 ctrl->quirks = quirks;
4460 ctrl->numa_node = NUMA_NO_NODE;
4461 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4462 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4463 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4464 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4465 init_waitqueue_head(&ctrl->state_wq);
4467 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4468 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4469 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4470 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4472 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4474 ctrl->discard_page = alloc_page(GFP_KERNEL);
4475 if (!ctrl->discard_page) {
4480 ret = ida_simple_get(&nvme_instance_ida, 0, 0, GFP_KERNEL);
4483 ctrl->instance = ret;
4485 device_initialize(&ctrl->ctrl_device);
4486 ctrl->device = &ctrl->ctrl_device;
4487 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4489 ctrl->device->class = nvme_class;
4490 ctrl->device->parent = ctrl->dev;
4491 ctrl->device->groups = nvme_dev_attr_groups;
4492 ctrl->device->release = nvme_free_ctrl;
4493 dev_set_drvdata(ctrl->device, ctrl);
4494 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4496 goto out_release_instance;
4498 nvme_get_ctrl(ctrl);
4499 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4500 ctrl->cdev.owner = ops->module;
4501 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4506 * Initialize latency tolerance controls. The sysfs files won't
4507 * be visible to userspace unless the device actually supports APST.
4509 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4510 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4511 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4513 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4514 nvme_mpath_init_ctrl(ctrl);
4518 nvme_put_ctrl(ctrl);
4519 kfree_const(ctrl->device->kobj.name);
4520 out_release_instance:
4521 ida_simple_remove(&nvme_instance_ida, ctrl->instance);
4523 if (ctrl->discard_page)
4524 __free_page(ctrl->discard_page);
4527 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4529 static void nvme_start_ns_queue(struct nvme_ns *ns)
4531 if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4532 blk_mq_unquiesce_queue(ns->queue);
4535 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4537 if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4538 blk_mq_quiesce_queue(ns->queue);
4540 blk_mq_wait_quiesce_done(ns->queue);
4544 * Prepare a queue for teardown.
4546 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4547 * the capacity to 0 after that to avoid blocking dispatchers that may be
4548 * holding bd_butex. This will end buffered writers dirtying pages that can't
4551 static void nvme_set_queue_dying(struct nvme_ns *ns)
4553 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4556 blk_set_queue_dying(ns->queue);
4557 nvme_start_ns_queue(ns);
4559 set_capacity_and_notify(ns->disk, 0);
4563 * nvme_kill_queues(): Ends all namespace queues
4564 * @ctrl: the dead controller that needs to end
4566 * Call this function when the driver determines it is unable to get the
4567 * controller in a state capable of servicing IO.
4569 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4573 down_read(&ctrl->namespaces_rwsem);
4575 /* Forcibly unquiesce queues to avoid blocking dispatch */
4576 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4577 nvme_start_admin_queue(ctrl);
4579 list_for_each_entry(ns, &ctrl->namespaces, list)
4580 nvme_set_queue_dying(ns);
4582 up_read(&ctrl->namespaces_rwsem);
4584 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4586 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4590 down_read(&ctrl->namespaces_rwsem);
4591 list_for_each_entry(ns, &ctrl->namespaces, list)
4592 blk_mq_unfreeze_queue(ns->queue);
4593 up_read(&ctrl->namespaces_rwsem);
4595 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4597 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4601 down_read(&ctrl->namespaces_rwsem);
4602 list_for_each_entry(ns, &ctrl->namespaces, list) {
4603 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4607 up_read(&ctrl->namespaces_rwsem);
4610 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4612 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4616 down_read(&ctrl->namespaces_rwsem);
4617 list_for_each_entry(ns, &ctrl->namespaces, list)
4618 blk_mq_freeze_queue_wait(ns->queue);
4619 up_read(&ctrl->namespaces_rwsem);
4621 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4623 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4627 down_read(&ctrl->namespaces_rwsem);
4628 list_for_each_entry(ns, &ctrl->namespaces, list)
4629 blk_freeze_queue_start(ns->queue);
4630 up_read(&ctrl->namespaces_rwsem);
4632 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4634 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4638 down_read(&ctrl->namespaces_rwsem);
4639 list_for_each_entry(ns, &ctrl->namespaces, list)
