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 static void nvme_log_error(struct request *req)
304 struct nvme_ns *ns = req->q->queuedata;
305 struct nvme_request *nr = nvme_req(req);
308 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
309 ns->disk ? ns->disk->disk_name : "?",
310 nvme_get_opcode_str(nr->cmd->common.opcode),
311 nr->cmd->common.opcode,
312 (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
313 (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
314 nvme_get_error_status_str(nr->status),
315 nr->status >> 8 & 7, /* Status Code Type */
316 nr->status & 0xff, /* Status Code */
317 nr->status & NVME_SC_MORE ? "MORE " : "",
318 nr->status & NVME_SC_DNR ? "DNR " : "");
322 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
323 dev_name(nr->ctrl->device),
324 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
325 nr->cmd->common.opcode,
326 nvme_get_error_status_str(nr->status),
327 nr->status >> 8 & 7, /* Status Code Type */
328 nr->status & 0xff, /* Status Code */
329 nr->status & NVME_SC_MORE ? "MORE " : "",
330 nr->status & NVME_SC_DNR ? "DNR " : "");
333 enum nvme_disposition {
339 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
341 if (likely(nvme_req(req)->status == 0))
344 if (blk_noretry_request(req) ||
345 (nvme_req(req)->status & NVME_SC_DNR) ||
346 nvme_req(req)->retries >= nvme_max_retries)
349 if (req->cmd_flags & REQ_NVME_MPATH) {
350 if (nvme_is_path_error(nvme_req(req)->status) ||
351 blk_queue_dying(req->q))
354 if (blk_queue_dying(req->q))
361 static inline void nvme_end_req_zoned(struct request *req)
363 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
364 req_op(req) == REQ_OP_ZONE_APPEND)
365 req->__sector = nvme_lba_to_sect(req->q->queuedata,
366 le64_to_cpu(nvme_req(req)->result.u64));
369 static inline void nvme_end_req(struct request *req)
371 blk_status_t status = nvme_error_status(nvme_req(req)->status);
373 if (unlikely(nvme_req(req)->status != NVME_SC_SUCCESS))
375 nvme_end_req_zoned(req);
376 nvme_trace_bio_complete(req);
377 blk_mq_end_request(req, status);
380 void nvme_complete_rq(struct request *req)
382 trace_nvme_complete_rq(req);
383 nvme_cleanup_cmd(req);
385 if (nvme_req(req)->ctrl->kas)
386 nvme_req(req)->ctrl->comp_seen = true;
388 switch (nvme_decide_disposition(req)) {
396 nvme_failover_req(req);
400 EXPORT_SYMBOL_GPL(nvme_complete_rq);
402 void nvme_complete_batch_req(struct request *req)
404 trace_nvme_complete_rq(req);
405 nvme_cleanup_cmd(req);
406 nvme_end_req_zoned(req);
408 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
411 * Called to unwind from ->queue_rq on a failed command submission so that the
412 * multipathing code gets called to potentially failover to another path.
413 * The caller needs to unwind all transport specific resource allocations and
414 * must return propagate the return value.
416 blk_status_t nvme_host_path_error(struct request *req)
418 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
419 blk_mq_set_request_complete(req);
420 nvme_complete_rq(req);
423 EXPORT_SYMBOL_GPL(nvme_host_path_error);
425 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
427 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
428 "Cancelling I/O %d", req->tag);
430 /* don't abort one completed request */
431 if (blk_mq_request_completed(req))
434 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
435 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
436 blk_mq_complete_request(req);
439 EXPORT_SYMBOL_GPL(nvme_cancel_request);
441 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
444 blk_mq_tagset_busy_iter(ctrl->tagset,
445 nvme_cancel_request, ctrl);
446 blk_mq_tagset_wait_completed_request(ctrl->tagset);
449 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
451 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
453 if (ctrl->admin_tagset) {
454 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
455 nvme_cancel_request, ctrl);
456 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
459 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
461 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
462 enum nvme_ctrl_state new_state)
464 enum nvme_ctrl_state old_state;
466 bool changed = false;
468 spin_lock_irqsave(&ctrl->lock, flags);
470 old_state = ctrl->state;
475 case NVME_CTRL_RESETTING:
476 case NVME_CTRL_CONNECTING:
483 case NVME_CTRL_RESETTING:
493 case NVME_CTRL_CONNECTING:
496 case NVME_CTRL_RESETTING:
503 case NVME_CTRL_DELETING:
506 case NVME_CTRL_RESETTING:
507 case NVME_CTRL_CONNECTING:
514 case NVME_CTRL_DELETING_NOIO:
516 case NVME_CTRL_DELETING:
526 case NVME_CTRL_DELETING:
538 ctrl->state = new_state;
539 wake_up_all(&ctrl->state_wq);
542 spin_unlock_irqrestore(&ctrl->lock, flags);
546 if (ctrl->state == NVME_CTRL_LIVE) {
547 if (old_state == NVME_CTRL_CONNECTING)
548 nvme_stop_failfast_work(ctrl);
549 nvme_kick_requeue_lists(ctrl);
550 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
551 old_state == NVME_CTRL_RESETTING) {
552 nvme_start_failfast_work(ctrl);
556 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
559 * Returns true for sink states that can't ever transition back to live.
561 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
563 switch (ctrl->state) {
566 case NVME_CTRL_RESETTING:
567 case NVME_CTRL_CONNECTING:
569 case NVME_CTRL_DELETING:
570 case NVME_CTRL_DELETING_NOIO:
574 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
580 * Waits for the controller state to be resetting, or returns false if it is
581 * not possible to ever transition to that state.
583 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
585 wait_event(ctrl->state_wq,
586 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
587 nvme_state_terminal(ctrl));
588 return ctrl->state == NVME_CTRL_RESETTING;
590 EXPORT_SYMBOL_GPL(nvme_wait_reset);
592 static void nvme_free_ns_head(struct kref *ref)
594 struct nvme_ns_head *head =
595 container_of(ref, struct nvme_ns_head, ref);
597 nvme_mpath_remove_disk(head);
598 ida_free(&head->subsys->ns_ida, head->instance);
599 cleanup_srcu_struct(&head->srcu);
600 nvme_put_subsystem(head->subsys);
604 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
606 return kref_get_unless_zero(&head->ref);
609 void nvme_put_ns_head(struct nvme_ns_head *head)
611 kref_put(&head->ref, nvme_free_ns_head);
614 static void nvme_free_ns(struct kref *kref)
616 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
619 nvme_put_ns_head(ns->head);
620 nvme_put_ctrl(ns->ctrl);
624 static inline bool nvme_get_ns(struct nvme_ns *ns)
626 return kref_get_unless_zero(&ns->kref);
629 void nvme_put_ns(struct nvme_ns *ns)
631 kref_put(&ns->kref, nvme_free_ns);
633 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
635 static inline void nvme_clear_nvme_request(struct request *req)
637 nvme_req(req)->status = 0;
638 nvme_req(req)->retries = 0;
639 nvme_req(req)->flags = 0;
640 req->rq_flags |= RQF_DONTPREP;
643 /* initialize a passthrough request */
644 void nvme_init_request(struct request *req, struct nvme_command *cmd)
646 if (req->q->queuedata)
647 req->timeout = NVME_IO_TIMEOUT;
648 else /* no queuedata implies admin queue */
649 req->timeout = NVME_ADMIN_TIMEOUT;
651 /* passthru commands should let the driver set the SGL flags */
652 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
654 req->cmd_flags |= REQ_FAILFAST_DRIVER;
655 if (req->mq_hctx->type == HCTX_TYPE_POLL)
656 req->cmd_flags |= REQ_POLLED;
657 nvme_clear_nvme_request(req);
658 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
660 EXPORT_SYMBOL_GPL(nvme_init_request);
663 * For something we're not in a state to send to the device the default action
664 * is to busy it and retry it after the controller state is recovered. However,
665 * if the controller is deleting or if anything is marked for failfast or
666 * nvme multipath it is immediately failed.
668 * Note: commands used to initialize the controller will be marked for failfast.
669 * Note: nvme cli/ioctl commands are marked for failfast.
671 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
674 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
675 ctrl->state != NVME_CTRL_DELETING &&
676 ctrl->state != NVME_CTRL_DEAD &&
677 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
678 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
679 return BLK_STS_RESOURCE;
680 return nvme_host_path_error(rq);
682 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
684 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
687 struct nvme_request *req = nvme_req(rq);
690 * currently we have a problem sending passthru commands
691 * on the admin_q if the controller is not LIVE because we can't
692 * make sure that they are going out after the admin connect,
693 * controller enable and/or other commands in the initialization
694 * sequence. until the controller will be LIVE, fail with
695 * BLK_STS_RESOURCE so that they will be rescheduled.
697 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
700 if (ctrl->ops->flags & NVME_F_FABRICS) {
702 * Only allow commands on a live queue, except for the connect
703 * command, which is require to set the queue live in the
704 * appropinquate states.
706 switch (ctrl->state) {
707 case NVME_CTRL_CONNECTING:
708 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
709 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
721 EXPORT_SYMBOL_GPL(__nvme_check_ready);
723 static int nvme_toggle_streams(struct nvme_ctrl *ctrl, bool enable)
725 struct nvme_command c = { };
727 c.directive.opcode = nvme_admin_directive_send;
728 c.directive.nsid = cpu_to_le32(NVME_NSID_ALL);
729 c.directive.doper = NVME_DIR_SND_ID_OP_ENABLE;
730 c.directive.dtype = NVME_DIR_IDENTIFY;
731 c.directive.tdtype = NVME_DIR_STREAMS;
732 c.directive.endir = enable ? NVME_DIR_ENDIR : 0;
734 return nvme_submit_sync_cmd(ctrl->admin_q, &c, NULL, 0);
737 static int nvme_disable_streams(struct nvme_ctrl *ctrl)
739 return nvme_toggle_streams(ctrl, false);
742 static int nvme_enable_streams(struct nvme_ctrl *ctrl)
744 return nvme_toggle_streams(ctrl, true);
747 static int nvme_get_stream_params(struct nvme_ctrl *ctrl,
748 struct streams_directive_params *s, u32 nsid)
750 struct nvme_command c = { };
752 memset(s, 0, sizeof(*s));
754 c.directive.opcode = nvme_admin_directive_recv;
755 c.directive.nsid = cpu_to_le32(nsid);
756 c.directive.numd = cpu_to_le32(nvme_bytes_to_numd(sizeof(*s)));
757 c.directive.doper = NVME_DIR_RCV_ST_OP_PARAM;
758 c.directive.dtype = NVME_DIR_STREAMS;
760 return nvme_submit_sync_cmd(ctrl->admin_q, &c, s, sizeof(*s));
763 static int nvme_configure_directives(struct nvme_ctrl *ctrl)
765 struct streams_directive_params s;
769 if (!(ctrl->oacs & NVME_CTRL_OACS_DIRECTIVES))
774 ret = nvme_enable_streams(ctrl);
778 ret = nvme_get_stream_params(ctrl, &s, NVME_NSID_ALL);
780 goto out_disable_stream;
782 nssa = le16_to_cpu(s.nssa);
783 if (nssa < BLK_MAX_WRITE_HINTS - 1) {
784 dev_info(ctrl->device, "too few streams (%u) available\n",
786 /* this condition is not an error: streams are optional */
788 goto out_disable_stream;
791 ctrl->nr_streams = min_t(u16, nssa, BLK_MAX_WRITE_HINTS - 1);
792 dev_info(ctrl->device, "Using %u streams\n", ctrl->nr_streams);
796 nvme_disable_streams(ctrl);
801 * Check if 'req' has a write hint associated with it. If it does, assign
802 * a valid namespace stream to the write.
804 static void nvme_assign_write_stream(struct nvme_ctrl *ctrl,
805 struct request *req, u16 *control,
808 enum rw_hint streamid = req->write_hint;
810 if (streamid == WRITE_LIFE_NOT_SET || streamid == WRITE_LIFE_NONE)
814 if (WARN_ON_ONCE(streamid > ctrl->nr_streams))
817 *control |= NVME_RW_DTYPE_STREAMS;
818 *dsmgmt |= streamid << 16;
821 if (streamid < ARRAY_SIZE(req->q->write_hints))
822 req->q->write_hints[streamid] += blk_rq_bytes(req) >> 9;
825 static inline void nvme_setup_flush(struct nvme_ns *ns,
826 struct nvme_command *cmnd)
828 memset(cmnd, 0, sizeof(*cmnd));
829 cmnd->common.opcode = nvme_cmd_flush;
830 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
833 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
834 struct nvme_command *cmnd)
836 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
837 struct nvme_dsm_range *range;
841 * Some devices do not consider the DSM 'Number of Ranges' field when
842 * determining how much data to DMA. Always allocate memory for maximum
843 * number of segments to prevent device reading beyond end of buffer.
