1 // SPDX-License-Identifier: GPL-2.0
3 * NVM Express device driver
4 * Copyright (c) 2011-2014, Intel Corporation.
7 #include <linux/blkdev.h>
8 #include <linux/blk-mq.h>
9 #include <linux/blk-integrity.h>
10 #include <linux/compat.h>
11 #include <linux/delay.h>
12 #include <linux/errno.h>
13 #include <linux/hdreg.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/backing-dev.h>
17 #include <linux/slab.h>
18 #include <linux/types.h>
20 #include <linux/ptrace.h>
21 #include <linux/nvme_ioctl.h>
22 #include <linux/pm_qos.h>
23 #include <asm/unaligned.h>
28 #define CREATE_TRACE_POINTS
31 #define NVME_MINORS (1U << MINORBITS)
33 unsigned int admin_timeout = 60;
34 module_param(admin_timeout, uint, 0644);
35 MODULE_PARM_DESC(admin_timeout, "timeout in seconds for admin commands");
36 EXPORT_SYMBOL_GPL(admin_timeout);
38 unsigned int nvme_io_timeout = 30;
39 module_param_named(io_timeout, nvme_io_timeout, uint, 0644);
40 MODULE_PARM_DESC(io_timeout, "timeout in seconds for I/O");
41 EXPORT_SYMBOL_GPL(nvme_io_timeout);
43 static unsigned char shutdown_timeout = 5;
44 module_param(shutdown_timeout, byte, 0644);
45 MODULE_PARM_DESC(shutdown_timeout, "timeout in seconds for controller shutdown");
47 static u8 nvme_max_retries = 5;
48 module_param_named(max_retries, nvme_max_retries, byte, 0644);
49 MODULE_PARM_DESC(max_retries, "max number of retries a command may have");
51 static unsigned long default_ps_max_latency_us = 100000;
52 module_param(default_ps_max_latency_us, ulong, 0644);
53 MODULE_PARM_DESC(default_ps_max_latency_us,
54 "max power saving latency for new devices; use PM QOS to change per device");
56 static bool force_apst;
57 module_param(force_apst, bool, 0644);
58 MODULE_PARM_DESC(force_apst, "allow APST for newly enumerated devices even if quirked off");
60 static unsigned long apst_primary_timeout_ms = 100;
61 module_param(apst_primary_timeout_ms, ulong, 0644);
62 MODULE_PARM_DESC(apst_primary_timeout_ms,
63 "primary APST timeout in ms");
65 static unsigned long apst_secondary_timeout_ms = 2000;
66 module_param(apst_secondary_timeout_ms, ulong, 0644);
67 MODULE_PARM_DESC(apst_secondary_timeout_ms,
68 "secondary APST timeout in ms");
70 static unsigned long apst_primary_latency_tol_us = 15000;
71 module_param(apst_primary_latency_tol_us, ulong, 0644);
72 MODULE_PARM_DESC(apst_primary_latency_tol_us,
73 "primary APST latency tolerance in us");
75 static unsigned long apst_secondary_latency_tol_us = 100000;
76 module_param(apst_secondary_latency_tol_us, ulong, 0644);
77 MODULE_PARM_DESC(apst_secondary_latency_tol_us,
78 "secondary APST latency tolerance in us");
81 * nvme_wq - hosts nvme related works that are not reset or delete
82 * nvme_reset_wq - hosts nvme reset works
83 * nvme_delete_wq - hosts nvme delete works
85 * nvme_wq will host works such as scan, aen handling, fw activation,
86 * keep-alive, periodic reconnects etc. nvme_reset_wq
87 * runs reset works which also flush works hosted on nvme_wq for
88 * serialization purposes. nvme_delete_wq host controller deletion
89 * works which flush reset works for serialization.
91 struct workqueue_struct *nvme_wq;
92 EXPORT_SYMBOL_GPL(nvme_wq);
94 struct workqueue_struct *nvme_reset_wq;
95 EXPORT_SYMBOL_GPL(nvme_reset_wq);
97 struct workqueue_struct *nvme_delete_wq;
98 EXPORT_SYMBOL_GPL(nvme_delete_wq);
100 static LIST_HEAD(nvme_subsystems);
101 static DEFINE_MUTEX(nvme_subsystems_lock);
103 static DEFINE_IDA(nvme_instance_ida);
104 static dev_t nvme_ctrl_base_chr_devt;
105 static struct class *nvme_class;
106 static struct class *nvme_subsys_class;
108 static DEFINE_IDA(nvme_ns_chr_minor_ida);
109 static dev_t nvme_ns_chr_devt;
110 static struct class *nvme_ns_chr_class;
112 static void nvme_put_subsystem(struct nvme_subsystem *subsys);
113 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
115 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
116 struct nvme_command *cmd);
118 void nvme_queue_scan(struct nvme_ctrl *ctrl)
121 * Only new queue scan work when admin and IO queues are both alive
123 if (ctrl->state == NVME_CTRL_LIVE && ctrl->tagset)
124 queue_work(nvme_wq, &ctrl->scan_work);
128 * Use this function to proceed with scheduling reset_work for a controller
129 * that had previously been set to the resetting state. This is intended for
130 * code paths that can't be interrupted by other reset attempts. A hot removal
131 * may prevent this from succeeding.
133 int nvme_try_sched_reset(struct nvme_ctrl *ctrl)
135 if (ctrl->state != NVME_CTRL_RESETTING)
137 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
141 EXPORT_SYMBOL_GPL(nvme_try_sched_reset);
143 static void nvme_failfast_work(struct work_struct *work)
145 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
146 struct nvme_ctrl, failfast_work);
148 if (ctrl->state != NVME_CTRL_CONNECTING)
151 set_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
152 dev_info(ctrl->device, "failfast expired\n");
153 nvme_kick_requeue_lists(ctrl);
156 static inline void nvme_start_failfast_work(struct nvme_ctrl *ctrl)
158 if (!ctrl->opts || ctrl->opts->fast_io_fail_tmo == -1)
161 schedule_delayed_work(&ctrl->failfast_work,
162 ctrl->opts->fast_io_fail_tmo * HZ);
165 static inline void nvme_stop_failfast_work(struct nvme_ctrl *ctrl)
170 cancel_delayed_work_sync(&ctrl->failfast_work);
171 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
175 int nvme_reset_ctrl(struct nvme_ctrl *ctrl)
177 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
179 if (!queue_work(nvme_reset_wq, &ctrl->reset_work))
183 EXPORT_SYMBOL_GPL(nvme_reset_ctrl);
185 int nvme_reset_ctrl_sync(struct nvme_ctrl *ctrl)
189 ret = nvme_reset_ctrl(ctrl);
191 flush_work(&ctrl->reset_work);
192 if (ctrl->state != NVME_CTRL_LIVE)
199 static void nvme_do_delete_ctrl(struct nvme_ctrl *ctrl)
201 dev_info(ctrl->device,
202 "Removing ctrl: NQN \"%s\"\n", nvmf_ctrl_subsysnqn(ctrl));
204 flush_work(&ctrl->reset_work);
205 nvme_stop_ctrl(ctrl);
206 nvme_remove_namespaces(ctrl);
207 ctrl->ops->delete_ctrl(ctrl);
208 nvme_uninit_ctrl(ctrl);
211 static void nvme_delete_ctrl_work(struct work_struct *work)
213 struct nvme_ctrl *ctrl =
214 container_of(work, struct nvme_ctrl, delete_work);
216 nvme_do_delete_ctrl(ctrl);
219 int nvme_delete_ctrl(struct nvme_ctrl *ctrl)
221 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
223 if (!queue_work(nvme_delete_wq, &ctrl->delete_work))
227 EXPORT_SYMBOL_GPL(nvme_delete_ctrl);
229 static void nvme_delete_ctrl_sync(struct nvme_ctrl *ctrl)
232 * Keep a reference until nvme_do_delete_ctrl() complete,
233 * since ->delete_ctrl can free the controller.
