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 <linux/ratelimit.h>
24 #include <asm/unaligned.h>
28 #include <linux/nvme-auth.h>
30 #define CREATE_TRACE_POINTS
33 #define NVME_MINORS (1U << MINORBITS)
36 struct nvme_ns_ids ids;
45 unsigned int admin_timeout = 60;
46 module_param(admin_timeout, uint, 0644);
47 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
48 EXPORT_SYMBOL_GPL(admin_timeout);
50 unsigned int nvme_io_timeout = 30;
51 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
52 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
53 EXPORT_SYMBOL_GPL(nvme_io_timeout);
55 static unsigned char shutdown_timeout = 5;
56 module_param(shutdown_timeout, byte, 0644);
57 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
59 static u8 nvme_max_retries = 5;
60 module_param_named(max_retries, nvme_max_retries, byte, 0644);
61 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
63 static unsigned long default_ps_max_latency_us = 100000;
64 module_param(default_ps_max_latency_us, ulong, 0644);
65 MODULE_PARM_DESC(default_ps_max_latency_us,
66 "max power saving latency for new devices; use PM QOS to change per device");
68 static bool force_apst;
69 module_param(force_apst, bool, 0644);
70 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
72 static unsigned long apst_primary_timeout_ms = 100;
73 module_param(apst_primary_timeout_ms, ulong, 0644);
74 MODULE_PARM_DESC(apst_primary_timeout_ms,
75 "primary APST timeout in ms");
77 static unsigned long apst_secondary_timeout_ms = 2000;
78 module_param(apst_secondary_timeout_ms, ulong, 0644);
79 MODULE_PARM_DESC(apst_secondary_timeout_ms,
80 "secondary APST timeout in ms");
82 static unsigned long apst_primary_latency_tol_us = 15000;
83 module_param(apst_primary_latency_tol_us, ulong, 0644);
84 MODULE_PARM_DESC(apst_primary_latency_tol_us,
85 "primary APST latency tolerance in us");
87 static unsigned long apst_secondary_latency_tol_us = 100000;
88 module_param(apst_secondary_latency_tol_us, ulong, 0644);
89 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
90 "secondary APST latency tolerance in us");
93 * nvme_wq - hosts nvme related works that are not reset or delete
94 * nvme_reset_wq - hosts nvme reset works
95 * nvme_delete_wq - hosts nvme delete works
97 * nvme_wq will host works such as scan, aen handling, fw activation,
98 * keep-alive, periodic reconnects etc. nvme_reset_wq
99 * runs reset works which also flush works hosted on nvme_wq for
100 * serialization purposes. nvme_delete_wq host controller deletion
101 * works which flush reset works for serialization.
103 struct workqueue_struct *nvme_wq;
104 EXPORT_SYMBOL_GPL(nvme_wq);
106 struct workqueue_struct *nvme_reset_wq;
107 EXPORT_SYMBOL_GPL(nvme_reset_wq);
109 struct workqueue_struct *nvme_delete_wq;
110 EXPORT_SYMBOL_GPL(nvme_delete_wq);
112 static LIST_HEAD(nvme_subsystems);
113 static DEFINE_MUTEX(nvme_subsystems_lock);
115 static DEFINE_IDA(nvme_instance_ida);
116 static dev_t nvme_ctrl_base_chr_devt;
117 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env);
118 static const struct class nvme_class = {
120 .dev_uevent = nvme_class_uevent,
123 static const struct class nvme_subsys_class = {
124 .name = "nvme-subsystem",
127 static DEFINE_IDA(nvme_ns_chr_minor_ida);
128 static dev_t nvme_ns_chr_devt;
129 static const struct class nvme_ns_chr_class = {
130 .name = "nvme-generic",
133 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
134 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
136 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
137 struct nvme_command *cmd);
139 void nvme_queue_scan(struct nvme_ctrl *ctrl)
142 * Only new queue scan work when admin and IO queues are both alive
144 if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE && ctrl->tagset)
145 queue_work(nvme_wq, &ctrl->scan_work);
149 * Use this function to proceed with scheduling reset_work for a controller
150 * that had previously been set to the resetting state. This is intended for
151 * code paths that can't be interrupted by other reset attempts. A hot removal
152 * may prevent this from succeeding.
154 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
156 if (nvme_ctrl_state(ctrl) != NVME_CTRL_RESETTING)
158 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
162 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
164 static void nvme_failfast_work(struct work_struct *work)
166 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
167 struct nvme_ctrl, failfast_work);
169 if (nvme_ctrl_state(ctrl) != NVME_CTRL_CONNECTING)
172 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
173 dev_info(ctrl->device, "failfast expired\n");
174 nvme_kick_requeue_lists(ctrl);
177 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
179 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
182 schedule_delayed_work(&ctrl->failfast_work,
183 ctrl->opts->fast_io_fail_tmo * HZ);
186 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
191 cancel_delayed_work_sync(&ctrl->failfast_work);
192 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
196 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
198 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
200 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
204 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
206 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
210 ret = nvme_reset_ctrl(ctrl);
212 flush_work(&ctrl->reset_work);
213 if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE)
220 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
222 dev_info(ctrl->device,
223 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
225 flush_work(&ctrl->reset_work);
226 nvme_stop_ctrl(ctrl);
227 nvme_remove_namespaces(ctrl);
228 ctrl->ops->delete_ctrl(ctrl);
229 nvme_uninit_ctrl(ctrl);
232 static void nvme_delete_ctrl_work(struct work_struct *work)
234 struct nvme_ctrl *ctrl =
235 container_of(work, struct nvme_ctrl, delete_work);
237 nvme_do_delete_ctrl(ctrl);
240 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
242 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
244 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
248 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
250 void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
253 * Keep a reference until nvme_do_delete_ctrl() complete,
254 * since ->delete_ctrl can free the controller.
257 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
258 nvme_do_delete_ctrl(ctrl);
262 static blk_status_t nvme_error_status(u16 status)
264 switch (status & 0x7ff) {
265 case NVME_SC_SUCCESS:
267 case NVME_SC_CAP_EXCEEDED:
268 return BLK_STS_NOSPC;
269 case NVME_SC_LBA_RANGE:
270 case NVME_SC_CMD_INTERRUPTED:
271 case NVME_SC_NS_NOT_READY:
272 return BLK_STS_TARGET;
273 case NVME_SC_BAD_ATTRIBUTES:
274 case NVME_SC_ONCS_NOT_SUPPORTED:
275 case NVME_SC_INVALID_OPCODE:
276 case NVME_SC_INVALID_FIELD:
277 case NVME_SC_INVALID_NS:
278 return BLK_STS_NOTSUPP;
279 case NVME_SC_WRITE_FAULT:
280 case NVME_SC_READ_ERROR:
281 case NVME_SC_UNWRITTEN_BLOCK:
282 case NVME_SC_ACCESS_DENIED:
283 case NVME_SC_READ_ONLY:
284 case NVME_SC_COMPARE_FAILED:
285 return BLK_STS_MEDIUM;
286 case NVME_SC_GUARD_CHECK:
287 case NVME_SC_APPTAG_CHECK:
288 case NVME_SC_REFTAG_CHECK:
289 case NVME_SC_INVALID_PI:
290 return BLK_STS_PROTECTION;
291 case NVME_SC_RESERVATION_CONFLICT:
292 return BLK_STS_RESV_CONFLICT;
293 case NVME_SC_HOST_PATH_ERROR:
294 return BLK_STS_TRANSPORT;
295 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
296 return BLK_STS_ZONE_ACTIVE_RESOURCE;
297 case NVME_SC_ZONE_TOO_MANY_OPEN:
298 return BLK_STS_ZONE_OPEN_RESOURCE;
300 return BLK_STS_IOERR;
304 static void nvme_retry_req(struct request *req)
306 unsigned long delay = 0;
309 /* The mask and shift result must be <= 3 */
310 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
312 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
314 nvme_req(req)->retries++;
315 blk_mq_requeue_request(req, false);
316 blk_mq_delay_kick_requeue_list(req->q, delay);
319 static void nvme_log_error(struct request *req)
321 struct nvme_ns *ns = req->q->queuedata;
322 struct nvme_request *nr = nvme_req(req);
325 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %u blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
326 ns->disk ? ns->disk->disk_name : "?",
327 nvme_get_opcode_str(nr->cmd->common.opcode),
328 nr->cmd->common.opcode,
329 nvme_sect_to_lba(ns->head, blk_rq_pos(req)),
330 blk_rq_bytes(req) >> ns->head->lba_shift,
331 nvme_get_error_status_str(nr->status),
332 nr->status >> 8 & 7, /* Status Code Type */
333 nr->status & 0xff, /* Status Code */
334 nr->status & NVME_SC_MORE ? "MORE " : "",
335 nr->status & NVME_SC_DNR ? "DNR " : "");
339 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
340 dev_name(nr->ctrl->device),
341 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
342 nr->cmd->common.opcode,
343 nvme_get_error_status_str(nr->status),
344 nr->status >> 8 & 7, /* Status Code Type */
345 nr->status & 0xff, /* Status Code */
346 nr->status & NVME_SC_MORE ? "MORE " : "",
347 nr->status & NVME_SC_DNR ? "DNR " : "");
350 static void nvme_log_err_passthru(struct request *req)
352 struct nvme_ns *ns = req->q->queuedata;
353 struct nvme_request *nr = nvme_req(req);
355 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s"
356 "cdw10=0x%x cdw11=0x%x cdw12=0x%x cdw13=0x%x cdw14=0x%x cdw15=0x%x\n",
357 ns ? ns->disk->disk_name : dev_name(nr->ctrl->device),
358 ns ? nvme_get_opcode_str(nr->cmd->common.opcode) :
359 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
360 nr->cmd->common.opcode,
361 nvme_get_error_status_str(nr->status),
362 nr->status >> 8 & 7, /* Status Code Type */
363 nr->status & 0xff, /* Status Code */
364 nr->status & NVME_SC_MORE ? "MORE " : "",
365 nr->status & NVME_SC_DNR ? "DNR " : "",
366 nr->cmd->common.cdw10,
367 nr->cmd->common.cdw11,
368 nr->cmd->common.cdw12,
369 nr->cmd->common.cdw13,
370 nr->cmd->common.cdw14,
371 nr->cmd->common.cdw14);
374 enum nvme_disposition {
381 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
383 if (likely(nvme_req(req)->status == 0))
386 if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
389 if (blk_noretry_request(req) ||
390 (nvme_req(req)->status & NVME_SC_DNR) ||
391 nvme_req(req)->retries >= nvme_max_retries)
394 if (req->cmd_flags & REQ_NVME_MPATH) {
395 if (nvme_is_path_error(nvme_req(req)->status) ||
396 blk_queue_dying(req->q))
399 if (blk_queue_dying(req->q))
406 static inline void nvme_end_req_zoned(struct request *req)
408 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
409 req_op(req) == REQ_OP_ZONE_APPEND) {
410 struct nvme_ns *ns = req->q->queuedata;
412 req->__sector = nvme_lba_to_sect(ns->head,
413 le64_to_cpu(nvme_req(req)->result.u64));
417 static inline void nvme_end_req(struct request *req)
419 blk_status_t status = nvme_error_status(nvme_req(req)->status);
421 if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET))) {
422 if (blk_rq_is_passthrough(req))
423 nvme_log_err_passthru(req);
427 nvme_end_req_zoned(req);
428 nvme_trace_bio_complete(req);
429 if (req->cmd_flags & REQ_NVME_MPATH)
430 nvme_mpath_end_request(req);
431 blk_mq_end_request(req, status);
434 void nvme_complete_rq(struct request *req)
436 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
438 trace_nvme_complete_rq(req);
439 nvme_cleanup_cmd(req);
442 * Completions of long-running commands should not be able to
443 * defer sending of periodic keep alives, since the controller
444 * may have completed processing such commands a long time ago
445 * (arbitrarily close to command submission time).
446 * req->deadline - req->timeout is the command submission time
450 req->deadline - req->timeout >= ctrl->ka_last_check_time)
451 ctrl->comp_seen = true;
453 switch (nvme_decide_disposition(req)) {
461 nvme_failover_req(req);
464 #ifdef CONFIG_NVME_HOST_AUTH
465 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
473 EXPORT_SYMBOL_GPL(nvme_complete_rq);
475 void nvme_complete_batch_req(struct request *req)
477 trace_nvme_complete_rq(req);
478 nvme_cleanup_cmd(req);
479 nvme_end_req_zoned(req);
481 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
484 * Called to unwind from ->queue_rq on a failed command submission so that the
485 * multipathing code gets called to potentially failover to another path.
486 * The caller needs to unwind all transport specific resource allocations and
487 * must return propagate the return value.
489 blk_status_t nvme_host_path_error(struct request *req)
491 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
492 blk_mq_set_request_complete(req);
493 nvme_complete_rq(req);
496 EXPORT_SYMBOL_GPL(nvme_host_path_error);
498 bool nvme_cancel_request(struct request *req, void *data)
500 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
501 "Cancelling I/O %d", req->tag);
503 /* don't abort one completed or idle request */
504 if (blk_mq_rq_state(req) != MQ_RQ_IN_FLIGHT)
507 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
508 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
509 blk_mq_complete_request(req);
512 EXPORT_SYMBOL_GPL(nvme_cancel_request);
514 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
517 blk_mq_tagset_busy_iter(ctrl->tagset,
518 nvme_cancel_request, ctrl);
519 blk_mq_tagset_wait_completed_request(ctrl->tagset);
522 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
524 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
526 if (ctrl->admin_tagset) {
527 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
528 nvme_cancel_request, ctrl);
529 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
532 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
534 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
535 enum nvme_ctrl_state new_state)
537 enum nvme_ctrl_state old_state;
539 bool changed = false;
541 spin_lock_irqsave(&ctrl->lock, flags);
543 old_state = nvme_ctrl_state(ctrl);
548 case NVME_CTRL_RESETTING:
549 case NVME_CTRL_CONNECTING:
556 case NVME_CTRL_RESETTING:
566 case NVME_CTRL_CONNECTING:
569 case NVME_CTRL_RESETTING:
576 case NVME_CTRL_DELETING:
579 case NVME_CTRL_RESETTING:
580 case NVME_CTRL_CONNECTING:
587 case NVME_CTRL_DELETING_NOIO:
589 case NVME_CTRL_DELETING:
599 case NVME_CTRL_DELETING:
611 WRITE_ONCE(ctrl->state, new_state);
612 wake_up_all(&ctrl->state_wq);
615 spin_unlock_irqrestore(&ctrl->lock, flags);
619 if (new_state == NVME_CTRL_LIVE) {
620 if (old_state == NVME_CTRL_CONNECTING)
621 nvme_stop_failfast_work(ctrl);
622 nvme_kick_requeue_lists(ctrl);
623 } else if (new_state == NVME_CTRL_CONNECTING &&
624 old_state == NVME_CTRL_RESETTING) {
625 nvme_start_failfast_work(ctrl);
629 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
632 * Returns true for sink states that can't ever transition back to live.
