1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
27 #include <linux/nvme-auth.h>
29 #define CREATE_TRACE_POINTS
32 #define NVME_MINORS (1U << MINORBITS)
35 struct nvme_ns_ids ids;
43 unsigned int admin_timeout = 60;
44 module_param(admin_timeout, uint, 0644);
45 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
46 EXPORT_SYMBOL_GPL(admin_timeout);
48 unsigned int nvme_io_timeout = 30;
49 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
50 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
51 EXPORT_SYMBOL_GPL(nvme_io_timeout);
53 static unsigned char shutdown_timeout = 5;
54 module_param(shutdown_timeout, byte, 0644);
55 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
57 static u8 nvme_max_retries = 5;
58 module_param_named(max_retries, nvme_max_retries, byte, 0644);
59 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
61 static unsigned long default_ps_max_latency_us = 100000;
62 module_param(default_ps_max_latency_us, ulong, 0644);
63 MODULE_PARM_DESC(default_ps_max_latency_us,
64 "max power saving latency for new devices; use PM QOS to change per device");
66 static bool force_apst;
67 module_param(force_apst, bool, 0644);
68 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
70 static unsigned long apst_primary_timeout_ms = 100;
71 module_param(apst_primary_timeout_ms, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_timeout_ms,
73 "primary APST timeout in ms");
75 static unsigned long apst_secondary_timeout_ms = 2000;
76 module_param(apst_secondary_timeout_ms, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_timeout_ms,
78 "secondary APST timeout in ms");
80 static unsigned long apst_primary_latency_tol_us = 15000;
81 module_param(apst_primary_latency_tol_us, ulong, 0644);
82 MODULE_PARM_DESC(apst_primary_latency_tol_us,
83 "primary APST latency tolerance in us");
85 static unsigned long apst_secondary_latency_tol_us = 100000;
86 module_param(apst_secondary_latency_tol_us, ulong, 0644);
87 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
88 "secondary APST latency tolerance in us");
91 * nvme_wq - hosts nvme related works that are not reset or delete
92 * nvme_reset_wq - hosts nvme reset works
93 * nvme_delete_wq - hosts nvme delete works
95 * nvme_wq will host works such as scan, aen handling, fw activation,
96 * keep-alive, periodic reconnects etc. nvme_reset_wq
97 * runs reset works which also flush works hosted on nvme_wq for
98 * serialization purposes. nvme_delete_wq host controller deletion
99 * works which flush reset works for serialization.
101 struct workqueue_struct *nvme_wq;
102 EXPORT_SYMBOL_GPL(nvme_wq);
104 struct workqueue_struct *nvme_reset_wq;
105 EXPORT_SYMBOL_GPL(nvme_reset_wq);
107 struct workqueue_struct *nvme_delete_wq;
108 EXPORT_SYMBOL_GPL(nvme_delete_wq);
110 static LIST_HEAD(nvme_subsystems);
111 static DEFINE_MUTEX(nvme_subsystems_lock);
113 static DEFINE_IDA(nvme_instance_ida);
114 static dev_t nvme_ctrl_base_chr_devt;
115 static struct class *nvme_class;
116 static struct class *nvme_subsys_class;
118 static DEFINE_IDA(nvme_ns_chr_minor_ida);
119 static dev_t nvme_ns_chr_devt;
120 static struct class *nvme_ns_chr_class;
122 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
123 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
125 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
126 struct nvme_command *cmd);
128 void nvme_queue_scan(struct nvme_ctrl *ctrl)
131 * Only new queue scan work when admin and IO queues are both alive
133 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
134 queue_work(nvme_wq, &ctrl->scan_work);
138 * Use this function to proceed with scheduling reset_work for a controller
139 * that had previously been set to the resetting state. This is intended for
140 * code paths that can't be interrupted by other reset attempts. A hot removal
141 * may prevent this from succeeding.
143 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
145 if (ctrl->state != NVME_CTRL_RESETTING)
147 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
151 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
153 static void nvme_failfast_work(struct work_struct *work)
155 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
156 struct nvme_ctrl, failfast_work);
158 if (ctrl->state != NVME_CTRL_CONNECTING)
161 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
162 dev_info(ctrl->device, "failfast expired\n");
163 nvme_kick_requeue_lists(ctrl);
166 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
168 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
171 schedule_delayed_work(&ctrl->failfast_work,
172 ctrl->opts->fast_io_fail_tmo * HZ);
175 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
180 cancel_delayed_work_sync(&ctrl->failfast_work);
181 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
185 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
187 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
189 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
193 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
195 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
199 ret = nvme_reset_ctrl(ctrl);
201 flush_work(&ctrl->reset_work);
202 if (ctrl->state != NVME_CTRL_LIVE)
209 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
211 dev_info(ctrl->device,
212 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
214 flush_work(&ctrl->reset_work);
215 nvme_stop_ctrl(ctrl);
216 nvme_remove_namespaces(ctrl);
217 ctrl->ops->delete_ctrl(ctrl);
218 nvme_uninit_ctrl(ctrl);
221 static void nvme_delete_ctrl_work(struct work_struct *work)
223 struct nvme_ctrl *ctrl =
224 container_of(work, struct nvme_ctrl, delete_work);
226 nvme_do_delete_ctrl(ctrl);
229 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
231 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
233 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
237 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
239 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
242 * Keep a reference until nvme_do_delete_ctrl() complete,
243 * since ->delete_ctrl can free the controller.
246 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
247 nvme_do_delete_ctrl(ctrl);
251 static blk_status_t nvme_error_status(u16 status)
253 switch (status & 0x7ff) {
254 case NVME_SC_SUCCESS:
256 case NVME_SC_CAP_EXCEEDED:
257 return BLK_STS_NOSPC;
258 case NVME_SC_LBA_RANGE:
259 case NVME_SC_CMD_INTERRUPTED:
260 case NVME_SC_NS_NOT_READY:
261 return BLK_STS_TARGET;
262 case NVME_SC_BAD_ATTRIBUTES:
263 case NVME_SC_ONCS_NOT_SUPPORTED:
264 case NVME_SC_INVALID_OPCODE:
265 case NVME_SC_INVALID_FIELD:
266 case NVME_SC_INVALID_NS:
267 return BLK_STS_NOTSUPP;
268 case NVME_SC_WRITE_FAULT:
269 case NVME_SC_READ_ERROR:
270 case NVME_SC_UNWRITTEN_BLOCK:
271 case NVME_SC_ACCESS_DENIED:
272 case NVME_SC_READ_ONLY:
273 case NVME_SC_COMPARE_FAILED:
274 return BLK_STS_MEDIUM;
275 case NVME_SC_GUARD_CHECK:
276 case NVME_SC_APPTAG_CHECK:
277 case NVME_SC_REFTAG_CHECK:
278 case NVME_SC_INVALID_PI:
279 return BLK_STS_PROTECTION;
280 case NVME_SC_RESERVATION_CONFLICT:
281 return BLK_STS_NEXUS;
282 case NVME_SC_HOST_PATH_ERROR:
283 return BLK_STS_TRANSPORT;
284 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
285 return BLK_STS_ZONE_ACTIVE_RESOURCE;
286 case NVME_SC_ZONE_TOO_MANY_OPEN:
287 return BLK_STS_ZONE_OPEN_RESOURCE;
289 return BLK_STS_IOERR;
293 static void nvme_retry_req(struct request *req)
295 unsigned long delay = 0;
298 /* The mask and shift result must be <= 3 */
299 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
301 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
303 nvme_req(req)->retries++;
304 blk_mq_requeue_request(req, false);
305 blk_mq_delay_kick_requeue_list(req->q, delay);
308 static void nvme_log_error(struct request *req)
310 struct nvme_ns *ns = req->q->queuedata;
311 struct nvme_request *nr = nvme_req(req);
314 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
315 ns->disk ? ns->disk->disk_name : "?",
316 nvme_get_opcode_str(nr->cmd->common.opcode),
317 nr->cmd->common.opcode,
318 (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
319 (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
320 nvme_get_error_status_str(nr->status),
321 nr->status >> 8 & 7, /* Status Code Type */
322 nr->status & 0xff, /* Status Code */
323 nr->status & NVME_SC_MORE ? "MORE " : "",
324 nr->status & NVME_SC_DNR ? "DNR " : "");
328 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
329 dev_name(nr->ctrl->device),
330 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
331 nr->cmd->common.opcode,
332 nvme_get_error_status_str(nr->status),
333 nr->status >> 8 & 7, /* Status Code Type */
334 nr->status & 0xff, /* Status Code */
335 nr->status & NVME_SC_MORE ? "MORE " : "",
336 nr->status & NVME_SC_DNR ? "DNR " : "");
339 enum nvme_disposition {
346 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
348 if (likely(nvme_req(req)->status == 0))
351 if ((nvme_req(req)->status & 0x7ff) == NVME_SC_AUTH_REQUIRED)
354 if (blk_noretry_request(req) ||
355 (nvme_req(req)->status & NVME_SC_DNR) ||
356 nvme_req(req)->retries >= nvme_max_retries)
359 if (req->cmd_flags & REQ_NVME_MPATH) {
360 if (nvme_is_path_error(nvme_req(req)->status) ||
361 blk_queue_dying(req->q))
364 if (blk_queue_dying(req->q))
371 static inline void nvme_end_req_zoned(struct request *req)
373 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
374 req_op(req) == REQ_OP_ZONE_APPEND)
375 req->__sector = nvme_lba_to_sect(req->q->queuedata,
376 le64_to_cpu(nvme_req(req)->result.u64));
379 static inline void nvme_end_req(struct request *req)
381 blk_status_t status = nvme_error_status(nvme_req(req)->status);
383 if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
385 nvme_end_req_zoned(req);
386 nvme_trace_bio_complete(req);
387 if (req->cmd_flags & REQ_NVME_MPATH)
388 nvme_mpath_end_request(req);
389 blk_mq_end_request(req, status);
392 void nvme_complete_rq(struct request *req)
394 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
396 trace_nvme_complete_rq(req);
397 nvme_cleanup_cmd(req);
400 ctrl->comp_seen = true;
402 switch (nvme_decide_disposition(req)) {
410 nvme_failover_req(req);
413 #ifdef CONFIG_NVME_AUTH
414 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
422 EXPORT_SYMBOL_GPL(nvme_complete_rq);
424 void nvme_complete_batch_req(struct request *req)
426 trace_nvme_complete_rq(req);
427 nvme_cleanup_cmd(req);
428 nvme_end_req_zoned(req);
430 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
433 * Called to unwind from ->queue_rq on a failed command submission so that the
434 * multipathing code gets called to potentially failover to another path.
435 * The caller needs to unwind all transport specific resource allocations and
436 * must return propagate the return value.
438 blk_status_t nvme_host_path_error(struct request *req)
440 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
441 blk_mq_set_request_complete(req);
442 nvme_complete_rq(req);
445 EXPORT_SYMBOL_GPL(nvme_host_path_error);
447 bool nvme_cancel_request(struct request *req, void *data)
449 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
450 "Cancelling I/O %d", req->tag);
452 /* don't abort one completed request */
453 if (blk_mq_request_completed(req))
456 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
457 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
458 blk_mq_complete_request(req);
461 EXPORT_SYMBOL_GPL(nvme_cancel_request);
463 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
466 blk_mq_tagset_busy_iter(ctrl->tagset,
467 nvme_cancel_request, ctrl);
468 blk_mq_tagset_wait_completed_request(ctrl->tagset);
471 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
473 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
475 if (ctrl->admin_tagset) {
476 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
477 nvme_cancel_request, ctrl);
478 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
481 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
483 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
484 enum nvme_ctrl_state new_state)
486 enum nvme_ctrl_state old_state;
488 bool changed = false;
490 spin_lock_irqsave(&ctrl->lock, flags);
492 old_state = ctrl->state;
497 case NVME_CTRL_RESETTING:
498 case NVME_CTRL_CONNECTING:
505 case NVME_CTRL_RESETTING:
515 case NVME_CTRL_CONNECTING:
518 case NVME_CTRL_RESETTING:
525 case NVME_CTRL_DELETING:
528 case NVME_CTRL_RESETTING:
529 case NVME_CTRL_CONNECTING:
536 case NVME_CTRL_DELETING_NOIO:
538 case NVME_CTRL_DELETING:
548 case NVME_CTRL_DELETING:
560 ctrl->state = new_state;
561 wake_up_all(&ctrl->state_wq);
564 spin_unlock_irqrestore(&ctrl->lock, flags);
568 if (ctrl->state == NVME_CTRL_LIVE) {
569 if (old_state == NVME_CTRL_CONNECTING)
570 nvme_stop_failfast_work(ctrl);
571 nvme_kick_requeue_lists(ctrl);
572 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
573 old_state == NVME_CTRL_RESETTING) {
574 nvme_start_failfast_work(ctrl);
578 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
581 * Returns true for sink states that can't ever transition back to live.
583 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
585 switch (ctrl->state) {
588 case NVME_CTRL_RESETTING:
589 case NVME_CTRL_CONNECTING:
591 case NVME_CTRL_DELETING:
592 case NVME_CTRL_DELETING_NOIO:
596 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
602 * Waits for the controller state to be resetting, or returns false if it is
603 * not possible to ever transition to that state.
605 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
607 wait_event(ctrl->state_wq,
608 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
609 nvme_state_terminal(ctrl));
610 return ctrl->state == NVME_CTRL_RESETTING;
612 EXPORT_SYMBOL_GPL(nvme_wait_reset);
614 static void nvme_free_ns_head(struct kref *ref)
616 struct nvme_ns_head *head =
617 container_of(ref, struct nvme_ns_head, ref);
619 nvme_mpath_remove_disk(head);
620 ida_free(&head->subsys->ns_ida, head->instance);
621 cleanup_srcu_struct(&head->srcu);
622 nvme_put_subsystem(head->subsys);
626 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
628 return kref_get_unless_zero(&head->ref);
631 void nvme_put_ns_head(struct nvme_ns_head *head)
633 kref_put(&head->ref, nvme_free_ns_head);
636 static void nvme_free_ns(struct kref *kref)
638 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
641 nvme_put_ns_head(ns->head);
642 nvme_put_ctrl(ns->ctrl);
646 static inline bool nvme_get_ns(struct nvme_ns *ns)
648 return kref_get_unless_zero(&ns->kref);
651 void nvme_put_ns(struct nvme_ns *ns)
653 kref_put(&ns->kref, nvme_free_ns);
655 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
657 static inline void nvme_clear_nvme_request(struct request *req)
659 nvme_req(req)->status = 0;
660 nvme_req(req)->retries = 0;
661 nvme_req(req)->flags = 0;
662 req->rq_flags |= RQF_DONTPREP;
665 /* initialize a passthrough request */
666 void nvme_init_request(struct request *req, struct nvme_command *cmd)
668 if (req->q->queuedata)
669 req->timeout = NVME_IO_TIMEOUT;
670 else /* no queuedata implies admin queue */
671 req->timeout = NVME_ADMIN_TIMEOUT;
673 /* passthru commands should let the driver set the SGL flags */
674 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
676 req->cmd_flags |= REQ_FAILFAST_DRIVER;
677 if (req->mq_hctx->type == HCTX_TYPE_POLL)
678 req->cmd_flags |= REQ_POLLED;
679 nvme_clear_nvme_request(req);
680 req->rq_flags |= RQF_QUIET;
681 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
683 EXPORT_SYMBOL_GPL(nvme_init_request);
686 * For something we're not in a state to send to the device the default action
687 * is to busy it and retry it after the controller state is recovered. However,
688 * if the controller is deleting or if anything is marked for failfast or
689 * nvme multipath it is immediately failed.
691 * Note: commands used to initialize the controller will be marked for failfast.
692 * Note: nvme cli/ioctl commands are marked for failfast.