236 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING))
237 nvme_do_delete_ctrl(ctrl);
241 static blk_status_t nvme_error_status(u16 status)
243 switch (status & 0x7ff) {
244 case NVME_SC_SUCCESS:
246 case NVME_SC_CAP_EXCEEDED:
247 return BLK_STS_NOSPC;
248 case NVME_SC_LBA_RANGE:
249 case NVME_SC_CMD_INTERRUPTED:
250 case NVME_SC_NS_NOT_READY:
251 return BLK_STS_TARGET;
252 case NVME_SC_BAD_ATTRIBUTES:
253 case NVME_SC_ONCS_NOT_SUPPORTED:
254 case NVME_SC_INVALID_OPCODE:
255 case NVME_SC_INVALID_FIELD:
256 case NVME_SC_INVALID_NS:
257 return BLK_STS_NOTSUPP;
258 case NVME_SC_WRITE_FAULT:
259 case NVME_SC_READ_ERROR:
260 case NVME_SC_UNWRITTEN_BLOCK:
261 case NVME_SC_ACCESS_DENIED:
262 case NVME_SC_READ_ONLY:
263 case NVME_SC_COMPARE_FAILED:
264 return BLK_STS_MEDIUM;
265 case NVME_SC_GUARD_CHECK:
266 case NVME_SC_APPTAG_CHECK:
267 case NVME_SC_REFTAG_CHECK:
268 case NVME_SC_INVALID_PI:
269 return BLK_STS_PROTECTION;
270 case NVME_SC_RESERVATION_CONFLICT:
271 return BLK_STS_NEXUS;
272 case NVME_SC_HOST_PATH_ERROR:
273 return BLK_STS_TRANSPORT;
274 case NVME_SC_ZONE_TOO_MANY_ACTIVE:
275 return BLK_STS_ZONE_ACTIVE_RESOURCE;
276 case NVME_SC_ZONE_TOO_MANY_OPEN:
277 return BLK_STS_ZONE_OPEN_RESOURCE;
279 return BLK_STS_IOERR;
283 static void nvme_retry_req(struct request *req)
285 unsigned long delay = 0;
288 /* The mask and shift result must be <= 3 */
289 crd = (nvme_req(req)->status & NVME_SC_CRD) >> 11;
291 delay = nvme_req(req)->ctrl->crdt[crd - 1] * 100;
293 nvme_req(req)->retries++;
294 blk_mq_requeue_request(req, false);
295 blk_mq_delay_kick_requeue_list(req->q, delay);
298 static void nvme_log_error(struct request *req)
300 struct nvme_ns *ns = req->q->queuedata;
301 struct nvme_request *nr = nvme_req(req);
304 pr_err_ratelimited("%s: %s(0x%x) @ LBA %llu, %llu blocks, %s (sct 0x%x / sc 0x%x) %s%s\n",
305 ns->disk ? ns->disk->disk_name : "?",
306 nvme_get_opcode_str(nr->cmd->common.opcode),
307 nr->cmd->common.opcode,
308 (unsigned long long)nvme_sect_to_lba(ns, blk_rq_pos(req)),
309 (unsigned long long)blk_rq_bytes(req) >> ns->lba_shift,
310 nvme_get_error_status_str(nr->status),
311 nr->status >> 8 & 7, /* Status Code Type */
312 nr->status & 0xff, /* Status Code */
313 nr->status & NVME_SC_MORE ? "MORE " : "",
314 nr->status & NVME_SC_DNR ? "DNR " : "");
318 pr_err_ratelimited("%s: %s(0x%x), %s (sct 0x%x / sc 0x%x) %s%s\n",
319 dev_name(nr->ctrl->device),
320 nvme_get_admin_opcode_str(nr->cmd->common.opcode),
321 nr->cmd->common.opcode,
322 nvme_get_error_status_str(nr->status),
323 nr->status >> 8 & 7, /* Status Code Type */
324 nr->status & 0xff, /* Status Code */
325 nr->status & NVME_SC_MORE ? "MORE " : "",
326 nr->status & NVME_SC_DNR ? "DNR " : "");
329 enum nvme_disposition {
335 static inline enum nvme_disposition nvme_decide_disposition(struct request *req)
337 if (likely(nvme_req(req)->status == 0))
340 if (blk_noretry_request(req) ||
341 (nvme_req(req)->status & NVME_SC_DNR) ||
342 nvme_req(req)->retries >= nvme_max_retries)
345 if (req->cmd_flags & REQ_NVME_MPATH) {
346 if (nvme_is_path_error(nvme_req(req)->status) ||
347 blk_queue_dying(req->q))
350 if (blk_queue_dying(req->q))
357 static inline void nvme_end_req_zoned(struct request *req)
359 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
360 req_op(req) == REQ_OP_ZONE_APPEND)
361 req->__sector = nvme_lba_to_sect(req->q->queuedata,
362 le64_to_cpu(nvme_req(req)->result.u64));
365 static inline void nvme_end_req(struct request *req)
367 blk_status_t status = nvme_error_status(nvme_req(req)->status);
369 if (unlikely(nvme_req(req)->status && !(req->rq_flags & RQF_QUIET)))
371 nvme_end_req_zoned(req);
372 nvme_trace_bio_complete(req);
373 blk_mq_end_request(req, status);
376 void nvme_complete_rq(struct request *req)
378 trace_nvme_complete_rq(req);
379 nvme_cleanup_cmd(req);
381 if (nvme_req(req)->ctrl->kas)
382 nvme_req(req)->ctrl->comp_seen = true;
384 switch (nvme_decide_disposition(req)) {
392 nvme_failover_req(req);
396 EXPORT_SYMBOL_GPL(nvme_complete_rq);
398 void nvme_complete_batch_req(struct request *req)
400 trace_nvme_complete_rq(req);
401 nvme_cleanup_cmd(req);
402 nvme_end_req_zoned(req);
404 EXPORT_SYMBOL_GPL(nvme_complete_batch_req);
407 * Called to unwind from ->queue_rq on a failed command submission so that the
408 * multipathing code gets called to potentially failover to another path.
409 * The caller needs to unwind all transport specific resource allocations and
410 * must return propagate the return value.
412 blk_status_t nvme_host_path_error(struct request *req)
414 nvme_req(req)->status = NVME_SC_HOST_PATH_ERROR;
415 blk_mq_set_request_complete(req);
416 nvme_complete_rq(req);
419 EXPORT_SYMBOL_GPL(nvme_host_path_error);
421 bool nvme_cancel_request(struct request *req, void *data, bool reserved)
423 dev_dbg_ratelimited(((struct nvme_ctrl *) data)->device,
424 "Cancelling I/O %d", req->tag);
426 /* don't abort one completed request */
427 if (blk_mq_request_completed(req))
430 nvme_req(req)->status = NVME_SC_HOST_ABORTED_CMD;
431 nvme_req(req)->flags |= NVME_REQ_CANCELLED;
432 blk_mq_complete_request(req);
435 EXPORT_SYMBOL_GPL(nvme_cancel_request);
437 void nvme_cancel_tagset(struct nvme_ctrl *ctrl)
440 blk_mq_tagset_busy_iter(ctrl->tagset,
441 nvme_cancel_request, ctrl);
442 blk_mq_tagset_wait_completed_request(ctrl->tagset);
445 EXPORT_SYMBOL_GPL(nvme_cancel_tagset);
447 void nvme_cancel_admin_tagset(struct nvme_ctrl *ctrl)
449 if (ctrl->admin_tagset) {
450 blk_mq_tagset_busy_iter(ctrl->admin_tagset,
451 nvme_cancel_request, ctrl);
452 blk_mq_tagset_wait_completed_request(ctrl->admin_tagset);
455 EXPORT_SYMBOL_GPL(nvme_cancel_admin_tagset);
457 bool nvme_change_ctrl_state(struct nvme_ctrl *ctrl,
458 enum nvme_ctrl_state new_state)
460 enum nvme_ctrl_state old_state;
462 bool changed = false;
464 spin_lock_irqsave(&ctrl->lock, flags);
466 old_state = ctrl->state;
471 case NVME_CTRL_RESETTING:
472 case NVME_CTRL_CONNECTING:
479 case NVME_CTRL_RESETTING:
489 case NVME_CTRL_CONNECTING:
492 case NVME_CTRL_RESETTING:
499 case NVME_CTRL_DELETING:
502 case NVME_CTRL_RESETTING:
503 case NVME_CTRL_CONNECTING:
510 case NVME_CTRL_DELETING_NOIO:
512 case NVME_CTRL_DELETING:
522 case NVME_CTRL_DELETING:
534 ctrl->state = new_state;
535 wake_up_all(&ctrl->state_wq);
538 spin_unlock_irqrestore(&ctrl->lock, flags);
542 if (ctrl->state == NVME_CTRL_LIVE) {
543 if (old_state == NVME_CTRL_CONNECTING)
544 nvme_stop_failfast_work(ctrl);
545 nvme_kick_requeue_lists(ctrl);
546 } else if (ctrl->state == NVME_CTRL_CONNECTING &&
547 old_state == NVME_CTRL_RESETTING) {
548 nvme_start_failfast_work(ctrl);
552 EXPORT_SYMBOL_GPL(nvme_change_ctrl_state);
555 * Returns true for sink states that can't ever transition back to live.
557 static bool nvme_state_terminal(struct nvme_ctrl *ctrl)
559 switch (ctrl->state) {
562 case NVME_CTRL_RESETTING:
563 case NVME_CTRL_CONNECTING:
565 case NVME_CTRL_DELETING:
566 case NVME_CTRL_DELETING_NOIO:
570 WARN_ONCE(1, "Unhandled ctrl state:%d", ctrl->state);
576 * Waits for the controller state to be resetting, or returns false if it is
577 * not possible to ever transition to that state.
579 bool nvme_wait_reset(struct nvme_ctrl *ctrl)
581 wait_event(ctrl->state_wq,
582 nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING) ||
583 nvme_state_terminal(ctrl));
584 return ctrl->state == NVME_CTRL_RESETTING;
586 EXPORT_SYMBOL_GPL(nvme_wait_reset);
588 static void nvme_free_ns_head(struct kref *ref)
590 struct nvme_ns_head *head =
591 container_of(ref, struct nvme_ns_head, ref);
593 nvme_mpath_remove_disk(head);
594 ida_free(&head->subsys->ns_ida, head->instance);
595 cleanup_srcu_struct(&head->srcu);
596 nvme_put_subsystem(head->subsys);
600 bool nvme_tryget_ns_head(struct nvme_ns_head *head)
602 return kref_get_unless_zero(&head->ref);
605 void nvme_put_ns_head(struct nvme_ns_head *head)
607 kref_put(&head->ref, nvme_free_ns_head);
610 static void nvme_free_ns(struct kref *kref)
612 struct nvme_ns *ns = container_of(kref, struct nvme_ns, kref);
615 nvme_put_ns_head(ns->head);
616 nvme_put_ctrl(ns->ctrl);
620 static inline bool nvme_get_ns(struct nvme_ns *ns)
622 return kref_get_unless_zero(&ns->kref);
625 void nvme_put_ns(struct nvme_ns *ns)
627 kref_put(&ns->kref, nvme_free_ns);
629 EXPORT_SYMBOL_NS_GPL(nvme_put_ns, NVME_TARGET_PASSTHRU);
631 static inline void nvme_clear_nvme_request(struct request *req)
633 nvme_req(req)->status = 0;
634 nvme_req(req)->retries = 0;
635 nvme_req(req)->flags = 0;
636 req->rq_flags |= RQF_DONTPREP;
639 /* initialize a passthrough request */
640 void nvme_init_request(struct request *req, struct nvme_command *cmd)
642 if (req->q->queuedata)
643 req->timeout = NVME_IO_TIMEOUT;
644 else /* no queuedata implies admin queue */
645 req->timeout = NVME_ADMIN_TIMEOUT;
647 /* passthru commands should let the driver set the SGL flags */
648 cmd->common.flags &= ~NVME_CMD_SGL_ALL;
650 req->cmd_flags |= REQ_FAILFAST_DRIVER;
651 if (req->mq_hctx->type == HCTX_TYPE_POLL)
652 req->cmd_flags |= REQ_POLLED;
653 nvme_clear_nvme_request(req);
654 memcpy(nvme_req(req)->cmd, cmd, sizeof(*cmd));
656 EXPORT_SYMBOL_GPL(nvme_init_request);
659 * For something we're not in a state to send to the device the default action
660 * is to busy it and retry it after the controller state is recovered. However,
661 * if the controller is deleting or if anything is marked for failfast or
662 * nvme multipath it is immediately failed.
664 * Note: commands used to initialize the controller will be marked for failfast.
665 * Note: nvme cli/ioctl commands are marked for failfast.
667 blk_status_t nvme_fail_nonready_command(struct nvme_ctrl *ctrl,
670 if (ctrl->state != NVME_CTRL_DELETING_NOIO &&
671 ctrl->state != NVME_CTRL_DELETING &&
672 ctrl->state != NVME_CTRL_DEAD &&
673 !test_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags) &&
674 !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
675 return BLK_STS_RESOURCE;
676 return nvme_host_path_error(rq);
678 EXPORT_SYMBOL_GPL(nvme_fail_nonready_command);
680 bool __nvme_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
683 struct nvme_request *req = nvme_req(rq);
686 * currently we have a problem sending passthru commands
687 * on the admin_q if the controller is not LIVE because we can't
688 * make sure that they are going out after the admin connect,
689 * controller enable and/or other commands in the initialization
690 * sequence. until the controller will be LIVE, fail with
691 * BLK_STS_RESOURCE so that they will be rescheduled.