634 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
636 switch (nvme_ctrl_state(ctrl)) {
639 case NVME_CTRL_RESETTING:
640 case NVME_CTRL_CONNECTING:
642 case NVME_CTRL_DELETING:
643 case NVME_CTRL_DELETING_NOIO:
647 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
653 * Waits for the controller state to be resetting, or returns false if it is
654 * not possible to ever transition to that state.
656 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
658 wait_event(ctrl->state_wq,
659 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
660 nvme_state_terminal(ctrl));
661 return nvme_ctrl_state(ctrl) == NVME_CTRL_RESETTING;
663 EXPORT_SYMBOL_GPL(nvme_wait_reset);
665 static void nvme_free_ns_head(struct kref *ref)
667 struct nvme_ns_head *head =
668 container_of(ref, struct nvme_ns_head, ref);
670 nvme_mpath_remove_disk(head);
671 ida_free(&head->subsys->ns_ida, head->instance);
672 cleanup_srcu_struct(&head->srcu);
673 nvme_put_subsystem(head->subsys);
677 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
679 return kref_get_unless_zero(&head->ref);
682 void nvme_put_ns_head(struct nvme_ns_head *head)
684 kref_put(&head->ref, nvme_free_ns_head);
687 static void nvme_free_ns(struct kref *kref)
689 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
692 nvme_put_ns_head(ns->head);
693 nvme_put_ctrl(ns->ctrl);
697 static inline bool nvme_get_ns(struct nvme_ns *ns)
699 return kref_get_unless_zero(&ns->kref);
702 void nvme_put_ns(struct nvme_ns *ns)
704 kref_put(&ns->kref, nvme_free_ns);
706 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
708 static inline void nvme_clear_nvme_request(struct request *req)
710 nvme_req(req)->status = 0;
711 nvme_req(req)->retries = 0;
712 nvme_req(req)->flags = 0;
713 req->rq_flags |= RQF_DONTPREP;
716 /* initialize a passthrough request */
717 void nvme_init_request(struct request *req, struct nvme_command *cmd)
719 struct nvme_request *nr = nvme_req(req);
720 bool logging_enabled;
722 if (req->q->queuedata) {
723 struct nvme_ns *ns = req->q->disk->private_data;
725 logging_enabled = ns->passthru_err_log_enabled;
726 req->timeout = NVME_IO_TIMEOUT;
727 } else { /* no queuedata implies admin queue */
728 logging_enabled = nr->ctrl->passthru_err_log_enabled;
729 req->timeout = NVME_ADMIN_TIMEOUT;
732 if (!logging_enabled)
733 req->rq_flags |= RQF_QUIET;
735 /* passthru commands should let the driver set the SGL flags */
736 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
738 req->cmd_flags |= REQ_FAILFAST_DRIVER;
739 if (req->mq_hctx->type == HCTX_TYPE_POLL)
740 req->cmd_flags |= REQ_POLLED;
741 nvme_clear_nvme_request(req);
742 memcpy(nr->cmd, cmd, sizeof(*cmd));
744 EXPORT_SYMBOL_GPL(nvme_init_request);
747 * For something we're not in a state to send to the device the default action
748 * is to busy it and retry it after the controller state is recovered. However,
749 * if the controller is deleting or if anything is marked for failfast or
750 * nvme multipath it is immediately failed.
752 * Note: commands used to initialize the controller will be marked for failfast.
753 * Note: nvme cli/ioctl commands are marked for failfast.
755 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
758 enum nvme_ctrl_state state = nvme_ctrl_state(ctrl);
760 if (state != NVME_CTRL_DELETING_NOIO &&
761 state != NVME_CTRL_DELETING &&
762 state != NVME_CTRL_DEAD &&
763 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
764 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
765 return BLK_STS_RESOURCE;
766 return nvme_host_path_error(rq);
768 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
770 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
771 bool queue_live, enum nvme_ctrl_state state)
773 struct nvme_request *req = nvme_req(rq);
776 * currently we have a problem sending passthru commands
777 * on the admin_q if the controller is not LIVE because we can't
778 * make sure that they are going out after the admin connect,
779 * controller enable and/or other commands in the initialization
780 * sequence. until the controller will be LIVE, fail with
781 * BLK_STS_RESOURCE so that they will be rescheduled.
783 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
786 if (ctrl->ops->flags & NVME_F_FABRICS) {
788 * Only allow commands on a live queue, except for the connect
789 * command, which is require to set the queue live in the
790 * appropinquate states.
793 case NVME_CTRL_CONNECTING:
794 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
795 (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
796 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
797 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
809 EXPORT_SYMBOL_GPL(__nvme_check_ready);
811 static inline void nvme_setup_flush(struct nvme_ns *ns,
812 struct nvme_command *cmnd)
814 memset(cmnd, 0, sizeof(*cmnd));
815 cmnd->common.opcode = nvme_cmd_flush;
816 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
819 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
820 struct nvme_command *cmnd)
822 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
823 struct nvme_dsm_range *range;
827 * Some devices do not consider the DSM 'Number of Ranges' field when
828 * determining how much data to DMA. Always allocate memory for maximum
829 * number of segments to prevent device reading beyond end of buffer.
831 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
833 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
836 * If we fail allocation our range, fallback to the controller
837 * discard page. If that's also busy, it's safe to return
838 * busy, as we know we can make progress once that's freed.
840 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
841 return BLK_STS_RESOURCE;
843 range = page_address(ns->ctrl->discard_page);
846 if (queue_max_discard_segments(req->q) == 1) {
847 u64 slba = nvme_sect_to_lba(ns->head, blk_rq_pos(req));
848 u32 nlb = blk_rq_sectors(req) >> (ns->head->lba_shift - 9);
850 range[0].cattr = cpu_to_le32(0);
851 range[0].nlb = cpu_to_le32(nlb);
852 range[0].slba = cpu_to_le64(slba);
855 __rq_for_each_bio(bio, req) {
856 u64 slba = nvme_sect_to_lba(ns->head,
857 bio->bi_iter.bi_sector);
858 u32 nlb = bio->bi_iter.bi_size >> ns->head->lba_shift;
861 range[n].cattr = cpu_to_le32(0);
862 range[n].nlb = cpu_to_le32(nlb);
863 range[n].slba = cpu_to_le64(slba);
869 if (WARN_ON_ONCE(n != segments)) {
870 if (virt_to_page(range) == ns->ctrl->discard_page)
871 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
874 return BLK_STS_IOERR;
877 memset(cmnd, 0, sizeof(*cmnd));
878 cmnd->dsm.opcode = nvme_cmd_dsm;
879 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
880 cmnd->dsm.nr = cpu_to_le32(segments - 1);
881 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
883 bvec_set_virt(&req->special_vec, range, alloc_size);
884 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
889 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
895 /* both rw and write zeroes share the same reftag format */
896 switch (ns->head->guard_type) {
897 case NVME_NVM_NS_16B_GUARD:
898 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
900 case NVME_NVM_NS_64B_GUARD:
901 ref48 = ext_pi_ref_tag(req);
902 lower = lower_32_bits(ref48);
903 upper = upper_32_bits(ref48);
905 cmnd->rw.reftag = cpu_to_le32(lower);
906 cmnd->rw.cdw3 = cpu_to_le32(upper);
913 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
914 struct request *req, struct nvme_command *cmnd)
916 memset(cmnd, 0, sizeof(*cmnd));
918 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
919 return nvme_setup_discard(ns, req, cmnd);
921 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
922 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
923 cmnd->write_zeroes.slba =
924 cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
925 cmnd->write_zeroes.length =
926 cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
928 if (!(req->cmd_flags & REQ_NOUNMAP) &&
929 (ns->head->features & NVME_NS_DEAC))
930 cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
932 if (nvme_ns_has_pi(ns->head)) {
933 cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
935 switch (ns->head->pi_type) {
936 case NVME_NS_DPS_PI_TYPE1:
937 case NVME_NS_DPS_PI_TYPE2:
938 nvme_set_ref_tag(ns, cmnd, req);
946 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
947 struct request *req, struct nvme_command *cmnd,
953 if (req->cmd_flags & REQ_FUA)
954 control |= NVME_RW_FUA;
955 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
956 control |= NVME_RW_LR;
958 if (req->cmd_flags & REQ_RAHEAD)
959 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
961 cmnd->rw.opcode = op;
963 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
966 cmnd->rw.metadata = 0;
968 cpu_to_le64(nvme_sect_to_lba(ns->head, blk_rq_pos(req)));
970 cpu_to_le16((blk_rq_bytes(req) >> ns->head->lba_shift) - 1);
973 cmnd->rw.appmask = 0;
977 * If formated with metadata, the block layer always provides a
978 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
979 * we enable the PRACT bit for protection information or set the
980 * namespace capacity to zero to prevent any I/O.
982 if (!blk_integrity_rq(req)) {
983 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns->head)))
984 return BLK_STS_NOTSUPP;
985 control |= NVME_RW_PRINFO_PRACT;
988 switch (ns->head->pi_type) {
989 case NVME_NS_DPS_PI_TYPE3:
990 control |= NVME_RW_PRINFO_PRCHK_GUARD;
992 case NVME_NS_DPS_PI_TYPE1:
993 case NVME_NS_DPS_PI_TYPE2:
994 control |= NVME_RW_PRINFO_PRCHK_GUARD |
995 NVME_RW_PRINFO_PRCHK_REF;
996 if (op == nvme_cmd_zone_append)
997 control |= NVME_RW_APPEND_PIREMAP;
998 nvme_set_ref_tag(ns, cmnd, req);
1003 cmnd->rw.control = cpu_to_le16(control);
1004 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
1008 void nvme_cleanup_cmd(struct request *req)
1010 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
1011 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
1013 if (req->special_vec.bv_page == ctrl->discard_page)
1014 clear_bit_unlock(0, &ctrl->discard_page_busy);
1016 kfree(bvec_virt(&req->special_vec));
1019 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
1021 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
1023 struct nvme_command *cmd = nvme_req(req)->cmd;
1024 blk_status_t ret = BLK_STS_OK;
1026 if (!(req->rq_flags & RQF_DONTPREP))
1027 nvme_clear_nvme_request(req);
1029 switch (req_op(req)) {
1031 case REQ_OP_DRV_OUT:
1032 /* these are setup prior to execution in nvme_init_request() */
1035 nvme_setup_flush(ns, cmd);
1037 case REQ_OP_ZONE_RESET_ALL:
1038 case REQ_OP_ZONE_RESET:
1039 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
1041 case REQ_OP_ZONE_OPEN:
1042 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
1044 case REQ_OP_ZONE_CLOSE:
1045 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
1047 case REQ_OP_ZONE_FINISH:
1048 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
1050 case REQ_OP_WRITE_ZEROES:
1051 ret = nvme_setup_write_zeroes(ns, req, cmd);
1053 case REQ_OP_DISCARD:
1054 ret = nvme_setup_discard(ns, req, cmd);
1057 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
1060 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
1062 case REQ_OP_ZONE_APPEND:
1063 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
1067 return BLK_STS_IOERR;
1070 cmd->common.command_id = nvme_cid(req);
1071 trace_nvme_setup_cmd(req, cmd);
1074 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1079 * >0: nvme controller's cqe status response
1080 * <0: kernel error in lieu of controller response
1082 int nvme_execute_rq(struct request *rq, bool at_head)
1084 blk_status_t status;
1086 status = blk_execute_rq(rq, at_head);
1087 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1089 if (nvme_req(rq)->status)
1090 return nvme_req(rq)->status;
1091 return blk_status_to_errno(status);
1093 EXPORT_SYMBOL_NS_GPL(nvme_execute_rq, NVME_TARGET_PASSTHRU);
1096 * Returns 0 on success. If the result is negative, it's a Linux error code;
1097 * if the result is positive, it's an NVM Express status code
1099 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1100 union nvme_result *result, void *buffer, unsigned bufflen,
1101 int qid, nvme_submit_flags_t flags)
1103 struct request *req;
1105 blk_mq_req_flags_t blk_flags = 0;
1107 if (flags & NVME_SUBMIT_NOWAIT)
1108 blk_flags |= BLK_MQ_REQ_NOWAIT;
1109 if (flags & NVME_SUBMIT_RESERVED)
1110 blk_flags |= BLK_MQ_REQ_RESERVED;
1111 if (qid == NVME_QID_ANY)
1112 req = blk_mq_alloc_request(q, nvme_req_op(cmd), blk_flags);
1114 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), blk_flags,
1118 return PTR_ERR(req);
1119 nvme_init_request(req, cmd);
1120 if (flags & NVME_SUBMIT_RETRY)
1121 req->cmd_flags &= ~REQ_FAILFAST_DRIVER;
1123 if (buffer && bufflen) {
1124 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1129 ret = nvme_execute_rq(req, flags & NVME_SUBMIT_AT_HEAD);
1130 if (result && ret >= 0)
1131 *result = nvme_req(req)->result;
1133 blk_mq_free_request(req);
1136 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1138 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1139 void *buffer, unsigned bufflen)
1141 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1144 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1146 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1151 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1152 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1153 dev_warn_once(ctrl->device,
1154 "IO command:%02x has unusual effects:%08x\n",
1158 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1159 * which would deadlock when done on an I/O command. Note that
1160 * We already warn about an unusual effect above.
1162 effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1164 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1169 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1171 u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1173 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1176 * For simplicity, IO to all namespaces is quiesced even if the command
1177 * effects say only one namespace is affected.