845 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
847 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
850 * If we fail allocation our range, fallback to the controller
851 * discard page. If that's also busy, it's safe to return
852 * busy, as we know we can make progress once that's freed.
854 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
855 return BLK_STS_RESOURCE;
857 range = page_address(ns->ctrl->discard_page);
860 __rq_for_each_bio(bio, req) {
861 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
862 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
865 range[n].cattr = cpu_to_le32(0);
866 range[n].nlb = cpu_to_le32(nlb);
867 range[n].slba = cpu_to_le64(slba);
872 if (WARN_ON_ONCE(n != segments)) {
873 if (virt_to_page(range) == ns->ctrl->discard_page)
874 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
877 return BLK_STS_IOERR;
880 memset(cmnd, 0, sizeof(*cmnd));
881 cmnd->dsm.opcode = nvme_cmd_dsm;
882 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
883 cmnd->dsm.nr = cpu_to_le32(segments - 1);
884 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
886 req->special_vec.bv_page = virt_to_page(range);
887 req->special_vec.bv_offset = offset_in_page(range);
888 req->special_vec.bv_len = alloc_size;
889 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
894 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
895 struct request *req, struct nvme_command *cmnd)
897 memset(cmnd, 0, sizeof(*cmnd));
899 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
900 return nvme_setup_discard(ns, req, cmnd);
902 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
903 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
904 cmnd->write_zeroes.slba =
905 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
906 cmnd->write_zeroes.length =
907 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
909 if (nvme_ns_has_pi(ns)) {
910 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
912 switch (ns->pi_type) {
913 case NVME_NS_DPS_PI_TYPE1:
914 case NVME_NS_DPS_PI_TYPE2:
915 cmnd->write_zeroes.reftag =
916 cpu_to_le32(t10_pi_ref_tag(req));
924 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
925 struct request *req, struct nvme_command *cmnd,
928 struct nvme_ctrl *ctrl = ns->ctrl;
932 if (req->cmd_flags & REQ_FUA)
933 control |= NVME_RW_FUA;
934 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
935 control |= NVME_RW_LR;
937 if (req->cmd_flags & REQ_RAHEAD)
938 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
940 cmnd->rw.opcode = op;
942 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
944 cmnd->rw.metadata = 0;
945 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
946 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
949 cmnd->rw.appmask = 0;
951 if (req_op(req) == REQ_OP_WRITE && ctrl->nr_streams)
952 nvme_assign_write_stream(ctrl, req, &control, &dsmgmt);
956 * If formated with metadata, the block layer always provides a
957 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
958 * we enable the PRACT bit for protection information or set the
959 * namespace capacity to zero to prevent any I/O.
961 if (!blk_integrity_rq(req)) {
962 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
963 return BLK_STS_NOTSUPP;
964 control |= NVME_RW_PRINFO_PRACT;
967 switch (ns->pi_type) {
968 case NVME_NS_DPS_PI_TYPE3:
969 control |= NVME_RW_PRINFO_PRCHK_GUARD;
971 case NVME_NS_DPS_PI_TYPE1:
972 case NVME_NS_DPS_PI_TYPE2:
973 control |= NVME_RW_PRINFO_PRCHK_GUARD |
974 NVME_RW_PRINFO_PRCHK_REF;
975 if (op == nvme_cmd_zone_append)
976 control |= NVME_RW_APPEND_PIREMAP;
977 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
982 cmnd->rw.control = cpu_to_le16(control);
983 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
987 void nvme_cleanup_cmd(struct request *req)
989 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
990 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
992 if (req->special_vec.bv_page == ctrl->discard_page)
993 clear_bit_unlock(0, &ctrl->discard_page_busy);
995 kfree(bvec_virt(&req->special_vec));
998 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
1000 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1002 struct nvme_command *cmd = nvme_req(req)->cmd;
1003 blk_status_t ret = BLK_STS_OK;
1005 if (!(req->rq_flags & RQF_DONTPREP))
1006 nvme_clear_nvme_request(req);
1008 switch (req_op(req)) {
1010 case REQ_OP_DRV_OUT:
1011 /* these are setup prior to execution in nvme_init_request() */
1014 nvme_setup_flush(ns, cmd);
1016 case REQ_OP_ZONE_RESET_ALL:
1017 case REQ_OP_ZONE_RESET:
1018 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1020 case REQ_OP_ZONE_OPEN:
1021 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1023 case REQ_OP_ZONE_CLOSE:
1024 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1026 case REQ_OP_ZONE_FINISH:
1027 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1029 case REQ_OP_WRITE_ZEROES:
1030 ret = nvme_setup_write_zeroes(ns, req, cmd);
1032 case REQ_OP_DISCARD:
1033 ret = nvme_setup_discard(ns, req, cmd);
1036 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1039 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1041 case REQ_OP_ZONE_APPEND:
1042 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1046 return BLK_STS_IOERR;
1049 cmd->common.command_id = nvme_cid(req);
1050 trace_nvme_setup_cmd(req, cmd);
1053 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1058 * >0: nvme controller's cqe status response
1059 * <0: kernel error in lieu of controller response
1061 static int nvme_execute_rq(struct request *rq, bool at_head)
1063 blk_status_t status;
1065 status = blk_execute_rq(rq, at_head);
1066 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1068 if (nvme_req(rq)->status)
1069 return nvme_req(rq)->status;
1070 return blk_status_to_errno(status);
1074 * Returns 0 on success. If the result is negative, it's a Linux error code;
1075 * if the result is positive, it's an NVM Express status code
1077 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1078 union nvme_result *result, void *buffer, unsigned bufflen,
1079 unsigned timeout, int qid, int at_head,
1080 blk_mq_req_flags_t flags)
1082 struct request *req;
1085 if (qid == NVME_QID_ANY)
1086 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1088 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1092 return PTR_ERR(req);
1093 nvme_init_request(req, cmd);
1096 req->timeout = timeout;
1098 if (buffer && bufflen) {
1099 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1104 ret = nvme_execute_rq(req, at_head);
1105 if (result && ret >= 0)
1106 *result = nvme_req(req)->result;
1108 blk_mq_free_request(req);
1111 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1113 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1114 void *buffer, unsigned bufflen)
1116 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1117 NVME_QID_ANY, 0, 0);
1119 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1121 static u32 nvme_known_admin_effects(u8 opcode)
1124 case nvme_admin_format_nvm:
1125 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1126 NVME_CMD_EFFECTS_CSE_MASK;
1127 case nvme_admin_sanitize_nvm:
1128 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1135 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1140 if (ns->head->effects)
1141 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1142 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1143 dev_warn_once(ctrl->device,
1144 "IO command:%02x has unhandled effects:%08x\n",
1150 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1151 effects |= nvme_known_admin_effects(opcode);
1155 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1157 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1160 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1163 * For simplicity, IO to all namespaces is quiesced even if the command
1164 * effects say only one namespace is affected.
1166 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1167 mutex_lock(&ctrl->scan_lock);
1168 mutex_lock(&ctrl->subsys->lock);
1169 nvme_mpath_start_freeze(ctrl->subsys);
1170 nvme_mpath_wait_freeze(ctrl->subsys);
1171 nvme_start_freeze(ctrl);
1172 nvme_wait_freeze(ctrl);
1177 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1178 struct nvme_command *cmd, int status)
1180 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1181 nvme_unfreeze(ctrl);
1182 nvme_mpath_unfreeze(ctrl->subsys);
1183 mutex_unlock(&ctrl->subsys->lock);
1184 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1185 mutex_unlock(&ctrl->scan_lock);
1187 if (effects & NVME_CMD_EFFECTS_CCC)
1188 nvme_init_ctrl_finish(ctrl);
1189 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1190 nvme_queue_scan(ctrl);
1191 flush_work(&ctrl->scan_work);
1194 switch (cmd->common.opcode) {
1195 case nvme_admin_set_features:
1196 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1197 case NVME_FEAT_KATO:
1199 * Keep alive commands interval on the host should be
1200 * updated when KATO is modified by Set Features
1204 nvme_update_keep_alive(ctrl, cmd);
1215 int nvme_execute_passthru_rq(struct request *rq)
1217 struct nvme_command *cmd = nvme_req(rq)->cmd;
1218 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1219 struct nvme_ns *ns = rq->q->queuedata;
1223 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1224 ret = nvme_execute_rq(rq, false);
1225 if (effects) /* nothing to be done for zero cmd effects */
1226 nvme_passthru_end(ctrl, effects, cmd, ret);
1230 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1233 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1235 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1236 * accounting for transport roundtrip times [..].
1238 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1240 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1243 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1245 struct nvme_ctrl *ctrl = rq->end_io_data;
1246 unsigned long flags;
1247 bool startka = false;
1249 blk_mq_free_request(rq);
1252 dev_err(ctrl->device,
1253 "failed nvme_keep_alive_end_io error=%d\n",
1258 ctrl->comp_seen = false;
1259 spin_lock_irqsave(&ctrl->lock, flags);
1260 if (ctrl->state == NVME_CTRL_LIVE ||
1261 ctrl->state == NVME_CTRL_CONNECTING)
1263 spin_unlock_irqrestore(&ctrl->lock, flags);
1265 nvme_queue_keep_alive_work(ctrl);
1268 static void nvme_keep_alive_work(struct work_struct *work)
1270 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1271 struct nvme_ctrl, ka_work);
1272 bool comp_seen = ctrl->comp_seen;
1275 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1276 dev_dbg(ctrl->device,
1277 "reschedule traffic based keep-alive timer\n");
1278 ctrl->comp_seen = false;
1279 nvme_queue_keep_alive_work(ctrl);
1283 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1284 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1286 /* allocation failure, reset the controller */
1287 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1288 nvme_reset_ctrl(ctrl);
1291 nvme_init_request(rq, &ctrl->ka_cmd);
1293 rq->timeout = ctrl->kato * HZ;
1294 rq->end_io_data = ctrl;
1295 blk_execute_rq_nowait(rq, false, nvme_keep_alive_end_io);
1298 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1300 if (unlikely(ctrl->kato == 0))
1303 nvme_queue_keep_alive_work(ctrl);
1306 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1308 if (unlikely(ctrl->kato == 0))
1311 cancel_delayed_work_sync(&ctrl->ka_work);
1313 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1315 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1316 struct nvme_command *cmd)
1318 unsigned int new_kato =
1319 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1321 dev_info(ctrl->device,
1322 "keep alive interval updated from %u ms to %u ms\n",
1323 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1325 nvme_stop_keep_alive(ctrl);
1326 ctrl->kato = new_kato;
1327 nvme_start_keep_alive(ctrl);
1331 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1332 * flag, thus sending any new CNS opcodes has a big chance of not working.
1333 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1334 * (but not for any later version).