693 if (rq->q == ctrl->admin_q && (req->flags & NVME_REQ_USERCMD))
696 if (ctrl->ops->flags & NVME_F_FABRICS) {
698 * Only allow commands on a live queue, except for the connect
699 * command, which is require to set the queue live in the
700 * appropinquate states.
702 switch (ctrl->state) {
703 case NVME_CTRL_CONNECTING:
704 if (blk_rq_is_passthrough(rq) && nvme_is_fabrics(req->cmd) &&
705 req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
717 EXPORT_SYMBOL_GPL(__nvme_check_ready);
719 static inline void nvme_setup_flush(struct nvme_ns *ns,
720 struct nvme_command *cmnd)
722 memset(cmnd, 0, sizeof(*cmnd));
723 cmnd->common.opcode = nvme_cmd_flush;
724 cmnd->common.nsid = cpu_to_le32(ns->head->ns_id);
727 static blk_status_t nvme_setup_discard(struct nvme_ns *ns, struct request *req,
728 struct nvme_command *cmnd)
730 unsigned short segments = blk_rq_nr_discard_segments(req), n = 0;
731 struct nvme_dsm_range *range;
735 * Some devices do not consider the DSM 'Number of Ranges' field when
736 * determining how much data to DMA. Always allocate memory for maximum
737 * number of segments to prevent device reading beyond end of buffer.
739 static const size_t alloc_size = sizeof(*range) * NVME_DSM_MAX_RANGES;
741 range = kzalloc(alloc_size, GFP_ATOMIC | __GFP_NOWARN);
744 * If we fail allocation our range, fallback to the controller
745 * discard page. If that's also busy, it's safe to return
746 * busy, as we know we can make progress once that's freed.
748 if (test_and_set_bit_lock(0, &ns->ctrl->discard_page_busy))
749 return BLK_STS_RESOURCE;
751 range = page_address(ns->ctrl->discard_page);
754 __rq_for_each_bio(bio, req) {
755 u64 slba = nvme_sect_to_lba(ns, bio->bi_iter.bi_sector);
756 u32 nlb = bio->bi_iter.bi_size >> ns->lba_shift;
759 range[n].cattr = cpu_to_le32(0);
760 range[n].nlb = cpu_to_le32(nlb);
761 range[n].slba = cpu_to_le64(slba);
766 if (WARN_ON_ONCE(n != segments)) {
767 if (virt_to_page(range) == ns->ctrl->discard_page)
768 clear_bit_unlock(0, &ns->ctrl->discard_page_busy);
771 return BLK_STS_IOERR;
774 memset(cmnd, 0, sizeof(*cmnd));
775 cmnd->dsm.opcode = nvme_cmd_dsm;
776 cmnd->dsm.nsid = cpu_to_le32(ns->head->ns_id);
777 cmnd->dsm.nr = cpu_to_le32(segments - 1);
778 cmnd->dsm.attributes = cpu_to_le32(NVME_DSMGMT_AD);
780 req->special_vec.bv_page = virt_to_page(range);
781 req->special_vec.bv_offset = offset_in_page(range);
782 req->special_vec.bv_len = alloc_size;
783 req->rq_flags |= RQF_SPECIAL_PAYLOAD;
788 static void nvme_set_ref_tag(struct nvme_ns *ns, struct nvme_command *cmnd,
794 /* both rw and write zeroes share the same reftag format */
795 switch (ns->guard_type) {
796 case NVME_NVM_NS_16B_GUARD:
797 cmnd->rw.reftag = cpu_to_le32(t10_pi_ref_tag(req));
799 case NVME_NVM_NS_64B_GUARD:
800 ref48 = ext_pi_ref_tag(req);
801 lower = lower_32_bits(ref48);
802 upper = upper_32_bits(ref48);
804 cmnd->rw.reftag = cpu_to_le32(lower);
805 cmnd->rw.cdw3 = cpu_to_le32(upper);
812 static inline blk_status_t nvme_setup_write_zeroes(struct nvme_ns *ns,
813 struct request *req, struct nvme_command *cmnd)
815 memset(cmnd, 0, sizeof(*cmnd));
817 if (ns->ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
818 return nvme_setup_discard(ns, req, cmnd);
820 cmnd->write_zeroes.opcode = nvme_cmd_write_zeroes;
821 cmnd->write_zeroes.nsid = cpu_to_le32(ns->head->ns_id);
822 cmnd->write_zeroes.slba =
823 cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
824 cmnd->write_zeroes.length =
825 cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
827 if (nvme_ns_has_pi(ns)) {
828 cmnd->write_zeroes.control = cpu_to_le16(NVME_RW_PRINFO_PRACT);
830 switch (ns->pi_type) {
831 case NVME_NS_DPS_PI_TYPE1:
832 case NVME_NS_DPS_PI_TYPE2:
833 nvme_set_ref_tag(ns, cmnd, req);
841 static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
842 struct request *req, struct nvme_command *cmnd,
848 if (req->cmd_flags & REQ_FUA)
849 control |= NVME_RW_FUA;
850 if (req->cmd_flags & (REQ_FAILFAST_DEV | REQ_RAHEAD))
851 control |= NVME_RW_LR;
853 if (req->cmd_flags & REQ_RAHEAD)
854 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
856 cmnd->rw.opcode = op;
858 cmnd->rw.nsid = cpu_to_le32(ns->head->ns_id);
861 cmnd->rw.metadata = 0;
862 cmnd->rw.slba = cpu_to_le64(nvme_sect_to_lba(ns, blk_rq_pos(req)));
863 cmnd->rw.length = cpu_to_le16((blk_rq_bytes(req) >> ns->lba_shift) - 1);
866 cmnd->rw.appmask = 0;
870 * If formated with metadata, the block layer always provides a
871 * metadata buffer if CONFIG_BLK_DEV_INTEGRITY is enabled. Else
872 * we enable the PRACT bit for protection information or set the
873 * namespace capacity to zero to prevent any I/O.
875 if (!blk_integrity_rq(req)) {
876 if (WARN_ON_ONCE(!nvme_ns_has_pi(ns)))
877 return BLK_STS_NOTSUPP;
878 control |= NVME_RW_PRINFO_PRACT;
881 switch (ns->pi_type) {
882 case NVME_NS_DPS_PI_TYPE3:
883 control |= NVME_RW_PRINFO_PRCHK_GUARD;
885 case NVME_NS_DPS_PI_TYPE1:
886 case NVME_NS_DPS_PI_TYPE2:
887 control |= NVME_RW_PRINFO_PRCHK_GUARD |
888 NVME_RW_PRINFO_PRCHK_REF;
889 if (op == nvme_cmd_zone_append)
890 control |= NVME_RW_APPEND_PIREMAP;
891 nvme_set_ref_tag(ns, cmnd, req);
896 cmnd->rw.control = cpu_to_le16(control);
897 cmnd->rw.dsmgmt = cpu_to_le32(dsmgmt);
901 void nvme_cleanup_cmd(struct request *req)
903 if (req->rq_flags & RQF_SPECIAL_PAYLOAD) {
904 struct nvme_ctrl *ctrl = nvme_req(req)->ctrl;
906 if (req->special_vec.bv_page == ctrl->discard_page)
907 clear_bit_unlock(0, &ctrl->discard_page_busy);
909 kfree(bvec_virt(&req->special_vec));
912 EXPORT_SYMBOL_GPL(nvme_cleanup_cmd);
914 blk_status_t nvme_setup_cmd(struct nvme_ns *ns, struct request *req)
916 struct nvme_command *cmd = nvme_req(req)->cmd;
917 blk_status_t ret = BLK_STS_OK;
919 if (!(req->rq_flags & RQF_DONTPREP))
920 nvme_clear_nvme_request(req);
922 switch (req_op(req)) {
925 /* these are setup prior to execution in nvme_init_request() */
928 nvme_setup_flush(ns, cmd);
930 case REQ_OP_ZONE_RESET_ALL:
931 case REQ_OP_ZONE_RESET:
932 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_RESET);
934 case REQ_OP_ZONE_OPEN:
935 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_OPEN);
937 case REQ_OP_ZONE_CLOSE:
938 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_CLOSE);
940 case REQ_OP_ZONE_FINISH:
941 ret = nvme_setup_zone_mgmt_send(ns, req, cmd, NVME_ZONE_FINISH);
943 case REQ_OP_WRITE_ZEROES:
944 ret = nvme_setup_write_zeroes(ns, req, cmd);
947 ret = nvme_setup_discard(ns, req, cmd);
950 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_read);
953 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_write);
955 case REQ_OP_ZONE_APPEND:
956 ret = nvme_setup_rw(ns, req, cmd, nvme_cmd_zone_append);
960 return BLK_STS_IOERR;
963 cmd->common.command_id = nvme_cid(req);
964 trace_nvme_setup_cmd(req, cmd);
967 EXPORT_SYMBOL_GPL(nvme_setup_cmd);
972 * >0: nvme controller's cqe status response
973 * <0: kernel error in lieu of controller response
975 static int nvme_execute_rq(struct request *rq, bool at_head)
979 status = blk_execute_rq(rq, at_head);
980 if (nvme_req(rq)->flags & NVME_REQ_CANCELLED)
982 if (nvme_req(rq)->status)
983 return nvme_req(rq)->status;
984 return blk_status_to_errno(status);
988 * Returns 0 on success. If the result is negative, it's a Linux error code;
989 * if the result is positive, it's an NVM Express status code
991 int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
992 union nvme_result *result, void *buffer, unsigned bufflen,
993 unsigned timeout, int qid, int at_head,
994 blk_mq_req_flags_t flags)
999 if (qid == NVME_QID_ANY)
1000 req = blk_mq_alloc_request(q, nvme_req_op(cmd), flags);
1002 req = blk_mq_alloc_request_hctx(q, nvme_req_op(cmd), flags,
1006 return PTR_ERR(req);
1007 nvme_init_request(req, cmd);
1010 req->timeout = timeout;
1012 if (buffer && bufflen) {
1013 ret = blk_rq_map_kern(q, req, buffer, bufflen, GFP_KERNEL);
1018 req->rq_flags |= RQF_QUIET;
1019 ret = nvme_execute_rq(req, at_head);
1020 if (result && ret >= 0)
1021 *result = nvme_req(req)->result;
1023 blk_mq_free_request(req);
1026 EXPORT_SYMBOL_GPL(__nvme_submit_sync_cmd);
1028 int nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
1029 void *buffer, unsigned bufflen)
1031 return __nvme_submit_sync_cmd(q, cmd, NULL, buffer, bufflen, 0,
1032 NVME_QID_ANY, 0, 0);
1034 EXPORT_SYMBOL_GPL(nvme_submit_sync_cmd);
1036 static u32 nvme_known_admin_effects(u8 opcode)
1039 case nvme_admin_format_nvm:
1040 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_NCC |
1041 NVME_CMD_EFFECTS_CSE_MASK;
1042 case nvme_admin_sanitize_nvm:
1043 return NVME_CMD_EFFECTS_LBCC | NVME_CMD_EFFECTS_CSE_MASK;
1050 u32 nvme_command_effects(struct nvme_ctrl *ctrl, struct nvme_ns *ns, u8 opcode)
1055 if (ns->head->effects)
1056 effects = le32_to_cpu(ns->head->effects->iocs[opcode]);
1057 if (effects & ~(NVME_CMD_EFFECTS_CSUPP | NVME_CMD_EFFECTS_LBCC))
1058 dev_warn_once(ctrl->device,
1059 "IO command:%02x has unhandled effects:%08x\n",
1065 effects = le32_to_cpu(ctrl->effects->acs[opcode]);
1066 effects |= nvme_known_admin_effects(opcode);
1070 EXPORT_SYMBOL_NS_GPL(nvme_command_effects, NVME_TARGET_PASSTHRU);
1072 static u32 nvme_passthru_start(struct nvme_ctrl *ctrl, struct nvme_ns *ns,
1075 u32 effects = nvme_command_effects(ctrl, ns, opcode);
1078 * For simplicity, IO to all namespaces is quiesced even if the command
1079 * effects say only one namespace is affected.