694 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
697 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
698 ctrl->state != NVME_CTRL_DELETING &&
699 ctrl->state != NVME_CTRL_DEAD &&
700 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
701 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
702 return BLK_STS_RESOURCE;
703 return nvme_host_path_error(rq);
705 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
707 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
710 struct nvme_request *req = nvme_req(rq);
713 * currently we have a problem sending passthru commands
714 * on the admin_q if the controller is not LIVE because we can't
715 * make sure that they are going out after the admin connect,
716 * controller enable and/or other commands in the initialization
717 * sequence. until the controller will be LIVE, fail with
718 * BLK_STS_RESOURCE so that they will be rescheduled.
720 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
723 if (ctrl->ops->flags & NVME_F_FABRICS) {
725 * Only allow commands on a live queue, except for the connect
726 * command, which is require to set the queue live in the
727 * appropinquate states.
729 switch (ctrl->state) {
730 case NVME_CTRL_CONNECTING:
731 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
732 (req->cmd->fabrics.fctype == nvme_fabrics_type_connect ||
733 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_send ||
734 req->cmd->fabrics.fctype == nvme_fabrics_type_auth_receive))
746 EXPORT_SYMBOL_GPL(__nvme_check_ready);
748 static inline void nvme_setup_flush(struct nvme_ns *ns,
749 struct nvme_command *cmnd)
751 memset(cmnd, 0, sizeof(*cmnd));
752 cmnd->common.opcode = nvme_cmd_flush;
753 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
756 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
757 struct nvme_command *cmnd)
759 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
760 struct nvme_dsm_range *range;
764 * Some devices do not consider the DSM 'Number of Ranges' field when
765 * determining how much data to DMA. Always allocate memory for maximum
766 * number of segments to prevent device reading beyond end of buffer.
768 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
770 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
773 * If we fail allocation our range, fallback to the controller
774 * discard page. If that's also busy, it's safe to return
775 * busy, as we know we can make progress once that's freed.
777 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
778 return BLK_STS_RESOURCE;
780 range = page_address(ns->ctrl->discard_page);
783 __rq_for_each_bio(bio, req) {
784 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
785 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
788 range[n].cattr = cpu_to_le32(0);
789 range[n].nlb = cpu_to_le32(nlb);
790 range[n].slba = cpu_to_le64(slba);
795 if (WARN_ON_ONCE(n != segments)) {
796 if (virt_to_page(range) == ns->ctrl->discard_page)
797 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
800 return BLK_STS_IOERR;
803 memset(cmnd, 0, sizeof(*cmnd));
804 cmnd->dsm.opcode = nvme_cmd_dsm;
805 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
806 cmnd->dsm.nr = cpu_to_le32(segments - 1);
807 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
809 req->special_vec.bv_page = virt_to_page(range);
810 req->special_vec.bv_offset = offset_in_page(range);
811 req->special_vec.bv_len = alloc_size;
812 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
817 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
823 /* both rw and write zeroes share the same reftag format */
824 switch (ns->guard_type) {
825 case NVME_NVM_NS_16B_GUARD:
826 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
828 case NVME_NVM_NS_64B_GUARD:
829 ref48 = ext_pi_ref_tag(req);
830 lower = lower_32_bits(ref48);
831 upper = upper_32_bits(ref48);
833 cmnd->rw.reftag = cpu_to_le32(lower);
834 cmnd->rw.cdw3 = cpu_to_le32(upper);
841 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
842 struct request *req, struct nvme_command *cmnd)
844 memset(cmnd, 0, sizeof(*cmnd));
846 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
847 return nvme_setup_discard(ns, req, cmnd);
849 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
850 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
851 cmnd->write_zeroes.slba =
852 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
853 cmnd->write_zeroes.length =
854 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
856 if (!(req->cmd_flags & REQ_NOUNMAP) && (ns->features & NVME_NS_DEAC))
857 cmnd->write_zeroes.control |= cpu_to_le16(NVME_WZ_DEAC);
859 if (nvme_ns_has_pi(ns)) {
860 cmnd->write_zeroes.control |= cpu_to_le16(NVME_RW_PRINFO_PRACT);
862 switch (ns->pi_type) {
863 case NVME_NS_DPS_PI_TYPE1:
864 case NVME_NS_DPS_PI_TYPE2:
865 nvme_set_ref_tag(ns, cmnd, req);
873 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
874 struct request *req, struct nvme_command *cmnd,
880 if (req->cmd_flags & REQ_FUA)
881 control |= NVME_RW_FUA;
882 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
883 control |= NVME_RW_LR;
885 if (req->cmd_flags & REQ_RAHEAD)
886 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
888 cmnd->rw.opcode = op;
890 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
893 cmnd->rw.metadata = 0;
894 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
895 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
898 cmnd->rw.appmask = 0;
902 * If formated with metadata, the block layer always provides a
903 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
904 * we enable the PRACT bit for protection information or set the
905 * namespace capacity to zero to prevent any I/O.
907 if (!blk_integrity_rq(req)) {
908 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
909 return BLK_STS_NOTSUPP;
910 control |= NVME_RW_PRINFO_PRACT;
913 switch (ns->pi_type) {
914 case NVME_NS_DPS_PI_TYPE3:
915 control |= NVME_RW_PRINFO_PRCHK_GUARD;
917 case NVME_NS_DPS_PI_TYPE1:
918 case NVME_NS_DPS_PI_TYPE2:
919 control |= NVME_RW_PRINFO_PRCHK_GUARD |
920 NVME_RW_PRINFO_PRCHK_REF;
921 if (op == nvme_cmd_zone_append)
922 control |= NVME_RW_APPEND_PIREMAP;
923 nvme_set_ref_tag(ns, cmnd, req);
928 cmnd->rw.control = cpu_to_le16(control);
929 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
933 void nvme_cleanup_cmd(struct request *req)
935 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
936 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
938 if (req->special_vec.bv_page == ctrl->discard_page)
939 clear_bit_unlock(0, &ctrl->discard_page_busy);
941 kfree(bvec_virt(&req->special_vec));
944 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
946 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
948 struct nvme_command *cmd = nvme_req(req)->cmd;
949 blk_status_t ret = BLK_STS_OK;
951 if (!(req->rq_flags & RQF_DONTPREP))
952 nvme_clear_nvme_request(req);
954 switch (req_op(req)) {
957 /* these are setup prior to execution in nvme_init_request() */
960 nvme_setup_flush(ns, cmd);
962 case REQ_OP_ZONE_RESET_ALL:
963 case REQ_OP_ZONE_RESET:
964 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
966 case REQ_OP_ZONE_OPEN:
967 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
969 case REQ_OP_ZONE_CLOSE:
970 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
972 case REQ_OP_ZONE_FINISH:
973 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
975 case REQ_OP_WRITE_ZEROES:
976 ret = nvme_setup_write_zeroes(ns, req, cmd);
979 ret = nvme_setup_discard(ns, req, cmd);
982 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
985 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
987 case REQ_OP_ZONE_APPEND:
988 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
992 return BLK_STS_IOERR;
995 cmd->common.command_id = nvme_cid(req);
996 trace_nvme_setup_cmd(req, cmd);
999 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
1004 * >0: nvme controller's cqe status response
1005 * <0: kernel error in lieu of controller response
1007 static int nvme_execute_rq(struct request *rq, bool at_head)
1009 blk_status_t status;
1011 status = blk_execute_rq(rq, at_head);
1012 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
1014 if (nvme_req(rq)->status)
1015 return nvme_req(rq)->status;
1016 return blk_status_to_errno(status);
1020 * Returns 0 on success. If the result is negative, it's a Linux error code;
1021 * if the result is positive, it's an NVM Express status code
1023 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1024 union nvme_result *result, void *buffer, unsigned bufflen,
1025 int qid, int at_head, blk_mq_req_flags_t flags)
1027 struct request *req;
1030 if (qid == NVME_QID_ANY)
1031 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1033 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1037 return PTR_ERR(req);
1038 nvme_init_request(req, cmd);
1040 if (buffer && bufflen) {
1041 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1046 ret = nvme_execute_rq(req, at_head);
1047 if (result && ret >= 0)
1048 *result = nvme_req(req)->result;
1050 blk_mq_free_request(req);
1053 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1055 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1056 void *buffer, unsigned bufflen)
1058 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen,
1059 NVME_QID_ANY, 0, 0);
1061 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1063 static u32 nvme_known_admin_effects(u8 opcode)
1066 case nvme_admin_format_nvm:
1067 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1068 NVME_CMD_EFFECTS_CSE_MASK;
1069 case nvme_admin_sanitize_nvm:
1070 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1077 static u32 nvme_known_nvm_effects(u8 opcode)
1080 case nvme_cmd_write:
1081 case nvme_cmd_write_zeroes:
1082 case nvme_cmd_write_uncor:
1083 return NVME_CMD_EFFECTS_LBCC;
1089 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1094 if (ns->head->effects)
1095 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1096 if (ns->head->ids.csi == NVME_CSI_NVM)
1097 effects |= nvme_known_nvm_effects(opcode);
1098 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1099 dev_warn_once(ctrl->device,
1100 "IO command:%02x has unusual effects:%08x\n",
1104 * NVME_CMD_EFFECTS_CSE_MASK causes a freeze all I/O queues,
1105 * which would deadlock when done on an I/O command. Note that
1106 * We already warn about an unusual effect above.
1108 effects &= ~NVME_CMD_EFFECTS_CSE_MASK;
1111 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1112 effects |= nvme_known_admin_effects(opcode);
1117 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1119 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1122 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1125 * For simplicity, IO to all namespaces is quiesced even if the command
1126 * effects say only one namespace is affected.
1128 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1129 mutex_lock(&ctrl->scan_lock);
1130 mutex_lock(&ctrl->subsys->lock);
1131 nvme_mpath_start_freeze(ctrl->subsys);
1132 nvme_mpath_wait_freeze(ctrl->subsys);
1133 nvme_start_freeze(ctrl);
1134 nvme_wait_freeze(ctrl);
1139 void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1140 struct nvme_command *cmd, int status)
1142 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1143 nvme_unfreeze(ctrl);
1144 nvme_mpath_unfreeze(ctrl->subsys);
1145 mutex_unlock(&ctrl->subsys->lock);
1146 mutex_unlock(&ctrl->scan_lock);
1148 if (effects & NVME_CMD_EFFECTS_CCC) {
1149 dev_info(ctrl->device,
1150 "controller capabilities changed, reset may be required to take effect.\n");
1152 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1153 nvme_queue_scan(ctrl);
1154 flush_work(&ctrl->scan_work);
1157 switch (cmd->common.opcode) {
1158 case nvme_admin_set_features:
1159 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1160 case NVME_FEAT_KATO:
1162 * Keep alive commands interval on the host should be
1163 * updated when KATO is modified by Set Features
1167 nvme_update_keep_alive(ctrl, cmd);
1177 EXPORT_SYMBOL_NS_GPL(nvme_passthru_end, NVME_TARGET_PASSTHRU);
1179 int nvme_execute_passthru_rq(struct request *rq, u32 *effects)
1181 struct nvme_command *cmd = nvme_req(rq)->cmd;
1182 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1183 struct nvme_ns *ns = rq->q->queuedata;
1185 *effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1186 return nvme_execute_rq(rq, false);
1188 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1191 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1193 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1194 * accounting for transport roundtrip times [..].
1196 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1198 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1201 static enum rq_end_io_ret nvme_keep_alive_end_io(struct request *rq,
1202 blk_status_t status)
1204 struct nvme_ctrl *ctrl = rq->end_io_data;
1205 unsigned long flags;
1206 bool startka = false;
1208 blk_mq_free_request(rq);
1211 dev_err(ctrl->device,
1212 "failed nvme_keep_alive_end_io error=%d\n",
1214 return RQ_END_IO_NONE;
1217 ctrl->comp_seen = false;
1218 spin_lock_irqsave(&ctrl->lock, flags);
1219 if (ctrl->state == NVME_CTRL_LIVE ||
1220 ctrl->state == NVME_CTRL_CONNECTING)
1222 spin_unlock_irqrestore(&ctrl->lock, flags);
1224 nvme_queue_keep_alive_work(ctrl);
1225 return RQ_END_IO_NONE;
1228 static void nvme_keep_alive_work(struct work_struct *work)
1230 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1231 struct nvme_ctrl, ka_work);
1232 bool comp_seen = ctrl->comp_seen;
1235 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1236 dev_dbg(ctrl->device,
1237 "reschedule traffic based keep-alive timer\n");
1238 ctrl->comp_seen = false;
1239 nvme_queue_keep_alive_work(ctrl);
1243 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1244 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1246 /* allocation failure, reset the controller */
1247 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1248 nvme_reset_ctrl(ctrl);
1251 nvme_init_request(rq, &ctrl->ka_cmd);
1253 rq->timeout = ctrl->kato * HZ;
1254 rq->end_io = nvme_keep_alive_end_io;
1255 rq->end_io_data = ctrl;
1256 blk_execute_rq_nowait(rq, false);
1259 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1261 if (unlikely(ctrl->kato == 0))
1264 nvme_queue_keep_alive_work(ctrl);
1267 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1269 if (unlikely(ctrl->kato == 0))
1272 cancel_delayed_work_sync(&ctrl->ka_work);
1274 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1276 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1277 struct nvme_command *cmd)
1279 unsigned int new_kato =
1280 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1282 dev_info(ctrl->device,
1283 "keep alive interval updated from %u ms to %u ms\n",
1284 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1286 nvme_stop_keep_alive(ctrl);
1287 ctrl->kato = new_kato;
1288 nvme_start_keep_alive(ctrl);
1292 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1293 * flag, thus sending any new CNS opcodes has a big chance of not working.
1294 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1295 * (but not for any later version).