1179 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1180 mutex_lock(&ctrl->scan_lock);
1181 mutex_lock(&ctrl->subsys->lock);
1182 nvme_mpath_start_freeze(ctrl->subsys);
1183 nvme_mpath_wait_freeze(ctrl->subsys);
1184 nvme_start_freeze(ctrl);
1185 nvme_wait_freeze(ctrl);
1189 EXPORT_SYMBOL_NS_GPL(nvme_passthru_start, NVME_TARGET_PASSTHRU);
1191 void nvme_passthru_end(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u32 effects,
1192 struct nvme_command *cmd, int status)
1194 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1195 nvme_unfreeze(ctrl);
1196 nvme_mpath_unfreeze(ctrl->subsys);
1197 mutex_unlock(&ctrl->subsys->lock);
1198 mutex_unlock(&ctrl->scan_lock);
1200 if (effects & NVME_CMD_EFFECTS_CCC) {
1201 if (!test_and_set_bit(NVME_CTRL_DIRTY_CAPABILITY,
1203 dev_info(ctrl->device,
1204 "controller capabilities changed, reset may be required to take effect.\n");
1207 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1208 nvme_queue_scan(ctrl);
1209 flush_work(&ctrl->scan_work);
1214 switch (cmd->common.opcode) {
1215 case nvme_admin_set_features:
1216 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1217 case NVME_FEAT_KATO:
1219 * Keep alive commands interval on the host should be
1220 * updated when KATO is modified by Set Features
1224 nvme_update_keep_alive(ctrl, cmd);
1234 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1237 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1239 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1240 * accounting for transport roundtrip times [..].
1242 static unsigned long nvme_keep_alive_work_period(struct nvme_ctrl *ctrl)
1244 unsigned long delay = ctrl->kato * HZ / 2;
1247 * When using Traffic Based Keep Alive, we need to run
1248 * nvme_keep_alive_work at twice the normal frequency, as one
1249 * command completion can postpone sending a keep alive command
1250 * by up to twice the delay between runs.
1252 if (ctrl->ctratt & NVME_CTRL_ATTR_TBKAS)
1257 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1259 unsigned long now = jiffies;
1260 unsigned long delay = nvme_keep_alive_work_period(ctrl);
1261 unsigned long ka_next_check_tm = ctrl->ka_last_check_time + delay;
1263 if (time_after(now, ka_next_check_tm))
1266 delay = ka_next_check_tm - now;
1268 queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1271 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1272 blk_status_t status)
1274 struct nvme_ctrl *ctrl = rq->end_io_data;
1275 unsigned long flags;
1276 bool startka = false;
1277 unsigned long rtt = jiffies - (rq->deadline - rq->timeout);
1278 unsigned long delay = nvme_keep_alive_work_period(ctrl);
1281 * Subtract off the keepalive RTT so nvme_keep_alive_work runs
1282 * at the desired frequency.
1287 dev_warn(ctrl->device, "long keepalive RTT (%u ms)\n",
1288 jiffies_to_msecs(rtt));
1292 blk_mq_free_request(rq);
1295 dev_err(ctrl->device,
1296 "failed nvme_keep_alive_end_io error=%d\n",
1298 return RQ_END_IO_NONE;
1301 ctrl->ka_last_check_time = jiffies;
1302 ctrl->comp_seen = false;
1303 spin_lock_irqsave(&ctrl->lock, flags);
1304 if (ctrl->state == NVME_CTRL_LIVE ||
1305 ctrl->state == NVME_CTRL_CONNECTING)
1307 spin_unlock_irqrestore(&ctrl->lock, flags);
1309 queue_delayed_work(nvme_wq, &ctrl->ka_work, delay);
1310 return RQ_END_IO_NONE;
1313 static void nvme_keep_alive_work(struct work_struct *work)
1315 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1316 struct nvme_ctrl, ka_work);
1317 bool comp_seen = ctrl->comp_seen;
1320 ctrl->ka_last_check_time = jiffies;
1322 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1323 dev_dbg(ctrl->device,
1324 "reschedule traffic based keep-alive timer\n");
1325 ctrl->comp_seen = false;
1326 nvme_queue_keep_alive_work(ctrl);
1330 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1331 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1333 /* allocation failure, reset the controller */
1334 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1335 nvme_reset_ctrl(ctrl);
1338 nvme_init_request(rq, &ctrl->ka_cmd);
1340 rq->timeout = ctrl->kato * HZ;
1341 rq->end_io = nvme_keep_alive_end_io;
1342 rq->end_io_data = ctrl;
1343 blk_execute_rq_nowait(rq, false);
1346 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1348 if (unlikely(ctrl->kato == 0))
1351 nvme_queue_keep_alive_work(ctrl);
1354 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1356 if (unlikely(ctrl->kato == 0))
1359 cancel_delayed_work_sync(&ctrl->ka_work);
1361 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1363 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1364 struct nvme_command *cmd)
1366 unsigned int new_kato =
1367 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1369 dev_info(ctrl->device,
1370 "keep alive interval updated from %u ms to %u ms\n",
1371 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1373 nvme_stop_keep_alive(ctrl);
1374 ctrl->kato = new_kato;
1375 nvme_start_keep_alive(ctrl);
1379 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1380 * flag, thus sending any new CNS opcodes has a big chance of not working.
1381 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1382 * (but not for any later version).
1384 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1386 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1387 return ctrl->vs < NVME_VS(1, 2, 0);
1388 return ctrl->vs < NVME_VS(1, 1, 0);
1391 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1393 struct nvme_command c = { };
1396 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1397 c.identify.opcode = nvme_admin_identify;
1398 c.identify.cns = NVME_ID_CNS_CTRL;
1400 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1404 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1405 sizeof(struct nvme_id_ctrl));
1413 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1414 struct nvme_ns_id_desc *cur, bool *csi_seen)
1416 const char *warn_str = "ctrl returned bogus length:";
1419 switch (cur->nidt) {
1420 case NVME_NIDT_EUI64:
1421 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1422 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1423 warn_str, cur->nidl);
1426 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1427 return NVME_NIDT_EUI64_LEN;
1428 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1429 return NVME_NIDT_EUI64_LEN;
1430 case NVME_NIDT_NGUID:
1431 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1432 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1433 warn_str, cur->nidl);
1436 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1437 return NVME_NIDT_NGUID_LEN;
1438 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1439 return NVME_NIDT_NGUID_LEN;
1440 case NVME_NIDT_UUID:
1441 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1442 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1443 warn_str, cur->nidl);
1446 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1447 return NVME_NIDT_UUID_LEN;
1448 uuid_copy(&ids->uuid, data + sizeof(*cur));
1449 return NVME_NIDT_UUID_LEN;
1451 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1452 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1453 warn_str, cur->nidl);
1456 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1458 return NVME_NIDT_CSI_LEN;
1460 /* Skip unknown types */
1465 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1466 struct nvme_ns_info *info)
1468 struct nvme_command c = { };
1469 bool csi_seen = false;
1470 int status, pos, len;
1473 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1475 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1478 c.identify.opcode = nvme_admin_identify;
1479 c.identify.nsid = cpu_to_le32(info->nsid);
1480 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1482 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1486 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1487 NVME_IDENTIFY_DATA_SIZE);
1489 dev_warn(ctrl->device,
1490 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1491 info->nsid, status);
1495 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1496 struct nvme_ns_id_desc *cur = data + pos;
1501 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1505 len += sizeof(*cur);
1508 if (nvme_multi_css(ctrl) && !csi_seen) {
1509 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1519 int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1520 struct nvme_id_ns **id)
1522 struct nvme_command c = { };
1525 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1526 c.identify.opcode = nvme_admin_identify;
1527 c.identify.nsid = cpu_to_le32(nsid);
1528 c.identify.cns = NVME_ID_CNS_NS;
1530 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1534 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1536 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1543 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1544 struct nvme_ns_info *info)
1546 struct nvme_ns_ids *ids = &info->ids;
1547 struct nvme_id_ns *id;
1550 ret = nvme_identify_ns(ctrl, info->nsid, &id);
1554 if (id->ncap == 0) {
1555 /* namespace not allocated or attached */
1556 info->is_removed = true;
1561 info->anagrpid = id->anagrpid;
1562 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1563 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1564 info->is_ready = true;
1565 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1566 dev_info(ctrl->device,
1567 "Ignoring bogus Namespace Identifiers\n");
1569 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1570 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1571 memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1572 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1573 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1574 memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1582 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1583 struct nvme_ns_info *info)
1585 struct nvme_id_ns_cs_indep *id;
1586 struct nvme_command c = {
1587 .identify.opcode = nvme_admin_identify,
1588 .identify.nsid = cpu_to_le32(info->nsid),
1589 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1593 id = kmalloc(sizeof(*id), GFP_KERNEL);
1597 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1599 info->anagrpid = id->anagrpid;
1600 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1601 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1602 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1608 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1609 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1611 union nvme_result res = { 0 };
1612 struct nvme_command c = { };
1615 c.features.opcode = op;
1616 c.features.fid = cpu_to_le32(fid);
1617 c.features.dword11 = cpu_to_le32(dword11);
1619 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1620 buffer, buflen, NVME_QID_ANY, 0);
1621 if (ret >= 0 && result)
1622 *result = le32_to_cpu(res.u32);
1626 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1627 unsigned int dword11, void *buffer, size_t buflen,
1630 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1633 EXPORT_SYMBOL_GPL(nvme_set_features);
1635 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1636 unsigned int dword11, void *buffer, size_t buflen,
1639 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1642 EXPORT_SYMBOL_GPL(nvme_get_features);
1644 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1646 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1648 int status, nr_io_queues;
1650 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1656 * Degraded controllers might return an error when setting the queue
1657 * count. We still want to be able to bring them online and offer
1658 * access to the admin queue, as that might be only way to fix them up.
1661 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1664 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1665 *count = min(*count, nr_io_queues);
1670 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1672 #define NVME_AEN_SUPPORTED \
1673 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1674 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1676 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1678 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1681 if (!supported_aens)
1684 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1687 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1690 queue_work(nvme_wq, &ctrl->async_event_work);
1693 static int nvme_ns_open(struct nvme_ns *ns)
1696 /* should never be called due to GENHD_FL_HIDDEN */
1697 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1699 if (!nvme_get_ns(ns))
1701 if (!try_module_get(ns->ctrl->ops->module))
1712 static void nvme_ns_release(struct nvme_ns *ns)
1715 module_put(ns->ctrl->ops->module);
1719 static int nvme_open(struct gendisk *disk, blk_mode_t mode)
1721 return nvme_ns_open(disk->private_data);
1724 static void nvme_release(struct gendisk *disk)
1726 nvme_ns_release(disk->private_data);
1729 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1731 /* some standard values */
1732 geo->heads = 1 << 6;
1733 geo->sectors = 1 << 5;
1734 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1738 static bool nvme_init_integrity(struct gendisk *disk, struct nvme_ns_head *head)
1740 struct blk_integrity integrity = { };
1742 blk_integrity_unregister(disk);
1748 * PI can always be supported as we can ask the controller to simply
1749 * insert/strip it, which is not possible for other kinds of metadata.
1751 if (!IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) ||
1752 !(head->features & NVME_NS_METADATA_SUPPORTED))
1753 return nvme_ns_has_pi(head);
1755 switch (head->pi_type) {
1756 case NVME_NS_DPS_PI_TYPE3:
1757 switch (head->guard_type) {
1758 case NVME_NVM_NS_16B_GUARD:
1759 integrity.profile = &t10_pi_type3_crc;
1760 integrity.tag_size = sizeof(u16) + sizeof(u32);
1761 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1763 case NVME_NVM_NS_64B_GUARD:
1764 integrity.profile = &ext_pi_type3_crc64;
1765 integrity.tag_size = sizeof(u16) + 6;
1766 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1769 integrity.profile = NULL;
1773 case NVME_NS_DPS_PI_TYPE1:
1774 case NVME_NS_DPS_PI_TYPE2:
1775 switch (head->guard_type) {
1776 case NVME_NVM_NS_16B_GUARD:
1777 integrity.profile = &t10_pi_type1_crc;
1778 integrity.tag_size = sizeof(u16);
1779 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1781 case NVME_NVM_NS_64B_GUARD:
1782 integrity.profile = &ext_pi_type1_crc64;
1783 integrity.tag_size = sizeof(u16);
1784 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1787 integrity.profile = NULL;
1792 integrity.profile = NULL;
1796 integrity.tuple_size = head->ms;
1797 integrity.pi_offset = head->pi_offset;
1798 blk_integrity_register(disk, &integrity);
1802 static void nvme_config_discard(struct nvme_ns *ns, struct queue_limits *lim)
1804 struct nvme_ctrl *ctrl = ns->ctrl;
1806 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1807 NVME_DSM_MAX_RANGES);
1809 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns->head, UINT_MAX))
1810 lim->max_hw_discard_sectors =
1811 nvme_lba_to_sect(ns->head, ctrl->dmrsl);
1812 else if (ctrl->oncs & NVME_CTRL_ONCS_DSM)
1813 lim->max_hw_discard_sectors = UINT_MAX;
1815 lim->max_hw_discard_sectors = 0;
1817 lim->discard_granularity = lim->logical_block_size;
1820 lim->max_discard_segments = ctrl->dmrl;
1822 lim->max_discard_segments = NVME_DSM_MAX_RANGES;
1825 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1827 return uuid_equal(&a->uuid, &b->uuid) &&
1828 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1829 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1833 static int nvme_identify_ns_nvm(struct nvme_ctrl *ctrl, unsigned int nsid,
1834 struct nvme_id_ns_nvm **nvmp)
1836 struct nvme_command c = {
1837 .identify.opcode = nvme_admin_identify,
1838 .identify.nsid = cpu_to_le32(nsid),
1839 .identify.cns = NVME_ID_CNS_CS_NS,
1840 .identify.csi = NVME_CSI_NVM,
1842 struct nvme_id_ns_nvm *nvm;
1845 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1849 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, nvm, sizeof(*nvm));
1857 static void nvme_configure_pi_elbas(struct nvme_ns_head *head,
1858 struct nvme_id_ns *id, struct nvme_id_ns_nvm *nvm)
1860 u32 elbaf = le32_to_cpu(nvm->elbaf[nvme_lbaf_index(id->flbas)]);
1862 /* no support for storage tag formats right now */
1863 if (nvme_elbaf_sts(elbaf))
1866 head->guard_type = nvme_elbaf_guard_type(elbaf);
1867 switch (head->guard_type) {
1868 case NVME_NVM_NS_64B_GUARD:
1869 head->pi_size = sizeof(struct crc64_pi_tuple);
1871 case NVME_NVM_NS_16B_GUARD:
1872 head->pi_size = sizeof(struct t10_pi_tuple);
1879 static void nvme_configure_metadata(struct nvme_ctrl *ctrl,
1880 struct nvme_ns_head *head, struct nvme_id_ns *id,
1881 struct nvme_id_ns_nvm *nvm)
1883 head->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1886 head->pi_offset = 0;
1887 head->ms = le16_to_cpu(id->lbaf[nvme_lbaf_index(id->flbas)].ms);
1888 if (!head->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1891 if (nvm && (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1892 nvme_configure_pi_elbas(head, id, nvm);
1894 head->pi_size = sizeof(struct t10_pi_tuple);
1895 head->guard_type = NVME_NVM_NS_16B_GUARD;
1898 if (head->pi_size && head->ms >= head->pi_size)
1899 head->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1900 if (!(id->dps & NVME_NS_DPS_PI_FIRST))
1901 head->pi_offset = head->ms - head->pi_size;
1903 if (ctrl->ops->flags & NVME_F_FABRICS) {
1905 * The NVMe over Fabrics specification only supports metadata as
1906 * part of the extended data LBA. We rely on HCA/HBA support to
1907 * remap the separate metadata buffer from the block layer.