1081 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1082 mutex_lock(&ctrl->scan_lock);
1083 mutex_lock(&ctrl->subsys->lock);
1084 nvme_mpath_start_freeze(ctrl->subsys);
1085 nvme_mpath_wait_freeze(ctrl->subsys);
1086 nvme_start_freeze(ctrl);
1087 nvme_wait_freeze(ctrl);
1092 static void nvme_passthru_end(struct nvme_ctrl *ctrl, u32 effects,
1093 struct nvme_command *cmd, int status)
1095 if (effects & NVME_CMD_EFFECTS_CSE_MASK) {
1096 nvme_unfreeze(ctrl);
1097 nvme_mpath_unfreeze(ctrl->subsys);
1098 mutex_unlock(&ctrl->subsys->lock);
1099 nvme_remove_invalid_namespaces(ctrl, NVME_NSID_ALL);
1100 mutex_unlock(&ctrl->scan_lock);
1102 if (effects & NVME_CMD_EFFECTS_CCC)
1103 nvme_init_ctrl_finish(ctrl);
1104 if (effects & (NVME_CMD_EFFECTS_NIC | NVME_CMD_EFFECTS_NCC)) {
1105 nvme_queue_scan(ctrl);
1106 flush_work(&ctrl->scan_work);
1109 switch (cmd->common.opcode) {
1110 case nvme_admin_set_features:
1111 switch (le32_to_cpu(cmd->common.cdw10) & 0xFF) {
1112 case NVME_FEAT_KATO:
1114 * Keep alive commands interval on the host should be
1115 * updated when KATO is modified by Set Features
1119 nvme_update_keep_alive(ctrl, cmd);
1130 int nvme_execute_passthru_rq(struct request *rq)
1132 struct nvme_command *cmd = nvme_req(rq)->cmd;
1133 struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
1134 struct nvme_ns *ns = rq->q->queuedata;
1138 effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
1139 ret = nvme_execute_rq(rq, false);
1140 if (effects) /* nothing to be done for zero cmd effects */
1141 nvme_passthru_end(ctrl, effects, cmd, ret);
1145 EXPORT_SYMBOL_NS_GPL(nvme_execute_passthru_rq, NVME_TARGET_PASSTHRU);
1148 * Recommended frequency for KATO commands per NVMe 1.4 section 7.12.1:
1150 * The host should send Keep Alive commands at half of the Keep Alive Timeout
1151 * accounting for transport roundtrip times [..].
1153 static void nvme_queue_keep_alive_work(struct nvme_ctrl *ctrl)
1155 queue_delayed_work(nvme_wq, &ctrl->ka_work, ctrl->kato * HZ / 2);
1158 static void nvme_keep_alive_end_io(struct request *rq, blk_status_t status)
1160 struct nvme_ctrl *ctrl = rq->end_io_data;
1161 unsigned long flags;
1162 bool startka = false;
1164 blk_mq_free_request(rq);
1167 dev_err(ctrl->device,
1168 "failed nvme_keep_alive_end_io error=%d\n",
1173 ctrl->comp_seen = false;
1174 spin_lock_irqsave(&ctrl->lock, flags);
1175 if (ctrl->state == NVME_CTRL_LIVE ||
1176 ctrl->state == NVME_CTRL_CONNECTING)
1178 spin_unlock_irqrestore(&ctrl->lock, flags);
1180 nvme_queue_keep_alive_work(ctrl);
1183 static void nvme_keep_alive_work(struct work_struct *work)
1185 struct nvme_ctrl *ctrl = container_of(to_delayed_work(work),
1186 struct nvme_ctrl, ka_work);
1187 bool comp_seen = ctrl->comp_seen;
1190 if ((ctrl->ctratt & NVME_CTRL_ATTR_TBKAS) && comp_seen) {
1191 dev_dbg(ctrl->device,
1192 "reschedule traffic based keep-alive timer\n");
1193 ctrl->comp_seen = false;
1194 nvme_queue_keep_alive_work(ctrl);
1198 rq = blk_mq_alloc_request(ctrl->admin_q, nvme_req_op(&ctrl->ka_cmd),
1199 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
1201 /* allocation failure, reset the controller */
1202 dev_err(ctrl->device, "keep-alive failed: %ld\n", PTR_ERR(rq));
1203 nvme_reset_ctrl(ctrl);
1206 nvme_init_request(rq, &ctrl->ka_cmd);
1208 rq->timeout = ctrl->kato * HZ;
1209 rq->end_io = nvme_keep_alive_end_io;
1210 rq->end_io_data = ctrl;
1211 rq->rq_flags |= RQF_QUIET;
1212 blk_execute_rq_nowait(rq, false);
1215 static void nvme_start_keep_alive(struct nvme_ctrl *ctrl)
1217 if (unlikely(ctrl->kato == 0))
1220 nvme_queue_keep_alive_work(ctrl);
1223 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl)
1225 if (unlikely(ctrl->kato == 0))
1228 cancel_delayed_work_sync(&ctrl->ka_work);
1230 EXPORT_SYMBOL_GPL(nvme_stop_keep_alive);
1232 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
1233 struct nvme_command *cmd)
1235 unsigned int new_kato =
1236 DIV_ROUND_UP(le32_to_cpu(cmd->common.cdw11), 1000);
1238 dev_info(ctrl->device,
1239 "keep alive interval updated from %u ms to %u ms\n",
1240 ctrl->kato * 1000 / 2, new_kato * 1000 / 2);
1242 nvme_stop_keep_alive(ctrl);
1243 ctrl->kato = new_kato;
1244 nvme_start_keep_alive(ctrl);
1248 * In NVMe 1.0 the CNS field was just a binary controller or namespace
1249 * flag, thus sending any new CNS opcodes has a big chance of not working.
1250 * Qemu unfortunately had that bug after reporting a 1.1 version compliance
1251 * (but not for any later version).