1297 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1299 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1300 return ctrl->vs < NVME_VS(1, 2, 0);
1301 return ctrl->vs < NVME_VS(1, 1, 0);
1304 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1306 struct nvme_command c = { };
1309 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1310 c.identify.opcode = nvme_admin_identify;
1311 c.identify.cns = NVME_ID_CNS_CTRL;
1313 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1317 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1318 sizeof(struct nvme_id_ctrl));
1324 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1325 struct nvme_ns_id_desc *cur, bool *csi_seen)
1327 const char *warn_str = "ctrl returned bogus length:";
1330 switch (cur->nidt) {
1331 case NVME_NIDT_EUI64:
1332 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1333 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1334 warn_str, cur->nidl);
1337 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1338 return NVME_NIDT_EUI64_LEN;
1339 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1340 return NVME_NIDT_EUI64_LEN;
1341 case NVME_NIDT_NGUID:
1342 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1343 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1344 warn_str, cur->nidl);
1347 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1348 return NVME_NIDT_NGUID_LEN;
1349 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1350 return NVME_NIDT_NGUID_LEN;
1351 case NVME_NIDT_UUID:
1352 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1353 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1354 warn_str, cur->nidl);
1357 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1358 return NVME_NIDT_UUID_LEN;
1359 uuid_copy(&ids->uuid, data + sizeof(*cur));
1360 return NVME_NIDT_UUID_LEN;
1362 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1363 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1364 warn_str, cur->nidl);
1367 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1369 return NVME_NIDT_CSI_LEN;
1371 /* Skip unknown types */
1376 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl,
1377 struct nvme_ns_info *info)
1379 struct nvme_command c = { };
1380 bool csi_seen = false;
1381 int status, pos, len;
1384 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1386 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1389 c.identify.opcode = nvme_admin_identify;
1390 c.identify.nsid = cpu_to_le32(info->nsid);
1391 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1393 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1397 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1398 NVME_IDENTIFY_DATA_SIZE);
1400 dev_warn(ctrl->device,
1401 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1402 info->nsid, status);
1406 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1407 struct nvme_ns_id_desc *cur = data + pos;
1412 len = nvme_process_ns_desc(ctrl, &info->ids, cur, &csi_seen);
1416 len += sizeof(*cur);
1419 if (nvme_multi_css(ctrl) && !csi_seen) {
1420 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1430 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1431 struct nvme_id_ns **id)
1433 struct nvme_command c = { };
1436 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1437 c.identify.opcode = nvme_admin_identify;
1438 c.identify.nsid = cpu_to_le32(nsid);
1439 c.identify.cns = NVME_ID_CNS_NS;
1441 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1445 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1447 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1451 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1452 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1461 static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
1462 struct nvme_ns_info *info)
1464 struct nvme_ns_ids *ids = &info->ids;
1465 struct nvme_id_ns *id;
1468 ret = nvme_identify_ns(ctrl, info->nsid, &id);
1471 info->anagrpid = id->anagrpid;
1472 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1473 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1474 info->is_ready = true;
1475 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1476 dev_info(ctrl->device,
1477 "Ignoring bogus Namespace Identifiers\n");
1479 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1480 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1481 memcpy(ids->eui64, id->eui64, sizeof(ids->eui64));
1482 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1483 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1484 memcpy(ids->nguid, id->nguid, sizeof(ids->nguid));
1490 static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
1491 struct nvme_ns_info *info)
1493 struct nvme_id_ns_cs_indep *id;
1494 struct nvme_command c = {
1495 .identify.opcode = nvme_admin_identify,
1496 .identify.nsid = cpu_to_le32(info->nsid),
1497 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1501 id = kmalloc(sizeof(*id), GFP_KERNEL);
1505 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
1507 info->anagrpid = id->anagrpid;
1508 info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
1509 info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
1510 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
1516 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1517 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1519 union nvme_result res = { 0 };
1520 struct nvme_command c = { };
1523 c.features.opcode = op;
1524 c.features.fid = cpu_to_le32(fid);
1525 c.features.dword11 = cpu_to_le32(dword11);
1527 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1528 buffer, buflen, NVME_QID_ANY, 0, 0);
1529 if (ret >= 0 && result)
1530 *result = le32_to_cpu(res.u32);
1534 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1535 unsigned int dword11, void *buffer, size_t buflen,
1538 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1541 EXPORT_SYMBOL_GPL(nvme_set_features);
1543 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1544 unsigned int dword11, void *buffer, size_t buflen,
1547 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1550 EXPORT_SYMBOL_GPL(nvme_get_features);
1552 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1554 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1556 int status, nr_io_queues;
1558 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1564 * Degraded controllers might return an error when setting the queue
1565 * count. We still want to be able to bring them online and offer
1566 * access to the admin queue, as that might be only way to fix them up.
1569 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1572 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1573 *count = min(*count, nr_io_queues);
1578 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1580 #define NVME_AEN_SUPPORTED \
1581 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1582 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1584 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1586 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1589 if (!supported_aens)
1592 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1595 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1598 queue_work(nvme_wq, &ctrl->async_event_work);
1601 static int nvme_ns_open(struct nvme_ns *ns)
1604 /* should never be called due to GENHD_FL_HIDDEN */
1605 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1607 if (!nvme_get_ns(ns))
1609 if (!try_module_get(ns->ctrl->ops->module))
1620 static void nvme_ns_release(struct nvme_ns *ns)
1623 module_put(ns->ctrl->ops->module);
1627 static int nvme_open(struct block_device *bdev, fmode_t mode)
1629 return nvme_ns_open(bdev->bd_disk->private_data);
1632 static void nvme_release(struct gendisk *disk, fmode_t mode)
1634 nvme_ns_release(disk->private_data);
1637 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1639 /* some standard values */
1640 geo->heads = 1 << 6;
1641 geo->sectors = 1 << 5;
1642 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1646 #ifdef CONFIG_BLK_DEV_INTEGRITY
1647 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1648 u32 max_integrity_segments)
1650 struct blk_integrity integrity = { };
1652 switch (ns->pi_type) {
1653 case NVME_NS_DPS_PI_TYPE3:
1654 switch (ns->guard_type) {
1655 case NVME_NVM_NS_16B_GUARD:
1656 integrity.profile = &t10_pi_type3_crc;
1657 integrity.tag_size = sizeof(u16) + sizeof(u32);
1658 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1660 case NVME_NVM_NS_64B_GUARD:
1661 integrity.profile = &ext_pi_type3_crc64;
1662 integrity.tag_size = sizeof(u16) + 6;
1663 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1666 integrity.profile = NULL;
1670 case NVME_NS_DPS_PI_TYPE1:
1671 case NVME_NS_DPS_PI_TYPE2:
1672 switch (ns->guard_type) {
1673 case NVME_NVM_NS_16B_GUARD:
1674 integrity.profile = &t10_pi_type1_crc;
1675 integrity.tag_size = sizeof(u16);
1676 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1678 case NVME_NVM_NS_64B_GUARD:
1679 integrity.profile = &ext_pi_type1_crc64;
1680 integrity.tag_size = sizeof(u16);
1681 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1684 integrity.profile = NULL;
1689 integrity.profile = NULL;
1693 integrity.tuple_size = ns->ms;
1694 blk_integrity_register(disk, &integrity);
1695 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1698 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1699 u32 max_integrity_segments)
1702 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1704 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1706 struct nvme_ctrl *ctrl = ns->ctrl;
1707 struct request_queue *queue = disk->queue;
1708 u32 size = queue_logical_block_size(queue);
1710 if (ctrl->max_discard_sectors == 0) {
1711 blk_queue_max_discard_sectors(queue, 0);
1715 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1716 NVME_DSM_MAX_RANGES);
1718 queue->limits.discard_granularity = size;
1720 /* If discard is already enabled, don't reset queue limits */
1721 if (queue->limits.max_discard_sectors)
1724 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1725 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1727 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1728 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1730 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1731 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1734 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1736 return uuid_equal(&a->uuid, &b->uuid) &&
1737 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1738 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1742 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1744 bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1745 unsigned lbaf = nvme_lbaf_index(id->flbas);
1746 struct nvme_ctrl *ctrl = ns->ctrl;
1747 struct nvme_command c = { };
1748 struct nvme_id_ns_nvm *nvm;
1753 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1754 if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1755 ns->pi_size = sizeof(struct t10_pi_tuple);
1756 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1760 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1764 c.identify.opcode = nvme_admin_identify;
1765 c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1766 c.identify.cns = NVME_ID_CNS_CS_NS;
1767 c.identify.csi = NVME_CSI_NVM;
1769 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1773 elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1775 /* no support for storage tag formats right now */
1776 if (nvme_elbaf_sts(elbaf))
1779 ns->guard_type = nvme_elbaf_guard_type(elbaf);
1780 switch (ns->guard_type) {
1781 case NVME_NVM_NS_64B_GUARD:
1782 ns->pi_size = sizeof(struct crc64_pi_tuple);
1784 case NVME_NVM_NS_16B_GUARD:
1785 ns->pi_size = sizeof(struct t10_pi_tuple);
1794 if (ns->pi_size && (first || ns->ms == ns->pi_size))
1795 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1802 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1804 struct nvme_ctrl *ctrl = ns->ctrl;
1806 if (nvme_init_ms(ns, id))
1809 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1810 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1813 if (ctrl->ops->flags & NVME_F_FABRICS) {
1815 * The NVMe over Fabrics specification only supports metadata as
1816 * part of the extended data LBA. We rely on HCA/HBA support to
1817 * remap the separate metadata buffer from the block layer.
1819 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1822 ns->features |= NVME_NS_EXT_LBAS;
1825 * The current fabrics transport drivers support namespace
1826 * metadata formats only if nvme_ns_has_pi() returns true.
1827 * Suppress support for all other formats so the namespace will
1828 * have a 0 capacity and not be usable through the block stack.
1830 * Note, this check will need to be modified if any drivers
1831 * gain the ability to use other metadata formats.
1833 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1834 ns->features |= NVME_NS_METADATA_SUPPORTED;
1837 * For PCIe controllers, we can't easily remap the separate
1838 * metadata buffer from the block layer and thus require a
1839 * separate metadata buffer for block layer metadata/PI support.
1840 * We allow extended LBAs for the passthrough interface, though.
1842 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1843 ns->features |= NVME_NS_EXT_LBAS;
1845 ns->features |= NVME_NS_METADATA_SUPPORTED;
1849 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1850 struct request_queue *q)
1852 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1854 if (ctrl->max_hw_sectors) {
1856 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1858 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1859 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1860 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1862 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1863 blk_queue_dma_alignment(q, 3);
1864 blk_queue_write_cache(q, vwc, vwc);
1867 static void nvme_update_disk_info(struct gendisk *disk,
1868 struct nvme_ns *ns, struct nvme_id_ns *id)
1870 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1871 unsigned short bs = 1 << ns->lba_shift;
1872 u32 atomic_bs, phys_bs, io_opt = 0;
1875 * The block layer can't support LBA sizes larger than the page size
1876 * yet, so catch this early and don't allow block I/O.
1878 if (ns->lba_shift > PAGE_SHIFT) {
1883 blk_integrity_unregister(disk);
1885 atomic_bs = phys_bs = bs;
1886 if (id->nabo == 0) {
1888 * Bit 1 indicates whether NAWUPF is defined for this namespace
1889 * and whether it should be used instead of AWUPF. If NAWUPF ==
1890 * 0 then AWUPF must be used instead.
1892 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1893 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1895 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1898 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1899 /* NPWG = Namespace Preferred Write Granularity */
1900 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1901 /* NOWS = Namespace Optimal Write Size */
1902 io_opt = bs * (1 + le16_to_cpu(id->nows));
1905 blk_queue_logical_block_size(disk->queue, bs);
1907 * Linux filesystems assume writing a single physical block is
1908 * an atomic operation. Hence limit the physical block size to the
1909 * value of the Atomic Write Unit Power Fail parameter.
1911 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1912 blk_queue_io_min(disk->queue, phys_bs);
1913 blk_queue_io_opt(disk->queue, io_opt);
1916 * Register a metadata profile for PI, or the plain non-integrity NVMe
1917 * metadata masquerading as Type 0 if supported, otherwise reject block
1918 * I/O to namespaces with metadata except when the namespace supports
1919 * PI, as it can strip/insert in that case.
1922 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1923 (ns->features & NVME_NS_METADATA_SUPPORTED))
1924 nvme_init_integrity(disk, ns,
1925 ns->ctrl->max_integrity_segments);
1926 else if (!nvme_ns_has_pi(ns))
1930 set_capacity_and_notify(disk, capacity);
1932 nvme_config_discard(disk, ns);
1933 blk_queue_max_write_zeroes_sectors(disk->queue,
1934 ns->ctrl->max_zeroes_sectors);
1937 static bool nvme_ns_is_readonly(struct nvme_ns *ns, struct nvme_ns_info *info)
1939 return info->is_readonly || test_bit(NVME_NS_FORCE_RO, &ns->flags);
1942 static inline bool nvme_first_scan(struct gendisk *disk)
1944 /* nvme_alloc_ns() scans the disk prior to adding it */
1945 return !disk_live(disk);
1948 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1950 struct nvme_ctrl *ctrl = ns->ctrl;
1953 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1954 is_power_of_2(ctrl->max_hw_sectors))
1955 iob = ctrl->max_hw_sectors;
1957 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1962 if (!is_power_of_2(iob)) {
1963 if (nvme_first_scan(ns->disk))
1964 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1965 ns->disk->disk_name, iob);
1969 if (blk_queue_is_zoned(ns->disk->queue)) {
1970 if (nvme_first_scan(ns->disk))
1971 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1972 ns->disk->disk_name);
1976 blk_queue_chunk_sectors(ns->queue, iob);
1979 static int nvme_update_ns_info_generic(struct nvme_ns *ns,
1980 struct nvme_ns_info *info)
1982 blk_mq_freeze_queue(ns->disk->queue);
1983 nvme_set_queue_limits(ns->ctrl, ns->queue);
1984 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
1985 blk_mq_unfreeze_queue(ns->disk->queue);
1987 if (nvme_ns_head_multipath(ns->head)) {
1988 blk_mq_freeze_queue(ns->head->disk->queue);
1989 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
1990 nvme_mpath_revalidate_paths(ns);
1991 blk_stack_limits(&ns->head->disk->queue->limits,
1992 &ns->queue->limits, 0);
1993 ns->head->disk->flags |= GENHD_FL_HIDDEN;
1994 blk_mq_unfreeze_queue(ns->head->disk->queue);
1997 /* Hide the block-interface for these devices */
1998 ns->disk->flags |= GENHD_FL_HIDDEN;
1999 set_bit(NVME_NS_READY, &ns->flags);
2004 static int nvme_update_ns_info_block(struct nvme_ns *ns,
2005 struct nvme_ns_info *info)
2007 struct nvme_id_ns *id;
2011 ret = nvme_identify_ns(ns->ctrl, info->nsid, &id);
2015 blk_mq_freeze_queue(ns->disk->queue);
2016 lbaf = nvme_lbaf_index(id->flbas);
2017 ns->lba_shift = id->lbaf[lbaf].ds;
2018 nvme_set_queue_limits(ns->ctrl, ns->queue);
2020 nvme_configure_metadata(ns, id);
2021 nvme_set_chunk_sectors(ns, id);
2022 nvme_update_disk_info(ns->disk, ns, id);
2024 if (ns->head->ids.csi == NVME_CSI_ZNS) {
2025 ret = nvme_update_zone_info(ns, lbaf);
2027 blk_mq_unfreeze_queue(ns->disk->queue);
2033 * Only set the DEAC bit if the device guarantees that reads from
2034 * deallocated data return zeroes. While the DEAC bit does not
2035 * require that, it must be a no-op if reads from deallocated data
2036 * do not return zeroes.
2038 if ((id->dlfeat & 0x7) == 0x1 && (id->dlfeat & (1 << 3)))
2039 ns->features |= NVME_NS_DEAC;
2040 set_disk_ro(ns->disk, nvme_ns_is_readonly(ns, info));
2041 set_bit(NVME_NS_READY, &ns->flags);
2042 blk_mq_unfreeze_queue(ns->disk->queue);
2044 if (blk_queue_is_zoned(ns->queue)) {
2045 ret = nvme_revalidate_zones(ns);
2046 if (ret && !nvme_first_scan(ns->disk))
2050 if (nvme_ns_head_multipath(ns->head)) {
2051 blk_mq_freeze_queue(ns->head->disk->queue);
2052 nvme_update_disk_info(ns->head->disk, ns, id);
2053 set_disk_ro(ns->head->disk, nvme_ns_is_readonly(ns, info));
2054 nvme_mpath_revalidate_paths(ns);
2055 blk_stack_limits(&ns->head->disk->queue->limits,
2056 &ns->queue->limits, 0);
2057 disk_update_readahead(ns->head->disk);
2058 blk_mq_unfreeze_queue(ns->head->disk->queue);
2064 * If probing fails due an unsupported feature, hide the block device,
2065 * but still allow other access.