1909 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1912 head->features |= NVME_NS_EXT_LBAS;
1915 * The current fabrics transport drivers support namespace
1916 * metadata formats only if nvme_ns_has_pi() returns true.
1917 * Suppress support for all other formats so the namespace will
1918 * have a 0 capacity and not be usable through the block stack.
1920 * Note, this check will need to be modified if any drivers
1921 * gain the ability to use other metadata formats.
1923 if (ctrl->max_integrity_segments && nvme_ns_has_pi(head))
1924 head->features |= NVME_NS_METADATA_SUPPORTED;
1927 * For PCIe controllers, we can't easily remap the separate
1928 * metadata buffer from the block layer and thus require a
1929 * separate metadata buffer for block layer metadata/PI support.
1930 * We allow extended LBAs for the passthrough interface, though.
1932 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1933 head->features |= NVME_NS_EXT_LBAS;
1935 head->features |= NVME_NS_METADATA_SUPPORTED;
1939 static u32 nvme_max_drv_segments(struct nvme_ctrl *ctrl)
1941 return ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> SECTOR_SHIFT) + 1;
1944 static void nvme_set_ctrl_limits(struct nvme_ctrl *ctrl,
1945 struct queue_limits *lim)
1947 lim->max_hw_sectors = ctrl->max_hw_sectors;
1948 lim->max_segments = min_t(u32, USHRT_MAX,
1949 min_not_zero(nvme_max_drv_segments(ctrl), ctrl->max_segments));
1950 lim->max_integrity_segments = ctrl->max_integrity_segments;
1951 lim->virt_boundary_mask = NVME_CTRL_PAGE_SIZE - 1;
1952 lim->max_segment_size = UINT_MAX;
1953 lim->dma_alignment = 3;
1956 static bool nvme_update_disk_info(struct nvme_ns *ns, struct nvme_id_ns *id,
1957 struct queue_limits *lim)
1959 struct nvme_ns_head *head = ns->head;
1960 u32 bs = 1U << head->lba_shift;
1961 u32 atomic_bs, phys_bs, io_opt = 0;
1965 * The block layer can't support LBA sizes larger than the page size
1966 * or smaller than a sector size yet, so catch this early and don't
1969 if (head->lba_shift > PAGE_SHIFT || head->lba_shift < SECTOR_SHIFT) {
1974 atomic_bs = phys_bs = bs;
1975 if (id->nabo == 0) {
1977 * Bit 1 indicates whether NAWUPF is defined for this namespace
1978 * and whether it should be used instead of AWUPF. If NAWUPF ==
1979 * 0 then AWUPF must be used instead.
1981 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1982 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1984 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1987 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1988 /* NPWG = Namespace Preferred Write Granularity */
1989 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1990 /* NOWS = Namespace Optimal Write Size */
1991 io_opt = bs * (1 + le16_to_cpu(id->nows));
1995 * Linux filesystems assume writing a single physical block is
1996 * an atomic operation. Hence limit the physical block size to the
1997 * value of the Atomic Write Unit Power Fail parameter.
1999 lim->logical_block_size = bs;
2000 lim->physical_block_size = min(phys_bs, atomic_bs);
2001 lim->io_min = phys_bs;
2002 lim->io_opt = io_opt;
2003 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
2004 lim->max_write_zeroes_sectors = UINT_MAX;
2006 lim->max_write_zeroes_sectors = ns->ctrl->max_zeroes_sectors;
2010 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
2012 return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
2015 static inline bool nvme_first_scan(struct gendisk *disk)
2017 /* nvme_alloc_ns() scans the disk prior to adding it */
2018 return !disk_live(disk);
2021 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id,
2022 struct queue_limits *lim)
2024 struct nvme_ctrl *ctrl = ns->ctrl;
2027 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
2028 is_power_of_2(ctrl->max_hw_sectors))
2029 iob = ctrl->max_hw_sectors;
2031 iob = nvme_lba_to_sect(ns->head, le16_to_cpu(id->noiob));
2036 if (!is_power_of_2(iob)) {
2037 if (nvme_first_scan(ns->disk))
2038 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
2039 ns->disk->disk_name, iob);
2043 if (blk_queue_is_zoned(ns->disk->queue)) {
2044 if (nvme_first_scan(ns->disk))
2045 pr_warn("%s: ignoring zoned namespace IO boundary\n",
2046 ns->disk->disk_name);
2050 lim->chunk_sectors = iob;
2053 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
2054 struct nvme_ns_info *info)
2056 struct queue_limits lim;
2059 blk_mq_freeze_queue(ns->disk->queue);
2060 lim = queue_limits_start_update(ns->disk->queue);
2061 nvme_set_ctrl_limits(ns->ctrl, &lim);
2062 ret = queue_limits_commit_update(ns->disk->queue, &lim);
2063 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2064 blk_mq_unfreeze_queue(ns->disk->queue);
2066 /* Hide the block-interface for these devices */
2072 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2073 struct nvme_ns_info *info)
2075 bool vwc = ns->ctrl->vwc & NVME_CTRL_VWC_PRESENT;
2076 struct queue_limits lim;
2077 struct nvme_id_ns_nvm *nvm = NULL;
2078 struct nvme_id_ns *id;
2083 ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2087 if (id->ncap == 0) {
2088 /* namespace not allocated or attached */
2089 info->is_removed = true;
2094 if (ns->ctrl->ctratt & NVME_CTRL_ATTR_ELBAS) {
2095 ret = nvme_identify_ns_nvm(ns->ctrl, info->nsid, &nvm);
2100 blk_mq_freeze_queue(ns->disk->queue);
2101 lbaf = nvme_lbaf_index(id->flbas);
2102 ns->head->lba_shift = id->lbaf[lbaf].ds;
2103 ns->head->nuse = le64_to_cpu(id->nuse);
2104 capacity = nvme_lba_to_sect(ns->head, le64_to_cpu(id->nsze));
2106 lim = queue_limits_start_update(ns->disk->queue);
2107 nvme_set_ctrl_limits(ns->ctrl, &lim);
2108 nvme_configure_metadata(ns->ctrl, ns->head, id, nvm);
2109 nvme_set_chunk_sectors(ns, id, &lim);
2110 if (!nvme_update_disk_info(ns, id, &lim))
2112 nvme_config_discard(ns, &lim);
2113 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
2114 ns->head->ids.csi == NVME_CSI_ZNS) {
2115 ret = nvme_update_zone_info(ns, lbaf, &lim);
2117 blk_mq_unfreeze_queue(ns->disk->queue);
2121 ret = queue_limits_commit_update(ns->disk->queue, &lim);
2123 blk_mq_unfreeze_queue(ns->disk->queue);
2128 * Register a metadata profile for PI, or the plain non-integrity NVMe
2129 * metadata masquerading as Type 0 if supported, otherwise reject block
2130 * I/O to namespaces with metadata except when the namespace supports
2131 * PI, as it can strip/insert in that case.
2133 if (!nvme_init_integrity(ns->disk, ns->head))
2136 set_capacity_and_notify(ns->disk, capacity);
2139 * Only set the DEAC bit if the device guarantees that reads from
2140 * deallocated data return zeroes. While the DEAC bit does not
2141 * require that, it must be a no-op if reads from deallocated data
2142 * do not return zeroes.
2144 if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2145 ns->head->features |= NVME_NS_DEAC;
2146 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2147 blk_queue_write_cache(ns->disk->queue, vwc, vwc);
2148 set_bit(NVME_NS_READY, &ns->flags);
2149 blk_mq_unfreeze_queue(ns->disk->queue);
2151 if (blk_queue_is_zoned(ns->queue)) {
2152 ret = blk_revalidate_disk_zones(ns->disk, NULL);
2153 if (ret && !nvme_first_scan(ns->disk))
2164 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2166 bool unsupported = false;
2169 switch (info->ids.csi) {
2171 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2172 dev_info(ns->ctrl->device,
2173 "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2175 ret = nvme_update_ns_info_generic(ns, info);
2178 ret = nvme_update_ns_info_block(ns, info);
2181 ret = nvme_update_ns_info_block(ns, info);
2184 dev_info(ns->ctrl->device,
2185 "block device for nsid %u not supported (csi %u)\n",
2186 info->nsid, info->ids.csi);
2187 ret = nvme_update_ns_info_generic(ns, info);
2192 * If probing fails due an unsupported feature, hide the block device,
2193 * but still allow other access.
2195 if (ret == -ENODEV) {
2196 ns->disk->flags |= GENHD_FL_HIDDEN;
2197 set_bit(NVME_NS_READY, &ns->flags);
2202 if (!ret && nvme_ns_head_multipath(ns->head)) {
2203 struct queue_limits lim;
2205 blk_mq_freeze_queue(ns->head->disk->queue);
2207 ns->head->disk->flags |= GENHD_FL_HIDDEN;
2209 nvme_init_integrity(ns->head->disk, ns->head);
2210 set_capacity_and_notify(ns->head->disk, get_capacity(ns->disk));
2211 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2212 nvme_mpath_revalidate_paths(ns);
2214 lim = queue_limits_start_update(ns->head->disk->queue);
2215 queue_limits_stack_bdev(&lim, ns->disk->part0, 0,
2216 ns->head->disk->disk_name);
2217 ret = queue_limits_commit_update(ns->head->disk->queue, &lim);
2218 blk_mq_unfreeze_queue(ns->head->disk->queue);
2224 #ifdef CONFIG_BLK_SED_OPAL
2225 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2228 struct nvme_ctrl *ctrl = data;
2229 struct nvme_command cmd = { };
2232 cmd.common.opcode = nvme_admin_security_send;
2234 cmd.common.opcode = nvme_admin_security_recv;
2235 cmd.common.nsid = 0;
2236 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2237 cmd.common.cdw11 = cpu_to_le32(len);
2239 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2240 NVME_QID_ANY, NVME_SUBMIT_AT_HEAD);
2243 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2245 if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2246 if (!ctrl->opal_dev)
2247 ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2248 else if (was_suspended)
2249 opal_unlock_from_suspend(ctrl->opal_dev);
2251 free_opal_dev(ctrl->opal_dev);
2252 ctrl->opal_dev = NULL;
2256 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2259 #endif /* CONFIG_BLK_SED_OPAL */
2261 #ifdef CONFIG_BLK_DEV_ZONED
2262 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2263 unsigned int nr_zones, report_zones_cb cb, void *data)
2265 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2269 #define nvme_report_zones NULL
2270 #endif /* CONFIG_BLK_DEV_ZONED */
2272 const struct block_device_operations nvme_bdev_ops = {
2273 .owner = THIS_MODULE,
2274 .ioctl = nvme_ioctl,
2275 .compat_ioctl = blkdev_compat_ptr_ioctl,
2277 .release = nvme_release,
2278 .getgeo = nvme_getgeo,
2279 .report_zones = nvme_report_zones,
2280 .pr_ops = &nvme_pr_ops,
2283 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2284 u32 timeout, const char *op)
2286 unsigned long timeout_jiffies = jiffies + timeout * HZ;
2290 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2293 if ((csts & mask) == val)
2296 usleep_range(1000, 2000);
2297 if (fatal_signal_pending(current))
2299 if (time_after(jiffies, timeout_jiffies)) {
2300 dev_err(ctrl->device,
2301 "Device not ready; aborting %s, CSTS=0x%x\n",
2310 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2314 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2316 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2318 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2320 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2325 return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2326 NVME_CSTS_SHST_CMPLT,
2327 ctrl->shutdown_timeout, "shutdown");
2329 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2330 msleep(NVME_QUIRK_DELAY_AMOUNT);
2331 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2332 (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2334 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2336 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2338 unsigned dev_page_min;
2342 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2344 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2347 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2349 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2350 dev_err(ctrl->device,
2351 "Minimum device page size %u too large for host (%u)\n",
2352 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2356 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2357 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2359 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2361 if (ctrl->cap & NVME_CAP_CRMS_CRWMS && ctrl->cap & NVME_CAP_CRMS_CRIMS)
2362 ctrl->ctrl_config |= NVME_CC_CRIME;
2364 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2365 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2366 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2367 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2371 /* Flush write to device (required if transport is PCI) */
2372 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2376 /* CAP value may change after initial CC write */
2377 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2381 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2382 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2383 u32 crto, ready_timeout;
2385 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2387 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2393 * CRTO should always be greater or equal to CAP.TO, but some
2394 * devices are known to get this wrong. Use the larger of the
2397 if (ctrl->ctrl_config & NVME_CC_CRIME)
2398 ready_timeout = NVME_CRTO_CRIMT(crto);
2400 ready_timeout = NVME_CRTO_CRWMT(crto);
2402 if (ready_timeout < timeout)
2403 dev_warn_once(ctrl->device, "bad crto:%x cap:%llx\n",
2406 timeout = ready_timeout;
2409 ctrl->ctrl_config |= NVME_CC_ENABLE;
2410 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2413 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2414 (timeout + 1) / 2, "initialisation");
2416 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2418 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2423 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2426 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2427 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2430 dev_warn_once(ctrl->device,
2431 "could not set timestamp (%d)\n", ret);
2435 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2437 struct nvme_feat_host_behavior *host;
2438 u8 acre = 0, lbafee = 0;
2441 /* Don't bother enabling the feature if retry delay is not reported */
2443 acre = NVME_ENABLE_ACRE;
2444 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2445 lbafee = NVME_ENABLE_LBAFEE;
2447 if (!acre && !lbafee)
2450 host = kzalloc(sizeof(*host), GFP_KERNEL);
2455 host->lbafee = lbafee;
2456 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2457 host, sizeof(*host), NULL);
2463 * The function checks whether the given total (exlat + enlat) latency of
2464 * a power state allows the latter to be used as an APST transition target.