1336 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1338 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1339 return ctrl->vs < NVME_VS(1, 2, 0);
1340 return ctrl->vs < NVME_VS(1, 1, 0);
1343 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1345 struct nvme_command c = { };
1348 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1349 c.identify.opcode = nvme_admin_identify;
1350 c.identify.cns = NVME_ID_CNS_CTRL;
1352 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1356 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1357 sizeof(struct nvme_id_ctrl));
1363 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1364 struct nvme_ns_id_desc *cur, bool *csi_seen)
1366 const char *warn_str = "ctrl returned bogus length:";
1369 switch (cur->nidt) {
1370 case NVME_NIDT_EUI64:
1371 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1372 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1373 warn_str, cur->nidl);
1376 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1377 return NVME_NIDT_EUI64_LEN;
1378 case NVME_NIDT_NGUID:
1379 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1380 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1381 warn_str, cur->nidl);
1384 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1385 return NVME_NIDT_NGUID_LEN;
1386 case NVME_NIDT_UUID:
1387 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1388 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1389 warn_str, cur->nidl);
1392 uuid_copy(&ids->uuid, data + sizeof(*cur));
1393 return NVME_NIDT_UUID_LEN;
1395 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1396 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1397 warn_str, cur->nidl);
1400 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1402 return NVME_NIDT_CSI_LEN;
1404 /* Skip unknown types */
1409 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1410 struct nvme_ns_ids *ids)
1412 struct nvme_command c = { };
1413 bool csi_seen = false;
1414 int status, pos, len;
1417 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1419 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1422 c.identify.opcode = nvme_admin_identify;
1423 c.identify.nsid = cpu_to_le32(nsid);
1424 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1426 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1430 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1431 NVME_IDENTIFY_DATA_SIZE);
1433 dev_warn(ctrl->device,
1434 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1439 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1440 struct nvme_ns_id_desc *cur = data + pos;
1445 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1449 len += sizeof(*cur);
1452 if (nvme_multi_css(ctrl) && !csi_seen) {
1453 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1463 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1464 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1466 struct nvme_command c = { };
1469 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1470 c.identify.opcode = nvme_admin_identify;
1471 c.identify.nsid = cpu_to_le32(nsid);
1472 c.identify.cns = NVME_ID_CNS_NS;
1474 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1478 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1480 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1484 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1485 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1488 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1489 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1490 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1491 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1492 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1493 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1502 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1503 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1505 union nvme_result res = { 0 };
1506 struct nvme_command c = { };
1509 c.features.opcode = op;
1510 c.features.fid = cpu_to_le32(fid);
1511 c.features.dword11 = cpu_to_le32(dword11);
1513 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1514 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1515 if (ret >= 0 && result)
1516 *result = le32_to_cpu(res.u32);
1520 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1521 unsigned int dword11, void *buffer, size_t buflen,
1524 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1527 EXPORT_SYMBOL_GPL(nvme_set_features);
1529 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1530 unsigned int dword11, void *buffer, size_t buflen,
1533 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1536 EXPORT_SYMBOL_GPL(nvme_get_features);
1538 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1540 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1542 int status, nr_io_queues;
1544 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1550 * Degraded controllers might return an error when setting the queue
1551 * count. We still want to be able to bring them online and offer
1552 * access to the admin queue, as that might be only way to fix them up.
1555 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1558 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1559 *count = min(*count, nr_io_queues);
1564 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1566 #define NVME_AEN_SUPPORTED \
1567 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1568 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1570 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1572 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1575 if (!supported_aens)
1578 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1581 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1584 queue_work(nvme_wq, &ctrl->async_event_work);
1587 static int nvme_ns_open(struct nvme_ns *ns)
1590 /* should never be called due to GENHD_FL_HIDDEN */
1591 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1593 if (!nvme_get_ns(ns))
1595 if (!try_module_get(ns->ctrl->ops->module))
1606 static void nvme_ns_release(struct nvme_ns *ns)
1609 module_put(ns->ctrl->ops->module);
1613 static int nvme_open(struct block_device *bdev, fmode_t mode)
1615 return nvme_ns_open(bdev->bd_disk->private_data);
1618 static void nvme_release(struct gendisk *disk, fmode_t mode)
1620 nvme_ns_release(disk->private_data);
1623 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1625 /* some standard values */
1626 geo->heads = 1 << 6;
1627 geo->sectors = 1 << 5;
1628 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1632 #ifdef CONFIG_BLK_DEV_INTEGRITY
1633 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1634 u32 max_integrity_segments)
1636 struct blk_integrity integrity = { };
1639 case NVME_NS_DPS_PI_TYPE3:
1640 integrity.profile = &t10_pi_type3_crc;
1641 integrity.tag_size = sizeof(u16) + sizeof(u32);
1642 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1644 case NVME_NS_DPS_PI_TYPE1:
1645 case NVME_NS_DPS_PI_TYPE2:
1646 integrity.profile = &t10_pi_type1_crc;
1647 integrity.tag_size = sizeof(u16);
1648 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1651 integrity.profile = NULL;
1654 integrity.tuple_size = ms;
1655 blk_integrity_register(disk, &integrity);
1656 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1659 static void nvme_init_integrity(struct gendisk *disk, u16 ms, u8 pi_type,
1660 u32 max_integrity_segments)
1663 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1665 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1667 struct nvme_ctrl *ctrl = ns->ctrl;
1668 struct request_queue *queue = disk->queue;
1669 u32 size = queue_logical_block_size(queue);
1671 if (ctrl->max_discard_sectors == 0) {
1672 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, queue);
1676 if (ctrl->nr_streams && ns->sws && ns->sgs)
1677 size *= ns->sws * ns->sgs;
1679 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1680 NVME_DSM_MAX_RANGES);
1682 queue->limits.discard_alignment = 0;
1683 queue->limits.discard_granularity = size;
1685 /* If discard is already enabled, don't reset queue limits */
1686 if (blk_queue_flag_test_and_set(QUEUE_FLAG_DISCARD, queue))
1689 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1690 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1692 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1693 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1696 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1698 return uuid_equal(&a->uuid, &b->uuid) &&
1699 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1700 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1704 static int nvme_setup_streams_ns(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1705 u32 *phys_bs, u32 *io_opt)
1707 struct streams_directive_params s;
1710 if (!ctrl->nr_streams)
1713 ret = nvme_get_stream_params(ctrl, &s, ns->head->ns_id);
1717 ns->sws = le32_to_cpu(s.sws);
1718 ns->sgs = le16_to_cpu(s.sgs);
1721 *phys_bs = ns->sws * (1 << ns->lba_shift);
1723 *io_opt = *phys_bs * ns->sgs;
1729 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1731 struct nvme_ctrl *ctrl = ns->ctrl;
1734 * The PI implementation requires the metadata size to be equal to the
1735 * t10 pi tuple size.
1737 ns->ms = le16_to_cpu(id->lbaf[id->flbas & NVME_NS_FLBAS_LBA_MASK].ms);
1738 if (ns->ms == sizeof(struct t10_pi_tuple))
1739 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1743 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1744 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1747 if (ctrl->ops->flags & NVME_F_FABRICS) {
1749 * The NVMe over Fabrics specification only supports metadata as
1750 * part of the extended data LBA. We rely on HCA/HBA support to
1751 * remap the separate metadata buffer from the block layer.
1753 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1756 ns->features |= NVME_NS_EXT_LBAS;
1759 * The current fabrics transport drivers support namespace
1760 * metadata formats only if nvme_ns_has_pi() returns true.
1761 * Suppress support for all other formats so the namespace will
1762 * have a 0 capacity and not be usable through the block stack.
1764 * Note, this check will need to be modified if any drivers
1765 * gain the ability to use other metadata formats.
1767 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1768 ns->features |= NVME_NS_METADATA_SUPPORTED;
1771 * For PCIe controllers, we can't easily remap the separate
1772 * metadata buffer from the block layer and thus require a
1773 * separate metadata buffer for block layer metadata/PI support.
1774 * We allow extended LBAs for the passthrough interface, though.
1776 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1777 ns->features |= NVME_NS_EXT_LBAS;
1779 ns->features |= NVME_NS_METADATA_SUPPORTED;
1783 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1784 struct request_queue *q)
1786 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1788 if (ctrl->max_hw_sectors) {
1790 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1792 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1793 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1794 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1796 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1797 blk_queue_dma_alignment(q, 7);
1798 blk_queue_write_cache(q, vwc, vwc);
1801 static void nvme_update_disk_info(struct gendisk *disk,
1802 struct nvme_ns *ns, struct nvme_id_ns *id)
1804 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1805 unsigned short bs = 1 << ns->lba_shift;
1806 u32 atomic_bs, phys_bs, io_opt = 0;
1809 * The block layer can't support LBA sizes larger than the page size
1810 * yet, so catch this early and don't allow block I/O.
1812 if (ns->lba_shift > PAGE_SHIFT) {
1817 blk_integrity_unregister(disk);
1819 atomic_bs = phys_bs = bs;
1820 nvme_setup_streams_ns(ns->ctrl, ns, &phys_bs, &io_opt);
1821 if (id->nabo == 0) {
1823 * Bit 1 indicates whether NAWUPF is defined for this namespace
1824 * and whether it should be used instead of AWUPF. If NAWUPF ==
1825 * 0 then AWUPF must be used instead.
1827 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1828 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1830 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1833 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1834 /* NPWG = Namespace Preferred Write Granularity */
1835 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1836 /* NOWS = Namespace Optimal Write Size */
1837 io_opt = bs * (1 + le16_to_cpu(id->nows));
1840 blk_queue_logical_block_size(disk->queue, bs);
1842 * Linux filesystems assume writing a single physical block is
1843 * an atomic operation. Hence limit the physical block size to the
1844 * value of the Atomic Write Unit Power Fail parameter.
1846 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1847 blk_queue_io_min(disk->queue, phys_bs);
1848 blk_queue_io_opt(disk->queue, io_opt);
1851 * Register a metadata profile for PI, or the plain non-integrity NVMe
1852 * metadata masquerading as Type 0 if supported, otherwise reject block
1853 * I/O to namespaces with metadata except when the namespace supports
1854 * PI, as it can strip/insert in that case.
1857 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1858 (ns->features & NVME_NS_METADATA_SUPPORTED))
1859 nvme_init_integrity(disk, ns->ms, ns->pi_type,
1860 ns->ctrl->max_integrity_segments);
1861 else if (!nvme_ns_has_pi(ns))
1865 set_capacity_and_notify(disk, capacity);
1867 nvme_config_discard(disk, ns);
1868 blk_queue_max_write_zeroes_sectors(disk->queue,
1869 ns->ctrl->max_zeroes_sectors);
1871 set_disk_ro(disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1872 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1875 static inline bool nvme_first_scan(struct gendisk *disk)
1877 /* nvme_alloc_ns() scans the disk prior to adding it */
1878 return !disk_live(disk);
1881 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1883 struct nvme_ctrl *ctrl = ns->ctrl;
1886 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1887 is_power_of_2(ctrl->max_hw_sectors))
1888 iob = ctrl->max_hw_sectors;
1890 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1895 if (!is_power_of_2(iob)) {
1896 if (nvme_first_scan(ns->disk))
1897 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1898 ns->disk->disk_name, iob);
1902 if (blk_queue_is_zoned(ns->disk->queue)) {
1903 if (nvme_first_scan(ns->disk))
1904 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1905 ns->disk->disk_name);
1909 blk_queue_chunk_sectors(ns->queue, iob);
1912 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1914 unsigned lbaf = id->flbas & NVME_NS_FLBAS_LBA_MASK;
1917 blk_mq_freeze_queue(ns->disk->queue);
1918 ns->lba_shift = id->lbaf[lbaf].ds;
1919 nvme_set_queue_limits(ns->ctrl, ns->queue);
1921 nvme_configure_metadata(ns, id);
1922 nvme_set_chunk_sectors(ns, id);
1923 nvme_update_disk_info(ns->disk, ns, id);
1925 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1926 ret = nvme_update_zone_info(ns, lbaf);
1931 set_bit(NVME_NS_READY, &ns->flags);
1932 blk_mq_unfreeze_queue(ns->disk->queue);
1934 if (blk_queue_is_zoned(ns->queue)) {
1935 ret = nvme_revalidate_zones(ns);
1936 if (ret && !nvme_first_scan(ns->disk))
1940 if (nvme_ns_head_multipath(ns->head)) {
1941 blk_mq_freeze_queue(ns->head->disk->queue);
1942 nvme_update_disk_info(ns->head->disk, ns, id);
1943 nvme_mpath_revalidate_paths(ns);
1944 blk_stack_limits(&ns->head->disk->queue->limits,
1945 &ns->queue->limits, 0);
1946 disk_update_readahead(ns->head->disk);
1947 blk_mq_unfreeze_queue(ns->head->disk->queue);
1953 * If probing fails due an unsupported feature, hide the block device,
1954 * but still allow other access.