2067 if (ret == -ENODEV) {
2068 ns->disk->flags |= GENHD_FL_HIDDEN;
2069 set_bit(NVME_NS_READY, &ns->flags);
2076 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
2078 switch (info->ids.csi) {
2080 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
2081 dev_info(ns->ctrl->device,
2082 "block device for nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
2084 return nvme_update_ns_info_generic(ns, info);
2086 return nvme_update_ns_info_block(ns, info);
2088 return nvme_update_ns_info_block(ns, info);
2090 dev_info(ns->ctrl->device,
2091 "block device for nsid %u not supported (csi %u)\n",
2092 info->nsid, info->ids.csi);
2093 return nvme_update_ns_info_generic(ns, info);
2097 static char nvme_pr_type(enum pr_type type)
2100 case PR_WRITE_EXCLUSIVE:
2102 case PR_EXCLUSIVE_ACCESS:
2104 case PR_WRITE_EXCLUSIVE_REG_ONLY:
2106 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
2108 case PR_WRITE_EXCLUSIVE_ALL_REGS:
2110 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
2117 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
2118 struct nvme_command *c, u8 data[16])
2120 struct nvme_ns_head *head = bdev->bd_disk->private_data;
2121 int srcu_idx = srcu_read_lock(&head->srcu);
2122 struct nvme_ns *ns = nvme_find_path(head);
2123 int ret = -EWOULDBLOCK;
2126 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2127 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2129 srcu_read_unlock(&head->srcu, srcu_idx);
2133 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2136 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2137 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2140 static int nvme_sc_to_pr_err(int nvme_sc)
2142 if (nvme_is_path_error(nvme_sc))
2143 return PR_STS_PATH_FAILED;
2146 case NVME_SC_SUCCESS:
2147 return PR_STS_SUCCESS;
2148 case NVME_SC_RESERVATION_CONFLICT:
2149 return PR_STS_RESERVATION_CONFLICT;
2150 case NVME_SC_ONCS_NOT_SUPPORTED:
2152 case NVME_SC_BAD_ATTRIBUTES:
2153 case NVME_SC_INVALID_OPCODE:
2154 case NVME_SC_INVALID_FIELD:
2155 case NVME_SC_INVALID_NS:
2158 return PR_STS_IOERR;
2162 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2163 u64 key, u64 sa_key, u8 op)
2165 struct nvme_command c = { };
2166 u8 data[16] = { 0, };
2169 put_unaligned_le64(key, &data[0]);
2170 put_unaligned_le64(sa_key, &data[8]);
2172 c.common.opcode = op;
2173 c.common.cdw10 = cpu_to_le32(cdw10);
2175 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2176 bdev->bd_disk->fops == &nvme_ns_head_ops)
2177 ret = nvme_send_ns_head_pr_command(bdev, &c, data);
2179 ret = nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c,
2184 return nvme_sc_to_pr_err(ret);
2187 static int nvme_pr_register(struct block_device *bdev, u64 old,
2188 u64 new, unsigned flags)
2192 if (flags & ~PR_FL_IGNORE_KEY)
2195 cdw10 = old ? 2 : 0;
2196 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2197 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2198 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2201 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2202 enum pr_type type, unsigned flags)
2206 if (flags & ~PR_FL_IGNORE_KEY)
2209 cdw10 = nvme_pr_type(type) << 8;
2210 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2211 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2214 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2215 enum pr_type type, bool abort)
2217 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2219 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2222 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2224 u32 cdw10 = 1 | (key ? 0 : 1 << 3);
2226 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2229 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2231 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 0 : 1 << 3);
2233 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2236 const struct pr_ops nvme_pr_ops = {
2237 .pr_register = nvme_pr_register,
2238 .pr_reserve = nvme_pr_reserve,
2239 .pr_release = nvme_pr_release,
2240 .pr_preempt = nvme_pr_preempt,
2241 .pr_clear = nvme_pr_clear,
2244 #ifdef CONFIG_BLK_SED_OPAL
2245 static int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2248 struct nvme_ctrl *ctrl = data;
2249 struct nvme_command cmd = { };
2252 cmd.common.opcode = nvme_admin_security_send;
2254 cmd.common.opcode = nvme_admin_security_recv;
2255 cmd.common.nsid = 0;
2256 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2257 cmd.common.cdw11 = cpu_to_le32(len);
2259 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len,
2260 NVME_QID_ANY, 1, 0);
2263 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2265 if (ctrl->oacs & NVME_CTRL_OACS_SEC_SUPP) {
2266 if (!ctrl->opal_dev)
2267 ctrl->opal_dev = init_opal_dev(ctrl, &nvme_sec_submit);
2268 else if (was_suspended)
2269 opal_unlock_from_suspend(ctrl->opal_dev);
2271 free_opal_dev(ctrl->opal_dev);
2272 ctrl->opal_dev = NULL;
2276 static void nvme_configure_opal(struct nvme_ctrl *ctrl, bool was_suspended)
2279 #endif /* CONFIG_BLK_SED_OPAL */
2281 #ifdef CONFIG_BLK_DEV_ZONED
2282 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2283 unsigned int nr_zones, report_zones_cb cb, void *data)
2285 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2289 #define nvme_report_zones NULL
2290 #endif /* CONFIG_BLK_DEV_ZONED */
2292 static const struct block_device_operations nvme_bdev_ops = {
2293 .owner = THIS_MODULE,
2294 .ioctl = nvme_ioctl,
2295 .compat_ioctl = blkdev_compat_ptr_ioctl,
2297 .release = nvme_release,
2298 .getgeo = nvme_getgeo,
2299 .report_zones = nvme_report_zones,
2300 .pr_ops = &nvme_pr_ops,
2303 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 mask, u32 val,
2304 u32 timeout, const char *op)
2306 unsigned long timeout_jiffies = jiffies + timeout * HZ;
2310 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2313 if ((csts & mask) == val)
2316 usleep_range(1000, 2000);
2317 if (fatal_signal_pending(current))
2319 if (time_after(jiffies, timeout_jiffies)) {
2320 dev_err(ctrl->device,
2321 "Device not ready; aborting %s, CSTS=0x%x\n",
2330 int nvme_disable_ctrl(struct nvme_ctrl *ctrl, bool shutdown)
2334 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2336 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2338 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2340 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2345 return nvme_wait_ready(ctrl, NVME_CSTS_SHST_MASK,
2346 NVME_CSTS_SHST_CMPLT,
2347 ctrl->shutdown_timeout, "shutdown");
2349 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2350 msleep(NVME_QUIRK_DELAY_AMOUNT);
2351 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, 0,
2352 (NVME_CAP_TIMEOUT(ctrl->cap) + 1) / 2, "reset");
2354 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2356 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2358 unsigned dev_page_min;
2362 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2364 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2367 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2369 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2370 dev_err(ctrl->device,
2371 "Minimum device page size %u too large for host (%u)\n",
2372 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2376 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2377 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2379 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2381 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2384 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2386 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2391 if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2392 ctrl->ctrl_config |= NVME_CC_CRIME;
2393 timeout = NVME_CRTO_CRIMT(crto);
2395 timeout = NVME_CRTO_CRWMT(crto);
2398 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2401 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2402 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2403 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2404 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2408 /* Flush write to device (required if transport is PCI) */
2409 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2413 ctrl->ctrl_config |= NVME_CC_ENABLE;
2414 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2417 return nvme_wait_ready(ctrl, NVME_CSTS_RDY, NVME_CSTS_RDY,
2418 (timeout + 1) / 2, "initialisation");
2420 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2422 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2427 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2430 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2431 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2434 dev_warn_once(ctrl->device,
2435 "could not set timestamp (%d)\n", ret);
2439 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2441 struct nvme_feat_host_behavior *host;
2442 u8 acre = 0, lbafee = 0;
2445 /* Don't bother enabling the feature if retry delay is not reported */
2447 acre = NVME_ENABLE_ACRE;
2448 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2449 lbafee = NVME_ENABLE_LBAFEE;
2451 if (!acre && !lbafee)
2454 host = kzalloc(sizeof(*host), GFP_KERNEL);
2459 host->lbafee = lbafee;
2460 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2461 host, sizeof(*host), NULL);
2467 * The function checks whether the given total (exlat + enlat) latency of
2468 * a power state allows the latter to be used as an APST transition target.
2469 * It does so by comparing the latency to the primary and secondary latency
2470 * tolerances defined by module params. If there's a match, the corresponding
2471 * timeout value is returned and the matching tolerance index (1 or 2) is
2474 static bool nvme_apst_get_transition_time(u64 total_latency,
2475 u64 *transition_time, unsigned *last_index)
2477 if (total_latency <= apst_primary_latency_tol_us) {
2478 if (*last_index == 1)
2481 *transition_time = apst_primary_timeout_ms;
2484 if (apst_secondary_timeout_ms &&
2485 total_latency <= apst_secondary_latency_tol_us) {
2486 if (*last_index <= 2)
2489 *transition_time = apst_secondary_timeout_ms;
2496 * APST (Autonomous Power State Transition) lets us program a table of power
2497 * state transitions that the controller will perform automatically.
2499 * Depending on module params, one of the two supported techniques will be used:
2501 * - If the parameters provide explicit timeouts and tolerances, they will be
2502 * used to build a table with up to 2 non-operational states to transition to.
2503 * The default parameter values were selected based on the values used by
2504 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2505 * regeneration of the APST table in the event of switching between external
2506 * and battery power, the timeouts and tolerances reflect a compromise
2507 * between values used by Microsoft for AC and battery scenarios.
2508 * - If not, we'll configure the table with a simple heuristic: we are willing
2509 * to spend at most 2% of the time transitioning between power states.
2510 * Therefore, when running in any given state, we will enter the next
2511 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2512 * microseconds, as long as that state's exit latency is under the requested
2515 * We will not autonomously enter any non-operational state for which the total
2516 * latency exceeds ps_max_latency_us.
2518 * Users can set ps_max_latency_us to zero to turn off APST.
2520 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2522 struct nvme_feat_auto_pst *table;
2529 unsigned last_lt_index = UINT_MAX;
2532 * If APST isn't supported or if we haven't been initialized yet,
2533 * then don't do anything.
2538 if (ctrl->npss > 31) {
2539 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2543 table = kzalloc(sizeof(*table), GFP_KERNEL);
2547 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2548 /* Turn off APST. */
2549 dev_dbg(ctrl->device, "APST disabled\n");
2554 * Walk through all states from lowest- to highest-power.
2555 * According to the spec, lower-numbered states use more power. NPSS,
2556 * despite the name, is the index of the lowest-power state, not the
2559 for (state = (int)ctrl->npss; state >= 0; state--) {
2560 u64 total_latency_us, exit_latency_us, transition_ms;
2563 table->entries[state] = target;
2566 * Don't allow transitions to the deepest state if it's quirked
2569 if (state == ctrl->npss &&
2570 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2574 * Is this state a useful non-operational state for higher-power
2575 * states to autonomously transition to?
2577 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2580 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2581 if (exit_latency_us > ctrl->ps_max_latency_us)
2584 total_latency_us = exit_latency_us +
2585 le32_to_cpu(ctrl->psd[state].entry_lat);
2588 * This state is good. It can be used as the APST idle target
2589 * for higher power states.
2591 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2592 if (!nvme_apst_get_transition_time(total_latency_us,
2593 &transition_ms, &last_lt_index))
2596 transition_ms = total_latency_us + 19;
2597 do_div(transition_ms, 20);
2598 if (transition_ms > (1 << 24) - 1)
2599 transition_ms = (1 << 24) - 1;
2602 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2605 if (total_latency_us > max_lat_us)
2606 max_lat_us = total_latency_us;
2610 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2612 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2613 max_ps, max_lat_us, (int)sizeof(*table), table);
2617 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2618 table, sizeof(*table), NULL);
2620 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2625 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2627 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2631 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2632 case PM_QOS_LATENCY_ANY:
2640 if (ctrl->ps_max_latency_us != latency) {
2641 ctrl->ps_max_latency_us = latency;
2642 if (ctrl->state == NVME_CTRL_LIVE)
2643 nvme_configure_apst(ctrl);
2647 struct nvme_core_quirk_entry {
2649 * NVMe model and firmware strings are padded with spaces. For
2650 * simplicity, strings in the quirk table are padded with NULLs
2656 unsigned long quirks;
2659 static const struct nvme_core_quirk_entry core_quirks[] = {
2662 * This Toshiba device seems to die using any APST states. See:
2663 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2666 .mn = "THNSF5256GPUK TOSHIBA",
2667 .quirks = NVME_QUIRK_NO_APST,
2671 * This LiteON CL1-3D*-Q11 firmware version has a race
2672 * condition associated with actions related to suspend to idle
2673 * LiteON has resolved the problem in future firmware
2677 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2681 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2682 * aborts I/O during any load, but more easily reproducible
2683 * with discards (fstrim).
2685 * The device is left in a state where it is also not possible
2686 * to use "nvme set-feature" to disable APST, but booting with
2687 * nvme_core.default_ps_max_latency=0 works.
2690 .mn = "KCD6XVUL6T40",
2691 .quirks = NVME_QUIRK_NO_APST,
2695 * The external Samsung X5 SSD fails initialization without a
2696 * delay before checking if it is ready and has a whole set of
2697 * other problems. To make this even more interesting, it
2698 * shares the PCI ID with internal Samsung 970 Evo Plus that
2699 * does not need or want these quirks.
2702 .mn = "Samsung Portable SSD X5",
2703 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2704 NVME_QUIRK_NO_DEEPEST_PS |
2705 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2709 /* match is null-terminated but idstr is space-padded. */
2710 static bool string_matches(const char *idstr, const char *match, size_t len)
2717 matchlen = strlen(match);
2718 WARN_ON_ONCE(matchlen > len);
2720 if (memcmp(idstr, match, matchlen))
2723 for (; matchlen < len; matchlen++)
2724 if (idstr[matchlen] != ' ')
2730 static bool quirk_matches(const struct nvme_id_ctrl *id,
2731 const struct nvme_core_quirk_entry *q)
2733 return q->vid == le16_to_cpu(id->vid) &&
2734 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2735 string_matches(id->fr, q->fr, sizeof(id->fr));
2738 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2739 struct nvme_id_ctrl *id)
2744 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2745 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2746 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2747 strscpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2751 if (ctrl->vs >= NVME_VS(1, 2, 1))
2752 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2756 * Generate a "fake" NQN similar to the one in Section 4.5 of the NVMe
2757 * Base Specification 2.0. It is slightly different from the format
2758 * specified there due to historic reasons, and we can't change it now.
2760 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2761 "nqn.2014.08.org.nvmexpress:%04x%04x",
2762 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2763 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2764 off += sizeof(id->sn);
2765 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2766 off += sizeof(id->mn);
2767 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2770 static void nvme_release_subsystem(struct device *dev)
2772 struct nvme_subsystem *subsys =
2773 container_of(dev, struct nvme_subsystem, dev);
2775 if (subsys->instance >= 0)
2776 ida_free(&nvme_instance_ida, subsys->instance);
2780 static void nvme_destroy_subsystem(struct kref *ref)
2782 struct nvme_subsystem *subsys =
2783 container_of(ref, struct nvme_subsystem, ref);
2785 mutex_lock(&nvme_subsystems_lock);
2786 list_del(&subsys->entry);
2787 mutex_unlock(&nvme_subsystems_lock);
2789 ida_destroy(&subsys->ns_ida);
2790 device_del(&subsys->dev);
2791 put_device(&subsys->dev);
2794 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2796 kref_put(&subsys->ref, nvme_destroy_subsystem);
2799 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2801 struct nvme_subsystem *subsys;
2803 lockdep_assert_held(&nvme_subsystems_lock);
2806 * Fail matches for discovery subsystems. This results
2807 * in each discovery controller bound to a unique subsystem.
2808 * This avoids issues with validating controller values
2809 * that can only be true when there is a single unique subsystem.
2810 * There may be multiple and completely independent entities
2811 * that provide discovery controllers.