2465 * It does so by comparing the latency to the primary and secondary latency
2466 * tolerances defined by module params. If there's a match, the corresponding
2467 * timeout value is returned and the matching tolerance index (1 or 2) is
2470 static bool nvme_apst_get_transition_time(u64 total_latency,
2471 u64 *transition_time, unsigned *last_index)
2473 if (total_latency <= apst_primary_latency_tol_us) {
2474 if (*last_index == 1)
2477 *transition_time = apst_primary_timeout_ms;
2480 if (apst_secondary_timeout_ms &&
2481 total_latency <= apst_secondary_latency_tol_us) {
2482 if (*last_index <= 2)
2485 *transition_time = apst_secondary_timeout_ms;
2492 * APST (Autonomous Power State Transition) lets us program a table of power
2493 * state transitions that the controller will perform automatically.
2495 * Depending on module params, one of the two supported techniques will be used:
2497 * - If the parameters provide explicit timeouts and tolerances, they will be
2498 * used to build a table with up to 2 non-operational states to transition to.
2499 * The default parameter values were selected based on the values used by
2500 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2501 * regeneration of the APST table in the event of switching between external
2502 * and battery power, the timeouts and tolerances reflect a compromise
2503 * between values used by Microsoft for AC and battery scenarios.
2504 * - If not, we'll configure the table with a simple heuristic: we are willing
2505 * to spend at most 2% of the time transitioning between power states.
2506 * Therefore, when running in any given state, we will enter the next
2507 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2508 * microseconds, as long as that state's exit latency is under the requested
2511 * We will not autonomously enter any non-operational state for which the total
2512 * latency exceeds ps_max_latency_us.
2514 * Users can set ps_max_latency_us to zero to turn off APST.
2516 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2518 struct nvme_feat_auto_pst *table;
2525 unsigned last_lt_index = UINT_MAX;
2528 * If APST isn't supported or if we haven't been initialized yet,
2529 * then don't do anything.
2534 if (ctrl->npss > 31) {
2535 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2539 table = kzalloc(sizeof(*table), GFP_KERNEL);
2543 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2544 /* Turn off APST. */
2545 dev_dbg(ctrl->device, "APST disabled\n");
2550 * Walk through all states from lowest- to highest-power.
2551 * According to the spec, lower-numbered states use more power. NPSS,
2552 * despite the name, is the index of the lowest-power state, not the
2555 for (state = (int)ctrl->npss; state >= 0; state--) {
2556 u64 total_latency_us, exit_latency_us, transition_ms;
2559 table->entries[state] = target;
2562 * Don't allow transitions to the deepest state if it's quirked
2565 if (state == ctrl->npss &&
2566 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2570 * Is this state a useful non-operational state for higher-power
2571 * states to autonomously transition to?
2573 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2576 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2577 if (exit_latency_us > ctrl->ps_max_latency_us)
2580 total_latency_us = exit_latency_us +
2581 le32_to_cpu(ctrl->psd[state].entry_lat);
2584 * This state is good. It can be used as the APST idle target
2585 * for higher power states.
2587 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2588 if (!nvme_apst_get_transition_time(total_latency_us,
2589 &transition_ms, &last_lt_index))
2592 transition_ms = total_latency_us + 19;
2593 do_div(transition_ms, 20);
2594 if (transition_ms > (1 << 24) - 1)
2595 transition_ms = (1 << 24) - 1;
2598 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2601 if (total_latency_us > max_lat_us)
2602 max_lat_us = total_latency_us;
2606 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2608 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2609 max_ps, max_lat_us, (int)sizeof(*table), table);
2613 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2614 table, sizeof(*table), NULL);
2616 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2621 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2623 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2627 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2628 case PM_QOS_LATENCY_ANY:
2636 if (ctrl->ps_max_latency_us != latency) {
2637 ctrl->ps_max_latency_us = latency;
2638 if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
2639 nvme_configure_apst(ctrl);
2643 struct nvme_core_quirk_entry {
2645 * NVMe model and firmware strings are padded with spaces. For
2646 * simplicity, strings in the quirk table are padded with NULLs
2652 unsigned long quirks;
2655 static const struct nvme_core_quirk_entry core_quirks[] = {
2658 * This Toshiba device seems to die using any APST states. See:
2659 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2662 .mn = "THNSF5256GPUK TOSHIBA",
2663 .quirks = NVME_QUIRK_NO_APST,
2667 * This LiteON CL1-3D*-Q11 firmware version has a race
2668 * condition associated with actions related to suspend to idle
2669 * LiteON has resolved the problem in future firmware
2673 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2677 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2678 * aborts I/O during any load, but more easily reproducible
2679 * with discards (fstrim).
2681 * The device is left in a state where it is also not possible
2682 * to use "nvme set-feature" to disable APST, but booting with
2683 * nvme_core.default_ps_max_latency=0 works.
2686 .mn = "KCD6XVUL6T40",
2687 .quirks = NVME_QUIRK_NO_APST,
2691 * The external Samsung X5 SSD fails initialization without a
2692 * delay before checking if it is ready and has a whole set of
2693 * other problems. To make this even more interesting, it
2694 * shares the PCI ID with internal Samsung 970 Evo Plus that
2695 * does not need or want these quirks.
2698 .mn = "Samsung Portable SSD X5",
2699 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2700 NVME_QUIRK_NO_DEEPEST_PS |
2701 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2705 /* match is null-terminated but idstr is space-padded. */
2706 static bool string_matches(const char *idstr, const char *match, size_t len)
2713 matchlen = strlen(match);
2714 WARN_ON_ONCE(matchlen > len);
2716 if (memcmp(idstr, match, matchlen))
2719 for (; matchlen < len; matchlen++)
2720 if (idstr[matchlen] != ' ')
2726 static bool quirk_matches(const struct nvme_id_ctrl *id,
2727 const struct nvme_core_quirk_entry *q)
2729 return q->vid == le16_to_cpu(id->vid) &&
2730 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2731 string_matches(id->fr, q->fr, sizeof(id->fr));
2734 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2735 struct nvme_id_ctrl *id)
2740 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2741 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2742 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2743 strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2747 if (ctrl->vs >= NVME_VS(1, 2, 1))
2748 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2752 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2753 * Base Specification 2.0. It is slightly different from the format
2754 * specified there due to historic reasons, and we can't change it now.
2756 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2757 "nqn.2014.08.org.nvmexpress:%04x%04x",
2758 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2759 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2760 off += sizeof(id->sn);
2761 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2762 off += sizeof(id->mn);
2763 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2766 static void nvme_release_subsystem(struct device *dev)
2768 struct nvme_subsystem *subsys =
2769 container_of(dev, struct nvme_subsystem, dev);
2771 if (subsys->instance >= 0)
2772 ida_free(&nvme_instance_ida, subsys->instance);
2776 static void nvme_destroy_subsystem(struct kref *ref)
2778 struct nvme_subsystem *subsys =
2779 container_of(ref, struct nvme_subsystem, ref);
2781 mutex_lock(&nvme_subsystems_lock);
2782 list_del(&subsys->entry);
2783 mutex_unlock(&nvme_subsystems_lock);
2785 ida_destroy(&subsys->ns_ida);
2786 device_del(&subsys->dev);
2787 put_device(&subsys->dev);
2790 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2792 kref_put(&subsys->ref, nvme_destroy_subsystem);
2795 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2797 struct nvme_subsystem *subsys;
2799 lockdep_assert_held(&nvme_subsystems_lock);
2802 * Fail matches for discovery subsystems. This results
2803 * in each discovery controller bound to a unique subsystem.
2804 * This avoids issues with validating controller values
2805 * that can only be true when there is a single unique subsystem.
2806 * There may be multiple and completely independent entities
2807 * that provide discovery controllers.
2809 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2812 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2813 if (strcmp(subsys->subnqn, subsysnqn))
2815 if (!kref_get_unless_zero(&subsys->ref))
2823 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2825 return ctrl->opts && ctrl->opts->discovery_nqn;
2828 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2829 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2831 struct nvme_ctrl *tmp;
2833 lockdep_assert_held(&nvme_subsystems_lock);
2835 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2836 if (nvme_state_terminal(tmp))
2839 if (tmp->cntlid == ctrl->cntlid) {
2840 dev_err(ctrl->device,
2841 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2842 ctrl->cntlid, dev_name(tmp->device),
2847 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2848 nvme_discovery_ctrl(ctrl))
2851 dev_err(ctrl->device,
2852 "Subsystem does not support multiple controllers\n");
2859 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2861 struct nvme_subsystem *subsys, *found;
2864 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2868 subsys->instance = -1;
2869 mutex_init(&subsys->lock);
2870 kref_init(&subsys->ref);
2871 INIT_LIST_HEAD(&subsys->ctrls);
2872 INIT_LIST_HEAD(&subsys->nsheads);
2873 nvme_init_subnqn(subsys, ctrl, id);
2874 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2875 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2876 subsys->vendor_id = le16_to_cpu(id->vid);
2877 subsys->cmic = id->cmic;
2879 /* Versions prior to 1.4 don't necessarily report a valid type */
2880 if (id->cntrltype == NVME_CTRL_DISC ||
2881 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2882 subsys->subtype = NVME_NQN_DISC;
2884 subsys->subtype = NVME_NQN_NVME;
2886 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2887 dev_err(ctrl->device,
2888 "Subsystem %s is not a discovery controller",
2893 subsys->awupf = le16_to_cpu(id->awupf);
2894 nvme_mpath_default_iopolicy(subsys);
2896 subsys->dev.class = &nvme_subsys_class;
2897 subsys->dev.release = nvme_release_subsystem;
2898 subsys->dev.groups = nvme_subsys_attrs_groups;
2899 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2900 device_initialize(&subsys->dev);
2902 mutex_lock(&nvme_subsystems_lock);
2903 found = __nvme_find_get_subsystem(subsys->subnqn);
2905 put_device(&subsys->dev);
2908 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2910 goto out_put_subsystem;
2913 ret = device_add(&subsys->dev);
2915 dev_err(ctrl->device,
2916 "failed to register subsystem device.\n");
2917 put_device(&subsys->dev);
2920 ida_init(&subsys->ns_ida);
2921 list_add_tail(&subsys->entry, &nvme_subsystems);
2924 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2925 dev_name(ctrl->device));
2927 dev_err(ctrl->device,
2928 "failed to create sysfs link from subsystem.\n");
2929 goto out_put_subsystem;
2933 subsys->instance = ctrl->instance;
2934 ctrl->subsys = subsys;
2935 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2936 mutex_unlock(&nvme_subsystems_lock);
2940 nvme_put_subsystem(subsys);
2942 mutex_unlock(&nvme_subsystems_lock);
2946 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2947 void *log, size_t size, u64 offset)
2949 struct nvme_command c = { };
2950 u32 dwlen = nvme_bytes_to_numd(size);
2952 c.get_log_page.opcode = nvme_admin_get_log_page;
2953 c.get_log_page.nsid = cpu_to_le32(nsid);
2954 c.get_log_page.lid = log_page;
2955 c.get_log_page.lsp = lsp;
2956 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2957 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2958 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2959 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2960 c.get_log_page.csi = csi;
2962 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2965 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2966 struct nvme_effects_log **log)
2968 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2974 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2978 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2979 cel, sizeof(*cel), 0);
2985 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2991 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2993 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2995 if (check_shl_overflow(1U, units + page_shift - 9, &val))
3000 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3002 struct nvme_command c = { };
3003 struct nvme_id_ctrl_nvm *id;
3007 * Even though NVMe spec explicitly states that MDTS is not applicable
3008 * to the write-zeroes, we are cautious and limit the size to the
3009 * controllers max_hw_sectors value, which is based on the MDTS field
3010 * and possibly other limiting factors.
3012 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3013 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3014 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3016 ctrl->max_zeroes_sectors = 0;
3018 if (ctrl->subsys->subtype != NVME_NQN_NVME ||
3019 nvme_ctrl_limited_cns(ctrl) ||
3020 test_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags))
3023 id = kzalloc(sizeof(*id), GFP_KERNEL);
3027 c.identify.opcode = nvme_admin_identify;
3028 c.identify.cns = NVME_ID_CNS_CS_CTRL;
3029 c.identify.csi = NVME_CSI_NVM;
3031 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3035 ctrl->dmrl = id->dmrl;
3036 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3038 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3042 set_bit(NVME_CTRL_SKIP_ID_CNS_CS, &ctrl->flags);
3047 static void nvme_init_known_nvm_effects(struct nvme_ctrl *ctrl)
3049 struct nvme_effects_log *log = ctrl->effects;
3051 log->acs[nvme_admin_format_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3052 NVME_CMD_EFFECTS_NCC |
3053 NVME_CMD_EFFECTS_CSE_MASK);
3054 log->acs[nvme_admin_sanitize_nvm] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC |
3055 NVME_CMD_EFFECTS_CSE_MASK);
3058 * The spec says the result of a security receive command depends on
3059 * the previous security send command. As such, many vendors log this
3060 * command as one to submitted only when no other commands to the same
3061 * namespace are outstanding. The intention is to tell the host to
3062 * prevent mixing security send and receive.
3064 * This driver can only enforce such exclusive access against IO
3065 * queues, though. We are not readily able to enforce such a rule for
3066 * two commands to the admin queue, which is the only queue that
3067 * matters for this command.
3069 * Rather than blindly freezing the IO queues for this effect that
3070 * doesn't even apply to IO, mask it off.