1956 if (ret == -ENODEV) {
1957 ns->disk->flags |= GENHD_FL_HIDDEN;
1958 set_bit(NVME_NS_READY, &ns->flags);
1961 blk_mq_unfreeze_queue(ns->disk->queue);
1965 static char nvme_pr_type(enum pr_type type)
1968 case PR_WRITE_EXCLUSIVE:
1970 case PR_EXCLUSIVE_ACCESS:
1972 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1974 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1976 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1978 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1985 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1986 struct nvme_command *c, u8 data[16])
1988 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1989 int srcu_idx = srcu_read_lock(&head->srcu);
1990 struct nvme_ns *ns = nvme_find_path(head);
1991 int ret = -EWOULDBLOCK;
1994 c->common.nsid = cpu_to_le32(ns->head->ns_id);
1995 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
1997 srcu_read_unlock(&head->srcu, srcu_idx);
2001 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2004 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2005 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2008 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2009 u64 key, u64 sa_key, u8 op)
2011 struct nvme_command c = { };
2012 u8 data[16] = { 0, };
2014 put_unaligned_le64(key, &data[0]);
2015 put_unaligned_le64(sa_key, &data[8]);
2017 c.common.opcode = op;
2018 c.common.cdw10 = cpu_to_le32(cdw10);
2020 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2021 bdev->bd_disk->fops == &nvme_ns_head_ops)
2022 return nvme_send_ns_head_pr_command(bdev, &c, data);
2023 return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2026 static int nvme_pr_register(struct block_device *bdev, u64 old,
2027 u64 new, unsigned flags)
2031 if (flags & ~PR_FL_IGNORE_KEY)
2034 cdw10 = old ? 2 : 0;
2035 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2036 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2037 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2040 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2041 enum pr_type type, unsigned flags)
2045 if (flags & ~PR_FL_IGNORE_KEY)
2048 cdw10 = nvme_pr_type(type) << 8;
2049 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2050 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2053 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2054 enum pr_type type, bool abort)
2056 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2058 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2061 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2063 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2065 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2068 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2070 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2072 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2075 const struct pr_ops nvme_pr_ops = {
2076 .pr_register = nvme_pr_register,
2077 .pr_reserve = nvme_pr_reserve,
2078 .pr_release = nvme_pr_release,
2079 .pr_preempt = nvme_pr_preempt,
2080 .pr_clear = nvme_pr_clear,
2083 #ifdef CONFIG_BLK_SED_OPAL
2084 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2087 struct nvme_ctrl *ctrl = data;
2088 struct nvme_command cmd = { };
2091 cmd.common.opcode = nvme_admin_security_send;
2093 cmd.common.opcode = nvme_admin_security_recv;
2094 cmd.common.nsid = 0;
2095 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2096 cmd.common.cdw11 = cpu_to_le32(len);
2098 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2099 NVME_QID_ANY, 1, 0);
2101 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2102 #endif /* CONFIG_BLK_SED_OPAL */
2104 #ifdef CONFIG_BLK_DEV_ZONED
2105 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2106 unsigned int nr_zones, report_zones_cb cb, void *data)
2108 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2112 #define nvme_report_zones NULL
2113 #endif /* CONFIG_BLK_DEV_ZONED */
2115 static const struct block_device_operations nvme_bdev_ops = {
2116 .owner = THIS_MODULE,
2117 .ioctl = nvme_ioctl,
2119 .release = nvme_release,
2120 .getgeo = nvme_getgeo,
2121 .report_zones = nvme_report_zones,
2122 .pr_ops = &nvme_pr_ops,
2125 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u64 cap, bool enabled)
2127 unsigned long timeout =
2128 ((NVME_CAP_TIMEOUT(cap) + 1) * HZ / 2) + jiffies;
2129 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2132 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2135 if ((csts & NVME_CSTS_RDY) == bit)
2138 usleep_range(1000, 2000);
2139 if (fatal_signal_pending(current))
2141 if (time_after(jiffies, timeout)) {
2142 dev_err(ctrl->device,
2143 "Device not ready; aborting %s, CSTS=0x%x\n",
2144 enabled ? "initialisation" : "reset", csts);
2153 * If the device has been passed off to us in an enabled state, just clear
2154 * the enabled bit. The spec says we should set the 'shutdown notification
2155 * bits', but doing so may cause the device to complete commands to the
2156 * admin queue ... and we don't know what memory that might be pointing at!
2158 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2162 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2163 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2165 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2169 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2170 msleep(NVME_QUIRK_DELAY_AMOUNT);
2172 return nvme_wait_ready(ctrl, ctrl->cap, false);
2174 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2176 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2178 unsigned dev_page_min;
2181 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2183 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2186 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2188 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2189 dev_err(ctrl->device,
2190 "Minimum device page size %u too large for host (%u)\n",
2191 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2195 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2196 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2198 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2199 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2200 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2201 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2202 ctrl->ctrl_config |= NVME_CC_ENABLE;
2204 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2207 return nvme_wait_ready(ctrl, ctrl->cap, true);
2209 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2211 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2213 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2217 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2218 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2220 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2224 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2225 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2229 if (fatal_signal_pending(current))
2231 if (time_after(jiffies, timeout)) {
2232 dev_err(ctrl->device,
2233 "Device shutdown incomplete; abort shutdown\n");
2240 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2242 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2247 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2250 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2251 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2254 dev_warn_once(ctrl->device,
2255 "could not set timestamp (%d)\n", ret);
2259 static int nvme_configure_acre(struct nvme_ctrl *ctrl)
2261 struct nvme_feat_host_behavior *host;
2264 /* Don't bother enabling the feature if retry delay is not reported */
2268 host = kzalloc(sizeof(*host), GFP_KERNEL);
2272 host->acre = NVME_ENABLE_ACRE;
2273 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2274 host, sizeof(*host), NULL);
2280 * The function checks whether the given total (exlat + enlat) latency of
2281 * a power state allows the latter to be used as an APST transition target.
2282 * It does so by comparing the latency to the primary and secondary latency
2283 * tolerances defined by module params. If there's a match, the corresponding
2284 * timeout value is returned and the matching tolerance index (1 or 2) is
2287 static bool nvme_apst_get_transition_time(u64 total_latency,
2288 u64 *transition_time, unsigned *last_index)
2290 if (total_latency <= apst_primary_latency_tol_us) {
2291 if (*last_index == 1)
2294 *transition_time = apst_primary_timeout_ms;
2297 if (apst_secondary_timeout_ms &&
2298 total_latency <= apst_secondary_latency_tol_us) {
2299 if (*last_index <= 2)
2302 *transition_time = apst_secondary_timeout_ms;
2309 * APST (Autonomous Power State Transition) lets us program a table of power
2310 * state transitions that the controller will perform automatically.
2312 * Depending on module params, one of the two supported techniques will be used:
2314 * - If the parameters provide explicit timeouts and tolerances, they will be
2315 * used to build a table with up to 2 non-operational states to transition to.
2316 * The default parameter values were selected based on the values used by
2317 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2318 * regeneration of the APST table in the event of switching between external
2319 * and battery power, the timeouts and tolerances reflect a compromise
2320 * between values used by Microsoft for AC and battery scenarios.
2321 * - If not, we'll configure the table with a simple heuristic: we are willing
2322 * to spend at most 2% of the time transitioning between power states.
2323 * Therefore, when running in any given state, we will enter the next
2324 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2325 * microseconds, as long as that state's exit latency is under the requested
2328 * We will not autonomously enter any non-operational state for which the total
2329 * latency exceeds ps_max_latency_us.
2331 * Users can set ps_max_latency_us to zero to turn off APST.
2333 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2335 struct nvme_feat_auto_pst *table;
2342 unsigned last_lt_index = UINT_MAX;
2345 * If APST isn't supported or if we haven't been initialized yet,
2346 * then don't do anything.
2351 if (ctrl->npss > 31) {
2352 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2356 table = kzalloc(sizeof(*table), GFP_KERNEL);
2360 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2361 /* Turn off APST. */
2362 dev_dbg(ctrl->device, "APST disabled\n");
2367 * Walk through all states from lowest- to highest-power.
2368 * According to the spec, lower-numbered states use more power. NPSS,
2369 * despite the name, is the index of the lowest-power state, not the
2372 for (state = (int)ctrl->npss; state >= 0; state--) {
2373 u64 total_latency_us, exit_latency_us, transition_ms;
2376 table->entries[state] = target;
2379 * Don't allow transitions to the deepest state if it's quirked
2382 if (state == ctrl->npss &&
2383 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2387 * Is this state a useful non-operational state for higher-power
2388 * states to autonomously transition to?
2390 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2393 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2394 if (exit_latency_us > ctrl->ps_max_latency_us)
2397 total_latency_us = exit_latency_us +
2398 le32_to_cpu(ctrl->psd[state].entry_lat);
2401 * This state is good. It can be used as the APST idle target
2402 * for higher power states.
2404 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2405 if (!nvme_apst_get_transition_time(total_latency_us,
2406 &transition_ms, &last_lt_index))
2409 transition_ms = total_latency_us + 19;
2410 do_div(transition_ms, 20);
2411 if (transition_ms > (1 << 24) - 1)
2412 transition_ms = (1 << 24) - 1;
2415 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2418 if (total_latency_us > max_lat_us)
2419 max_lat_us = total_latency_us;
2423 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2425 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2426 max_ps, max_lat_us, (int)sizeof(*table), table);
2430 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2431 table, sizeof(*table), NULL);
2433 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2438 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2440 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2444 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2445 case PM_QOS_LATENCY_ANY:
2453 if (ctrl->ps_max_latency_us != latency) {
2454 ctrl->ps_max_latency_us = latency;
2455 if (ctrl->state == NVME_CTRL_LIVE)
2456 nvme_configure_apst(ctrl);
2460 struct nvme_core_quirk_entry {
2462 * NVMe model and firmware strings are padded with spaces. For
2463 * simplicity, strings in the quirk table are padded with NULLs
2469 unsigned long quirks;
2472 static const struct nvme_core_quirk_entry core_quirks[] = {
2475 * This Toshiba device seems to die using any APST states. See:
2476 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2479 .mn = "THNSF5256GPUK TOSHIBA",
2480 .quirks = NVME_QUIRK_NO_APST,
2484 * This LiteON CL1-3D*-Q11 firmware version has a race
2485 * condition associated with actions related to suspend to idle
2486 * LiteON has resolved the problem in future firmware
2490 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2494 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2495 * aborts I/O during any load, but more easily reproducible
2496 * with discards (fstrim).
2498 * The device is left in a state where it is also not possible
2499 * to use "nvme set-feature" to disable APST, but booting with
2500 * nvme_core.default_ps_max_latency=0 works.
2503 .mn = "KCD6XVUL6T40",
2504 .quirks = NVME_QUIRK_NO_APST,
2508 /* match is null-terminated but idstr is space-padded. */
2509 static bool string_matches(const char *idstr, const char *match, size_t len)
2516 matchlen = strlen(match);
2517 WARN_ON_ONCE(matchlen > len);
2519 if (memcmp(idstr, match, matchlen))
2522 for (; matchlen < len; matchlen++)
2523 if (idstr[matchlen] != ' ')
2529 static bool quirk_matches(const struct nvme_id_ctrl *id,
2530 const struct nvme_core_quirk_entry *q)
2532 return q->vid == le16_to_cpu(id->vid) &&
2533 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2534 string_matches(id->fr, q->fr, sizeof(id->fr));
2537 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2538 struct nvme_id_ctrl *id)
2543 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2544 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2545 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2546 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2550 if (ctrl->vs >= NVME_VS(1, 2, 1))
2551 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2554 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2555 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2556 "nqn.2014.08.org.nvmexpress:%04x%04x",
2557 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2558 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2559 off += sizeof(id->sn);
2560 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2561 off += sizeof(id->mn);
2562 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2565 static void nvme_release_subsystem(struct device *dev)
2567 struct nvme_subsystem *subsys =
2568 container_of(dev, struct nvme_subsystem, dev);
2570 if (subsys->instance >= 0)
2571 ida_free(&nvme_instance_ida, subsys->instance);
2575 static void nvme_destroy_subsystem(struct kref *ref)
2577 struct nvme_subsystem *subsys =
2578 container_of(ref, struct nvme_subsystem, ref);
2580 mutex_lock(&nvme_subsystems_lock);
2581 list_del(&subsys->entry);
2582 mutex_unlock(&nvme_subsystems_lock);
2584 ida_destroy(&subsys->ns_ida);
2585 device_del(&subsys->dev);
2586 put_device(&subsys->dev);
2589 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2591 kref_put(&subsys->ref, nvme_destroy_subsystem);
2594 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2596 struct nvme_subsystem *subsys;
2598 lockdep_assert_held(&nvme_subsystems_lock);
2601 * Fail matches for discovery subsystems. This results
2602 * in each discovery controller bound to a unique subsystem.
2603 * This avoids issues with validating controller values
2604 * that can only be true when there is a single unique subsystem.
2605 * There may be multiple and completely independent entities
2606 * that provide discovery controllers.