1253 static bool nvme_ctrl_limited_cns(struct nvme_ctrl *ctrl)
1255 if (ctrl->quirks & NVME_QUIRK_IDENTIFY_CNS)
1256 return ctrl->vs < NVME_VS(1, 2, 0);
1257 return ctrl->vs < NVME_VS(1, 1, 0);
1260 static int nvme_identify_ctrl(struct nvme_ctrl *dev, struct nvme_id_ctrl **id)
1262 struct nvme_command c = { };
1265 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1266 c.identify.opcode = nvme_admin_identify;
1267 c.identify.cns = NVME_ID_CNS_CTRL;
1269 *id = kmalloc(sizeof(struct nvme_id_ctrl), GFP_KERNEL);
1273 error = nvme_submit_sync_cmd(dev->admin_q, &c, *id,
1274 sizeof(struct nvme_id_ctrl));
1280 static int nvme_process_ns_desc(struct nvme_ctrl *ctrl, struct nvme_ns_ids *ids,
1281 struct nvme_ns_id_desc *cur, bool *csi_seen)
1283 const char *warn_str = "ctrl returned bogus length:";
1286 switch (cur->nidt) {
1287 case NVME_NIDT_EUI64:
1288 if (cur->nidl != NVME_NIDT_EUI64_LEN) {
1289 dev_warn(ctrl->device, "%s %d for NVME_NIDT_EUI64\n",
1290 warn_str, cur->nidl);
1293 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1294 return NVME_NIDT_EUI64_LEN;
1295 memcpy(ids->eui64, data + sizeof(*cur), NVME_NIDT_EUI64_LEN);
1296 return NVME_NIDT_EUI64_LEN;
1297 case NVME_NIDT_NGUID:
1298 if (cur->nidl != NVME_NIDT_NGUID_LEN) {
1299 dev_warn(ctrl->device, "%s %d for NVME_NIDT_NGUID\n",
1300 warn_str, cur->nidl);
1303 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1304 return NVME_NIDT_NGUID_LEN;
1305 memcpy(ids->nguid, data + sizeof(*cur), NVME_NIDT_NGUID_LEN);
1306 return NVME_NIDT_NGUID_LEN;
1307 case NVME_NIDT_UUID:
1308 if (cur->nidl != NVME_NIDT_UUID_LEN) {
1309 dev_warn(ctrl->device, "%s %d for NVME_NIDT_UUID\n",
1310 warn_str, cur->nidl);
1313 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID)
1314 return NVME_NIDT_UUID_LEN;
1315 uuid_copy(&ids->uuid, data + sizeof(*cur));
1316 return NVME_NIDT_UUID_LEN;
1318 if (cur->nidl != NVME_NIDT_CSI_LEN) {
1319 dev_warn(ctrl->device, "%s %d for NVME_NIDT_CSI\n",
1320 warn_str, cur->nidl);
1323 memcpy(&ids->csi, data + sizeof(*cur), NVME_NIDT_CSI_LEN);
1325 return NVME_NIDT_CSI_LEN;
1327 /* Skip unknown types */
1332 static int nvme_identify_ns_descs(struct nvme_ctrl *ctrl, unsigned nsid,
1333 struct nvme_ns_ids *ids)
1335 struct nvme_command c = { };
1336 bool csi_seen = false;
1337 int status, pos, len;
1340 if (ctrl->vs < NVME_VS(1, 3, 0) && !nvme_multi_css(ctrl))
1342 if (ctrl->quirks & NVME_QUIRK_NO_NS_DESC_LIST)
1345 c.identify.opcode = nvme_admin_identify;
1346 c.identify.nsid = cpu_to_le32(nsid);
1347 c.identify.cns = NVME_ID_CNS_NS_DESC_LIST;
1349 data = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
1353 status = nvme_submit_sync_cmd(ctrl->admin_q, &c, data,
1354 NVME_IDENTIFY_DATA_SIZE);
1356 dev_warn(ctrl->device,
1357 "Identify Descriptors failed (nsid=%u, status=0x%x)\n",
1362 for (pos = 0; pos < NVME_IDENTIFY_DATA_SIZE; pos += len) {
1363 struct nvme_ns_id_desc *cur = data + pos;
1368 len = nvme_process_ns_desc(ctrl, ids, cur, &csi_seen);
1372 len += sizeof(*cur);
1375 if (nvme_multi_css(ctrl) && !csi_seen) {
1376 dev_warn(ctrl->device, "Command set not reported for nsid:%d\n",
1386 static int nvme_identify_ns(struct nvme_ctrl *ctrl, unsigned nsid,
1387 struct nvme_ns_ids *ids, struct nvme_id_ns **id)
1389 struct nvme_command c = { };
1392 /* gcc-4.4.4 (at least) has issues with initializers and anon unions */
1393 c.identify.opcode = nvme_admin_identify;
1394 c.identify.nsid = cpu_to_le32(nsid);
1395 c.identify.cns = NVME_ID_CNS_NS;
1397 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1401 error = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1403 dev_warn(ctrl->device, "Identify namespace failed (%d)\n", error);
1407 error = NVME_SC_INVALID_NS | NVME_SC_DNR;
1408 if ((*id)->ncap == 0) /* namespace not allocated or attached */
1412 if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
1413 dev_info(ctrl->device,
1414 "Ignoring bogus Namespace Identifiers\n");
1416 if (ctrl->vs >= NVME_VS(1, 1, 0) &&
1417 !memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
1418 memcpy(ids->eui64, (*id)->eui64, sizeof(ids->eui64));
1419 if (ctrl->vs >= NVME_VS(1, 2, 0) &&
1420 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
1421 memcpy(ids->nguid, (*id)->nguid, sizeof(ids->nguid));
1431 static int nvme_identify_ns_cs_indep(struct nvme_ctrl *ctrl, unsigned nsid,
1432 struct nvme_id_ns_cs_indep **id)
1434 struct nvme_command c = {
1435 .identify.opcode = nvme_admin_identify,
1436 .identify.nsid = cpu_to_le32(nsid),
1437 .identify.cns = NVME_ID_CNS_NS_CS_INDEP,
1441 *id = kmalloc(sizeof(**id), GFP_KERNEL);
1445 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, *id, sizeof(**id));
1447 dev_warn(ctrl->device,
1448 "Identify namespace (CS independent) failed (%d)\n",
1457 static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
1458 unsigned int dword11, void *buffer, size_t buflen, u32 *result)
1460 union nvme_result res = { 0 };
1461 struct nvme_command c = { };
1464 c.features.opcode = op;
1465 c.features.fid = cpu_to_le32(fid);
1466 c.features.dword11 = cpu_to_le32(dword11);
1468 ret = __nvme_submit_sync_cmd(dev->admin_q, &c, &res,
1469 buffer, buflen, 0, NVME_QID_ANY, 0, 0);
1470 if (ret >= 0 && result)
1471 *result = le32_to_cpu(res.u32);
1475 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
1476 unsigned int dword11, void *buffer, size_t buflen,
1479 return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
1482 EXPORT_SYMBOL_GPL(nvme_set_features);
1484 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
1485 unsigned int dword11, void *buffer, size_t buflen,
1488 return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
1491 EXPORT_SYMBOL_GPL(nvme_get_features);
1493 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count)
1495 u32 q_count = (*count - 1) | ((*count - 1) << 16);
1497 int status, nr_io_queues;
1499 status = nvme_set_features(ctrl, NVME_FEAT_NUM_QUEUES, q_count, NULL, 0,
1505 * Degraded controllers might return an error when setting the queue
1506 * count. We still want to be able to bring them online and offer
1507 * access to the admin queue, as that might be only way to fix them up.
1510 dev_err(ctrl->device, "Could not set queue count (%d)\n", status);
1513 nr_io_queues = min(result & 0xffff, result >> 16) + 1;
1514 *count = min(*count, nr_io_queues);
1519 EXPORT_SYMBOL_GPL(nvme_set_queue_count);
1521 #define NVME_AEN_SUPPORTED \
1522 (NVME_AEN_CFG_NS_ATTR | NVME_AEN_CFG_FW_ACT | \
1523 NVME_AEN_CFG_ANA_CHANGE | NVME_AEN_CFG_DISC_CHANGE)
1525 static void nvme_enable_aen(struct nvme_ctrl *ctrl)
1527 u32 result, supported_aens = ctrl->oaes & NVME_AEN_SUPPORTED;
1530 if (!supported_aens)
1533 status = nvme_set_features(ctrl, NVME_FEAT_ASYNC_EVENT, supported_aens,
1536 dev_warn(ctrl->device, "Failed to configure AEN (cfg %x)\n",
1539 queue_work(nvme_wq, &ctrl->async_event_work);
1542 static int nvme_ns_open(struct nvme_ns *ns)
1545 /* should never be called due to GENHD_FL_HIDDEN */
1546 if (WARN_ON_ONCE(nvme_ns_head_multipath(ns->head)))
1548 if (!nvme_get_ns(ns))
1550 if (!try_module_get(ns->ctrl->ops->module))
1561 static void nvme_ns_release(struct nvme_ns *ns)
1564 module_put(ns->ctrl->ops->module);
1568 static int nvme_open(struct block_device *bdev, fmode_t mode)
1570 return nvme_ns_open(bdev->bd_disk->private_data);
1573 static void nvme_release(struct gendisk *disk, fmode_t mode)
1575 nvme_ns_release(disk->private_data);
1578 int nvme_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1580 /* some standard values */
1581 geo->heads = 1 << 6;
1582 geo->sectors = 1 << 5;
1583 geo->cylinders = get_capacity(bdev->bd_disk) >> 11;
1587 #ifdef CONFIG_BLK_DEV_INTEGRITY
1588 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1589 u32 max_integrity_segments)
1591 struct blk_integrity integrity = { };
1593 switch (ns->pi_type) {
1594 case NVME_NS_DPS_PI_TYPE3:
1595 switch (ns->guard_type) {
1596 case NVME_NVM_NS_16B_GUARD:
1597 integrity.profile = &t10_pi_type3_crc;
1598 integrity.tag_size = sizeof(u16) + sizeof(u32);
1599 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1601 case NVME_NVM_NS_64B_GUARD:
1602 integrity.profile = &ext_pi_type3_crc64;
1603 integrity.tag_size = sizeof(u16) + 6;
1604 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1607 integrity.profile = NULL;
1611 case NVME_NS_DPS_PI_TYPE1:
1612 case NVME_NS_DPS_PI_TYPE2:
1613 switch (ns->guard_type) {
1614 case NVME_NVM_NS_16B_GUARD:
1615 integrity.profile = &t10_pi_type1_crc;
1616 integrity.tag_size = sizeof(u16);
1617 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1619 case NVME_NVM_NS_64B_GUARD:
1620 integrity.profile = &ext_pi_type1_crc64;
1621 integrity.tag_size = sizeof(u16);
1622 integrity.flags |= BLK_INTEGRITY_DEVICE_CAPABLE;
1625 integrity.profile = NULL;
1630 integrity.profile = NULL;
1634 integrity.tuple_size = ns->ms;
1635 blk_integrity_register(disk, &integrity);
1636 blk_queue_max_integrity_segments(disk->queue, max_integrity_segments);
1639 static void nvme_init_integrity(struct gendisk *disk, struct nvme_ns *ns,
1640 u32 max_integrity_segments)
1643 #endif /* CONFIG_BLK_DEV_INTEGRITY */
1645 static void nvme_config_discard(struct gendisk *disk, struct nvme_ns *ns)
1647 struct nvme_ctrl *ctrl = ns->ctrl;
1648 struct request_queue *queue = disk->queue;
1649 u32 size = queue_logical_block_size(queue);
1651 if (ctrl->max_discard_sectors == 0) {
1652 blk_queue_max_discard_sectors(queue, 0);
1656 BUILD_BUG_ON(PAGE_SIZE / sizeof(struct nvme_dsm_range) <
1657 NVME_DSM_MAX_RANGES);
1659 queue->limits.discard_granularity = size;
1661 /* If discard is already enabled, don't reset queue limits */
1662 if (queue->limits.max_discard_sectors)
1665 if (ctrl->dmrsl && ctrl->dmrsl <= nvme_sect_to_lba(ns, UINT_MAX))
1666 ctrl->max_discard_sectors = nvme_lba_to_sect(ns, ctrl->dmrsl);
1668 blk_queue_max_discard_sectors(queue, ctrl->max_discard_sectors);
1669 blk_queue_max_discard_segments(queue, ctrl->max_discard_segments);
1671 if (ctrl->quirks & NVME_QUIRK_DEALLOCATE_ZEROES)
1672 blk_queue_max_write_zeroes_sectors(queue, UINT_MAX);
1675 static bool nvme_ns_ids_equal(struct nvme_ns_ids *a, struct nvme_ns_ids *b)
1677 return uuid_equal(&a->uuid, &b->uuid) &&
1678 memcmp(&a->nguid, &b->nguid, sizeof(a->nguid)) == 0 &&
1679 memcmp(&a->eui64, &b->eui64, sizeof(a->eui64)) == 0 &&
1683 static int nvme_init_ms(struct nvme_ns *ns, struct nvme_id_ns *id)
1685 bool first = id->dps & NVME_NS_DPS_PI_FIRST;
1686 unsigned lbaf = nvme_lbaf_index(id->flbas);
1687 struct nvme_ctrl *ctrl = ns->ctrl;
1688 struct nvme_command c = { };
1689 struct nvme_id_ns_nvm *nvm;
1694 ns->ms = le16_to_cpu(id->lbaf[lbaf].ms);
1695 if (!(ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)) {
1696 ns->pi_size = sizeof(struct t10_pi_tuple);
1697 ns->guard_type = NVME_NVM_NS_16B_GUARD;
1701 nvm = kzalloc(sizeof(*nvm), GFP_KERNEL);
1705 c.identify.opcode = nvme_admin_identify;
1706 c.identify.nsid = cpu_to_le32(ns->head->ns_id);
1707 c.identify.cns = NVME_ID_CNS_CS_NS;
1708 c.identify.csi = NVME_CSI_NVM;
1710 ret = nvme_submit_sync_cmd(ns->ctrl->admin_q, &c, nvm, sizeof(*nvm));
1714 elbaf = le32_to_cpu(nvm->elbaf[lbaf]);
1716 /* no support for storage tag formats right now */
1717 if (nvme_elbaf_sts(elbaf))
1720 ns->guard_type = nvme_elbaf_guard_type(elbaf);
1721 switch (ns->guard_type) {
1722 case NVME_NVM_NS_64B_GUARD:
1723 ns->pi_size = sizeof(struct crc64_pi_tuple);
1725 case NVME_NVM_NS_16B_GUARD:
1726 ns->pi_size = sizeof(struct t10_pi_tuple);
1735 if (ns->pi_size && (first || ns->ms == ns->pi_size))
1736 ns->pi_type = id->dps & NVME_NS_DPS_PI_MASK;
1743 static void nvme_configure_metadata(struct nvme_ns *ns, struct nvme_id_ns *id)
1745 struct nvme_ctrl *ctrl = ns->ctrl;
1747 if (nvme_init_ms(ns, id))
1750 ns->features &= ~(NVME_NS_METADATA_SUPPORTED | NVME_NS_EXT_LBAS);
1751 if (!ns->ms || !(ctrl->ops->flags & NVME_F_METADATA_SUPPORTED))
1754 if (ctrl->ops->flags & NVME_F_FABRICS) {
1756 * The NVMe over Fabrics specification only supports metadata as
1757 * part of the extended data LBA. We rely on HCA/HBA support to
1758 * remap the separate metadata buffer from the block layer.