3072 log->acs[nvme_admin_security_recv] &= cpu_to_le32(~NVME_CMD_EFFECTS_CSE_MASK);
3074 log->iocs[nvme_cmd_write] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3075 log->iocs[nvme_cmd_write_zeroes] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3076 log->iocs[nvme_cmd_write_uncor] |= cpu_to_le32(NVME_CMD_EFFECTS_LBCC);
3079 static int nvme_init_effects(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3086 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3087 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3092 if (!ctrl->effects) {
3093 ctrl->effects = kzalloc(sizeof(*ctrl->effects), GFP_KERNEL);
3096 xa_store(&ctrl->cels, NVME_CSI_NVM, ctrl->effects, GFP_KERNEL);
3099 nvme_init_known_nvm_effects(ctrl);
3103 static int nvme_check_ctrl_fabric_info(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
3106 * In fabrics we need to verify the cntlid matches the
3109 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3110 dev_err(ctrl->device,
3111 "Mismatching cntlid: Connect %u vs Identify %u, rejecting\n",
3112 ctrl->cntlid, le16_to_cpu(id->cntlid));
3116 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3117 dev_err(ctrl->device,
3118 "keep-alive support is mandatory for fabrics\n");
3122 if (!nvme_discovery_ctrl(ctrl) && ctrl->ioccsz < 4) {
3123 dev_err(ctrl->device,
3124 "I/O queue command capsule supported size %d < 4\n",
3129 if (!nvme_discovery_ctrl(ctrl) && ctrl->iorcsz < 1) {
3130 dev_err(ctrl->device,
3131 "I/O queue response capsule supported size %d < 1\n",
3136 if (!ctrl->maxcmd) {
3137 dev_err(ctrl->device, "Maximum outstanding commands is 0\n");
3144 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3146 struct queue_limits lim;
3147 struct nvme_id_ctrl *id;
3149 bool prev_apst_enabled;
3152 ret = nvme_identify_ctrl(ctrl, &id);
3154 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3158 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3159 ctrl->cntlid = le16_to_cpu(id->cntlid);
3161 if (!ctrl->identified) {
3165 * Check for quirks. Quirk can depend on firmware version,
3166 * so, in principle, the set of quirks present can change
3167 * across a reset. As a possible future enhancement, we
3168 * could re-scan for quirks every time we reinitialize
3169 * the device, but we'd have to make sure that the driver
3170 * behaves intelligently if the quirks change.
3172 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3173 if (quirk_matches(id, &core_quirks[i]))
3174 ctrl->quirks |= core_quirks[i].quirks;
3177 ret = nvme_init_subsystem(ctrl, id);
3181 ret = nvme_init_effects(ctrl, id);
3185 memcpy(ctrl->subsys->firmware_rev, id->fr,
3186 sizeof(ctrl->subsys->firmware_rev));
3188 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3189 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3190 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3193 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3194 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3195 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3197 ctrl->oacs = le16_to_cpu(id->oacs);
3198 ctrl->oncs = le16_to_cpu(id->oncs);
3199 ctrl->mtfa = le16_to_cpu(id->mtfa);
3200 ctrl->oaes = le32_to_cpu(id->oaes);
3201 ctrl->wctemp = le16_to_cpu(id->wctemp);
3202 ctrl->cctemp = le16_to_cpu(id->cctemp);
3204 atomic_set(&ctrl->abort_limit, id->acl + 1);
3205 ctrl->vwc = id->vwc;
3207 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3209 max_hw_sectors = UINT_MAX;
3210 ctrl->max_hw_sectors =
3211 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3213 lim = queue_limits_start_update(ctrl->admin_q);
3214 nvme_set_ctrl_limits(ctrl, &lim);
3215 ret = queue_limits_commit_update(ctrl->admin_q, &lim);
3219 ctrl->sgls = le32_to_cpu(id->sgls);
3220 ctrl->kas = le16_to_cpu(id->kas);
3221 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3222 ctrl->ctratt = le32_to_cpu(id->ctratt);
3224 ctrl->cntrltype = id->cntrltype;
3225 ctrl->dctype = id->dctype;
3229 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3231 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3232 shutdown_timeout, 60);
3234 if (ctrl->shutdown_timeout != shutdown_timeout)
3235 dev_info(ctrl->device,
3236 "D3 entry latency set to %u seconds\n",
3237 ctrl->shutdown_timeout);
3239 ctrl->shutdown_timeout = shutdown_timeout;
3241 ctrl->npss = id->npss;
3242 ctrl->apsta = id->apsta;
3243 prev_apst_enabled = ctrl->apst_enabled;
3244 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3245 if (force_apst && id->apsta) {
3246 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3247 ctrl->apst_enabled = true;
3249 ctrl->apst_enabled = false;
3252 ctrl->apst_enabled = id->apsta;
3254 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3256 if (ctrl->ops->flags & NVME_F_FABRICS) {
3257 ctrl->icdoff = le16_to_cpu(id->icdoff);
3258 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3259 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3260 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3262 ret = nvme_check_ctrl_fabric_info(ctrl, id);
3266 ctrl->hmpre = le32_to_cpu(id->hmpre);
3267 ctrl->hmmin = le32_to_cpu(id->hmmin);
3268 ctrl->hmminds = le32_to_cpu(id->hmminds);
3269 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3272 ret = nvme_mpath_init_identify(ctrl, id);
3276 if (ctrl->apst_enabled && !prev_apst_enabled)
3277 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3278 else if (!ctrl->apst_enabled && prev_apst_enabled)
3279 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3287 * Initialize the cached copies of the Identify data and various controller
3288 * register in our nvme_ctrl structure. This should be called as soon as
3289 * the admin queue is fully up and running.
3291 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3295 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3297 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3301 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3303 if (ctrl->vs >= NVME_VS(1, 1, 0))
3304 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3306 ret = nvme_init_identify(ctrl);
3310 ret = nvme_configure_apst(ctrl);
3314 ret = nvme_configure_timestamp(ctrl);
3318 ret = nvme_configure_host_options(ctrl);
3322 nvme_configure_opal(ctrl, was_suspended);
3324 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3326 * Do not return errors unless we are in a controller reset,
3327 * the controller works perfectly fine without hwmon.
3329 ret = nvme_hwmon_init(ctrl);
3334 clear_bit(NVME_CTRL_DIRTY_CAPABILITY, &ctrl->flags);
3335 ctrl->identified = true;
3337 nvme_start_keep_alive(ctrl);
3341 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3343 static int nvme_dev_open(struct inode *inode, struct file *file)
3345 struct nvme_ctrl *ctrl =
3346 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3348 switch (nvme_ctrl_state(ctrl)) {
3349 case NVME_CTRL_LIVE:
3352 return -EWOULDBLOCK;
3355 nvme_get_ctrl(ctrl);
3356 if (!try_module_get(ctrl->ops->module)) {
3357 nvme_put_ctrl(ctrl);
3361 file->private_data = ctrl;
3365 static int nvme_dev_release(struct inode *inode, struct file *file)
3367 struct nvme_ctrl *ctrl =
3368 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3370 module_put(ctrl->ops->module);
3371 nvme_put_ctrl(ctrl);
3375 static const struct file_operations nvme_dev_fops = {
3376 .owner = THIS_MODULE,
3377 .open = nvme_dev_open,
3378 .release = nvme_dev_release,
3379 .unlocked_ioctl = nvme_dev_ioctl,
3380 .compat_ioctl = compat_ptr_ioctl,
3381 .uring_cmd = nvme_dev_uring_cmd,
3384 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3387 struct nvme_ns_head *h;
3389 lockdep_assert_held(&ctrl->subsys->lock);
3391 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3393 * Private namespaces can share NSIDs under some conditions.
3394 * In that case we can't use the same ns_head for namespaces
3395 * with the same NSID.
3397 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3399 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3406 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3407 struct nvme_ns_ids *ids)
3409 bool has_uuid = !uuid_is_null(&ids->uuid);
3410 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3411 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3412 struct nvme_ns_head *h;
3414 lockdep_assert_held(&subsys->lock);
3416 list_for_each_entry(h, &subsys->nsheads, entry) {
3417 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3420 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3423 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3430 static void nvme_cdev_rel(struct device *dev)
3432 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3435 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3437 cdev_device_del(cdev, cdev_device);
3438 put_device(cdev_device);
3441 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3442 const struct file_operations *fops, struct module *owner)
3446 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3449 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3450 cdev_device->class = &nvme_ns_chr_class;
3451 cdev_device->release = nvme_cdev_rel;
3452 device_initialize(cdev_device);
3453 cdev_init(cdev, fops);
3454 cdev->owner = owner;
3455 ret = cdev_device_add(cdev, cdev_device);
3457 put_device(cdev_device);
3462 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3464 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3467 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3469 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3473 static const struct file_operations nvme_ns_chr_fops = {
3474 .owner = THIS_MODULE,
3475 .open = nvme_ns_chr_open,
3476 .release = nvme_ns_chr_release,
3477 .unlocked_ioctl = nvme_ns_chr_ioctl,
3478 .compat_ioctl = compat_ptr_ioctl,
3479 .uring_cmd = nvme_ns_chr_uring_cmd,
3480 .uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
3483 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3487 ns->cdev_device.parent = ns->ctrl->device;
3488 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3489 ns->ctrl->instance, ns->head->instance);
3493 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3494 ns->ctrl->ops->module);
3497 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3498 struct nvme_ns_info *info)
3500 struct nvme_ns_head *head;
3501 size_t size = sizeof(*head);
3504 #ifdef CONFIG_NVME_MULTIPATH
3505 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3508 head = kzalloc(size, GFP_KERNEL);
3511 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3514 head->instance = ret;
3515 INIT_LIST_HEAD(&head->list);
3516 ret = init_srcu_struct(&head->srcu);
3518 goto out_ida_remove;
3519 head->subsys = ctrl->subsys;
3520 head->ns_id = info->nsid;
3521 head->ids = info->ids;
3522 head->shared = info->is_shared;
3523 ratelimit_state_init(&head->rs_nuse, 5 * HZ, 1);
3524 ratelimit_set_flags(&head->rs_nuse, RATELIMIT_MSG_ON_RELEASE);
3525 kref_init(&head->ref);
3527 if (head->ids.csi) {
3528 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3530 goto out_cleanup_srcu;
3532 head->effects = ctrl->effects;
3534 ret = nvme_mpath_alloc_disk(ctrl, head);
3536 goto out_cleanup_srcu;
3538 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3540 kref_get(&ctrl->subsys->ref);
3544 cleanup_srcu_struct(&head->srcu);
3546 ida_free(&ctrl->subsys->ns_ida, head->instance);
3551 ret = blk_status_to_errno(nvme_error_status(ret));
3552 return ERR_PTR(ret);
3555 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3556 struct nvme_ns_ids *ids)
3558 struct nvme_subsystem *s;
3562 * Note that this check is racy as we try to avoid holding the global
3563 * lock over the whole ns_head creation. But it is only intended as
3564 * a sanity check anyway.
3566 mutex_lock(&nvme_subsystems_lock);
3567 list_for_each_entry(s, &nvme_subsystems, entry) {
3570 mutex_lock(&s->lock);
3571 ret = nvme_subsys_check_duplicate_ids(s, ids);
3572 mutex_unlock(&s->lock);
3576 mutex_unlock(&nvme_subsystems_lock);
3581 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
3583 struct nvme_ctrl *ctrl = ns->ctrl;
3584 struct nvme_ns_head *head = NULL;
3587 ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
3590 * We've found two different namespaces on two different
3591 * subsystems that report the same ID. This is pretty nasty
3592 * for anything that actually requires unique device
3593 * identification. In the kernel we need this for multipathing,
3594 * and in user space the /dev/disk/by-id/ links rely on it.
3596 * If the device also claims to be multi-path capable back off
3597 * here now and refuse the probe the second device as this is a
3598 * recipe for data corruption. If not this is probably a
3599 * cheap consumer device if on the PCIe bus, so let the user
3600 * proceed and use the shiny toy, but warn that with changing
3601 * probing order (which due to our async probing could just be
3602 * device taking longer to startup) the other device could show
3605 nvme_print_device_info(ctrl);
3606 if ((ns->ctrl->ops->flags & NVME_F_FABRICS) || /* !PCIe */
3607 ((ns->ctrl->subsys->cmic & NVME_CTRL_CMIC_MULTI_CTRL) &&
3609 dev_err(ctrl->device,
3610 "ignoring nsid %d because of duplicate IDs\n",
3615 dev_err(ctrl->device,
3616 "clearing duplicate IDs for nsid %d\n", info->nsid);
3617 dev_err(ctrl->device,
3618 "use of /dev/disk/by-id/ may cause data corruption\n");
3619 memset(&info->ids.nguid, 0, sizeof(info->ids.nguid));
3620 memset(&info->ids.uuid, 0, sizeof(info->ids.uuid));
3621 memset(&info->ids.eui64, 0, sizeof(info->ids.eui64));
3622 ctrl->quirks |= NVME_QUIRK_BOGUS_NID;
3625 mutex_lock(&ctrl->subsys->lock);
3626 head = nvme_find_ns_head(ctrl, info->nsid);
3628 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
3630 dev_err(ctrl->device,
3631 "duplicate IDs in subsystem for nsid %d\n",
3635 head = nvme_alloc_ns_head(ctrl, info);
3637 ret = PTR_ERR(head);
3642 if (!info->is_shared || !head->shared) {
3643 dev_err(ctrl->device,
3644 "Duplicate unshared namespace %d\n",
3646 goto out_put_ns_head;
3648 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
3649 dev_err(ctrl->device,
3650 "IDs don't match for shared namespace %d\n",
3652 goto out_put_ns_head;
3656 dev_warn(ctrl->device,
3657 "Found shared namespace %d, but multipathing not supported.\n",
3659 dev_warn_once(ctrl->device,
3660 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3664 list_add_tail_rcu(&ns->siblings, &head->list);
3666 mutex_unlock(&ctrl->subsys->lock);
3670 nvme_put_ns_head(head);
3672 mutex_unlock(&ctrl->subsys->lock);
3676 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3678 struct nvme_ns *ns, *ret = NULL;
3680 down_read(&ctrl->namespaces_rwsem);
3681 list_for_each_entry(ns, &ctrl->namespaces, list) {
3682 if (ns->head->ns_id == nsid) {
3683 if (!nvme_get_ns(ns))
3688 if (ns->head->ns_id > nsid)
3691 up_read(&ctrl->namespaces_rwsem);
3694 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3697 * Add the namespace to the controller list while keeping the list ordered.
3699 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3701 struct nvme_ns *tmp;
3703 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3704 if (tmp->head->ns_id < ns->head->ns_id) {
3705 list_add(&ns->list, &tmp->list);
3709 list_add(&ns->list, &ns->ctrl->namespaces);
3712 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
3715 struct gendisk *disk;
3716 int node = ctrl->numa_node;
3718 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3722 disk = blk_mq_alloc_disk(ctrl->tagset, NULL, ns);
3725 disk->fops = &nvme_bdev_ops;
3726 disk->private_data = ns;
3729 ns->queue = disk->queue;
3730 ns->passthru_err_log_enabled = false;
3732 if (ctrl->opts && ctrl->opts->data_digest)
3733 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3735 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3736 if (ctrl->ops->supports_pci_p2pdma &&
3737 ctrl->ops->supports_pci_p2pdma(ctrl))
3738 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
3741 kref_init(&ns->kref);
3743 if (nvme_init_ns_head(ns, info))
3744 goto out_cleanup_disk;
3747 * If multipathing is enabled, the device name for all disks and not
3748 * just those that represent shared namespaces needs to be based on the
3749 * subsystem instance. Using the controller instance for private
3750 * namespaces could lead to naming collisions between shared and private
3751 * namespaces if they don't use a common numbering scheme.