2608 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2611 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2612 if (strcmp(subsys->subnqn, subsysnqn))
2614 if (!kref_get_unless_zero(&subsys->ref))
2622 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2623 struct device_attribute subsys_attr_##_name = \
2624 __ATTR(_name, _mode, _show, NULL)
2626 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2627 struct device_attribute *attr,
2630 struct nvme_subsystem *subsys =
2631 container_of(dev, struct nvme_subsystem, dev);
2633 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2635 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2637 static ssize_t nvme_subsys_show_type(struct device *dev,
2638 struct device_attribute *attr,
2641 struct nvme_subsystem *subsys =
2642 container_of(dev, struct nvme_subsystem, dev);
2644 switch (subsys->subtype) {
2646 return sysfs_emit(buf, "discovery\n");
2648 return sysfs_emit(buf, "nvm\n");
2650 return sysfs_emit(buf, "reserved\n");
2653 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2655 #define nvme_subsys_show_str_function(field) \
2656 static ssize_t subsys_##field##_show(struct device *dev, \
2657 struct device_attribute *attr, char *buf) \
2659 struct nvme_subsystem *subsys = \
2660 container_of(dev, struct nvme_subsystem, dev); \
2661 return sysfs_emit(buf, "%.*s\n", \
2662 (int)sizeof(subsys->field), subsys->field); \
2664 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2666 nvme_subsys_show_str_function(model);
2667 nvme_subsys_show_str_function(serial);
2668 nvme_subsys_show_str_function(firmware_rev);
2670 static struct attribute *nvme_subsys_attrs[] = {
2671 &subsys_attr_model.attr,
2672 &subsys_attr_serial.attr,
2673 &subsys_attr_firmware_rev.attr,
2674 &subsys_attr_subsysnqn.attr,
2675 &subsys_attr_subsystype.attr,
2676 #ifdef CONFIG_NVME_MULTIPATH
2677 &subsys_attr_iopolicy.attr,
2682 static const struct attribute_group nvme_subsys_attrs_group = {
2683 .attrs = nvme_subsys_attrs,
2686 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2687 &nvme_subsys_attrs_group,
2691 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2693 return ctrl->opts && ctrl->opts->discovery_nqn;
2696 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2697 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2699 struct nvme_ctrl *tmp;
2701 lockdep_assert_held(&nvme_subsystems_lock);
2703 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2704 if (nvme_state_terminal(tmp))
2707 if (tmp->cntlid == ctrl->cntlid) {
2708 dev_err(ctrl->device,
2709 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2710 ctrl->cntlid, dev_name(tmp->device),
2715 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2716 nvme_discovery_ctrl(ctrl))
2719 dev_err(ctrl->device,
2720 "Subsystem does not support multiple controllers\n");
2727 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2729 struct nvme_subsystem *subsys, *found;
2732 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2736 subsys->instance = -1;
2737 mutex_init(&subsys->lock);
2738 kref_init(&subsys->ref);
2739 INIT_LIST_HEAD(&subsys->ctrls);
2740 INIT_LIST_HEAD(&subsys->nsheads);
2741 nvme_init_subnqn(subsys, ctrl, id);
2742 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2743 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2744 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2745 subsys->vendor_id = le16_to_cpu(id->vid);
2746 subsys->cmic = id->cmic;
2748 /* Versions prior to 1.4 don't necessarily report a valid type */
2749 if (id->cntrltype == NVME_CTRL_DISC ||
2750 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2751 subsys->subtype = NVME_NQN_DISC;
2753 subsys->subtype = NVME_NQN_NVME;
2755 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2756 dev_err(ctrl->device,
2757 "Subsystem %s is not a discovery controller",
2762 subsys->awupf = le16_to_cpu(id->awupf);
2763 nvme_mpath_default_iopolicy(subsys);
2765 subsys->dev.class = nvme_subsys_class;
2766 subsys->dev.release = nvme_release_subsystem;
2767 subsys->dev.groups = nvme_subsys_attrs_groups;
2768 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2769 device_initialize(&subsys->dev);
2771 mutex_lock(&nvme_subsystems_lock);
2772 found = __nvme_find_get_subsystem(subsys->subnqn);
2774 put_device(&subsys->dev);
2777 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2779 goto out_put_subsystem;
2782 ret = device_add(&subsys->dev);
2784 dev_err(ctrl->device,
2785 "failed to register subsystem device.\n");
2786 put_device(&subsys->dev);
2789 ida_init(&subsys->ns_ida);
2790 list_add_tail(&subsys->entry, &nvme_subsystems);
2793 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2794 dev_name(ctrl->device));
2796 dev_err(ctrl->device,
2797 "failed to create sysfs link from subsystem.\n");
2798 goto out_put_subsystem;
2802 subsys->instance = ctrl->instance;
2803 ctrl->subsys = subsys;
2804 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2805 mutex_unlock(&nvme_subsystems_lock);
2809 nvme_put_subsystem(subsys);
2811 mutex_unlock(&nvme_subsystems_lock);
2815 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2816 void *log, size_t size, u64 offset)
2818 struct nvme_command c = { };
2819 u32 dwlen = nvme_bytes_to_numd(size);
2821 c.get_log_page.opcode = nvme_admin_get_log_page;
2822 c.get_log_page.nsid = cpu_to_le32(nsid);
2823 c.get_log_page.lid = log_page;
2824 c.get_log_page.lsp = lsp;
2825 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2826 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2827 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2828 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2829 c.get_log_page.csi = csi;
2831 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2834 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2835 struct nvme_effects_log **log)
2837 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2843 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2847 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2848 cel, sizeof(*cel), 0);
2854 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2860 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2862 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2864 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2869 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2871 struct nvme_command c = { };
2872 struct nvme_id_ctrl_nvm *id;
2875 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2876 ctrl->max_discard_sectors = UINT_MAX;
2877 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2879 ctrl->max_discard_sectors = 0;
2880 ctrl->max_discard_segments = 0;
2884 * Even though NVMe spec explicitly states that MDTS is not applicable
2885 * to the write-zeroes, we are cautious and limit the size to the
2886 * controllers max_hw_sectors value, which is based on the MDTS field
2887 * and possibly other limiting factors.
2889 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2890 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2891 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2893 ctrl->max_zeroes_sectors = 0;
2895 if (nvme_ctrl_limited_cns(ctrl))
2898 id = kzalloc(sizeof(*id), GFP_KERNEL);
2902 c.identify.opcode = nvme_admin_identify;
2903 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2904 c.identify.csi = NVME_CSI_NVM;
2906 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2911 ctrl->max_discard_segments = id->dmrl;
2913 ctrl->max_discard_sectors = le32_to_cpu(id->dmrsl);
2915 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2922 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2924 struct nvme_id_ctrl *id;
2926 bool prev_apst_enabled;
2929 ret = nvme_identify_ctrl(ctrl, &id);
2931 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2935 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2936 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2941 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2942 ctrl->cntlid = le16_to_cpu(id->cntlid);
2944 if (!ctrl->identified) {
2947 ret = nvme_init_subsystem(ctrl, id);
2952 * Check for quirks. Quirk can depend on firmware version,
2953 * so, in principle, the set of quirks present can change
2954 * across a reset. As a possible future enhancement, we
2955 * could re-scan for quirks every time we reinitialize
2956 * the device, but we'd have to make sure that the driver
2957 * behaves intelligently if the quirks change.
2959 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
2960 if (quirk_matches(id, &core_quirks[i]))
2961 ctrl->quirks |= core_quirks[i].quirks;
2965 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
2966 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
2967 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
2970 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
2971 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
2972 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
2974 ctrl->oacs = le16_to_cpu(id->oacs);
2975 ctrl->oncs = le16_to_cpu(id->oncs);
2976 ctrl->mtfa = le16_to_cpu(id->mtfa);
2977 ctrl->oaes = le32_to_cpu(id->oaes);
2978 ctrl->wctemp = le16_to_cpu(id->wctemp);
2979 ctrl->cctemp = le16_to_cpu(id->cctemp);
2981 atomic_set(&ctrl->abort_limit, id->acl + 1);
2982 ctrl->vwc = id->vwc;
2984 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
2986 max_hw_sectors = UINT_MAX;
2987 ctrl->max_hw_sectors =
2988 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
2990 nvme_set_queue_limits(ctrl, ctrl->admin_q);
2991 ctrl->sgls = le32_to_cpu(id->sgls);
2992 ctrl->kas = le16_to_cpu(id->kas);
2993 ctrl->max_namespaces = le32_to_cpu(id->mnan);
2994 ctrl->ctratt = le32_to_cpu(id->ctratt);
2996 ctrl->cntrltype = id->cntrltype;
2997 ctrl->dctype = id->dctype;
3001 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3003 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3004 shutdown_timeout, 60);
3006 if (ctrl->shutdown_timeout != shutdown_timeout)
3007 dev_info(ctrl->device,
3008 "Shutdown timeout set to %u seconds\n",
3009 ctrl->shutdown_timeout);
3011 ctrl->shutdown_timeout = shutdown_timeout;
3013 ctrl->npss = id->npss;
3014 ctrl->apsta = id->apsta;
3015 prev_apst_enabled = ctrl->apst_enabled;
3016 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3017 if (force_apst && id->apsta) {
3018 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3019 ctrl->apst_enabled = true;
3021 ctrl->apst_enabled = false;
3024 ctrl->apst_enabled = id->apsta;
3026 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3028 if (ctrl->ops->flags & NVME_F_FABRICS) {
3029 ctrl->icdoff = le16_to_cpu(id->icdoff);
3030 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3031 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3032 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3035 * In fabrics we need to verify the cntlid matches the
3038 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3039 dev_err(ctrl->device,
3040 "Mismatching cntlid: Connect %u vs Identify "
3042 ctrl->cntlid, le16_to_cpu(id->cntlid));
3047 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3048 dev_err(ctrl->device,
3049 "keep-alive support is mandatory for fabrics\n");
3054 ctrl->hmpre = le32_to_cpu(id->hmpre);
3055 ctrl->hmmin = le32_to_cpu(id->hmmin);
3056 ctrl->hmminds = le32_to_cpu(id->hmminds);
3057 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3060 ret = nvme_mpath_init_identify(ctrl, id);
3064 if (ctrl->apst_enabled && !prev_apst_enabled)
3065 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3066 else if (!ctrl->apst_enabled && prev_apst_enabled)
3067 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3075 * Initialize the cached copies of the Identify data and various controller
3076 * register in our nvme_ctrl structure. This should be called as soon as
3077 * the admin queue is fully up and running.