1760 if (WARN_ON_ONCE(!(id->flbas & NVME_NS_FLBAS_META_EXT)))
1763 ns->features |= NVME_NS_EXT_LBAS;
1766 * The current fabrics transport drivers support namespace
1767 * metadata formats only if nvme_ns_has_pi() returns true.
1768 * Suppress support for all other formats so the namespace will
1769 * have a 0 capacity and not be usable through the block stack.
1771 * Note, this check will need to be modified if any drivers
1772 * gain the ability to use other metadata formats.
1774 if (ctrl->max_integrity_segments && nvme_ns_has_pi(ns))
1775 ns->features |= NVME_NS_METADATA_SUPPORTED;
1778 * For PCIe controllers, we can't easily remap the separate
1779 * metadata buffer from the block layer and thus require a
1780 * separate metadata buffer for block layer metadata/PI support.
1781 * We allow extended LBAs for the passthrough interface, though.
1783 if (id->flbas & NVME_NS_FLBAS_META_EXT)
1784 ns->features |= NVME_NS_EXT_LBAS;
1786 ns->features |= NVME_NS_METADATA_SUPPORTED;
1790 static void nvme_set_queue_limits(struct nvme_ctrl *ctrl,
1791 struct request_queue *q)
1793 bool vwc = ctrl->vwc & NVME_CTRL_VWC_PRESENT;
1795 if (ctrl->max_hw_sectors) {
1797 (ctrl->max_hw_sectors / (NVME_CTRL_PAGE_SIZE >> 9)) + 1;
1799 max_segments = min_not_zero(max_segments, ctrl->max_segments);
1800 blk_queue_max_hw_sectors(q, ctrl->max_hw_sectors);
1801 blk_queue_max_segments(q, min_t(u32, max_segments, USHRT_MAX));
1803 blk_queue_virt_boundary(q, NVME_CTRL_PAGE_SIZE - 1);
1804 blk_queue_dma_alignment(q, 3);
1805 blk_queue_write_cache(q, vwc, vwc);
1808 static void nvme_update_disk_info(struct gendisk *disk,
1809 struct nvme_ns *ns, struct nvme_id_ns *id)
1811 sector_t capacity = nvme_lba_to_sect(ns, le64_to_cpu(id->nsze));
1812 unsigned short bs = 1 << ns->lba_shift;
1813 u32 atomic_bs, phys_bs, io_opt = 0;
1816 * The block layer can't support LBA sizes larger than the page size
1817 * yet, so catch this early and don't allow block I/O.
1819 if (ns->lba_shift > PAGE_SHIFT) {
1824 blk_integrity_unregister(disk);
1826 atomic_bs = phys_bs = bs;
1827 if (id->nabo == 0) {
1829 * Bit 1 indicates whether NAWUPF is defined for this namespace
1830 * and whether it should be used instead of AWUPF. If NAWUPF ==
1831 * 0 then AWUPF must be used instead.
1833 if (id->nsfeat & NVME_NS_FEAT_ATOMICS && id->nawupf)
1834 atomic_bs = (1 + le16_to_cpu(id->nawupf)) * bs;
1836 atomic_bs = (1 + ns->ctrl->subsys->awupf) * bs;
1839 if (id->nsfeat & NVME_NS_FEAT_IO_OPT) {
1840 /* NPWG = Namespace Preferred Write Granularity */
1841 phys_bs = bs * (1 + le16_to_cpu(id->npwg));
1842 /* NOWS = Namespace Optimal Write Size */
1843 io_opt = bs * (1 + le16_to_cpu(id->nows));
1846 blk_queue_logical_block_size(disk->queue, bs);
1848 * Linux filesystems assume writing a single physical block is
1849 * an atomic operation. Hence limit the physical block size to the
1850 * value of the Atomic Write Unit Power Fail parameter.
1852 blk_queue_physical_block_size(disk->queue, min(phys_bs, atomic_bs));
1853 blk_queue_io_min(disk->queue, phys_bs);
1854 blk_queue_io_opt(disk->queue, io_opt);
1857 * Register a metadata profile for PI, or the plain non-integrity NVMe
1858 * metadata masquerading as Type 0 if supported, otherwise reject block
1859 * I/O to namespaces with metadata except when the namespace supports
1860 * PI, as it can strip/insert in that case.
1863 if (IS_ENABLED(CONFIG_BLK_DEV_INTEGRITY) &&
1864 (ns->features & NVME_NS_METADATA_SUPPORTED))
1865 nvme_init_integrity(disk, ns,
1866 ns->ctrl->max_integrity_segments);
1867 else if (!nvme_ns_has_pi(ns))
1871 set_capacity_and_notify(disk, capacity);
1873 nvme_config_discard(disk, ns);
1874 blk_queue_max_write_zeroes_sectors(disk->queue,
1875 ns->ctrl->max_zeroes_sectors);
1878 static inline bool nvme_first_scan(struct gendisk *disk)
1880 /* nvme_alloc_ns() scans the disk prior to adding it */
1881 return !disk_live(disk);
1884 static void nvme_set_chunk_sectors(struct nvme_ns *ns, struct nvme_id_ns *id)
1886 struct nvme_ctrl *ctrl = ns->ctrl;
1889 if ((ctrl->quirks & NVME_QUIRK_STRIPE_SIZE) &&
1890 is_power_of_2(ctrl->max_hw_sectors))
1891 iob = ctrl->max_hw_sectors;
1893 iob = nvme_lba_to_sect(ns, le16_to_cpu(id->noiob));
1898 if (!is_power_of_2(iob)) {
1899 if (nvme_first_scan(ns->disk))
1900 pr_warn("%s: ignoring unaligned IO boundary:%u\n",
1901 ns->disk->disk_name, iob);
1905 if (blk_queue_is_zoned(ns->disk->queue)) {
1906 if (nvme_first_scan(ns->disk))
1907 pr_warn("%s: ignoring zoned namespace IO boundary\n",
1908 ns->disk->disk_name);
1912 blk_queue_chunk_sectors(ns->queue, iob);
1915 static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_id_ns *id)
1917 unsigned lbaf = nvme_lbaf_index(id->flbas);
1920 blk_mq_freeze_queue(ns->disk->queue);
1921 ns->lba_shift = id->lbaf[lbaf].ds;
1922 nvme_set_queue_limits(ns->ctrl, ns->queue);
1924 nvme_configure_metadata(ns, id);
1925 nvme_set_chunk_sectors(ns, id);
1926 nvme_update_disk_info(ns->disk, ns, id);
1928 if (ns->head->ids.csi == NVME_CSI_ZNS) {
1929 ret = nvme_update_zone_info(ns, lbaf);
1934 set_disk_ro(ns->disk, (id->nsattr & NVME_NS_ATTR_RO) ||
1935 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1936 set_bit(NVME_NS_READY, &ns->flags);
1937 blk_mq_unfreeze_queue(ns->disk->queue);
1939 if (blk_queue_is_zoned(ns->queue)) {
1940 ret = nvme_revalidate_zones(ns);
1941 if (ret && !nvme_first_scan(ns->disk))
1945 if (nvme_ns_head_multipath(ns->head)) {
1946 blk_mq_freeze_queue(ns->head->disk->queue);
1947 nvme_update_disk_info(ns->head->disk, ns, id);
1948 set_disk_ro(ns->head->disk,
1949 (id->nsattr & NVME_NS_ATTR_RO) ||
1950 test_bit(NVME_NS_FORCE_RO, &ns->flags));
1951 nvme_mpath_revalidate_paths(ns);
1952 blk_stack_limits(&ns->head->disk->queue->limits,
1953 &ns->queue->limits, 0);
1954 disk_update_readahead(ns->head->disk);
1955 blk_mq_unfreeze_queue(ns->head->disk->queue);
1961 * If probing fails due an unsupported feature, hide the block device,
1962 * but still allow other access.