4640 nvme_stop_ns_queue(ns);
4641 up_read(&ctrl->namespaces_rwsem);
4643 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4645 void nvme_start_queues(struct nvme_ctrl *ctrl)
4649 down_read(&ctrl->namespaces_rwsem);
4650 list_for_each_entry(ns, &ctrl->namespaces, list)
4651 nvme_start_ns_queue(ns);
4652 up_read(&ctrl->namespaces_rwsem);
4654 EXPORT_SYMBOL_GPL(nvme_start_queues);
4656 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4658 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4659 blk_mq_quiesce_queue(ctrl->admin_q);
4661 blk_mq_wait_quiesce_done(ctrl->admin_q);
4663 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4665 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4667 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4668 blk_mq_unquiesce_queue(ctrl->admin_q);
4670 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4672 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4676 down_read(&ctrl->namespaces_rwsem);
4677 list_for_each_entry(ns, &ctrl->namespaces, list)
4678 blk_sync_queue(ns->queue);
4679 up_read(&ctrl->namespaces_rwsem);
4681 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4683 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4685 nvme_sync_io_queues(ctrl);
4687 blk_sync_queue(ctrl->admin_q);
4689 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4691 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4693 if (file->f_op != &nvme_dev_fops)
4695 return file->private_data;
4697 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4700 * Check we didn't inadvertently grow the command structure sizes:
4702 static inline void _nvme_check_size(void)
4704 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4705 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4706 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4707 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4708 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4709 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4710 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4711 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4712 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4713 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4714 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4715 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4716 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4717 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4718 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4719 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4720 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4721 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4722 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4723 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4727 static int __init nvme_core_init(void)
4729 int result = -ENOMEM;
4733 nvme_wq = alloc_workqueue("nvme-wq",
4734 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4738 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4739 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4743 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4744 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4745 if (!nvme_delete_wq)
4746 goto destroy_reset_wq;
4748 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4749 NVME_MINORS, "nvme");
4751 goto destroy_delete_wq;
4753 nvme_class = class_create(THIS_MODULE, "nvme");
4754 if (IS_ERR(nvme_class)) {
4755 result = PTR_ERR(nvme_class);
4756 goto unregister_chrdev;
4758 nvme_class->dev_uevent = nvme_class_uevent;
4760 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4761 if (IS_ERR(nvme_subsys_class)) {
4762 result = PTR_ERR(nvme_subsys_class);
4766 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4769 goto destroy_subsys_class;
4771 nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4772 if (IS_ERR(nvme_ns_chr_class)) {
4773 result = PTR_ERR(nvme_ns_chr_class);
4774 goto unregister_generic_ns;
4779 unregister_generic_ns:
4780 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4781 destroy_subsys_class:
4782 class_destroy(nvme_subsys_class);
4784 class_destroy(nvme_class);
4786 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4788 destroy_workqueue(nvme_delete_wq);
4790 destroy_workqueue(nvme_reset_wq);
4792 destroy_workqueue(nvme_wq);
4797 static void __exit nvme_core_exit(void)
4799 class_destroy(nvme_ns_chr_class);
4800 class_destroy(nvme_subsys_class);
4801 class_destroy(nvme_class);
4802 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4803 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4804 destroy_workqueue(nvme_delete_wq);
4805 destroy_workqueue(nvme_reset_wq);
4806 destroy_workqueue(nvme_wq);
4807 ida_destroy(&nvme_ns_chr_minor_ida);
4808 ida_destroy(&nvme_instance_ida);
4811 MODULE_LICENSE("GPL");
4812 MODULE_VERSION("1.0");
4813 module_init(nvme_core_init);
4814 module_exit(nvme_core_exit);