3079 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3083 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3085 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3089 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3091 if (ctrl->vs >= NVME_VS(1, 1, 0))
3092 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3094 ret = nvme_init_identify(ctrl);
3098 ret = nvme_init_non_mdts_limits(ctrl);
3102 ret = nvme_configure_apst(ctrl);
3106 ret = nvme_configure_timestamp(ctrl);
3110 ret = nvme_configure_directives(ctrl);
3114 ret = nvme_configure_acre(ctrl);
3118 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3119 ret = nvme_hwmon_init(ctrl);
3124 ctrl->identified = true;
3128 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3130 static int nvme_dev_open(struct inode *inode, struct file *file)
3132 struct nvme_ctrl *ctrl =
3133 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3135 switch (ctrl->state) {
3136 case NVME_CTRL_LIVE:
3139 return -EWOULDBLOCK;
3142 nvme_get_ctrl(ctrl);
3143 if (!try_module_get(ctrl->ops->module)) {
3144 nvme_put_ctrl(ctrl);
3148 file->private_data = ctrl;
3152 static int nvme_dev_release(struct inode *inode, struct file *file)
3154 struct nvme_ctrl *ctrl =
3155 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3157 module_put(ctrl->ops->module);
3158 nvme_put_ctrl(ctrl);
3162 static const struct file_operations nvme_dev_fops = {
3163 .owner = THIS_MODULE,
3164 .open = nvme_dev_open,
3165 .release = nvme_dev_release,
3166 .unlocked_ioctl = nvme_dev_ioctl,
3167 .compat_ioctl = compat_ptr_ioctl,
3170 static ssize_t nvme_sysfs_reset(struct device *dev,
3171 struct device_attribute *attr, const char *buf,
3174 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3177 ret = nvme_reset_ctrl_sync(ctrl);
3182 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3184 static ssize_t nvme_sysfs_rescan(struct device *dev,
3185 struct device_attribute *attr, const char *buf,
3188 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3190 nvme_queue_scan(ctrl);
3193 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3195 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3197 struct gendisk *disk = dev_to_disk(dev);
3199 if (disk->fops == &nvme_bdev_ops)
3200 return nvme_get_ns_from_dev(dev)->head;
3202 return disk->private_data;
3205 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3208 struct nvme_ns_head *head = dev_to_ns_head(dev);
3209 struct nvme_ns_ids *ids = &head->ids;
3210 struct nvme_subsystem *subsys = head->subsys;
3211 int serial_len = sizeof(subsys->serial);
3212 int model_len = sizeof(subsys->model);
3214 if (!uuid_is_null(&ids->uuid))
3215 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3217 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3218 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3220 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3221 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3223 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3224 subsys->serial[serial_len - 1] == '\0'))
3226 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3227 subsys->model[model_len - 1] == '\0'))
3230 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3231 serial_len, subsys->serial, model_len, subsys->model,
3234 static DEVICE_ATTR_RO(wwid);
3236 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3239 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3241 static DEVICE_ATTR_RO(nguid);
3243 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3246 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3248 /* For backward compatibility expose the NGUID to userspace if
3249 * we have no UUID set
3251 if (uuid_is_null(&ids->uuid)) {
3252 printk_ratelimited(KERN_WARNING
3253 "No UUID available providing old NGUID\n");
3254 return sysfs_emit(buf, "%pU\n", ids->nguid);
3256 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3258 static DEVICE_ATTR_RO(uuid);
3260 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3263 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3265 static DEVICE_ATTR_RO(eui);
3267 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3270 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3272 static DEVICE_ATTR_RO(nsid);
3274 static struct attribute *nvme_ns_id_attrs[] = {
3275 &dev_attr_wwid.attr,
3276 &dev_attr_uuid.attr,
3277 &dev_attr_nguid.attr,
3279 &dev_attr_nsid.attr,
3280 #ifdef CONFIG_NVME_MULTIPATH
3281 &dev_attr_ana_grpid.attr,
3282 &dev_attr_ana_state.attr,
3287 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3288 struct attribute *a, int n)
3290 struct device *dev = container_of(kobj, struct device, kobj);
3291 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3293 if (a == &dev_attr_uuid.attr) {
3294 if (uuid_is_null(&ids->uuid) &&
3295 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3298 if (a == &dev_attr_nguid.attr) {
3299 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3302 if (a == &dev_attr_eui.attr) {
3303 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3306 #ifdef CONFIG_NVME_MULTIPATH
3307 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3308 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3310 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3317 static const struct attribute_group nvme_ns_id_attr_group = {
3318 .attrs = nvme_ns_id_attrs,
3319 .is_visible = nvme_ns_id_attrs_are_visible,
3322 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3323 &nvme_ns_id_attr_group,
3327 #define nvme_show_str_function(field) \
3328 static ssize_t field##_show(struct device *dev, \
3329 struct device_attribute *attr, char *buf) \
3331 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3332 return sysfs_emit(buf, "%.*s\n", \
3333 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3335 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3337 nvme_show_str_function(model);
3338 nvme_show_str_function(serial);
3339 nvme_show_str_function(firmware_rev);
3341 #define nvme_show_int_function(field) \
3342 static ssize_t field##_show(struct device *dev, \
3343 struct device_attribute *attr, char *buf) \
3345 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3346 return sysfs_emit(buf, "%d\n", ctrl->field); \
3348 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3350 nvme_show_int_function(cntlid);
3351 nvme_show_int_function(numa_node);
3352 nvme_show_int_function(queue_count);
3353 nvme_show_int_function(sqsize);
3354 nvme_show_int_function(kato);
3356 static ssize_t nvme_sysfs_delete(struct device *dev,
3357 struct device_attribute *attr, const char *buf,
3360 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3362 if (device_remove_file_self(dev, attr))
3363 nvme_delete_ctrl_sync(ctrl);
3366 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3368 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3369 struct device_attribute *attr,
3372 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3374 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3376 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3378 static ssize_t nvme_sysfs_show_state(struct device *dev,
3379 struct device_attribute *attr,
3382 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3383 static const char *const state_name[] = {
3384 [NVME_CTRL_NEW] = "new",
3385 [NVME_CTRL_LIVE] = "live",
3386 [NVME_CTRL_RESETTING] = "resetting",
3387 [NVME_CTRL_CONNECTING] = "connecting",
3388 [NVME_CTRL_DELETING] = "deleting",
3389 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3390 [NVME_CTRL_DEAD] = "dead",
3393 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3394 state_name[ctrl->state])
3395 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3397 return sysfs_emit(buf, "unknown state\n");
3400 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3402 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3403 struct device_attribute *attr,
3406 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3408 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3410 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3412 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3413 struct device_attribute *attr,
3416 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3418 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3420 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3422 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3423 struct device_attribute *attr,
3426 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3428 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3430 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3432 static ssize_t nvme_sysfs_show_address(struct device *dev,
3433 struct device_attribute *attr,
3436 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3438 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3440 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3442 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3443 struct device_attribute *attr, char *buf)
3445 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3446 struct nvmf_ctrl_options *opts = ctrl->opts;
3448 if (ctrl->opts->max_reconnects == -1)
3449 return sysfs_emit(buf, "off\n");
3450 return sysfs_emit(buf, "%d\n",
3451 opts->max_reconnects * opts->reconnect_delay);
3454 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3455 struct device_attribute *attr, const char *buf, size_t count)
3457 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3458 struct nvmf_ctrl_options *opts = ctrl->opts;
3459 int ctrl_loss_tmo, err;
3461 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3465 if (ctrl_loss_tmo < 0)
3466 opts->max_reconnects = -1;
3468 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3469 opts->reconnect_delay);
3472 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3473 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3475 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3476 struct device_attribute *attr, char *buf)
3478 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3480 if (ctrl->opts->reconnect_delay == -1)
3481 return sysfs_emit(buf, "off\n");
3482 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3485 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3486 struct device_attribute *attr, const char *buf, size_t count)
3488 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3492 err = kstrtou32(buf, 10, &v);
3496 ctrl->opts->reconnect_delay = v;
3499 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3500 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3502 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3503 struct device_attribute *attr, char *buf)
3505 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3507 if (ctrl->opts->fast_io_fail_tmo == -1)
3508 return sysfs_emit(buf, "off\n");
3509 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3512 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3513 struct device_attribute *attr, const char *buf, size_t count)
3515 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3516 struct nvmf_ctrl_options *opts = ctrl->opts;
3517 int fast_io_fail_tmo, err;
3519 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3523 if (fast_io_fail_tmo < 0)
3524 opts->fast_io_fail_tmo = -1;
3526 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3529 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3530 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3532 static ssize_t cntrltype_show(struct device *dev,
3533 struct device_attribute *attr, char *buf)
3535 static const char * const type[] = {
3536 [NVME_CTRL_IO] = "io\n",
3537 [NVME_CTRL_DISC] = "discovery\n",
3538 [NVME_CTRL_ADMIN] = "admin\n",
3540 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3542 if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3543 return sysfs_emit(buf, "reserved\n");
3545 return sysfs_emit(buf, type[ctrl->cntrltype]);
3547 static DEVICE_ATTR_RO(cntrltype);
3549 static ssize_t dctype_show(struct device *dev,
3550 struct device_attribute *attr, char *buf)
3552 static const char * const type[] = {
3553 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3554 [NVME_DCTYPE_DDC] = "ddc\n",
3555 [NVME_DCTYPE_CDC] = "cdc\n",
3557 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3559 if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3560 return sysfs_emit(buf, "reserved\n");
3562 return sysfs_emit(buf, type[ctrl->dctype]);
3564 static DEVICE_ATTR_RO(dctype);
3566 static struct attribute *nvme_dev_attrs[] = {
3567 &dev_attr_reset_controller.attr,
3568 &dev_attr_rescan_controller.attr,
3569 &dev_attr_model.attr,
3570 &dev_attr_serial.attr,
3571 &dev_attr_firmware_rev.attr,
3572 &dev_attr_cntlid.attr,
3573 &dev_attr_delete_controller.attr,
3574 &dev_attr_transport.attr,
3575 &dev_attr_subsysnqn.attr,
3576 &dev_attr_address.attr,
3577 &dev_attr_state.attr,
3578 &dev_attr_numa_node.attr,
3579 &dev_attr_queue_count.attr,
3580 &dev_attr_sqsize.attr,
3581 &dev_attr_hostnqn.attr,
3582 &dev_attr_hostid.attr,
3583 &dev_attr_ctrl_loss_tmo.attr,
3584 &dev_attr_reconnect_delay.attr,
3585 &dev_attr_fast_io_fail_tmo.attr,
3586 &dev_attr_kato.attr,
3587 &dev_attr_cntrltype.attr,
3588 &dev_attr_dctype.attr,
3592 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3593 struct attribute *a, int n)
3595 struct device *dev = container_of(kobj, struct device, kobj);
3596 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3598 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3600 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3602 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3604 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3606 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3608 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3610 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3616 static const struct attribute_group nvme_dev_attrs_group = {
3617 .attrs = nvme_dev_attrs,
3618 .is_visible = nvme_dev_attrs_are_visible,
3621 static const struct attribute_group *nvme_dev_attr_groups[] = {
3622 &nvme_dev_attrs_group,
3626 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_subsystem *subsys,
3629 struct nvme_ns_head *h;
3631 lockdep_assert_held(&subsys->lock);
3633 list_for_each_entry(h, &subsys->nsheads, entry) {
3634 if (h->ns_id != nsid)
3636 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3643 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3644 struct nvme_ns_ids *ids)
3646 bool has_uuid = !uuid_is_null(&ids->uuid);
3647 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3648 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3649 struct nvme_ns_head *h;
3651 lockdep_assert_held(&subsys->lock);
3653 list_for_each_entry(h, &subsys->nsheads, entry) {
3654 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3657 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3660 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3667 static void nvme_cdev_rel(struct device *dev)
3669 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3672 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3674 cdev_device_del(cdev, cdev_device);
3675 put_device(cdev_device);
3678 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3679 const struct file_operations *fops, struct module *owner)
3683 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3686 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3687 cdev_device->class = nvme_ns_chr_class;
3688 cdev_device->release = nvme_cdev_rel;
3689 device_initialize(cdev_device);
3690 cdev_init(cdev, fops);
3691 cdev->owner = owner;
3692 ret = cdev_device_add(cdev, cdev_device);
3694 put_device(cdev_device);
3699 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3701 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3704 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3706 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3710 static const struct file_operations nvme_ns_chr_fops = {
3711 .owner = THIS_MODULE,
3712 .open = nvme_ns_chr_open,
3713 .release = nvme_ns_chr_release,
3714 .unlocked_ioctl = nvme_ns_chr_ioctl,
3715 .compat_ioctl = compat_ptr_ioctl,
3718 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3722 ns->cdev_device.parent = ns->ctrl->device;
3723 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3724 ns->ctrl->instance, ns->head->instance);
3728 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3729 ns->ctrl->ops->module);
3732 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3733 unsigned nsid, struct nvme_ns_ids *ids)
3735 struct nvme_ns_head *head;
3736 size_t size = sizeof(*head);
3739 #ifdef CONFIG_NVME_MULTIPATH
3740 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3743 head = kzalloc(size, GFP_KERNEL);
3746 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3749 head->instance = ret;
3750 INIT_LIST_HEAD(&head->list);
3751 ret = init_srcu_struct(&head->srcu);
3753 goto out_ida_remove;
3754 head->subsys = ctrl->subsys;
3757 kref_init(&head->ref);
3759 if (head->ids.csi) {
3760 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3762 goto out_cleanup_srcu;
3764 head->effects = ctrl->effects;
3766 ret = nvme_mpath_alloc_disk(ctrl, head);
3768 goto out_cleanup_srcu;
3770 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3772 kref_get(&ctrl->subsys->ref);
3776 cleanup_srcu_struct(&head->srcu);
3778 ida_free(&ctrl->subsys->ns_ida, head->instance);
3783 ret = blk_status_to_errno(nvme_error_status(ret));
3784 return ERR_PTR(ret);
3787 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3788 struct nvme_ns_ids *ids)
3790 struct nvme_subsystem *s;
3794 * Note that this check is racy as we try to avoid holding the global
3795 * lock over the whole ns_head creation. But it is only intended as
3796 * a sanity check anyway.