1964 if (ret == -ENODEV) {
1965 ns->disk->flags |= GENHD_FL_HIDDEN;
1966 set_bit(NVME_NS_READY, &ns->flags);
1969 blk_mq_unfreeze_queue(ns->disk->queue);
1973 static char nvme_pr_type(enum pr_type type)
1976 case PR_WRITE_EXCLUSIVE:
1978 case PR_EXCLUSIVE_ACCESS:
1980 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1982 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1984 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1986 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1993 static int nvme_send_ns_head_pr_command(struct block_device *bdev,
1994 struct nvme_command *c, u8 data[16])
1996 struct nvme_ns_head *head = bdev->bd_disk->private_data;
1997 int srcu_idx = srcu_read_lock(&head->srcu);
1998 struct nvme_ns *ns = nvme_find_path(head);
1999 int ret = -EWOULDBLOCK;
2002 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2003 ret = nvme_submit_sync_cmd(ns->queue, c, data, 16);
2005 srcu_read_unlock(&head->srcu, srcu_idx);
2009 static int nvme_send_ns_pr_command(struct nvme_ns *ns, struct nvme_command *c,
2012 c->common.nsid = cpu_to_le32(ns->head->ns_id);
2013 return nvme_submit_sync_cmd(ns->queue, c, data, 16);
2016 static int nvme_pr_command(struct block_device *bdev, u32 cdw10,
2017 u64 key, u64 sa_key, u8 op)
2019 struct nvme_command c = { };
2020 u8 data[16] = { 0, };
2022 put_unaligned_le64(key, &data[0]);
2023 put_unaligned_le64(sa_key, &data[8]);
2025 c.common.opcode = op;
2026 c.common.cdw10 = cpu_to_le32(cdw10);
2028 if (IS_ENABLED(CONFIG_NVME_MULTIPATH) &&
2029 bdev->bd_disk->fops == &nvme_ns_head_ops)
2030 return nvme_send_ns_head_pr_command(bdev, &c, data);
2031 return nvme_send_ns_pr_command(bdev->bd_disk->private_data, &c, data);
2034 static int nvme_pr_register(struct block_device *bdev, u64 old,
2035 u64 new, unsigned flags)
2039 if (flags & ~PR_FL_IGNORE_KEY)
2042 cdw10 = old ? 2 : 0;
2043 cdw10 |= (flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0;
2044 cdw10 |= (1 << 30) | (1 << 31); /* PTPL=1 */
2045 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_register);
2048 static int nvme_pr_reserve(struct block_device *bdev, u64 key,
2049 enum pr_type type, unsigned flags)
2053 if (flags & ~PR_FL_IGNORE_KEY)
2056 cdw10 = nvme_pr_type(type) << 8;
2057 cdw10 |= ((flags & PR_FL_IGNORE_KEY) ? 1 << 3 : 0);
2058 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_acquire);
2061 static int nvme_pr_preempt(struct block_device *bdev, u64 old, u64 new,
2062 enum pr_type type, bool abort)
2064 u32 cdw10 = nvme_pr_type(type) << 8 | (abort ? 2 : 1);
2066 return nvme_pr_command(bdev, cdw10, old, new, nvme_cmd_resv_acquire);
2069 static int nvme_pr_clear(struct block_device *bdev, u64 key)
2071 u32 cdw10 = 1 | (key ? 1 << 3 : 0);
2073 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_register);
2076 static int nvme_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2078 u32 cdw10 = nvme_pr_type(type) << 8 | (key ? 1 << 3 : 0);
2080 return nvme_pr_command(bdev, cdw10, key, 0, nvme_cmd_resv_release);
2083 const struct pr_ops nvme_pr_ops = {
2084 .pr_register = nvme_pr_register,
2085 .pr_reserve = nvme_pr_reserve,
2086 .pr_release = nvme_pr_release,
2087 .pr_preempt = nvme_pr_preempt,
2088 .pr_clear = nvme_pr_clear,
2091 #ifdef CONFIG_BLK_SED_OPAL
2092 int nvme_sec_submit(void *data, u16 spsp, u8 secp, void *buffer, size_t len,
2095 struct nvme_ctrl *ctrl = data;
2096 struct nvme_command cmd = { };
2099 cmd.common.opcode = nvme_admin_security_send;
2101 cmd.common.opcode = nvme_admin_security_recv;
2102 cmd.common.nsid = 0;
2103 cmd.common.cdw10 = cpu_to_le32(((u32)secp) << 24 | ((u32)spsp) << 8);
2104 cmd.common.cdw11 = cpu_to_le32(len);
2106 return __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, buffer, len, 0,
2107 NVME_QID_ANY, 1, 0);
2109 EXPORT_SYMBOL_GPL(nvme_sec_submit);
2110 #endif /* CONFIG_BLK_SED_OPAL */
2112 #ifdef CONFIG_BLK_DEV_ZONED
2113 static int nvme_report_zones(struct gendisk *disk, sector_t sector,
2114 unsigned int nr_zones, report_zones_cb cb, void *data)
2116 return nvme_ns_report_zones(disk->private_data, sector, nr_zones, cb,
2120 #define nvme_report_zones NULL
2121 #endif /* CONFIG_BLK_DEV_ZONED */
2123 static const struct block_device_operations nvme_bdev_ops = {
2124 .owner = THIS_MODULE,
2125 .ioctl = nvme_ioctl,
2127 .release = nvme_release,
2128 .getgeo = nvme_getgeo,
2129 .report_zones = nvme_report_zones,
2130 .pr_ops = &nvme_pr_ops,
2133 static int nvme_wait_ready(struct nvme_ctrl *ctrl, u32 timeout, bool enabled)
2135 unsigned long timeout_jiffies = ((timeout + 1) * HZ / 2) + jiffies;
2136 u32 csts, bit = enabled ? NVME_CSTS_RDY : 0;
2139 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2142 if ((csts & NVME_CSTS_RDY) == bit)
2145 usleep_range(1000, 2000);
2146 if (fatal_signal_pending(current))
2148 if (time_after(jiffies, timeout_jiffies)) {
2149 dev_err(ctrl->device,
2150 "Device not ready; aborting %s, CSTS=0x%x\n",
2151 enabled ? "initialisation" : "reset", csts);
2160 * If the device has been passed off to us in an enabled state, just clear
2161 * the enabled bit. The spec says we should set the 'shutdown notification
2162 * bits', but doing so may cause the device to complete commands to the
2163 * admin queue ... and we don't know what memory that might be pointing at!
2165 int nvme_disable_ctrl(struct nvme_ctrl *ctrl)
2169 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2170 ctrl->ctrl_config &= ~NVME_CC_ENABLE;
2172 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2176 if (ctrl->quirks & NVME_QUIRK_DELAY_BEFORE_CHK_RDY)
2177 msleep(NVME_QUIRK_DELAY_AMOUNT);
2179 return nvme_wait_ready(ctrl, NVME_CAP_TIMEOUT(ctrl->cap), false);
2181 EXPORT_SYMBOL_GPL(nvme_disable_ctrl);
2183 int nvme_enable_ctrl(struct nvme_ctrl *ctrl)
2185 unsigned dev_page_min;
2189 ret = ctrl->ops->reg_read64(ctrl, NVME_REG_CAP, &ctrl->cap);
2191 dev_err(ctrl->device, "Reading CAP failed (%d)\n", ret);
2194 dev_page_min = NVME_CAP_MPSMIN(ctrl->cap) + 12;
2196 if (NVME_CTRL_PAGE_SHIFT < dev_page_min) {
2197 dev_err(ctrl->device,
2198 "Minimum device page size %u too large for host (%u)\n",
2199 1 << dev_page_min, 1 << NVME_CTRL_PAGE_SHIFT);
2203 if (NVME_CAP_CSS(ctrl->cap) & NVME_CAP_CSS_CSI)
2204 ctrl->ctrl_config = NVME_CC_CSS_CSI;
2206 ctrl->ctrl_config = NVME_CC_CSS_NVM;
2208 if (ctrl->cap & NVME_CAP_CRMS_CRWMS) {
2211 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CRTO, &crto);
2213 dev_err(ctrl->device, "Reading CRTO failed (%d)\n",
2218 if (ctrl->cap & NVME_CAP_CRMS_CRIMS) {
2219 ctrl->ctrl_config |= NVME_CC_CRIME;
2220 timeout = NVME_CRTO_CRIMT(crto);
2222 timeout = NVME_CRTO_CRWMT(crto);
2225 timeout = NVME_CAP_TIMEOUT(ctrl->cap);
2228 ctrl->ctrl_config |= (NVME_CTRL_PAGE_SHIFT - 12) << NVME_CC_MPS_SHIFT;
2229 ctrl->ctrl_config |= NVME_CC_AMS_RR | NVME_CC_SHN_NONE;
2230 ctrl->ctrl_config |= NVME_CC_IOSQES | NVME_CC_IOCQES;
2231 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2235 /* Flush write to device (required if transport is PCI) */
2236 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CC, &ctrl->ctrl_config);
2240 ctrl->ctrl_config |= NVME_CC_ENABLE;
2241 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2244 return nvme_wait_ready(ctrl, timeout, true);
2246 EXPORT_SYMBOL_GPL(nvme_enable_ctrl);
2248 int nvme_shutdown_ctrl(struct nvme_ctrl *ctrl)
2250 unsigned long timeout = jiffies + (ctrl->shutdown_timeout * HZ);
2254 ctrl->ctrl_config &= ~NVME_CC_SHN_MASK;
2255 ctrl->ctrl_config |= NVME_CC_SHN_NORMAL;
2257 ret = ctrl->ops->reg_write32(ctrl, NVME_REG_CC, ctrl->ctrl_config);
2261 while ((ret = ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts)) == 0) {
2262 if ((csts & NVME_CSTS_SHST_MASK) == NVME_CSTS_SHST_CMPLT)
2266 if (fatal_signal_pending(current))
2268 if (time_after(jiffies, timeout)) {
2269 dev_err(ctrl->device,
2270 "Device shutdown incomplete; abort shutdown\n");
2277 EXPORT_SYMBOL_GPL(nvme_shutdown_ctrl);
2279 static int nvme_configure_timestamp(struct nvme_ctrl *ctrl)
2284 if (!(ctrl->oncs & NVME_CTRL_ONCS_TIMESTAMP))
2287 ts = cpu_to_le64(ktime_to_ms(ktime_get_real()));
2288 ret = nvme_set_features(ctrl, NVME_FEAT_TIMESTAMP, 0, &ts, sizeof(ts),
2291 dev_warn_once(ctrl->device,
2292 "could not set timestamp (%d)\n", ret);
2296 static int nvme_configure_host_options(struct nvme_ctrl *ctrl)
2298 struct nvme_feat_host_behavior *host;
2299 u8 acre = 0, lbafee = 0;
2302 /* Don't bother enabling the feature if retry delay is not reported */
2304 acre = NVME_ENABLE_ACRE;
2305 if (ctrl->ctratt & NVME_CTRL_ATTR_ELBAS)
2306 lbafee = NVME_ENABLE_LBAFEE;
2308 if (!acre && !lbafee)
2311 host = kzalloc(sizeof(*host), GFP_KERNEL);
2316 host->lbafee = lbafee;
2317 ret = nvme_set_features(ctrl, NVME_FEAT_HOST_BEHAVIOR, 0,
2318 host, sizeof(*host), NULL);
2324 * The function checks whether the given total (exlat + enlat) latency of
2325 * a power state allows the latter to be used as an APST transition target.