3753 * If multipathing is not enabled, disk names must use the controller
3754 * instance as shared namespaces will show up as multiple block
3757 if (nvme_ns_head_multipath(ns->head)) {
3758 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
3759 ctrl->instance, ns->head->instance);
3760 disk->flags |= GENHD_FL_HIDDEN;
3761 } else if (multipath) {
3762 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
3763 ns->head->instance);
3765 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
3766 ns->head->instance);
3769 if (nvme_update_ns_info(ns, info))
3772 down_write(&ctrl->namespaces_rwsem);
3774 * Ensure that no namespaces are added to the ctrl list after the queues
3775 * are frozen, thereby avoiding a deadlock between scan and reset.
3777 if (test_bit(NVME_CTRL_FROZEN, &ctrl->flags)) {
3778 up_write(&ctrl->namespaces_rwsem);
3781 nvme_ns_add_to_ctrl_list(ns);
3782 up_write(&ctrl->namespaces_rwsem);
3783 nvme_get_ctrl(ctrl);
3785 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_attr_groups))
3786 goto out_cleanup_ns_from_list;
3788 if (!nvme_ns_head_multipath(ns->head))
3789 nvme_add_ns_cdev(ns);
3791 nvme_mpath_add_disk(ns, info->anagrpid);
3792 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
3795 * Set ns->disk->device->driver_data to ns so we can access
3796 * ns->logging_enabled in nvme_passthru_err_log_enabled_store() and
3797 * nvme_passthru_err_log_enabled_show().
3799 dev_set_drvdata(disk_to_dev(ns->disk), ns);
3803 out_cleanup_ns_from_list:
3804 nvme_put_ctrl(ctrl);
3805 down_write(&ctrl->namespaces_rwsem);
3806 list_del_init(&ns->list);
3807 up_write(&ctrl->namespaces_rwsem);
3809 mutex_lock(&ctrl->subsys->lock);
3810 list_del_rcu(&ns->siblings);
3811 if (list_empty(&ns->head->list))
3812 list_del_init(&ns->head->entry);
3813 mutex_unlock(&ctrl->subsys->lock);
3814 nvme_put_ns_head(ns->head);
3821 static void nvme_ns_remove(struct nvme_ns *ns)
3823 bool last_path = false;
3825 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
3828 clear_bit(NVME_NS_READY, &ns->flags);
3829 set_capacity(ns->disk, 0);
3830 nvme_fault_inject_fini(&ns->fault_inject);
3833 * Ensure that !NVME_NS_READY is seen by other threads to prevent
3834 * this ns going back into current_path.
3836 synchronize_srcu(&ns->head->srcu);
3838 /* wait for concurrent submissions */
3839 if (nvme_mpath_clear_current_path(ns))
3840 synchronize_srcu(&ns->head->srcu);
3842 mutex_lock(&ns->ctrl->subsys->lock);
3843 list_del_rcu(&ns->siblings);
3844 if (list_empty(&ns->head->list)) {
3845 list_del_init(&ns->head->entry);
3848 mutex_unlock(&ns->ctrl->subsys->lock);
3850 /* guarantee not available in head->list */
3851 synchronize_srcu(&ns->head->srcu);
3853 if (!nvme_ns_head_multipath(ns->head))
3854 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
3855 del_gendisk(ns->disk);
3857 down_write(&ns->ctrl->namespaces_rwsem);
3858 list_del_init(&ns->list);
3859 up_write(&ns->ctrl->namespaces_rwsem);
3862 nvme_mpath_shutdown_disk(ns->head);
3866 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
3868 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
3876 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
3878 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
3880 if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
3881 dev_err(ns->ctrl->device,
3882 "identifiers changed for nsid %d\n", ns->head->ns_id);
3886 ret = nvme_update_ns_info(ns, info);
3889 * Only remove the namespace if we got a fatal error back from the
3890 * device, otherwise ignore the error and just move on.
3892 * TODO: we should probably schedule a delayed retry here.
3894 if (ret > 0 && (ret & NVME_SC_DNR))
3898 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3900 struct nvme_ns_info info = { .nsid = nsid };
3904 if (nvme_identify_ns_descs(ctrl, &info))
3907 if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
3908 dev_warn(ctrl->device,
3909 "command set not reported for nsid: %d\n", nsid);
3914 * If available try to use the Command Set Idependent Identify Namespace
3915 * data structure to find all the generic information that is needed to
3916 * set up a namespace. If not fall back to the legacy version.
3918 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
3919 (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS))
3920 ret = nvme_ns_info_from_id_cs_indep(ctrl, &info);
3922 ret = nvme_ns_info_from_identify(ctrl, &info);
3924 if (info.is_removed)
3925 nvme_ns_remove_by_nsid(ctrl, nsid);
3928 * Ignore the namespace if it is not ready. We will get an AEN once it
3929 * becomes ready and restart the scan.
3931 if (ret || !info.is_ready)
3934 ns = nvme_find_get_ns(ctrl, nsid);
3936 nvme_validate_ns(ns, &info);
3939 nvme_alloc_ns(ctrl, &info);
3943 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
3946 struct nvme_ns *ns, *next;
3949 down_write(&ctrl->namespaces_rwsem);
3950 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
3951 if (ns->head->ns_id > nsid)
3952 list_move_tail(&ns->list, &rm_list);
3954 up_write(&ctrl->namespaces_rwsem);
3956 list_for_each_entry_safe(ns, next, &rm_list, list)
3961 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
3963 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
3968 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
3973 struct nvme_command cmd = {
3974 .identify.opcode = nvme_admin_identify,
3975 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
3976 .identify.nsid = cpu_to_le32(prev),
3979 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
3980 NVME_IDENTIFY_DATA_SIZE);
3982 dev_warn(ctrl->device,
3983 "Identify NS List failed (status=0x%x)\n", ret);
3987 for (i = 0; i < nr_entries; i++) {
3988 u32 nsid = le32_to_cpu(ns_list[i]);
3990 if (!nsid) /* end of the list? */
3992 nvme_scan_ns(ctrl, nsid);
3993 while (++prev < nsid)
3994 nvme_ns_remove_by_nsid(ctrl, prev);
3998 nvme_remove_invalid_namespaces(ctrl, prev);
4004 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4006 struct nvme_id_ctrl *id;
4009 if (nvme_identify_ctrl(ctrl, &id))
4011 nn = le32_to_cpu(id->nn);
4014 for (i = 1; i <= nn; i++)
4015 nvme_scan_ns(ctrl, i);
4017 nvme_remove_invalid_namespaces(ctrl, nn);
4020 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4022 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4026 log = kzalloc(log_size, GFP_KERNEL);
4031 * We need to read the log to clear the AEN, but we don't want to rely
4032 * on it for the changed namespace information as userspace could have
4033 * raced with us in reading the log page, which could cause us to miss
4036 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4037 NVME_CSI_NVM, log, log_size, 0);
4039 dev_warn(ctrl->device,
4040 "reading changed ns log failed: %d\n", error);
4045 static void nvme_scan_work(struct work_struct *work)
4047 struct nvme_ctrl *ctrl =
4048 container_of(work, struct nvme_ctrl, scan_work);
4051 /* No tagset on a live ctrl means IO queues could not created */
4052 if (nvme_ctrl_state(ctrl) != NVME_CTRL_LIVE || !ctrl->tagset)
4056 * Identify controller limits can change at controller reset due to
4057 * new firmware download, even though it is not common we cannot ignore
4058 * such scenario. Controller's non-mdts limits are reported in the unit
4059 * of logical blocks that is dependent on the format of attached
4060 * namespace. Hence re-read the limits at the time of ns allocation.
4062 ret = nvme_init_non_mdts_limits(ctrl);
4064 dev_warn(ctrl->device,
4065 "reading non-mdts-limits failed: %d\n", ret);
4069 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4070 dev_info(ctrl->device, "rescanning namespaces.\n");
4071 nvme_clear_changed_ns_log(ctrl);
4074 mutex_lock(&ctrl->scan_lock);
4075 if (nvme_ctrl_limited_cns(ctrl)) {
4076 nvme_scan_ns_sequential(ctrl);
4079 * Fall back to sequential scan if DNR is set to handle broken
4080 * devices which should support Identify NS List (as per the VS
4081 * they report) but don't actually support it.
4083 ret = nvme_scan_ns_list(ctrl);
4084 if (ret > 0 && ret & NVME_SC_DNR)
4085 nvme_scan_ns_sequential(ctrl);
4087 mutex_unlock(&ctrl->scan_lock);
4091 * This function iterates the namespace list unlocked to allow recovery from
4092 * controller failure. It is up to the caller to ensure the namespace list is
4093 * not modified by scan work while this function is executing.
4095 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4097 struct nvme_ns *ns, *next;
4101 * make sure to requeue I/O to all namespaces as these
4102 * might result from the scan itself and must complete
4103 * for the scan_work to make progress
4105 nvme_mpath_clear_ctrl_paths(ctrl);
4108 * Unquiesce io queues so any pending IO won't hang, especially
4109 * those submitted from scan work
4111 nvme_unquiesce_io_queues(ctrl);
4113 /* prevent racing with ns scanning */
4114 flush_work(&ctrl->scan_work);
4117 * The dead states indicates the controller was not gracefully
4118 * disconnected. In that case, we won't be able to flush any data while
4119 * removing the namespaces' disks; fail all the queues now to avoid
4120 * potentially having to clean up the failed sync later.
4122 if (nvme_ctrl_state(ctrl) == NVME_CTRL_DEAD)
4123 nvme_mark_namespaces_dead(ctrl);
4125 /* this is a no-op when called from the controller reset handler */
4126 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4128 down_write(&ctrl->namespaces_rwsem);
4129 list_splice_init(&ctrl->namespaces, &ns_list);
4130 up_write(&ctrl->namespaces_rwsem);
4132 list_for_each_entry_safe(ns, next, &ns_list, list)
4135 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4137 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4139 const struct nvme_ctrl *ctrl =
4140 container_of(dev, struct nvme_ctrl, ctrl_device);
4141 struct nvmf_ctrl_options *opts = ctrl->opts;
4144 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4149 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4153 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4154 opts->trsvcid ?: "none");
4158 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4159 opts->host_traddr ?: "none");
4163 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4164 opts->host_iface ?: "none");
4169 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4171 char *envp[2] = { envdata, NULL };
4173 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4176 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4178 char *envp[2] = { NULL, NULL };
4179 u32 aen_result = ctrl->aen_result;
4181 ctrl->aen_result = 0;
4185 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4188 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4192 static void nvme_async_event_work(struct work_struct *work)
4194 struct nvme_ctrl *ctrl =
4195 container_of(work, struct nvme_ctrl, async_event_work);
4197 nvme_aen_uevent(ctrl);
4200 * The transport drivers must guarantee AER submission here is safe by
4201 * flushing ctrl async_event_work after changing the controller state
4202 * from LIVE and before freeing the admin queue.
4204 if (nvme_ctrl_state(ctrl) == NVME_CTRL_LIVE)
4205 ctrl->ops->submit_async_event(ctrl);
4208 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4213 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4219 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4222 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4224 struct nvme_fw_slot_info_log *log;
4226 log = kmalloc(sizeof(*log), GFP_KERNEL);
4230 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4231 log, sizeof(*log), 0)) {
4232 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4236 if (log->afi & 0x70 || !(log->afi & 0x7)) {
4237 dev_info(ctrl->device,
4238 "Firmware is activated after next Controller Level Reset\n");
4242 memcpy(ctrl->subsys->firmware_rev, &log->frs[(log->afi & 0x7) - 1],
4243 sizeof(ctrl->subsys->firmware_rev));
4249 static void nvme_fw_act_work(struct work_struct *work)
4251 struct nvme_ctrl *ctrl = container_of(work,
4252 struct nvme_ctrl, fw_act_work);
4253 unsigned long fw_act_timeout;
4255 nvme_auth_stop(ctrl);
4258 fw_act_timeout = jiffies +
4259 msecs_to_jiffies(ctrl->mtfa * 100);
4261 fw_act_timeout = jiffies +
4262 msecs_to_jiffies(admin_timeout * 1000);
4264 nvme_quiesce_io_queues(ctrl);
4265 while (nvme_ctrl_pp_status(ctrl)) {
4266 if (time_after(jiffies, fw_act_timeout)) {
4267 dev_warn(ctrl->device,
4268 "Fw activation timeout, reset controller\n");
4269 nvme_try_sched_reset(ctrl);
4275 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4278 nvme_unquiesce_io_queues(ctrl);
4279 /* read FW slot information to clear the AER */
4280 nvme_get_fw_slot_info(ctrl);
4282 queue_work(nvme_wq, &ctrl->async_event_work);
4285 static u32 nvme_aer_type(u32 result)
4287 return result & 0x7;
4290 static u32 nvme_aer_subtype(u32 result)
4292 return (result & 0xff00) >> 8;
4295 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4297 u32 aer_notice_type = nvme_aer_subtype(result);
4298 bool requeue = true;
4300 switch (aer_notice_type) {
4301 case NVME_AER_NOTICE_NS_CHANGED:
4302 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4303 nvme_queue_scan(ctrl);
4305 case NVME_AER_NOTICE_FW_ACT_STARTING:
4307 * We are (ab)using the RESETTING state to prevent subsequent
4308 * recovery actions from interfering with the controller's
4309 * firmware activation.
4311 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4313 queue_work(nvme_wq, &ctrl->fw_act_work);
4316 #ifdef CONFIG_NVME_MULTIPATH
4317 case NVME_AER_NOTICE_ANA:
4318 if (!ctrl->ana_log_buf)
4320 queue_work(nvme_wq, &ctrl->ana_work);
4323 case NVME_AER_NOTICE_DISC_CHANGED:
4324 ctrl->aen_result = result;
4327 dev_warn(ctrl->device, "async event result %08x\n", result);
4332 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4334 dev_warn(ctrl->device, "resetting controller due to AER\n");
4335 nvme_reset_ctrl(ctrl);
4338 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4339 volatile union nvme_result *res)
4341 u32 result = le32_to_cpu(res->u32);
4342 u32 aer_type = nvme_aer_type(result);
4343 u32 aer_subtype = nvme_aer_subtype(result);
4344 bool requeue = true;
4346 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4349 trace_nvme_async_event(ctrl, result);
4351 case NVME_AER_NOTICE:
4352 requeue = nvme_handle_aen_notice(ctrl, result);
4354 case NVME_AER_ERROR:
4356 * For a persistent internal error, don't run async_event_work
4357 * to submit a new AER. The controller reset will do it.