3798 mutex_lock(&nvme_subsystems_lock);
3799 list_for_each_entry(s, &nvme_subsystems, entry) {
3802 mutex_lock(&s->lock);
3803 ret = nvme_subsys_check_duplicate_ids(s, ids);
3804 mutex_unlock(&s->lock);
3808 mutex_unlock(&nvme_subsystems_lock);
3813 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3814 struct nvme_ns_ids *ids, bool is_shared)
3816 struct nvme_ctrl *ctrl = ns->ctrl;
3817 struct nvme_ns_head *head = NULL;
3820 ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3822 dev_err(ctrl->device,
3823 "globally duplicate IDs for nsid %d\n", nsid);
3827 mutex_lock(&ctrl->subsys->lock);
3828 head = nvme_find_ns_head(ctrl->subsys, nsid);
3830 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3832 dev_err(ctrl->device,
3833 "duplicate IDs in subsystem for nsid %d\n",
3837 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3839 ret = PTR_ERR(head);
3842 head->shared = is_shared;
3845 if (!is_shared || !head->shared) {
3846 dev_err(ctrl->device,
3847 "Duplicate unshared namespace %d\n", nsid);
3848 goto out_put_ns_head;
3850 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3851 dev_err(ctrl->device,
3852 "IDs don't match for shared namespace %d\n",
3854 goto out_put_ns_head;
3857 if (!multipath && !list_empty(&head->list)) {
3858 dev_warn(ctrl->device,
3859 "Found shared namespace %d, but multipathing not supported.\n",
3861 dev_warn_once(ctrl->device,
3862 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3866 list_add_tail_rcu(&ns->siblings, &head->list);
3868 mutex_unlock(&ctrl->subsys->lock);
3872 nvme_put_ns_head(head);
3874 mutex_unlock(&ctrl->subsys->lock);
3878 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3880 struct nvme_ns *ns, *ret = NULL;
3882 down_read(&ctrl->namespaces_rwsem);
3883 list_for_each_entry(ns, &ctrl->namespaces, list) {
3884 if (ns->head->ns_id == nsid) {
3885 if (!nvme_get_ns(ns))
3890 if (ns->head->ns_id > nsid)
3893 up_read(&ctrl->namespaces_rwsem);
3896 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3899 * Add the namespace to the controller list while keeping the list ordered.
3901 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3903 struct nvme_ns *tmp;
3905 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3906 if (tmp->head->ns_id < ns->head->ns_id) {
3907 list_add(&ns->list, &tmp->list);
3911 list_add(&ns->list, &ns->ctrl->namespaces);
3914 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3915 struct nvme_ns_ids *ids)
3918 struct gendisk *disk;
3919 struct nvme_id_ns *id;
3920 int node = ctrl->numa_node;
3922 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3925 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3929 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3932 disk->fops = &nvme_bdev_ops;
3933 disk->private_data = ns;
3936 ns->queue = disk->queue;
3938 if (ctrl->opts && ctrl->opts->data_digest)
3939 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3941 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3942 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
3943 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3946 kref_init(&ns->kref);
3948 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
3949 goto out_cleanup_disk;
3952 * If multipathing is enabled, the device name for all disks and not
3953 * just those that represent shared namespaces needs to be based on the
3954 * subsystem instance. Using the controller instance for private
3955 * namespaces could lead to naming collisions between shared and private
3956 * namespaces if they don't use a common numbering scheme.
3958 * If multipathing is not enabled, disk names must use the controller
3959 * instance as shared namespaces will show up as multiple block
3962 if (ns->head->disk) {
3963 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3964 ctrl->instance, ns->head->instance);
3965 disk->flags |= GENHD_FL_HIDDEN;
3966 } else if (multipath) {
3967 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3968 ns->head->instance);
3970 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3971 ns->head->instance);
3974 if (nvme_update_ns_info(ns, id))
3977 down_write(&ctrl->namespaces_rwsem);
3978 nvme_ns_add_to_ctrl_list(ns);
3979 up_write(&ctrl->namespaces_rwsem);
3980 nvme_get_ctrl(ctrl);
3982 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
3983 goto out_cleanup_ns_from_list;
3985 if (!nvme_ns_head_multipath(ns->head))
3986 nvme_add_ns_cdev(ns);
3988 nvme_mpath_add_disk(ns, id);
3989 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3994 out_cleanup_ns_from_list:
3995 nvme_put_ctrl(ctrl);
3996 down_write(&ctrl->namespaces_rwsem);
3997 list_del_init(&ns->list);
3998 up_write(&ctrl->namespaces_rwsem);
4000 mutex_lock(&ctrl->subsys->lock);
4001 list_del_rcu(&ns->siblings);
4002 if (list_empty(&ns->head->list))
4003 list_del_init(&ns->head->entry);
4004 mutex_unlock(&ctrl->subsys->lock);
4005 nvme_put_ns_head(ns->head);
4007 blk_cleanup_disk(disk);
4014 static void nvme_ns_remove(struct nvme_ns *ns)
4016 bool last_path = false;
4018 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4021 clear_bit(NVME_NS_READY, &ns->flags);
4022 set_capacity(ns->disk, 0);
4023 nvme_fault_inject_fini(&ns->fault_inject);
4025 mutex_lock(&ns->ctrl->subsys->lock);
4026 list_del_rcu(&ns->siblings);
4027 if (list_empty(&ns->head->list)) {
4028 list_del_init(&ns->head->entry);
4031 mutex_unlock(&ns->ctrl->subsys->lock);
4033 /* guarantee not available in head->list */
4036 /* wait for concurrent submissions */
4037 if (nvme_mpath_clear_current_path(ns))
4038 synchronize_srcu(&ns->head->srcu);
4040 if (!nvme_ns_head_multipath(ns->head))
4041 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4042 del_gendisk(ns->disk);
4043 blk_cleanup_queue(ns->queue);
4045 down_write(&ns->ctrl->namespaces_rwsem);
4046 list_del_init(&ns->list);
4047 up_write(&ns->ctrl->namespaces_rwsem);
4050 nvme_mpath_shutdown_disk(ns->head);
4054 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4056 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4064 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4066 struct nvme_id_ns *id;
4067 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4069 if (test_bit(NVME_NS_DEAD, &ns->flags))
4072 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4076 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4077 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4078 dev_err(ns->ctrl->device,
4079 "identifiers changed for nsid %d\n", ns->head->ns_id);
4083 ret = nvme_update_ns_info(ns, id);
4089 * Only remove the namespace if we got a fatal error back from the
4090 * device, otherwise ignore the error and just move on.
4092 * TODO: we should probably schedule a delayed retry here.
4094 if (ret > 0 && (ret & NVME_SC_DNR))
4098 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4100 struct nvme_ns_ids ids = { };
4103 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4106 ns = nvme_find_get_ns(ctrl, nsid);
4108 nvme_validate_ns(ns, &ids);
4115 nvme_alloc_ns(ctrl, nsid, &ids);
4118 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4119 dev_warn(ctrl->device,
4120 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4124 if (!nvme_multi_css(ctrl)) {
4125 dev_warn(ctrl->device,
4126 "command set not reported for nsid: %d\n",
4130 nvme_alloc_ns(ctrl, nsid, &ids);
4133 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4139 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4142 struct nvme_ns *ns, *next;
4145 down_write(&ctrl->namespaces_rwsem);
4146 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4147 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4148 list_move_tail(&ns->list, &rm_list);
4150 up_write(&ctrl->namespaces_rwsem);
4152 list_for_each_entry_safe(ns, next, &rm_list, list)
4157 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4159 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4164 if (nvme_ctrl_limited_cns(ctrl))
4167 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4172 struct nvme_command cmd = {
4173 .identify.opcode = nvme_admin_identify,
4174 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4175 .identify.nsid = cpu_to_le32(prev),
4178 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4179 NVME_IDENTIFY_DATA_SIZE);
4181 dev_warn(ctrl->device,
4182 "Identify NS List failed (status=0x%x)\n", ret);
4186 for (i = 0; i < nr_entries; i++) {
4187 u32 nsid = le32_to_cpu(ns_list[i]);
4189 if (!nsid) /* end of the list? */
4191 nvme_validate_or_alloc_ns(ctrl, nsid);
4192 while (++prev < nsid)
4193 nvme_ns_remove_by_nsid(ctrl, prev);
4197 nvme_remove_invalid_namespaces(ctrl, prev);
4203 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4205 struct nvme_id_ctrl *id;
4208 if (nvme_identify_ctrl(ctrl, &id))
4210 nn = le32_to_cpu(id->nn);
4213 for (i = 1; i <= nn; i++)
4214 nvme_validate_or_alloc_ns(ctrl, i);
4216 nvme_remove_invalid_namespaces(ctrl, nn);
4219 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4221 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4225 log = kzalloc(log_size, GFP_KERNEL);
4230 * We need to read the log to clear the AEN, but we don't want to rely
4231 * on it for the changed namespace information as userspace could have
4232 * raced with us in reading the log page, which could cause us to miss
4235 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4236 NVME_CSI_NVM, log, log_size, 0);
4238 dev_warn(ctrl->device,
4239 "reading changed ns log failed: %d\n", error);
4244 static void nvme_scan_work(struct work_struct *work)
4246 struct nvme_ctrl *ctrl =
4247 container_of(work, struct nvme_ctrl, scan_work);
4249 /* No tagset on a live ctrl means IO queues could not created */
4250 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4253 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4254 dev_info(ctrl->device, "rescanning namespaces.\n");
4255 nvme_clear_changed_ns_log(ctrl);
4258 mutex_lock(&ctrl->scan_lock);
4259 if (nvme_scan_ns_list(ctrl) != 0)
4260 nvme_scan_ns_sequential(ctrl);
4261 mutex_unlock(&ctrl->scan_lock);
4265 * This function iterates the namespace list unlocked to allow recovery from
4266 * controller failure. It is up to the caller to ensure the namespace list is
4267 * not modified by scan work while this function is executing.
4269 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4271 struct nvme_ns *ns, *next;
4275 * make sure to requeue I/O to all namespaces as these
4276 * might result from the scan itself and must complete
4277 * for the scan_work to make progress
4279 nvme_mpath_clear_ctrl_paths(ctrl);
4281 /* prevent racing with ns scanning */
4282 flush_work(&ctrl->scan_work);
4285 * The dead states indicates the controller was not gracefully
4286 * disconnected. In that case, we won't be able to flush any data while
4287 * removing the namespaces' disks; fail all the queues now to avoid
4288 * potentially having to clean up the failed sync later.
4290 if (ctrl->state == NVME_CTRL_DEAD)
4291 nvme_kill_queues(ctrl);
4293 /* this is a no-op when called from the controller reset handler */
4294 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4296 down_write(&ctrl->namespaces_rwsem);
4297 list_splice_init(&ctrl->namespaces, &ns_list);
4298 up_write(&ctrl->namespaces_rwsem);
4300 list_for_each_entry_safe(ns, next, &ns_list, list)
4303 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4305 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4307 struct nvme_ctrl *ctrl =
4308 container_of(dev, struct nvme_ctrl, ctrl_device);
4309 struct nvmf_ctrl_options *opts = ctrl->opts;
4312 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4317 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4321 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4322 opts->trsvcid ?: "none");
4326 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4327 opts->host_traddr ?: "none");
4331 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4332 opts->host_iface ?: "none");
4337 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4339 char *envp[2] = { envdata, NULL };
4341 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4344 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4346 char *envp[2] = { NULL, NULL };
4347 u32 aen_result = ctrl->aen_result;
4349 ctrl->aen_result = 0;
4353 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4356 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4360 static void nvme_async_event_work(struct work_struct *work)
4362 struct nvme_ctrl *ctrl =
4363 container_of(work, struct nvme_ctrl, async_event_work);
4365 nvme_aen_uevent(ctrl);
4368 * The transport drivers must guarantee AER submission here is safe by
4369 * flushing ctrl async_event_work after changing the controller state
4370 * from LIVE and before freeing the admin queue.