2813 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2816 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2817 if (strcmp(subsys->subnqn, subsysnqn))
2819 if (!kref_get_unless_zero(&subsys->ref))
2827 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2828 struct device_attribute subsys_attr_##_name = \
2829 __ATTR(_name, _mode, _show, NULL)
2831 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2832 struct device_attribute *attr,
2835 struct nvme_subsystem *subsys =
2836 container_of(dev, struct nvme_subsystem, dev);
2838 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2840 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2842 static ssize_t nvme_subsys_show_type(struct device *dev,
2843 struct device_attribute *attr,
2846 struct nvme_subsystem *subsys =
2847 container_of(dev, struct nvme_subsystem, dev);
2849 switch (subsys->subtype) {
2851 return sysfs_emit(buf, "discovery\n");
2853 return sysfs_emit(buf, "nvm\n");
2855 return sysfs_emit(buf, "reserved\n");
2858 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2860 #define nvme_subsys_show_str_function(field) \
2861 static ssize_t subsys_##field##_show(struct device *dev, \
2862 struct device_attribute *attr, char *buf) \
2864 struct nvme_subsystem *subsys = \
2865 container_of(dev, struct nvme_subsystem, dev); \
2866 return sysfs_emit(buf, "%.*s\n", \
2867 (int)sizeof(subsys->field), subsys->field); \
2869 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2871 nvme_subsys_show_str_function(model);
2872 nvme_subsys_show_str_function(serial);
2873 nvme_subsys_show_str_function(firmware_rev);
2875 static struct attribute *nvme_subsys_attrs[] = {
2876 &subsys_attr_model.attr,
2877 &subsys_attr_serial.attr,
2878 &subsys_attr_firmware_rev.attr,
2879 &subsys_attr_subsysnqn.attr,
2880 &subsys_attr_subsystype.attr,
2881 #ifdef CONFIG_NVME_MULTIPATH
2882 &subsys_attr_iopolicy.attr,
2887 static const struct attribute_group nvme_subsys_attrs_group = {
2888 .attrs = nvme_subsys_attrs,
2891 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2892 &nvme_subsys_attrs_group,
2896 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2898 return ctrl->opts && ctrl->opts->discovery_nqn;
2901 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2902 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2904 struct nvme_ctrl *tmp;
2906 lockdep_assert_held(&nvme_subsystems_lock);
2908 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2909 if (nvme_state_terminal(tmp))
2912 if (tmp->cntlid == ctrl->cntlid) {
2913 dev_err(ctrl->device,
2914 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2915 ctrl->cntlid, dev_name(tmp->device),
2920 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2921 nvme_discovery_ctrl(ctrl))
2924 dev_err(ctrl->device,
2925 "Subsystem does not support multiple controllers\n");
2932 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2934 struct nvme_subsystem *subsys, *found;
2937 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2941 subsys->instance = -1;
2942 mutex_init(&subsys->lock);
2943 kref_init(&subsys->ref);
2944 INIT_LIST_HEAD(&subsys->ctrls);
2945 INIT_LIST_HEAD(&subsys->nsheads);
2946 nvme_init_subnqn(subsys, ctrl, id);
2947 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2948 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2949 subsys->vendor_id = le16_to_cpu(id->vid);
2950 subsys->cmic = id->cmic;
2952 /* Versions prior to 1.4 don't necessarily report a valid type */
2953 if (id->cntrltype == NVME_CTRL_DISC ||
2954 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2955 subsys->subtype = NVME_NQN_DISC;
2957 subsys->subtype = NVME_NQN_NVME;
2959 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2960 dev_err(ctrl->device,
2961 "Subsystem %s is not a discovery controller",
2966 subsys->awupf = le16_to_cpu(id->awupf);
2967 nvme_mpath_default_iopolicy(subsys);
2969 subsys->dev.class = nvme_subsys_class;
2970 subsys->dev.release = nvme_release_subsystem;
2971 subsys->dev.groups = nvme_subsys_attrs_groups;
2972 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2973 device_initialize(&subsys->dev);
2975 mutex_lock(&nvme_subsystems_lock);
2976 found = __nvme_find_get_subsystem(subsys->subnqn);
2978 put_device(&subsys->dev);
2981 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2983 goto out_put_subsystem;
2986 ret = device_add(&subsys->dev);
2988 dev_err(ctrl->device,
2989 "failed to register subsystem device.\n");
2990 put_device(&subsys->dev);
2993 ida_init(&subsys->ns_ida);
2994 list_add_tail(&subsys->entry, &nvme_subsystems);
2997 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2998 dev_name(ctrl->device));
3000 dev_err(ctrl->device,
3001 "failed to create sysfs link from subsystem.\n");
3002 goto out_put_subsystem;
3006 subsys->instance = ctrl->instance;
3007 ctrl->subsys = subsys;
3008 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
3009 mutex_unlock(&nvme_subsystems_lock);
3013 nvme_put_subsystem(subsys);
3015 mutex_unlock(&nvme_subsystems_lock);
3019 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
3020 void *log, size_t size, u64 offset)
3022 struct nvme_command c = { };
3023 u32 dwlen = nvme_bytes_to_numd(size);
3025 c.get_log_page.opcode = nvme_admin_get_log_page;
3026 c.get_log_page.nsid = cpu_to_le32(nsid);
3027 c.get_log_page.lid = log_page;
3028 c.get_log_page.lsp = lsp;
3029 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
3030 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
3031 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
3032 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
3033 c.get_log_page.csi = csi;
3035 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
3038 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
3039 struct nvme_effects_log **log)
3041 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
3047 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
3051 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
3052 cel, sizeof(*cel), 0);
3058 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
3064 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
3066 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
3068 if (check_shl_overflow(1U, units + page_shift - 9, &val))
3073 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
3075 struct nvme_command c = { };
3076 struct nvme_id_ctrl_nvm *id;
3079 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
3080 ctrl->max_discard_sectors = UINT_MAX;
3081 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
3083 ctrl->max_discard_sectors = 0;
3084 ctrl->max_discard_segments = 0;
3088 * Even though NVMe spec explicitly states that MDTS is not applicable
3089 * to the write-zeroes, we are cautious and limit the size to the
3090 * controllers max_hw_sectors value, which is based on the MDTS field
3091 * and possibly other limiting factors.
3093 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
3094 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
3095 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
3097 ctrl->max_zeroes_sectors = 0;
3099 if (nvme_ctrl_limited_cns(ctrl))
3102 id = kzalloc(sizeof(*id), GFP_KERNEL);
3106 c.identify.opcode = nvme_admin_identify;
3107 c.identify.cns = NVME_ID_CNS_CS_CTRL;
3108 c.identify.csi = NVME_CSI_NVM;
3110 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
3115 ctrl->max_discard_segments = id->dmrl;
3116 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
3118 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
3125 static int nvme_init_identify(struct nvme_ctrl *ctrl)
3127 struct nvme_id_ctrl *id;
3129 bool prev_apst_enabled;
3132 ret = nvme_identify_ctrl(ctrl, &id);
3134 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
3138 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
3139 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
3144 if (!(ctrl->ops->flags & NVME_F_FABRICS))
3145 ctrl->cntlid = le16_to_cpu(id->cntlid);
3147 if (!ctrl->identified) {
3151 * Check for quirks. Quirk can depend on firmware version,
3152 * so, in principle, the set of quirks present can change
3153 * across a reset. As a possible future enhancement, we
3154 * could re-scan for quirks every time we reinitialize
3155 * the device, but we'd have to make sure that the driver
3156 * behaves intelligently if the quirks change.
3158 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3159 if (quirk_matches(id, &core_quirks[i]))
3160 ctrl->quirks |= core_quirks[i].quirks;
3163 ret = nvme_init_subsystem(ctrl, id);
3167 memcpy(ctrl->subsys->firmware_rev, id->fr,
3168 sizeof(ctrl->subsys->firmware_rev));
3170 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3171 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3172 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3175 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3176 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3177 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3179 ctrl->oacs = le16_to_cpu(id->oacs);
3180 ctrl->oncs = le16_to_cpu(id->oncs);
3181 ctrl->mtfa = le16_to_cpu(id->mtfa);
3182 ctrl->oaes = le32_to_cpu(id->oaes);
3183 ctrl->wctemp = le16_to_cpu(id->wctemp);
3184 ctrl->cctemp = le16_to_cpu(id->cctemp);
3186 atomic_set(&ctrl->abort_limit, id->acl + 1);
3187 ctrl->vwc = id->vwc;
3189 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3191 max_hw_sectors = UINT_MAX;
3192 ctrl->max_hw_sectors =
3193 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3195 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3196 ctrl->sgls = le32_to_cpu(id->sgls);
3197 ctrl->kas = le16_to_cpu(id->kas);
3198 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3199 ctrl->ctratt = le32_to_cpu(id->ctratt);
3201 ctrl->cntrltype = id->cntrltype;
3202 ctrl->dctype = id->dctype;
3206 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3208 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3209 shutdown_timeout, 60);
3211 if (ctrl->shutdown_timeout != shutdown_timeout)
3212 dev_info(ctrl->device,
3213 "Shutdown timeout set to %u seconds\n",
3214 ctrl->shutdown_timeout);
3216 ctrl->shutdown_timeout = shutdown_timeout;
3218 ctrl->npss = id->npss;
3219 ctrl->apsta = id->apsta;
3220 prev_apst_enabled = ctrl->apst_enabled;
3221 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3222 if (force_apst && id->apsta) {
3223 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3224 ctrl->apst_enabled = true;
3226 ctrl->apst_enabled = false;
3229 ctrl->apst_enabled = id->apsta;
3231 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3233 if (ctrl->ops->flags & NVME_F_FABRICS) {
3234 ctrl->icdoff = le16_to_cpu(id->icdoff);
3235 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3236 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3237 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3240 * In fabrics we need to verify the cntlid matches the
3243 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3244 dev_err(ctrl->device,
3245 "Mismatching cntlid: Connect %u vs Identify "
3247 ctrl->cntlid, le16_to_cpu(id->cntlid));
3252 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3253 dev_err(ctrl->device,
3254 "keep-alive support is mandatory for fabrics\n");
3259 ctrl->hmpre = le32_to_cpu(id->hmpre);
3260 ctrl->hmmin = le32_to_cpu(id->hmmin);
3261 ctrl->hmminds = le32_to_cpu(id->hmminds);
3262 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3265 ret = nvme_mpath_init_identify(ctrl, id);
3269 if (ctrl->apst_enabled && !prev_apst_enabled)
3270 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3271 else if (!ctrl->apst_enabled && prev_apst_enabled)
3272 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3280 * Initialize the cached copies of the Identify data and various controller
3281 * register in our nvme_ctrl structure. This should be called as soon as
3282 * the admin queue is fully up and running.
3284 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl, bool was_suspended)
3288 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3290 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3294 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3296 if (ctrl->vs >= NVME_VS(1, 1, 0))
3297 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3299 ret = nvme_init_identify(ctrl);
3303 ret = nvme_configure_apst(ctrl);
3307 ret = nvme_configure_timestamp(ctrl);
3311 ret = nvme_configure_host_options(ctrl);
3315 nvme_configure_opal(ctrl, was_suspended);
3317 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3319 * Do not return errors unless we are in a controller reset,
3320 * the controller works perfectly fine without hwmon.
3322 ret = nvme_hwmon_init(ctrl);
3327 ctrl->identified = true;
3331 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3333 static int nvme_dev_open(struct inode *inode, struct file *file)
3335 struct nvme_ctrl *ctrl =
3336 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3338 switch (ctrl->state) {
3339 case NVME_CTRL_LIVE:
3342 return -EWOULDBLOCK;
3345 nvme_get_ctrl(ctrl);
3346 if (!try_module_get(ctrl->ops->module)) {
3347 nvme_put_ctrl(ctrl);
3351 file->private_data = ctrl;
3355 static int nvme_dev_release(struct inode *inode, struct file *file)
3357 struct nvme_ctrl *ctrl =
3358 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3360 module_put(ctrl->ops->module);
3361 nvme_put_ctrl(ctrl);
3365 static const struct file_operations nvme_dev_fops = {
3366 .owner = THIS_MODULE,
3367 .open = nvme_dev_open,
3368 .release = nvme_dev_release,
3369 .unlocked_ioctl = nvme_dev_ioctl,
3370 .compat_ioctl = compat_ptr_ioctl,
3371 .uring_cmd = nvme_dev_uring_cmd,
3374 static ssize_t nvme_sysfs_reset(struct device *dev,
3375 struct device_attribute *attr, const char *buf,
3378 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3381 ret = nvme_reset_ctrl_sync(ctrl);
3386 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3388 static ssize_t nvme_sysfs_rescan(struct device *dev,
3389 struct device_attribute *attr, const char *buf,
3392 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3394 nvme_queue_scan(ctrl);
3397 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3399 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3401 struct gendisk *disk = dev_to_disk(dev);
3403 if (disk->fops == &nvme_bdev_ops)
3404 return nvme_get_ns_from_dev(dev)->head;
3406 return disk->private_data;
3409 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3412 struct nvme_ns_head *head = dev_to_ns_head(dev);
3413 struct nvme_ns_ids *ids = &head->ids;
3414 struct nvme_subsystem *subsys = head->subsys;
3415 int serial_len = sizeof(subsys->serial);
3416 int model_len = sizeof(subsys->model);
3418 if (!uuid_is_null(&ids->uuid))
3419 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3421 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3422 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3424 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3425 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3427 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3428 subsys->serial[serial_len - 1] == '\0'))
3430 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3431 subsys->model[model_len - 1] == '\0'))
3434 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3435 serial_len, subsys->serial, model_len, subsys->model,
3438 static DEVICE_ATTR_RO(wwid);
3440 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3443 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3445 static DEVICE_ATTR_RO(nguid);
3447 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3450 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3452 /* For backward compatibility expose the NGUID to userspace if
3453 * we have no UUID set
3455 if (uuid_is_null(&ids->uuid)) {
3456 dev_warn_ratelimited(dev,
3457 "No UUID available providing old NGUID\n");
3458 return sysfs_emit(buf, "%pU\n", ids->nguid);
3460 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3462 static DEVICE_ATTR_RO(uuid);
3464 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3467 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3469 static DEVICE_ATTR_RO(eui);
3471 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3474 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3476 static DEVICE_ATTR_RO(nsid);
3478 static struct attribute *nvme_ns_id_attrs[] = {
3479 &dev_attr_wwid.attr,
3480 &dev_attr_uuid.attr,
3481 &dev_attr_nguid.attr,
3483 &dev_attr_nsid.attr,
3484 #ifdef CONFIG_NVME_MULTIPATH
3485 &dev_attr_ana_grpid.attr,
3486 &dev_attr_ana_state.attr,
3491 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3492 struct attribute *a, int n)
3494 struct device *dev = container_of(kobj, struct device, kobj);
3495 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3497 if (a == &dev_attr_uuid.attr) {
3498 if (uuid_is_null(&ids->uuid) &&
3499 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3502 if (a == &dev_attr_nguid.attr) {
3503 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3506 if (a == &dev_attr_eui.attr) {
3507 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3510 #ifdef CONFIG_NVME_MULTIPATH
3511 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3512 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3514 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3521 static const struct attribute_group nvme_ns_id_attr_group = {
3522 .attrs = nvme_ns_id_attrs,
3523 .is_visible = nvme_ns_id_attrs_are_visible,
3526 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3527 &nvme_ns_id_attr_group,
3531 #define nvme_show_str_function(field) \
3532 static ssize_t field##_show(struct device *dev, \
3533 struct device_attribute *attr, char *buf) \
3535 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3536 return sysfs_emit(buf, "%.*s\n", \