4359 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4360 nvme_handle_aer_persistent_error(ctrl);
4364 case NVME_AER_SMART:
4367 ctrl->aen_result = result;
4374 queue_work(nvme_wq, &ctrl->async_event_work);
4376 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4378 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4379 const struct blk_mq_ops *ops, unsigned int cmd_size)
4381 struct queue_limits lim = {};
4384 memset(set, 0, sizeof(*set));
4386 set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4387 if (ctrl->ops->flags & NVME_F_FABRICS)
4388 /* Reserved for fabric connect and keep alive */
4389 set->reserved_tags = 2;
4390 set->numa_node = ctrl->numa_node;
4391 set->flags = BLK_MQ_F_NO_SCHED;
4392 if (ctrl->ops->flags & NVME_F_BLOCKING)
4393 set->flags |= BLK_MQ_F_BLOCKING;
4394 set->cmd_size = cmd_size;
4395 set->driver_data = ctrl;
4396 set->nr_hw_queues = 1;
4397 set->timeout = NVME_ADMIN_TIMEOUT;
4398 ret = blk_mq_alloc_tag_set(set);
4402 ctrl->admin_q = blk_mq_alloc_queue(set, &lim, NULL);
4403 if (IS_ERR(ctrl->admin_q)) {
4404 ret = PTR_ERR(ctrl->admin_q);
4405 goto out_free_tagset;
4408 if (ctrl->ops->flags & NVME_F_FABRICS) {
4409 ctrl->fabrics_q = blk_mq_alloc_queue(set, NULL, NULL);
4410 if (IS_ERR(ctrl->fabrics_q)) {
4411 ret = PTR_ERR(ctrl->fabrics_q);
4412 goto out_cleanup_admin_q;
4416 ctrl->admin_tagset = set;
4419 out_cleanup_admin_q:
4420 blk_mq_destroy_queue(ctrl->admin_q);
4421 blk_put_queue(ctrl->admin_q);
4423 blk_mq_free_tag_set(set);
4424 ctrl->admin_q = NULL;
4425 ctrl->fabrics_q = NULL;
4428 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4430 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4432 blk_mq_destroy_queue(ctrl->admin_q);
4433 blk_put_queue(ctrl->admin_q);
4434 if (ctrl->ops->flags & NVME_F_FABRICS) {
4435 blk_mq_destroy_queue(ctrl->fabrics_q);
4436 blk_put_queue(ctrl->fabrics_q);
4438 blk_mq_free_tag_set(ctrl->admin_tagset);
4440 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4442 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4443 const struct blk_mq_ops *ops, unsigned int nr_maps,
4444 unsigned int cmd_size)
4448 memset(set, 0, sizeof(*set));
4450 set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4452 * Some Apple controllers requires tags to be unique across admin and
4453 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4455 if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4456 set->reserved_tags = NVME_AQ_DEPTH;
4457 else if (ctrl->ops->flags & NVME_F_FABRICS)
4458 /* Reserved for fabric connect */
4459 set->reserved_tags = 1;
4460 set->numa_node = ctrl->numa_node;
4461 set->flags = BLK_MQ_F_SHOULD_MERGE;
4462 if (ctrl->ops->flags & NVME_F_BLOCKING)
4463 set->flags |= BLK_MQ_F_BLOCKING;
4464 set->cmd_size = cmd_size,
4465 set->driver_data = ctrl;
4466 set->nr_hw_queues = ctrl->queue_count - 1;
4467 set->timeout = NVME_IO_TIMEOUT;
4468 set->nr_maps = nr_maps;
4469 ret = blk_mq_alloc_tag_set(set);
4473 if (ctrl->ops->flags & NVME_F_FABRICS) {
4474 ctrl->connect_q = blk_mq_alloc_queue(set, NULL, NULL);
4475 if (IS_ERR(ctrl->connect_q)) {
4476 ret = PTR_ERR(ctrl->connect_q);
4477 goto out_free_tag_set;
4479 blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4487 blk_mq_free_tag_set(set);
4488 ctrl->connect_q = NULL;
4491 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4493 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4495 if (ctrl->ops->flags & NVME_F_FABRICS) {
4496 blk_mq_destroy_queue(ctrl->connect_q);
4497 blk_put_queue(ctrl->connect_q);
4499 blk_mq_free_tag_set(ctrl->tagset);
4501 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
4503 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4505 nvme_mpath_stop(ctrl);
4506 nvme_auth_stop(ctrl);
4507 nvme_stop_keep_alive(ctrl);
4508 nvme_stop_failfast_work(ctrl);
4509 flush_work(&ctrl->async_event_work);
4510 cancel_work_sync(&ctrl->fw_act_work);
4511 if (ctrl->ops->stop_ctrl)
4512 ctrl->ops->stop_ctrl(ctrl);
4514 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4516 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4518 nvme_enable_aen(ctrl);
4521 * persistent discovery controllers need to send indication to userspace
4522 * to re-read the discovery log page to learn about possible changes
4523 * that were missed. We identify persistent discovery controllers by
4524 * checking that they started once before, hence are reconnecting back.
4526 if (test_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
4527 nvme_discovery_ctrl(ctrl))
4528 nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
4530 if (ctrl->queue_count > 1) {
4531 nvme_queue_scan(ctrl);
4532 nvme_unquiesce_io_queues(ctrl);
4533 nvme_mpath_update(ctrl);
4536 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4537 set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags);
4539 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4541 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4543 nvme_hwmon_exit(ctrl);
4544 nvme_fault_inject_fini(&ctrl->fault_inject);
4545 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4546 cdev_device_del(&ctrl->cdev, ctrl->device);
4547 nvme_put_ctrl(ctrl);
4549 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4551 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4553 struct nvme_effects_log *cel;
4556 xa_for_each(&ctrl->cels, i, cel) {
4557 xa_erase(&ctrl->cels, i);
4561 xa_destroy(&ctrl->cels);
4564 static void nvme_free_ctrl(struct device *dev)
4566 struct nvme_ctrl *ctrl =
4567 container_of(dev, struct nvme_ctrl, ctrl_device);
4568 struct nvme_subsystem *subsys = ctrl->subsys;
4570 if (!subsys || ctrl->instance != subsys->instance)
4571 ida_free(&nvme_instance_ida, ctrl->instance);
4572 key_put(ctrl->tls_key);
4573 nvme_free_cels(ctrl);
4574 nvme_mpath_uninit(ctrl);
4575 nvme_auth_stop(ctrl);
4576 nvme_auth_free(ctrl);
4577 __free_page(ctrl->discard_page);
4578 free_opal_dev(ctrl->opal_dev);
4581 mutex_lock(&nvme_subsystems_lock);
4582 list_del(&ctrl->subsys_entry);
4583 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4584 mutex_unlock(&nvme_subsystems_lock);
4587 ctrl->ops->free_ctrl(ctrl);
4590 nvme_put_subsystem(subsys);
4594 * Initialize a NVMe controller structures. This needs to be called during
4595 * earliest initialization so that we have the initialized structured around
4598 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4599 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4603 WRITE_ONCE(ctrl->state, NVME_CTRL_NEW);
4604 ctrl->passthru_err_log_enabled = false;
4605 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4606 spin_lock_init(&ctrl->lock);
4607 mutex_init(&ctrl->scan_lock);
4608 INIT_LIST_HEAD(&ctrl->namespaces);
4609 xa_init(&ctrl->cels);
4610 init_rwsem(&ctrl->namespaces_rwsem);
4613 ctrl->quirks = quirks;
4614 ctrl->numa_node = NUMA_NO_NODE;
4615 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4616 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4617 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4618 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4619 init_waitqueue_head(&ctrl->state_wq);
4621 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4622 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4623 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4624 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4625 ctrl->ka_last_check_time = jiffies;
4627 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4629 ctrl->discard_page = alloc_page(GFP_KERNEL);
4630 if (!ctrl->discard_page) {
4635 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4638 ctrl->instance = ret;
4640 device_initialize(&ctrl->ctrl_device);
4641 ctrl->device = &ctrl->ctrl_device;
4642 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4644 ctrl->device->class = &nvme_class;
4645 ctrl->device->parent = ctrl->dev;
4646 if (ops->dev_attr_groups)
4647 ctrl->device->groups = ops->dev_attr_groups;
4649 ctrl->device->groups = nvme_dev_attr_groups;
4650 ctrl->device->release = nvme_free_ctrl;
4651 dev_set_drvdata(ctrl->device, ctrl);
4652 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4654 goto out_release_instance;
4656 nvme_get_ctrl(ctrl);
4657 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4658 ctrl->cdev.owner = ops->module;
4659 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4664 * Initialize latency tolerance controls. The sysfs files won't
4665 * be visible to userspace unless the device actually supports APST.
4667 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4668 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4669 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4671 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4672 nvme_mpath_init_ctrl(ctrl);
4673 ret = nvme_auth_init_ctrl(ctrl);
4679 nvme_fault_inject_fini(&ctrl->fault_inject);
4680 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4681 cdev_device_del(&ctrl->cdev, ctrl->device);
4683 nvme_put_ctrl(ctrl);
4684 kfree_const(ctrl->device->kobj.name);
4685 out_release_instance:
4686 ida_free(&nvme_instance_ida, ctrl->instance);
4688 if (ctrl->discard_page)
4689 __free_page(ctrl->discard_page);
4692 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4694 /* let I/O to all namespaces fail in preparation for surprise removal */
4695 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
4699 down_read(&ctrl->namespaces_rwsem);
4700 list_for_each_entry(ns, &ctrl->namespaces, list)
4701 blk_mark_disk_dead(ns->disk);
4702 up_read(&ctrl->namespaces_rwsem);
4704 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
4706 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4710 down_read(&ctrl->namespaces_rwsem);
4711 list_for_each_entry(ns, &ctrl->namespaces, list)
4712 blk_mq_unfreeze_queue(ns->queue);
4713 up_read(&ctrl->namespaces_rwsem);
4714 clear_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4716 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4718 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4722 down_read(&ctrl->namespaces_rwsem);
4723 list_for_each_entry(ns, &ctrl->namespaces, list) {
4724 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4728 up_read(&ctrl->namespaces_rwsem);
4731 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4733 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4737 down_read(&ctrl->namespaces_rwsem);
4738 list_for_each_entry(ns, &ctrl->namespaces, list)
4739 blk_mq_freeze_queue_wait(ns->queue);
4740 up_read(&ctrl->namespaces_rwsem);
4742 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4744 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4748 set_bit(NVME_CTRL_FROZEN, &ctrl->flags);
4749 down_read(&ctrl->namespaces_rwsem);
4750 list_for_each_entry(ns, &ctrl->namespaces, list)
4751 blk_freeze_queue_start(ns->queue);
4752 up_read(&ctrl->namespaces_rwsem);
4754 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4756 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
4760 if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4761 blk_mq_quiesce_tagset(ctrl->tagset);
4763 blk_mq_wait_quiesce_done(ctrl->tagset);
4765 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
4767 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
4771 if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
4772 blk_mq_unquiesce_tagset(ctrl->tagset);
4774 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
4776 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
4778 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4779 blk_mq_quiesce_queue(ctrl->admin_q);
4781 blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
4783 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
4785 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
4787 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4788 blk_mq_unquiesce_queue(ctrl->admin_q);
4790 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
4792 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4796 down_read(&ctrl->namespaces_rwsem);
4797 list_for_each_entry(ns, &ctrl->namespaces, list)
4798 blk_sync_queue(ns->queue);
4799 up_read(&ctrl->namespaces_rwsem);
4801 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4803 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4805 nvme_sync_io_queues(ctrl);
4807 blk_sync_queue(ctrl->admin_q);
4809 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4811 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4813 if (file->f_op != &nvme_dev_fops)
4815 return file->private_data;
4817 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4820 * Check we didn't inadvertently grow the command structure sizes:
4822 static inline void _nvme_check_size(void)
4824 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4825 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4826 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4827 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4828 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4829 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4830 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4831 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4832 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4833 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4834 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4835 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4836 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4837 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4838 NVME_IDENTIFY_DATA_SIZE);
4839 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4840 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4841 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4842 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4843 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4844 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4845 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4846 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4847 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4851 static int __init nvme_core_init(void)
4853 int result = -ENOMEM;
4857 nvme_wq = alloc_workqueue("nvme-wq",
4858 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4862 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4863 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4867 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4868 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4869 if (!nvme_delete_wq)
4870 goto destroy_reset_wq;
4872 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4873 NVME_MINORS, "nvme");
4875 goto destroy_delete_wq;
4877 result = class_register(&nvme_class);
4879 goto unregister_chrdev;
4881 result = class_register(&nvme_subsys_class);
4885 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
4888 goto destroy_subsys_class;
4890 result = class_register(&nvme_ns_chr_class);
4892 goto unregister_generic_ns;
4894 result = nvme_init_auth();
4896 goto destroy_ns_chr;
4900 class_unregister(&nvme_ns_chr_class);
4901 unregister_generic_ns:
4902 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4903 destroy_subsys_class:
4904 class_unregister(&nvme_subsys_class);
4906 class_unregister(&nvme_class);
4908 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4910 destroy_workqueue(nvme_delete_wq);
4912 destroy_workqueue(nvme_reset_wq);
4914 destroy_workqueue(nvme_wq);
4919 static void __exit nvme_core_exit(void)
4922 class_unregister(&nvme_ns_chr_class);
4923 class_unregister(&nvme_subsys_class);
4924 class_unregister(&nvme_class);
4925 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
4926 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
4927 destroy_workqueue(nvme_delete_wq);
4928 destroy_workqueue(nvme_reset_wq);
4929 destroy_workqueue(nvme_wq);
4930 ida_destroy(&nvme_ns_chr_minor_ida);
4931 ida_destroy(&nvme_instance_ida);
4934 MODULE_LICENSE("GPL");
4935 MODULE_VERSION("1.0");
4936 MODULE_DESCRIPTION("NVMe host core framework");
4937 module_init(nvme_core_init);
4938 module_exit(nvme_core_exit);