4372 if (ctrl->state == NVME_CTRL_LIVE)
4373 ctrl->ops->submit_async_event(ctrl);
4376 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4381 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4387 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4390 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4392 struct nvme_fw_slot_info_log *log;
4394 log = kmalloc(sizeof(*log), GFP_KERNEL);
4398 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4399 log, sizeof(*log), 0))
4400 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4404 static void nvme_fw_act_work(struct work_struct *work)
4406 struct nvme_ctrl *ctrl = container_of(work,
4407 struct nvme_ctrl, fw_act_work);
4408 unsigned long fw_act_timeout;
4411 fw_act_timeout = jiffies +
4412 msecs_to_jiffies(ctrl->mtfa * 100);
4414 fw_act_timeout = jiffies +
4415 msecs_to_jiffies(admin_timeout * 1000);
4417 nvme_stop_queues(ctrl);
4418 while (nvme_ctrl_pp_status(ctrl)) {
4419 if (time_after(jiffies, fw_act_timeout)) {
4420 dev_warn(ctrl->device,
4421 "Fw activation timeout, reset controller\n");
4422 nvme_try_sched_reset(ctrl);
4428 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4431 nvme_start_queues(ctrl);
4432 /* read FW slot information to clear the AER */
4433 nvme_get_fw_slot_info(ctrl);
4436 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4438 u32 aer_notice_type = (result & 0xff00) >> 8;
4440 trace_nvme_async_event(ctrl, aer_notice_type);
4442 switch (aer_notice_type) {
4443 case NVME_AER_NOTICE_NS_CHANGED:
4444 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4445 nvme_queue_scan(ctrl);
4447 case NVME_AER_NOTICE_FW_ACT_STARTING:
4449 * We are (ab)using the RESETTING state to prevent subsequent
4450 * recovery actions from interfering with the controller's
4451 * firmware activation.
4453 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4454 queue_work(nvme_wq, &ctrl->fw_act_work);
4456 #ifdef CONFIG_NVME_MULTIPATH
4457 case NVME_AER_NOTICE_ANA:
4458 if (!ctrl->ana_log_buf)
4460 queue_work(nvme_wq, &ctrl->ana_work);
4463 case NVME_AER_NOTICE_DISC_CHANGED:
4464 ctrl->aen_result = result;
4467 dev_warn(ctrl->device, "async event result %08x\n", result);
4471 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4472 volatile union nvme_result *res)
4474 u32 result = le32_to_cpu(res->u32);
4475 u32 aer_type = result & 0x07;
4477 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4481 case NVME_AER_NOTICE:
4482 nvme_handle_aen_notice(ctrl, result);
4484 case NVME_AER_ERROR:
4485 case NVME_AER_SMART:
4488 trace_nvme_async_event(ctrl, aer_type);
4489 ctrl->aen_result = result;
4494 queue_work(nvme_wq, &ctrl->async_event_work);
4496 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4498 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4500 nvme_mpath_stop(ctrl);
4501 nvme_stop_keep_alive(ctrl);
4502 nvme_stop_failfast_work(ctrl);
4503 flush_work(&ctrl->async_event_work);
4504 cancel_work_sync(&ctrl->fw_act_work);
4506 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4508 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4510 nvme_start_keep_alive(ctrl);
4512 nvme_enable_aen(ctrl);
4514 if (ctrl->queue_count > 1) {
4515 nvme_queue_scan(ctrl);
4516 nvme_start_queues(ctrl);
4519 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4521 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4523 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4525 nvme_hwmon_exit(ctrl);
4526 nvme_fault_inject_fini(&ctrl->fault_inject);
4527 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4528 cdev_device_del(&ctrl->cdev, ctrl->device);
4529 nvme_put_ctrl(ctrl);
4531 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4533 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4535 struct nvme_effects_log *cel;
4538 xa_for_each(&ctrl->cels, i, cel) {
4539 xa_erase(&ctrl->cels, i);
4543 xa_destroy(&ctrl->cels);
4546 static void nvme_free_ctrl(struct device *dev)
4548 struct nvme_ctrl *ctrl =
4549 container_of(dev, struct nvme_ctrl, ctrl_device);
4550 struct nvme_subsystem *subsys = ctrl->subsys;
4552 if (!subsys || ctrl->instance != subsys->instance)
4553 ida_free(&nvme_instance_ida, ctrl->instance);
4555 nvme_free_cels(ctrl);
4556 nvme_mpath_uninit(ctrl);
4557 __free_page(ctrl->discard_page);
4560 mutex_lock(&nvme_subsystems_lock);
4561 list_del(&ctrl->subsys_entry);
4562 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4563 mutex_unlock(&nvme_subsystems_lock);
4566 ctrl->ops->free_ctrl(ctrl);
4569 nvme_put_subsystem(subsys);
4573 * Initialize a NVMe controller structures. This needs to be called during
4574 * earliest initialization so that we have the initialized structured around
4577 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4578 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4582 ctrl->state = NVME_CTRL_NEW;
4583 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4584 spin_lock_init(&ctrl->lock);
4585 mutex_init(&ctrl->scan_lock);
4586 INIT_LIST_HEAD(&ctrl->namespaces);
4587 xa_init(&ctrl->cels);
4588 init_rwsem(&ctrl->namespaces_rwsem);
4591 ctrl->quirks = quirks;
4592 ctrl->numa_node = NUMA_NO_NODE;
4593 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4594 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4595 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4596 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4597 init_waitqueue_head(&ctrl->state_wq);
4599 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4600 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4601 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4602 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4604 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4606 ctrl->discard_page = alloc_page(GFP_KERNEL);
4607 if (!ctrl->discard_page) {
4612 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4615 ctrl->instance = ret;
4617 device_initialize(&ctrl->ctrl_device);
4618 ctrl->device = &ctrl->ctrl_device;
4619 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4621 ctrl->device->class = nvme_class;
4622 ctrl->device->parent = ctrl->dev;
4623 ctrl->device->groups = nvme_dev_attr_groups;
4624 ctrl->device->release = nvme_free_ctrl;
4625 dev_set_drvdata(ctrl->device, ctrl);
4626 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4628 goto out_release_instance;
4630 nvme_get_ctrl(ctrl);
4631 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4632 ctrl->cdev.owner = ops->module;
4633 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4638 * Initialize latency tolerance controls. The sysfs files won't
4639 * be visible to userspace unless the device actually supports APST.
4641 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4642 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4643 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4645 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4646 nvme_mpath_init_ctrl(ctrl);
4650 nvme_put_ctrl(ctrl);
4651 kfree_const(ctrl->device->kobj.name);
4652 out_release_instance:
4653 ida_free(&nvme_instance_ida, ctrl->instance);
4655 if (ctrl->discard_page)
4656 __free_page(ctrl->discard_page);
4659 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4661 static void nvme_start_ns_queue(struct nvme_ns *ns)
4663 if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4664 blk_mq_unquiesce_queue(ns->queue);
4667 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4669 if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4670 blk_mq_quiesce_queue(ns->queue);
4672 blk_mq_wait_quiesce_done(ns->queue);
4676 * Prepare a queue for teardown.
4678 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4679 * the capacity to 0 after that to avoid blocking dispatchers that may be
4680 * holding bd_butex. This will end buffered writers dirtying pages that can't
4683 static void nvme_set_queue_dying(struct nvme_ns *ns)
4685 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4688 blk_mark_disk_dead(ns->disk);
4689 nvme_start_ns_queue(ns);
4691 set_capacity_and_notify(ns->disk, 0);
4695 * nvme_kill_queues(): Ends all namespace queues
4696 * @ctrl: the dead controller that needs to end
4698 * Call this function when the driver determines it is unable to get the
4699 * controller in a state capable of servicing IO.
4701 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4705 down_read(&ctrl->namespaces_rwsem);
4707 /* Forcibly unquiesce queues to avoid blocking dispatch */
4708 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4709 nvme_start_admin_queue(ctrl);
4711 list_for_each_entry(ns, &ctrl->namespaces, list)
4712 nvme_set_queue_dying(ns);
4714 up_read(&ctrl->namespaces_rwsem);
4716 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4718 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4722 down_read(&ctrl->namespaces_rwsem);
4723 list_for_each_entry(ns, &ctrl->namespaces, list)
4724 blk_mq_unfreeze_queue(ns->queue);
4725 up_read(&ctrl->namespaces_rwsem);
4727 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4729 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4733 down_read(&ctrl->namespaces_rwsem);
4734 list_for_each_entry(ns, &ctrl->namespaces, list) {
4735 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4739 up_read(&ctrl->namespaces_rwsem);
4742 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4744 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4748 down_read(&ctrl->namespaces_rwsem);
4749 list_for_each_entry(ns, &ctrl->namespaces, list)
4750 blk_mq_freeze_queue_wait(ns->queue);
4751 up_read(&ctrl->namespaces_rwsem);
4753 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4755 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4759 down_read(&ctrl->namespaces_rwsem);
4760 list_for_each_entry(ns, &ctrl->namespaces, list)
4761 blk_freeze_queue_start(ns->queue);
4762 up_read(&ctrl->namespaces_rwsem);
4764 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4766 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4770 down_read(&ctrl->namespaces_rwsem);
4771 list_for_each_entry(ns, &ctrl->namespaces, list)
4772 nvme_stop_ns_queue(ns);
4773 up_read(&ctrl->namespaces_rwsem);
4775 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4777 void nvme_start_queues(struct nvme_ctrl *ctrl)
4781 down_read(&ctrl->namespaces_rwsem);
4782 list_for_each_entry(ns, &ctrl->namespaces, list)
4783 nvme_start_ns_queue(ns);
4784 up_read(&ctrl->namespaces_rwsem);
4786 EXPORT_SYMBOL_GPL(nvme_start_queues);
4788 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4790 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4791 blk_mq_quiesce_queue(ctrl->admin_q);
4793 blk_mq_wait_quiesce_done(ctrl->admin_q);
4795 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4797 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4799 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4800 blk_mq_unquiesce_queue(ctrl->admin_q);
4802 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4804 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4808 down_read(&ctrl->namespaces_rwsem);
4809 list_for_each_entry(ns, &ctrl->namespaces, list)
4810 blk_sync_queue(ns->queue);
4811 up_read(&ctrl->namespaces_rwsem);
4813 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4815 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4817 nvme_sync_io_queues(ctrl);
4819 blk_sync_queue(ctrl->admin_q);
4821 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4823 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4825 if (file->f_op != &nvme_dev_fops)
4827 return file->private_data;
4829 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4832 * Check we didn't inadvertently grow the command structure sizes:
4834 static inline void _nvme_check_size(void)
4836 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4837 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4838 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4839 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4840 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4841 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4842 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4843 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4844 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4845 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4846 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4847 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4848 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4849 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4850 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4851 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4852 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4853 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4854 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4855 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4859 static int __init nvme_core_init(void)
4861 int result = -ENOMEM;
4865 nvme_wq = alloc_workqueue("nvme-wq",
4866 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4870 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4871 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4875 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4876 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4877 if (!nvme_delete_wq)
4878 goto destroy_reset_wq;
4880 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4881 NVME_MINORS, "nvme");
4883 goto destroy_delete_wq;
4885 nvme_class = class_create(THIS_MODULE, "nvme");
4886 if (IS_ERR(nvme_class)) {
4887 result = PTR_ERR(nvme_class);
4888 goto unregister_chrdev;
4890 nvme_class->dev_uevent = nvme_class_uevent;
4892 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4893 if (IS_ERR(nvme_subsys_class)) {
4894 result = PTR_ERR(nvme_subsys_class);
4898 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4901 goto destroy_subsys_class;
4903 nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
4904 if (IS_ERR(nvme_ns_chr_class)) {
4905 result = PTR_ERR(nvme_ns_chr_class);
4906 goto unregister_generic_ns;
4911 unregister_generic_ns:
4912 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4913 destroy_subsys_class:
4914 class_destroy(nvme_subsys_class);
4916 class_destroy(nvme_class);
4918 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4920 destroy_workqueue(nvme_delete_wq);
4922 destroy_workqueue(nvme_reset_wq);
4924 destroy_workqueue(nvme_wq);
4929 static void __exit nvme_core_exit(void)
4931 class_destroy(nvme_ns_chr_class);
4932 class_destroy(nvme_subsys_class);
4933 class_destroy(nvme_class);
4934 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4935 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4936 destroy_workqueue(nvme_delete_wq);
4937 destroy_workqueue(nvme_reset_wq);
4938 destroy_workqueue(nvme_wq);
4939 ida_destroy(&nvme_ns_chr_minor_ida);
4940 ida_destroy(&nvme_instance_ida);
4943 MODULE_LICENSE("GPL");
4944 MODULE_VERSION("1.0");
4945 module_init(nvme_core_init);
4946 module_exit(nvme_core_exit);