3537 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3539 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3541 nvme_show_str_function(model);
3542 nvme_show_str_function(serial);
3543 nvme_show_str_function(firmware_rev);
3545 #define nvme_show_int_function(field) \
3546 static ssize_t field##_show(struct device *dev, \
3547 struct device_attribute *attr, char *buf) \
3549 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3550 return sysfs_emit(buf, "%d\n", ctrl->field); \
3552 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3554 nvme_show_int_function(cntlid);
3555 nvme_show_int_function(numa_node);
3556 nvme_show_int_function(queue_count);
3557 nvme_show_int_function(sqsize);
3558 nvme_show_int_function(kato);
3560 static ssize_t nvme_sysfs_delete(struct device *dev,
3561 struct device_attribute *attr, const char *buf,
3564 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3566 if (device_remove_file_self(dev, attr))
3567 nvme_delete_ctrl_sync(ctrl);
3570 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3572 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3573 struct device_attribute *attr,
3576 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3578 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3580 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3582 static ssize_t nvme_sysfs_show_state(struct device *dev,
3583 struct device_attribute *attr,
3586 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3587 static const char *const state_name[] = {
3588 [NVME_CTRL_NEW] = "new",
3589 [NVME_CTRL_LIVE] = "live",
3590 [NVME_CTRL_RESETTING] = "resetting",
3591 [NVME_CTRL_CONNECTING] = "connecting",
3592 [NVME_CTRL_DELETING] = "deleting",
3593 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3594 [NVME_CTRL_DEAD] = "dead",
3597 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3598 state_name[ctrl->state])
3599 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3601 return sysfs_emit(buf, "unknown state\n");
3604 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3606 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3607 struct device_attribute *attr,
3610 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3612 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3614 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3616 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3617 struct device_attribute *attr,
3620 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3622 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3624 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3626 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3627 struct device_attribute *attr,
3630 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3632 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3634 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3636 static ssize_t nvme_sysfs_show_address(struct device *dev,
3637 struct device_attribute *attr,
3640 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3642 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3644 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3646 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3647 struct device_attribute *attr, char *buf)
3649 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3650 struct nvmf_ctrl_options *opts = ctrl->opts;
3652 if (ctrl->opts->max_reconnects == -1)
3653 return sysfs_emit(buf, "off\n");
3654 return sysfs_emit(buf, "%d\n",
3655 opts->max_reconnects * opts->reconnect_delay);
3658 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3659 struct device_attribute *attr, const char *buf, size_t count)
3661 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3662 struct nvmf_ctrl_options *opts = ctrl->opts;
3663 int ctrl_loss_tmo, err;
3665 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3669 if (ctrl_loss_tmo < 0)
3670 opts->max_reconnects = -1;
3672 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3673 opts->reconnect_delay);
3676 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3677 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3679 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3680 struct device_attribute *attr, char *buf)
3682 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3684 if (ctrl->opts->reconnect_delay == -1)
3685 return sysfs_emit(buf, "off\n");
3686 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3689 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3690 struct device_attribute *attr, const char *buf, size_t count)
3692 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3696 err = kstrtou32(buf, 10, &v);
3700 ctrl->opts->reconnect_delay = v;
3703 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3704 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3706 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3707 struct device_attribute *attr, char *buf)
3709 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3711 if (ctrl->opts->fast_io_fail_tmo == -1)
3712 return sysfs_emit(buf, "off\n");
3713 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3716 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3717 struct device_attribute *attr, const char *buf, size_t count)
3719 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3720 struct nvmf_ctrl_options *opts = ctrl->opts;
3721 int fast_io_fail_tmo, err;
3723 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3727 if (fast_io_fail_tmo < 0)
3728 opts->fast_io_fail_tmo = -1;
3730 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3733 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3734 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3736 static ssize_t cntrltype_show(struct device *dev,
3737 struct device_attribute *attr, char *buf)
3739 static const char * const type[] = {
3740 [NVME_CTRL_IO] = "io\n",
3741 [NVME_CTRL_DISC] = "discovery\n",
3742 [NVME_CTRL_ADMIN] = "admin\n",
3744 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3746 if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3747 return sysfs_emit(buf, "reserved\n");
3749 return sysfs_emit(buf, type[ctrl->cntrltype]);
3751 static DEVICE_ATTR_RO(cntrltype);
3753 static ssize_t dctype_show(struct device *dev,
3754 struct device_attribute *attr, char *buf)
3756 static const char * const type[] = {
3757 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3758 [NVME_DCTYPE_DDC] = "ddc\n",
3759 [NVME_DCTYPE_CDC] = "cdc\n",
3761 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3763 if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3764 return sysfs_emit(buf, "reserved\n");
3766 return sysfs_emit(buf, type[ctrl->dctype]);
3768 static DEVICE_ATTR_RO(dctype);
3770 #ifdef CONFIG_NVME_AUTH
3771 static ssize_t nvme_ctrl_dhchap_secret_show(struct device *dev,
3772 struct device_attribute *attr, char *buf)
3774 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3775 struct nvmf_ctrl_options *opts = ctrl->opts;
3777 if (!opts->dhchap_secret)
3778 return sysfs_emit(buf, "none\n");
3779 return sysfs_emit(buf, "%s\n", opts->dhchap_secret);
3782 static ssize_t nvme_ctrl_dhchap_secret_store(struct device *dev,
3783 struct device_attribute *attr, const char *buf, size_t count)
3785 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3786 struct nvmf_ctrl_options *opts = ctrl->opts;
3787 char *dhchap_secret;
3789 if (!ctrl->opts->dhchap_secret)
3793 if (memcmp(buf, "DHHC-1:", 7))
3796 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3799 memcpy(dhchap_secret, buf, count);
3800 nvme_auth_stop(ctrl);
3801 if (strcmp(dhchap_secret, opts->dhchap_secret)) {
3802 struct nvme_dhchap_key *key, *host_key;
3805 ret = nvme_auth_generate_key(dhchap_secret, &key);
3808 kfree(opts->dhchap_secret);
3809 opts->dhchap_secret = dhchap_secret;
3810 host_key = ctrl->host_key;
3811 mutex_lock(&ctrl->dhchap_auth_mutex);
3812 ctrl->host_key = key;
3813 mutex_unlock(&ctrl->dhchap_auth_mutex);
3814 nvme_auth_free_key(host_key);
3816 /* Start re-authentication */
3817 dev_info(ctrl->device, "re-authenticating controller\n");
3818 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3822 static DEVICE_ATTR(dhchap_secret, S_IRUGO | S_IWUSR,
3823 nvme_ctrl_dhchap_secret_show, nvme_ctrl_dhchap_secret_store);
3825 static ssize_t nvme_ctrl_dhchap_ctrl_secret_show(struct device *dev,
3826 struct device_attribute *attr, char *buf)
3828 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3829 struct nvmf_ctrl_options *opts = ctrl->opts;
3831 if (!opts->dhchap_ctrl_secret)
3832 return sysfs_emit(buf, "none\n");
3833 return sysfs_emit(buf, "%s\n", opts->dhchap_ctrl_secret);
3836 static ssize_t nvme_ctrl_dhchap_ctrl_secret_store(struct device *dev,
3837 struct device_attribute *attr, const char *buf, size_t count)
3839 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3840 struct nvmf_ctrl_options *opts = ctrl->opts;
3841 char *dhchap_secret;
3843 if (!ctrl->opts->dhchap_ctrl_secret)
3847 if (memcmp(buf, "DHHC-1:", 7))
3850 dhchap_secret = kzalloc(count + 1, GFP_KERNEL);
3853 memcpy(dhchap_secret, buf, count);
3854 nvme_auth_stop(ctrl);
3855 if (strcmp(dhchap_secret, opts->dhchap_ctrl_secret)) {
3856 struct nvme_dhchap_key *key, *ctrl_key;
3859 ret = nvme_auth_generate_key(dhchap_secret, &key);
3862 kfree(opts->dhchap_ctrl_secret);
3863 opts->dhchap_ctrl_secret = dhchap_secret;
3864 ctrl_key = ctrl->ctrl_key;
3865 mutex_lock(&ctrl->dhchap_auth_mutex);
3866 ctrl->ctrl_key = key;
3867 mutex_unlock(&ctrl->dhchap_auth_mutex);
3868 nvme_auth_free_key(ctrl_key);
3870 /* Start re-authentication */
3871 dev_info(ctrl->device, "re-authenticating controller\n");
3872 queue_work(nvme_wq, &ctrl->dhchap_auth_work);
3876 static DEVICE_ATTR(dhchap_ctrl_secret, S_IRUGO | S_IWUSR,
3877 nvme_ctrl_dhchap_ctrl_secret_show, nvme_ctrl_dhchap_ctrl_secret_store);
3880 static struct attribute *nvme_dev_attrs[] = {
3881 &dev_attr_reset_controller.attr,
3882 &dev_attr_rescan_controller.attr,
3883 &dev_attr_model.attr,
3884 &dev_attr_serial.attr,
3885 &dev_attr_firmware_rev.attr,
3886 &dev_attr_cntlid.attr,
3887 &dev_attr_delete_controller.attr,
3888 &dev_attr_transport.attr,
3889 &dev_attr_subsysnqn.attr,
3890 &dev_attr_address.attr,
3891 &dev_attr_state.attr,
3892 &dev_attr_numa_node.attr,
3893 &dev_attr_queue_count.attr,
3894 &dev_attr_sqsize.attr,
3895 &dev_attr_hostnqn.attr,
3896 &dev_attr_hostid.attr,
3897 &dev_attr_ctrl_loss_tmo.attr,
3898 &dev_attr_reconnect_delay.attr,
3899 &dev_attr_fast_io_fail_tmo.attr,
3900 &dev_attr_kato.attr,
3901 &dev_attr_cntrltype.attr,
3902 &dev_attr_dctype.attr,
3903 #ifdef CONFIG_NVME_AUTH
3904 &dev_attr_dhchap_secret.attr,
3905 &dev_attr_dhchap_ctrl_secret.attr,
3910 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3911 struct attribute *a, int n)
3913 struct device *dev = container_of(kobj, struct device, kobj);
3914 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3916 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3918 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3920 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3922 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3924 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3926 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3928 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3930 #ifdef CONFIG_NVME_AUTH
3931 if (a == &dev_attr_dhchap_secret.attr && !ctrl->opts)
3933 if (a == &dev_attr_dhchap_ctrl_secret.attr && !ctrl->opts)
3940 const struct attribute_group nvme_dev_attrs_group = {
3941 .attrs = nvme_dev_attrs,
3942 .is_visible = nvme_dev_attrs_are_visible,
3944 EXPORT_SYMBOL_GPL(nvme_dev_attrs_group);
3946 static const struct attribute_group *nvme_dev_attr_groups[] = {
3947 &nvme_dev_attrs_group,
3951 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3954 struct nvme_ns_head *h;
3956 lockdep_assert_held(&ctrl->subsys->lock);
3958 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3960 * Private namespaces can share NSIDs under some conditions.
3961 * In that case we can't use the same ns_head for namespaces
3962 * with the same NSID.
3964 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3966 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3973 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3974 struct nvme_ns_ids *ids)
3976 bool has_uuid = !uuid_is_null(&ids->uuid);
3977 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3978 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3979 struct nvme_ns_head *h;
3981 lockdep_assert_held(&subsys->lock);
3983 list_for_each_entry(h, &subsys->nsheads, entry) {
3984 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3987 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3990 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3997 static void nvme_cdev_rel(struct device *dev)
3999 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
4002 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
4004 cdev_device_del(cdev, cdev_device);
4005 put_device(cdev_device);
4008 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
4009 const struct file_operations *fops, struct module *owner)
4013 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
4016 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
4017 cdev_device->class = nvme_ns_chr_class;
4018 cdev_device->release = nvme_cdev_rel;
4019 device_initialize(cdev_device);
4020 cdev_init(cdev, fops);
4021 cdev->owner = owner;
4022 ret = cdev_device_add(cdev, cdev_device);
4024 put_device(cdev_device);
4029 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
4031 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
4034 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
4036 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
4040 static const struct file_operations nvme_ns_chr_fops = {
4041 .owner = THIS_MODULE,
4042 .open = nvme_ns_chr_open,
4043 .release = nvme_ns_chr_release,
4044 .unlocked_ioctl = nvme_ns_chr_ioctl,
4045 .compat_ioctl = compat_ptr_ioctl,
4046 .uring_cmd = nvme_ns_chr_uring_cmd,
4047 .uring_cmd_iopoll = nvme_ns_chr_uring_cmd_iopoll,
4050 static int nvme_add_ns_cdev(struct nvme_ns *ns)
4054 ns->cdev_device.parent = ns->ctrl->device;
4055 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
4056 ns->ctrl->instance, ns->head->instance);
4060 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
4061 ns->ctrl->ops->module);
4064 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
4065 struct nvme_ns_info *info)
4067 struct nvme_ns_head *head;
4068 size_t size = sizeof(*head);
4071 #ifdef CONFIG_NVME_MULTIPATH
4072 size += num_possible_nodes() * sizeof(struct nvme_ns *);
4075 head = kzalloc(size, GFP_KERNEL);
4078 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
4081 head->instance = ret;
4082 INIT_LIST_HEAD(&head->list);
4083 ret = init_srcu_struct(&head->srcu);
4085 goto out_ida_remove;
4086 head->subsys = ctrl->subsys;
4087 head->ns_id = info->nsid;
4088 head->ids = info->ids;
4089 head->shared = info->is_shared;
4090 kref_init(&head->ref);
4092 if (head->ids.csi) {
4093 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
4095 goto out_cleanup_srcu;
4097 head->effects = ctrl->effects;
4099 ret = nvme_mpath_alloc_disk(ctrl, head);
4101 goto out_cleanup_srcu;
4103 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
4105 kref_get(&ctrl->subsys->ref);
4109 cleanup_srcu_struct(&head->srcu);
4111 ida_free(&ctrl->subsys->ns_ida, head->instance);
4116 ret = blk_status_to_errno(nvme_error_status(ret));
4117 return ERR_PTR(ret);
4120 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
4121 struct nvme_ns_ids *ids)
4123 struct nvme_subsystem *s;
4127 * Note that this check is racy as we try to avoid holding the global
4128 * lock over the whole ns_head creation. But it is only intended as
4129 * a sanity check anyway.
4131 mutex_lock(&nvme_subsystems_lock);
4132 list_for_each_entry(s, &nvme_subsystems, entry) {
4135 mutex_lock(&s->lock);
4136 ret = nvme_subsys_check_duplicate_ids(s, ids);
4137 mutex_unlock(&s->lock);
4141 mutex_unlock(&nvme_subsystems_lock);
4146 static int nvme_init_ns_head(struct nvme_ns *ns, struct nvme_ns_info *info)
4148 struct nvme_ctrl *ctrl = ns->ctrl;
4149 struct nvme_ns_head *head = NULL;
4152 ret = nvme_global_check_duplicate_ids(ctrl->subsys, &info->ids);
4154 dev_err(ctrl->device,
4155 "globally duplicate IDs for nsid %d\n", info->nsid);
4156 nvme_print_device_info(ctrl);
4160 mutex_lock(&ctrl->subsys->lock);
4161 head = nvme_find_ns_head(ctrl, info->nsid);
4163 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, &info->ids);
4165 dev_err(ctrl->device,
4166 "duplicate IDs in subsystem for nsid %d\n",
4170 head = nvme_alloc_ns_head(ctrl, info);
4172 ret = PTR_ERR(head);
4177 if (!info->is_shared || !head->shared) {
4178 dev_err(ctrl->device,
4179 "Duplicate unshared namespace %d\n",
4181 goto out_put_ns_head;
4183 if (!nvme_ns_ids_equal(&head->ids, &info->ids)) {
4184 dev_err(ctrl->device,
4185 "IDs don't match for shared namespace %d\n",
4187 goto out_put_ns_head;
4190 if (!multipath && !list_empty(&head->list)) {
4191 dev_warn(ctrl->device,
4192 "Found shared namespace %d, but multipathing not supported.\n",
4194 dev_warn_once(ctrl->device,
4195 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
4199 list_add_tail_rcu(&ns->siblings, &head->list);
4201 mutex_unlock(&ctrl->subsys->lock);
4205 nvme_put_ns_head(head);
4207 mutex_unlock(&ctrl->subsys->lock);
4211 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4213 struct nvme_ns *ns, *ret = NULL;
4215 down_read(&ctrl->namespaces_rwsem);
4216 list_for_each_entry(ns, &ctrl->namespaces, list) {
4217 if (ns->head->ns_id == nsid) {
4218 if (!nvme_get_ns(ns))
4223 if (ns->head->ns_id > nsid)
4226 up_read(&ctrl->namespaces_rwsem);
4229 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
4232 * Add the namespace to the controller list while keeping the list ordered.
4234 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
4236 struct nvme_ns *tmp;
4238 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
4239 if (tmp->head->ns_id < ns->head->ns_id) {
4240 list_add(&ns->list, &tmp->list);
4244 list_add(&ns->list, &ns->ctrl->namespaces);
4247 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, struct nvme_ns_info *info)
4250 struct gendisk *disk;
4251 int node = ctrl->numa_node;
4253 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
4257 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
4260 disk->fops = &nvme_bdev_ops;
4261 disk->private_data = ns;
4264 ns->queue = disk->queue;
4266 if (ctrl->opts && ctrl->opts->data_digest)
4267 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
4269 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
4270 if (ctrl->ops->supports_pci_p2pdma &&
4271 ctrl->ops->supports_pci_p2pdma(ctrl))
4272 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4275 kref_init(&ns->kref);
4277 if (nvme_init_ns_head(ns, info))
4278 goto out_cleanup_disk;
4281 * If multipathing is enabled, the device name for all disks and not
4282 * just those that represent shared namespaces needs to be based on the
4283 * subsystem instance. Using the controller instance for private
4284 * namespaces could lead to naming collisions between shared and private
4285 * namespaces if they don't use a common numbering scheme.
4287 * If multipathing is not enabled, disk names must use the controller
4288 * instance as shared namespaces will show up as multiple block
4291 if (ns->head->disk) {
4292 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4293 ctrl->instance, ns->head->instance);
4294 disk->flags |= GENHD_FL_HIDDEN;
4295 } else if (multipath) {
4296 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4297 ns->head->instance);
4299 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4300 ns->head->instance);
4303 if (nvme_update_ns_info(ns, info))
4306 down_write(&ctrl->namespaces_rwsem);
4307 nvme_ns_add_to_ctrl_list(ns);
4308 up_write(&ctrl->namespaces_rwsem);
4309 nvme_get_ctrl(ctrl);
4311 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4312 goto out_cleanup_ns_from_list;
4314 if (!nvme_ns_head_multipath(ns->head))
4315 nvme_add_ns_cdev(ns);
4317 nvme_mpath_add_disk(ns, info->anagrpid);
4318 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4322 out_cleanup_ns_from_list:
4323 nvme_put_ctrl(ctrl);
4324 down_write(&ctrl->namespaces_rwsem);
4325 list_del_init(&ns->list);
4326 up_write(&ctrl->namespaces_rwsem);
4328 mutex_lock(&ctrl->subsys->lock);
4329 list_del_rcu(&ns->siblings);
4330 if (list_empty(&ns->head->list))
4331 list_del_init(&ns->head->entry);
4332 mutex_unlock(&ctrl->subsys->lock);
4333 nvme_put_ns_head(ns->head);
4340 static void nvme_ns_remove(struct nvme_ns *ns)
4342 bool last_path = false;
4344 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4347 clear_bit(NVME_NS_READY, &ns->flags);
4348 set_capacity(ns->disk, 0);
4349 nvme_fault_inject_fini(&ns->fault_inject);
4352 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4353 * this ns going back into current_path.
4355 synchronize_srcu(&ns->head->srcu);
4357 /* wait for concurrent submissions */
4358 if (nvme_mpath_clear_current_path(ns))
4359 synchronize_srcu(&ns->head->srcu);
4361 mutex_lock(&ns->ctrl->subsys->lock);
4362 list_del_rcu(&ns->siblings);
4363 if (list_empty(&ns->head->list)) {
4364 list_del_init(&ns->head->entry);
4367 mutex_unlock(&ns->ctrl->subsys->lock);
4369 /* guarantee not available in head->list */
4370 synchronize_srcu(&ns->head->srcu);
4372 if (!nvme_ns_head_multipath(ns->head))
4373 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4374 del_gendisk(ns->disk);
4376 down_write(&ns->ctrl->namespaces_rwsem);
4377 list_del_init(&ns->list);
4378 up_write(&ns->ctrl->namespaces_rwsem);
4381 nvme_mpath_shutdown_disk(ns->head);
4385 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4387 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4395 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_info *info)
4397 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4399 if (!nvme_ns_ids_equal(&ns->head->ids, &info->ids)) {
4400 dev_err(ns->ctrl->device,
4401 "identifiers changed for nsid %d\n", ns->head->ns_id);
4405 ret = nvme_update_ns_info(ns, info);
4408 * Only remove the namespace if we got a fatal error back from the
4409 * device, otherwise ignore the error and just move on.
4411 * TODO: we should probably schedule a delayed retry here.
4413 if (ret > 0 && (ret & NVME_SC_DNR))
4417 static void nvme_scan_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4419 struct nvme_ns_info info = { .nsid = nsid };
4422 if (nvme_identify_ns_descs(ctrl, &info))
4425 if (info.ids.csi != NVME_CSI_NVM && !nvme_multi_css(ctrl)) {
4426 dev_warn(ctrl->device,
4427 "command set not reported for nsid: %d\n", nsid);
4432 * If available try to use the Command Set Idependent Identify Namespace
4433 * data structure to find all the generic information that is needed to
4434 * set up a namespace. If not fall back to the legacy version.
4436 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) ||
4437 (info.ids.csi != NVME_CSI_NVM && info.ids.csi != NVME_CSI_ZNS)) {
4438 if (nvme_ns_info_from_id_cs_indep(ctrl, &info))
4441 if (nvme_ns_info_from_identify(ctrl, &info))
4446 * Ignore the namespace if it is not ready. We will get an AEN once it
4447 * becomes ready and restart the scan.
4452 ns = nvme_find_get_ns(ctrl, nsid);
4454 nvme_validate_ns(ns, &info);
4457 nvme_alloc_ns(ctrl, &info);
4461 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4464 struct nvme_ns *ns, *next;
4467 down_write(&ctrl->namespaces_rwsem);
4468 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4469 if (ns->head->ns_id > nsid)
4470 list_move_tail(&ns->list, &rm_list);
4472 up_write(&ctrl->namespaces_rwsem);
4474 list_for_each_entry_safe(ns, next, &rm_list, list)
4479 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4481 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4486 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4491 struct nvme_command cmd = {
4492 .identify.opcode = nvme_admin_identify,
4493 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4494 .identify.nsid = cpu_to_le32(prev),
4497 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4498 NVME_IDENTIFY_DATA_SIZE);
4500 dev_warn(ctrl->device,
4501 "Identify NS List failed (status=0x%x)\n", ret);
4505 for (i = 0; i < nr_entries; i++) {
4506 u32 nsid = le32_to_cpu(ns_list[i]);
4508 if (!nsid) /* end of the list? */
4510 nvme_scan_ns(ctrl, nsid);
4511 while (++prev < nsid)
4512 nvme_ns_remove_by_nsid(ctrl, prev);
4516 nvme_remove_invalid_namespaces(ctrl, prev);
4522 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4524 struct nvme_id_ctrl *id;
4527 if (nvme_identify_ctrl(ctrl, &id))
4529 nn = le32_to_cpu(id->nn);
4532 for (i = 1; i <= nn; i++)
4533 nvme_scan_ns(ctrl, i);
4535 nvme_remove_invalid_namespaces(ctrl, nn);
4538 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4540 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4544 log = kzalloc(log_size, GFP_KERNEL);
4549 * We need to read the log to clear the AEN, but we don't want to rely
4550 * on it for the changed namespace information as userspace could have
4551 * raced with us in reading the log page, which could cause us to miss
4554 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4555 NVME_CSI_NVM, log, log_size, 0);
4557 dev_warn(ctrl->device,
4558 "reading changed ns log failed: %d\n", error);
4563 static void nvme_scan_work(struct work_struct *work)
4565 struct nvme_ctrl *ctrl =
4566 container_of(work, struct nvme_ctrl, scan_work);
4569 /* No tagset on a live ctrl means IO queues could not created */
4570 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4574 * Identify controller limits can change at controller reset due to
4575 * new firmware download, even though it is not common we cannot ignore
4576 * such scenario. Controller's non-mdts limits are reported in the unit
4577 * of logical blocks that is dependent on the format of attached
4578 * namespace. Hence re-read the limits at the time of ns allocation.
4580 ret = nvme_init_non_mdts_limits(ctrl);
4582 dev_warn(ctrl->device,
4583 "reading non-mdts-limits failed: %d\n", ret);
4587 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4588 dev_info(ctrl->device, "rescanning namespaces.\n");
4589 nvme_clear_changed_ns_log(ctrl);
4592 mutex_lock(&ctrl->scan_lock);
4593 if (nvme_ctrl_limited_cns(ctrl)) {
4594 nvme_scan_ns_sequential(ctrl);
4597 * Fall back to sequential scan if DNR is set to handle broken
4598 * devices which should support Identify NS List (as per the VS
4599 * they report) but don't actually support it.
4601 ret = nvme_scan_ns_list(ctrl);
4602 if (ret > 0 && ret & NVME_SC_DNR)
4603 nvme_scan_ns_sequential(ctrl);
4605 mutex_unlock(&ctrl->scan_lock);
4609 * This function iterates the namespace list unlocked to allow recovery from
4610 * controller failure. It is up to the caller to ensure the namespace list is
4611 * not modified by scan work while this function is executing.
4613 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4615 struct nvme_ns *ns, *next;
4619 * make sure to requeue I/O to all namespaces as these
4620 * might result from the scan itself and must complete
4621 * for the scan_work to make progress
4623 nvme_mpath_clear_ctrl_paths(ctrl);
4625 /* prevent racing with ns scanning */
4626 flush_work(&ctrl->scan_work);
4629 * The dead states indicates the controller was not gracefully
4630 * disconnected. In that case, we won't be able to flush any data while
4631 * removing the namespaces' disks; fail all the queues now to avoid
4632 * potentially having to clean up the failed sync later.
4634 if (ctrl->state == NVME_CTRL_DEAD) {
4635 nvme_mark_namespaces_dead(ctrl);
4636 nvme_unquiesce_io_queues(ctrl);
4639 /* this is a no-op when called from the controller reset handler */
4640 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4642 down_write(&ctrl->namespaces_rwsem);
4643 list_splice_init(&ctrl->namespaces, &ns_list);
4644 up_write(&ctrl->namespaces_rwsem);
4646 list_for_each_entry_safe(ns, next, &ns_list, list)
4649 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4651 static int nvme_class_uevent(const struct device *dev, struct kobj_uevent_env *env)
4653 const struct nvme_ctrl *ctrl =
4654 container_of(dev, struct nvme_ctrl, ctrl_device);
4655 struct nvmf_ctrl_options *opts = ctrl->opts;
4658 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4663 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4667 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4668 opts->trsvcid ?: "none");
4672 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4673 opts->host_traddr ?: "none");
4677 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4678 opts->host_iface ?: "none");
4683 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4685 char *envp[2] = { envdata, NULL };
4687 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4690 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4692 char *envp[2] = { NULL, NULL };
4693 u32 aen_result = ctrl->aen_result;
4695 ctrl->aen_result = 0;
4699 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4702 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4706 static void nvme_async_event_work(struct work_struct *work)
4708 struct nvme_ctrl *ctrl =
4709 container_of(work, struct nvme_ctrl, async_event_work);
4711 nvme_aen_uevent(ctrl);
4714 * The transport drivers must guarantee AER submission here is safe by
4715 * flushing ctrl async_event_work after changing the controller state
4716 * from LIVE and before freeing the admin queue.
4718 if (ctrl->state == NVME_CTRL_LIVE)
4719 ctrl->ops->submit_async_event(ctrl);
4722 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4727 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4733 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4736 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4738 struct nvme_fw_slot_info_log *log;
4740 log = kmalloc(sizeof(*log), GFP_KERNEL);
4744 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4745 log, sizeof(*log), 0))
4746 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4750 static void nvme_fw_act_work(struct work_struct *work)
4752 struct nvme_ctrl *ctrl = container_of(work,
4753 struct nvme_ctrl, fw_act_work);
4754 unsigned long fw_act_timeout;
4757 fw_act_timeout = jiffies +
4758 msecs_to_jiffies(ctrl->mtfa * 100);
4760 fw_act_timeout = jiffies +
4761 msecs_to_jiffies(admin_timeout * 1000);
4763 nvme_quiesce_io_queues(ctrl);
4764 while (nvme_ctrl_pp_status(ctrl)) {
4765 if (time_after(jiffies, fw_act_timeout)) {
4766 dev_warn(ctrl->device,
4767 "Fw activation timeout, reset controller\n");
4768 nvme_try_sched_reset(ctrl);
4774 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4777 nvme_unquiesce_io_queues(ctrl);
4778 /* read FW slot information to clear the AER */
4779 nvme_get_fw_slot_info(ctrl);
4781 queue_work(nvme_wq, &ctrl->async_event_work);
4784 static u32 nvme_aer_type(u32 result)
4786 return result & 0x7;
4789 static u32 nvme_aer_subtype(u32 result)
4791 return (result & 0xff00) >> 8;
4794 static bool nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4796 u32 aer_notice_type = nvme_aer_subtype(result);
4797 bool requeue = true;
4799 trace_nvme_async_event(ctrl, aer_notice_type);
4801 switch (aer_notice_type) {
4802 case NVME_AER_NOTICE_NS_CHANGED:
4803 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4804 nvme_queue_scan(ctrl);
4806 case NVME_AER_NOTICE_FW_ACT_STARTING:
4808 * We are (ab)using the RESETTING state to prevent subsequent
4809 * recovery actions from interfering with the controller's
4810 * firmware activation.
4812 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING)) {
4813 nvme_auth_stop(ctrl);
4815 queue_work(nvme_wq, &ctrl->fw_act_work);
4818 #ifdef CONFIG_NVME_MULTIPATH
4819 case NVME_AER_NOTICE_ANA:
4820 if (!ctrl->ana_log_buf)
4822 queue_work(nvme_wq, &ctrl->ana_work);
4825 case NVME_AER_NOTICE_DISC_CHANGED:
4826 ctrl->aen_result = result;
4829 dev_warn(ctrl->device, "async event result %08x\n", result);
4834 static void nvme_handle_aer_persistent_error(struct nvme_ctrl *ctrl)
4836 trace_nvme_async_event(ctrl, NVME_AER_ERROR);
4837 dev_warn(ctrl->device, "resetting controller due to AER\n");
4838 nvme_reset_ctrl(ctrl);
4841 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4842 volatile union nvme_result *res)
4844 u32 result = le32_to_cpu(res->u32);
4845 u32 aer_type = nvme_aer_type(result);
4846 u32 aer_subtype = nvme_aer_subtype(result);
4847 bool requeue = true;
4849 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4853 case NVME_AER_NOTICE:
4854 requeue = nvme_handle_aen_notice(ctrl, result);
4856 case NVME_AER_ERROR:
4858 * For a persistent internal error, don't run async_event_work
4859 * to submit a new AER. The controller reset will do it.
4861 if (aer_subtype == NVME_AER_ERROR_PERSIST_INT_ERR) {
4862 nvme_handle_aer_persistent_error(ctrl);
4866 case NVME_AER_SMART:
4869 trace_nvme_async_event(ctrl, aer_type);
4870 ctrl->aen_result = result;
4877 queue_work(nvme_wq, &ctrl->async_event_work);
4879 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4881 int nvme_alloc_admin_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4882 const struct blk_mq_ops *ops, unsigned int cmd_size)
4886 memset(set, 0, sizeof(*set));
4888 set->queue_depth = NVME_AQ_MQ_TAG_DEPTH;
4889 if (ctrl->ops->flags & NVME_F_FABRICS)
4890 set->reserved_tags = NVMF_RESERVED_TAGS;
4891 set->numa_node = ctrl->numa_node;
4892 set->flags = BLK_MQ_F_NO_SCHED;
4893 if (ctrl->ops->flags & NVME_F_BLOCKING)
4894 set->flags |= BLK_MQ_F_BLOCKING;
4895 set->cmd_size = cmd_size;
4896 set->driver_data = ctrl;
4897 set->nr_hw_queues = 1;
4898 set->timeout = NVME_ADMIN_TIMEOUT;
4899 ret = blk_mq_alloc_tag_set(set);
4903 ctrl->admin_q = blk_mq_init_queue(set);
4904 if (IS_ERR(ctrl->admin_q)) {
4905 ret = PTR_ERR(ctrl->admin_q);
4906 goto out_free_tagset;
4909 if (ctrl->ops->flags & NVME_F_FABRICS) {
4910 ctrl->fabrics_q = blk_mq_init_queue(set);
4911 if (IS_ERR(ctrl->fabrics_q)) {
4912 ret = PTR_ERR(ctrl->fabrics_q);
4913 goto out_cleanup_admin_q;
4917 ctrl->admin_tagset = set;
4920 out_cleanup_admin_q:
4921 blk_mq_destroy_queue(ctrl->admin_q);
4922 blk_put_queue(ctrl->admin_q);
4924 blk_mq_free_tag_set(set);
4925 ctrl->admin_q = NULL;
4926 ctrl->fabrics_q = NULL;
4929 EXPORT_SYMBOL_GPL(nvme_alloc_admin_tag_set);
4931 void nvme_remove_admin_tag_set(struct nvme_ctrl *ctrl)
4933 blk_mq_destroy_queue(ctrl->admin_q);
4934 blk_put_queue(ctrl->admin_q);
4935 if (ctrl->ops->flags & NVME_F_FABRICS) {
4936 blk_mq_destroy_queue(ctrl->fabrics_q);
4937 blk_put_queue(ctrl->fabrics_q);
4939 blk_mq_free_tag_set(ctrl->admin_tagset);
4941 EXPORT_SYMBOL_GPL(nvme_remove_admin_tag_set);
4943 int nvme_alloc_io_tag_set(struct nvme_ctrl *ctrl, struct blk_mq_tag_set *set,
4944 const struct blk_mq_ops *ops, unsigned int nr_maps,
4945 unsigned int cmd_size)
4949 memset(set, 0, sizeof(*set));
4951 set->queue_depth = min_t(unsigned, ctrl->sqsize, BLK_MQ_MAX_DEPTH - 1);
4953 * Some Apple controllers requires tags to be unique across admin and
4954 * the (only) I/O queue, so reserve the first 32 tags of the I/O queue.
4956 if (ctrl->quirks & NVME_QUIRK_SHARED_TAGS)
4957 set->reserved_tags = NVME_AQ_DEPTH;
4958 else if (ctrl->ops->flags & NVME_F_FABRICS)
4959 set->reserved_tags = NVMF_RESERVED_TAGS;
4960 set->numa_node = ctrl->numa_node;
4961 set->flags = BLK_MQ_F_SHOULD_MERGE;
4962 if (ctrl->ops->flags & NVME_F_BLOCKING)
4963 set->flags |= BLK_MQ_F_BLOCKING;
4964 set->cmd_size = cmd_size,
4965 set->driver_data = ctrl;
4966 set->nr_hw_queues = ctrl->queue_count - 1;
4967 set->timeout = NVME_IO_TIMEOUT;
4968 set->nr_maps = nr_maps;
4969 ret = blk_mq_alloc_tag_set(set);
4973 if (ctrl->ops->flags & NVME_F_FABRICS) {
4974 ctrl->connect_q = blk_mq_init_queue(set);
4975 if (IS_ERR(ctrl->connect_q)) {
4976 ret = PTR_ERR(ctrl->connect_q);
4977 goto out_free_tag_set;
4979 blk_queue_flag_set(QUEUE_FLAG_SKIP_TAGSET_QUIESCE,
4987 blk_mq_free_tag_set(set);
4988 ctrl->connect_q = NULL;
4991 EXPORT_SYMBOL_GPL(nvme_alloc_io_tag_set);
4993 void nvme_remove_io_tag_set(struct nvme_ctrl *ctrl)
4995 if (ctrl->ops->flags & NVME_F_FABRICS) {
4996 blk_mq_destroy_queue(ctrl->connect_q);
4997 blk_put_queue(ctrl->connect_q);
4999 blk_mq_free_tag_set(ctrl->tagset);
5001 EXPORT_SYMBOL_GPL(nvme_remove_io_tag_set);
5003 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
5005 nvme_mpath_stop(ctrl);
5006 nvme_auth_stop(ctrl);
5007 nvme_stop_keep_alive(ctrl);
5008 nvme_stop_failfast_work(ctrl);
5009 flush_work(&ctrl->async_event_work);
5010 cancel_work_sync(&ctrl->fw_act_work);
5011 if (ctrl->ops->stop_ctrl)
5012 ctrl->ops->stop_ctrl(ctrl);
5014 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
5016 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
5018 nvme_start_keep_alive(ctrl);
5020 nvme_enable_aen(ctrl);
5023 * persistent discovery controllers need to send indication to userspace
5024 * to re-read the discovery log page to learn about possible changes
5025 * that were missed. We identify persistent discovery controllers by
5026 * checking that they started once before, hence are reconnecting back.
5028 if (test_and_set_bit(NVME_CTRL_STARTED_ONCE, &ctrl->flags) &&
5029 nvme_discovery_ctrl(ctrl))
5030 nvme_change_uevent(ctrl, "NVME_EVENT=rediscover");
5032 if (ctrl->queue_count > 1) {
5033 nvme_queue_scan(ctrl);
5034 nvme_unquiesce_io_queues(ctrl);
5035 nvme_mpath_update(ctrl);
5038 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
5040 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
5042 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
5044 nvme_hwmon_exit(ctrl);
5045 nvme_fault_inject_fini(&ctrl->fault_inject);
5046 dev_pm_qos_hide_latency_tolerance(ctrl->device);
5047 cdev_device_del(&ctrl->cdev, ctrl->device);
5048 nvme_put_ctrl(ctrl);
5050 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
5052 static void nvme_free_cels(struct nvme_ctrl *ctrl)
5054 struct nvme_effects_log *cel;
5057 xa_for_each(&ctrl->cels, i, cel) {
5058 xa_erase(&ctrl->cels, i);
5062 xa_destroy(&ctrl->cels);
5065 static void nvme_free_ctrl(struct device *dev)
5067 struct nvme_ctrl *ctrl =
5068 container_of(dev, struct nvme_ctrl, ctrl_device);
5069 struct nvme_subsystem *subsys = ctrl->subsys;
5071 if (!subsys || ctrl->instance != subsys->instance)
5072 ida_free(&nvme_instance_ida, ctrl->instance);
5074 nvme_free_cels(ctrl);
5075 nvme_mpath_uninit(ctrl);
5076 nvme_auth_stop(ctrl);
5077 nvme_auth_free(ctrl);
5078 __free_page(ctrl->discard_page);
5079 free_opal_dev(ctrl->opal_dev);
5082 mutex_lock(&nvme_subsystems_lock);
5083 list_del(&ctrl->subsys_entry);
5084 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
5085 mutex_unlock(&nvme_subsystems_lock);
5088 ctrl->ops->free_ctrl(ctrl);
5091 nvme_put_subsystem(subsys);
5095 * Initialize a NVMe controller structures. This needs to be called during
5096 * earliest initialization so that we have the initialized structured around
5099 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
5100 const struct nvme_ctrl_ops *ops, unsigned long quirks)
5104 ctrl->state = NVME_CTRL_NEW;
5105 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
5106 spin_lock_init(&ctrl->lock);
5107 mutex_init(&ctrl->scan_lock);
5108 INIT_LIST_HEAD(&ctrl->namespaces);
5109 xa_init(&ctrl->cels);
5110 init_rwsem(&ctrl->namespaces_rwsem);
5113 ctrl->quirks = quirks;
5114 ctrl->numa_node = NUMA_NO_NODE;
5115 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
5116 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
5117 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
5118 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
5119 init_waitqueue_head(&ctrl->state_wq);
5121 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
5122 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
5123 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
5124 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
5126 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
5128 ctrl->discard_page = alloc_page(GFP_KERNEL);
5129 if (!ctrl->discard_page) {
5134 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
5137 ctrl->instance = ret;
5139 device_initialize(&ctrl->ctrl_device);
5140 ctrl->device = &ctrl->ctrl_device;
5141 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
5143 ctrl->device->class = nvme_class;
5144 ctrl->device->parent = ctrl->dev;
5145 if (ops->dev_attr_groups)
5146 ctrl->device->groups = ops->dev_attr_groups;
5148 ctrl->device->groups = nvme_dev_attr_groups;
5149 ctrl->device->release = nvme_free_ctrl;
5150 dev_set_drvdata(ctrl->device, ctrl);
5151 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
5153 goto out_release_instance;
5155 nvme_get_ctrl(ctrl);
5156 cdev_init(&ctrl->cdev, &nvme_dev_fops);
5157 ctrl->cdev.owner = ops->module;
5158 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
5163 * Initialize latency tolerance controls. The sysfs files won't
5164 * be visible to userspace unless the device actually supports APST.
5166 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
5167 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
5168 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
5170 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
5171 nvme_mpath_init_ctrl(ctrl);
5172 ret = nvme_auth_init_ctrl(ctrl);
5178 cdev_device_del(&ctrl->cdev, ctrl->device);
5180 nvme_put_ctrl(ctrl);
5181 kfree_const(ctrl->device->kobj.name);
5182 out_release_instance:
5183 ida_free(&nvme_instance_ida, ctrl->instance);
5185 if (ctrl->discard_page)
5186 __free_page(ctrl->discard_page);
5189 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
5191 /* let I/O to all namespaces fail in preparation for surprise removal */
5192 void nvme_mark_namespaces_dead(struct nvme_ctrl *ctrl)
5196 down_read(&ctrl->namespaces_rwsem);
5197 list_for_each_entry(ns, &ctrl->namespaces, list)
5198 blk_mark_disk_dead(ns->disk);
5199 up_read(&ctrl->namespaces_rwsem);
5201 EXPORT_SYMBOL_GPL(nvme_mark_namespaces_dead);
5203 void nvme_unfreeze(struct nvme_ctrl *ctrl)
5207 down_read(&ctrl->namespaces_rwsem);
5208 list_for_each_entry(ns, &ctrl->namespaces, list)
5209 blk_mq_unfreeze_queue(ns->queue);
5210 up_read(&ctrl->namespaces_rwsem);
5212 EXPORT_SYMBOL_GPL(nvme_unfreeze);
5214 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
5218 down_read(&ctrl->namespaces_rwsem);
5219 list_for_each_entry(ns, &ctrl->namespaces, list) {
5220 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
5224 up_read(&ctrl->namespaces_rwsem);
5227 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
5229 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
5233 down_read(&ctrl->namespaces_rwsem);
5234 list_for_each_entry(ns, &ctrl->namespaces, list)
5235 blk_mq_freeze_queue_wait(ns->queue);
5236 up_read(&ctrl->namespaces_rwsem);
5238 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
5240 void nvme_start_freeze(struct nvme_ctrl *ctrl)
5244 down_read(&ctrl->namespaces_rwsem);
5245 list_for_each_entry(ns, &ctrl->namespaces, list)
5246 blk_freeze_queue_start(ns->queue);
5247 up_read(&ctrl->namespaces_rwsem);
5249 EXPORT_SYMBOL_GPL(nvme_start_freeze);
5251 void nvme_quiesce_io_queues(struct nvme_ctrl *ctrl)
5255 if (!test_and_set_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5256 blk_mq_quiesce_tagset(ctrl->tagset);
5258 blk_mq_wait_quiesce_done(ctrl->tagset);
5260 EXPORT_SYMBOL_GPL(nvme_quiesce_io_queues);
5262 void nvme_unquiesce_io_queues(struct nvme_ctrl *ctrl)
5266 if (test_and_clear_bit(NVME_CTRL_STOPPED, &ctrl->flags))
5267 blk_mq_unquiesce_tagset(ctrl->tagset);
5269 EXPORT_SYMBOL_GPL(nvme_unquiesce_io_queues);
5271 void nvme_quiesce_admin_queue(struct nvme_ctrl *ctrl)
5273 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5274 blk_mq_quiesce_queue(ctrl->admin_q);
5276 blk_mq_wait_quiesce_done(ctrl->admin_q->tag_set);
5278 EXPORT_SYMBOL_GPL(nvme_quiesce_admin_queue);
5280 void nvme_unquiesce_admin_queue(struct nvme_ctrl *ctrl)
5282 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
5283 blk_mq_unquiesce_queue(ctrl->admin_q);
5285 EXPORT_SYMBOL_GPL(nvme_unquiesce_admin_queue);
5287 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
5291 down_read(&ctrl->namespaces_rwsem);
5292 list_for_each_entry(ns, &ctrl->namespaces, list)
5293 blk_sync_queue(ns->queue);
5294 up_read(&ctrl->namespaces_rwsem);
5296 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
5298 void nvme_sync_queues(struct nvme_ctrl *ctrl)
5300 nvme_sync_io_queues(ctrl);
5302 blk_sync_queue(ctrl->admin_q);
5304 EXPORT_SYMBOL_GPL(nvme_sync_queues);
5306 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
5308 if (file->f_op != &nvme_dev_fops)
5310 return file->private_data;
5312 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
5315 * Check we didn't inadvertently grow the command structure sizes:
5317 static inline void _nvme_check_size(void)
5319 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
5320 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
5321 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
5322 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
5323 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
5324 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
5325 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
5326 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
5327 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
5328 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
5329 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
5330 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
5331 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
5332 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
5333 NVME_IDENTIFY_DATA_SIZE);
5334 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
5335 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
5336 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
5337 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
5338 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
5339 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
5340 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
5341 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
5342 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
5346 static int __init nvme_core_init(void)
5348 int result = -ENOMEM;
5352 nvme_wq = alloc_workqueue("nvme-wq",
5353 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5357 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
5358 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5362 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
5363 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
5364 if (!nvme_delete_wq)
5365 goto destroy_reset_wq;
5367 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
5368 NVME_MINORS, "nvme");
5370 goto destroy_delete_wq;
5372 nvme_class = class_create(THIS_MODULE, "nvme");
5373 if (IS_ERR(nvme_class)) {
5374 result = PTR_ERR(nvme_class);
5375 goto unregister_chrdev;
5377 nvme_class->dev_uevent = nvme_class_uevent;
5379 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
5380 if (IS_ERR(nvme_subsys_class)) {
5381 result = PTR_ERR(nvme_subsys_class);
5385 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5388 goto destroy_subsys_class;
5390 nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5391 if (IS_ERR(nvme_ns_chr_class)) {
5392 result = PTR_ERR(nvme_ns_chr_class);
5393 goto unregister_generic_ns;
5396 result = nvme_init_auth();
5398 goto destroy_ns_chr;
5402 class_destroy(nvme_ns_chr_class);
5403 unregister_generic_ns:
5404 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5405 destroy_subsys_class:
5406 class_destroy(nvme_subsys_class);
5408 class_destroy(nvme_class);
5410 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5412 destroy_workqueue(nvme_delete_wq);
5414 destroy_workqueue(nvme_reset_wq);
5416 destroy_workqueue(nvme_wq);
5421 static void __exit nvme_core_exit(void)
5424 class_destroy(nvme_ns_chr_class);
5425 class_destroy(nvme_subsys_class);
5426 class_destroy(nvme_class);
5427 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5428 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5429 destroy_workqueue(nvme_delete_wq);
5430 destroy_workqueue(nvme_reset_wq);
5431 destroy_workqueue(nvme_wq);
5432 ida_destroy(&nvme_ns_chr_minor_ida);
5433 ida_destroy(&nvme_instance_ida);
5436 MODULE_LICENSE("GPL");
5437 MODULE_VERSION("1.0");
5438 module_init(nvme_core_init);
5439 module_exit(nvme_core_exit);