2326 * It does so by comparing the latency to the primary and secondary latency
2327 * tolerances defined by module params. If there's a match, the corresponding
2328 * timeout value is returned and the matching tolerance index (1 or 2) is
2331 static bool nvme_apst_get_transition_time(u64 total_latency,
2332 u64 *transition_time, unsigned *last_index)
2334 if (total_latency <= apst_primary_latency_tol_us) {
2335 if (*last_index == 1)
2338 *transition_time = apst_primary_timeout_ms;
2341 if (apst_secondary_timeout_ms &&
2342 total_latency <= apst_secondary_latency_tol_us) {
2343 if (*last_index <= 2)
2346 *transition_time = apst_secondary_timeout_ms;
2353 * APST (Autonomous Power State Transition) lets us program a table of power
2354 * state transitions that the controller will perform automatically.
2356 * Depending on module params, one of the two supported techniques will be used:
2358 * - If the parameters provide explicit timeouts and tolerances, they will be
2359 * used to build a table with up to 2 non-operational states to transition to.
2360 * The default parameter values were selected based on the values used by
2361 * Microsoft's and Intel's NVMe drivers. Yet, since we don't implement dynamic
2362 * regeneration of the APST table in the event of switching between external
2363 * and battery power, the timeouts and tolerances reflect a compromise
2364 * between values used by Microsoft for AC and battery scenarios.
2365 * - If not, we'll configure the table with a simple heuristic: we are willing
2366 * to spend at most 2% of the time transitioning between power states.
2367 * Therefore, when running in any given state, we will enter the next
2368 * lower-power non-operational state after waiting 50 * (enlat + exlat)
2369 * microseconds, as long as that state's exit latency is under the requested
2372 * We will not autonomously enter any non-operational state for which the total
2373 * latency exceeds ps_max_latency_us.
2375 * Users can set ps_max_latency_us to zero to turn off APST.
2377 static int nvme_configure_apst(struct nvme_ctrl *ctrl)
2379 struct nvme_feat_auto_pst *table;
2386 unsigned last_lt_index = UINT_MAX;
2389 * If APST isn't supported or if we haven't been initialized yet,
2390 * then don't do anything.
2395 if (ctrl->npss > 31) {
2396 dev_warn(ctrl->device, "NPSS is invalid; not using APST\n");
2400 table = kzalloc(sizeof(*table), GFP_KERNEL);
2404 if (!ctrl->apst_enabled || ctrl->ps_max_latency_us == 0) {
2405 /* Turn off APST. */
2406 dev_dbg(ctrl->device, "APST disabled\n");
2411 * Walk through all states from lowest- to highest-power.
2412 * According to the spec, lower-numbered states use more power. NPSS,
2413 * despite the name, is the index of the lowest-power state, not the
2416 for (state = (int)ctrl->npss; state >= 0; state--) {
2417 u64 total_latency_us, exit_latency_us, transition_ms;
2420 table->entries[state] = target;
2423 * Don't allow transitions to the deepest state if it's quirked
2426 if (state == ctrl->npss &&
2427 (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS))
2431 * Is this state a useful non-operational state for higher-power
2432 * states to autonomously transition to?
2434 if (!(ctrl->psd[state].flags & NVME_PS_FLAGS_NON_OP_STATE))
2437 exit_latency_us = (u64)le32_to_cpu(ctrl->psd[state].exit_lat);
2438 if (exit_latency_us > ctrl->ps_max_latency_us)
2441 total_latency_us = exit_latency_us +
2442 le32_to_cpu(ctrl->psd[state].entry_lat);
2445 * This state is good. It can be used as the APST idle target
2446 * for higher power states.
2448 if (apst_primary_timeout_ms && apst_primary_latency_tol_us) {
2449 if (!nvme_apst_get_transition_time(total_latency_us,
2450 &transition_ms, &last_lt_index))
2453 transition_ms = total_latency_us + 19;
2454 do_div(transition_ms, 20);
2455 if (transition_ms > (1 << 24) - 1)
2456 transition_ms = (1 << 24) - 1;
2459 target = cpu_to_le64((state << 3) | (transition_ms << 8));
2462 if (total_latency_us > max_lat_us)
2463 max_lat_us = total_latency_us;
2467 dev_dbg(ctrl->device, "APST enabled but no non-operational states are available\n");
2469 dev_dbg(ctrl->device, "APST enabled: max PS = %d, max round-trip latency = %lluus, table = %*phN\n",
2470 max_ps, max_lat_us, (int)sizeof(*table), table);
2474 ret = nvme_set_features(ctrl, NVME_FEAT_AUTO_PST, apste,
2475 table, sizeof(*table), NULL);
2477 dev_err(ctrl->device, "failed to set APST feature (%d)\n", ret);
2482 static void nvme_set_latency_tolerance(struct device *dev, s32 val)
2484 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
2488 case PM_QOS_LATENCY_TOLERANCE_NO_CONSTRAINT:
2489 case PM_QOS_LATENCY_ANY:
2497 if (ctrl->ps_max_latency_us != latency) {
2498 ctrl->ps_max_latency_us = latency;
2499 if (ctrl->state == NVME_CTRL_LIVE)
2500 nvme_configure_apst(ctrl);
2504 struct nvme_core_quirk_entry {
2506 * NVMe model and firmware strings are padded with spaces. For
2507 * simplicity, strings in the quirk table are padded with NULLs
2513 unsigned long quirks;
2516 static const struct nvme_core_quirk_entry core_quirks[] = {
2519 * This Toshiba device seems to die using any APST states. See:
2520 * https://bugs.launchpad.net/ubuntu/+source/linux/+bug/1678184/comments/11
2523 .mn = "THNSF5256GPUK TOSHIBA",
2524 .quirks = NVME_QUIRK_NO_APST,
2528 * This LiteON CL1-3D*-Q11 firmware version has a race
2529 * condition associated with actions related to suspend to idle
2530 * LiteON has resolved the problem in future firmware
2534 .quirks = NVME_QUIRK_SIMPLE_SUSPEND,
2538 * This Kioxia CD6-V Series / HPE PE8030 device times out and
2539 * aborts I/O during any load, but more easily reproducible
2540 * with discards (fstrim).
2542 * The device is left in a state where it is also not possible
2543 * to use "nvme set-feature" to disable APST, but booting with
2544 * nvme_core.default_ps_max_latency=0 works.
2547 .mn = "KCD6XVUL6T40",
2548 .quirks = NVME_QUIRK_NO_APST,
2552 * The external Samsung X5 SSD fails initialization without a
2553 * delay before checking if it is ready and has a whole set of
2554 * other problems. To make this even more interesting, it
2555 * shares the PCI ID with internal Samsung 970 Evo Plus that
2556 * does not need or want these quirks.
2559 .mn = "Samsung Portable SSD X5",
2560 .quirks = NVME_QUIRK_DELAY_BEFORE_CHK_RDY |
2561 NVME_QUIRK_NO_DEEPEST_PS |
2562 NVME_QUIRK_IGNORE_DEV_SUBNQN,
2566 /* match is null-terminated but idstr is space-padded. */
2567 static bool string_matches(const char *idstr, const char *match, size_t len)
2574 matchlen = strlen(match);
2575 WARN_ON_ONCE(matchlen > len);
2577 if (memcmp(idstr, match, matchlen))
2580 for (; matchlen < len; matchlen++)
2581 if (idstr[matchlen] != ' ')
2587 static bool quirk_matches(const struct nvme_id_ctrl *id,
2588 const struct nvme_core_quirk_entry *q)
2590 return q->vid == le16_to_cpu(id->vid) &&
2591 string_matches(id->mn, q->mn, sizeof(id->mn)) &&
2592 string_matches(id->fr, q->fr, sizeof(id->fr));
2595 static void nvme_init_subnqn(struct nvme_subsystem *subsys, struct nvme_ctrl *ctrl,
2596 struct nvme_id_ctrl *id)
2601 if(!(ctrl->quirks & NVME_QUIRK_IGNORE_DEV_SUBNQN)) {
2602 nqnlen = strnlen(id->subnqn, NVMF_NQN_SIZE);
2603 if (nqnlen > 0 && nqnlen < NVMF_NQN_SIZE) {
2604 strlcpy(subsys->subnqn, id->subnqn, NVMF_NQN_SIZE);
2608 if (ctrl->vs >= NVME_VS(1, 2, 1))
2609 dev_warn(ctrl->device, "missing or invalid SUBNQN field.\n");
2612 /* Generate a "fake" NQN per Figure 254 in NVMe 1.3 + ECN 001 */
2613 off = snprintf(subsys->subnqn, NVMF_NQN_SIZE,
2614 "nqn.2014.08.org.nvmexpress:%04x%04x",
2615 le16_to_cpu(id->vid), le16_to_cpu(id->ssvid));
2616 memcpy(subsys->subnqn + off, id->sn, sizeof(id->sn));
2617 off += sizeof(id->sn);
2618 memcpy(subsys->subnqn + off, id->mn, sizeof(id->mn));
2619 off += sizeof(id->mn);
2620 memset(subsys->subnqn + off, 0, sizeof(subsys->subnqn) - off);
2623 static void nvme_release_subsystem(struct device *dev)
2625 struct nvme_subsystem *subsys =
2626 container_of(dev, struct nvme_subsystem, dev);
2628 if (subsys->instance >= 0)
2629 ida_free(&nvme_instance_ida, subsys->instance);
2633 static void nvme_destroy_subsystem(struct kref *ref)
2635 struct nvme_subsystem *subsys =
2636 container_of(ref, struct nvme_subsystem, ref);
2638 mutex_lock(&nvme_subsystems_lock);
2639 list_del(&subsys->entry);
2640 mutex_unlock(&nvme_subsystems_lock);
2642 ida_destroy(&subsys->ns_ida);
2643 device_del(&subsys->dev);
2644 put_device(&subsys->dev);
2647 static void nvme_put_subsystem(struct nvme_subsystem *subsys)
2649 kref_put(&subsys->ref, nvme_destroy_subsystem);
2652 static struct nvme_subsystem *__nvme_find_get_subsystem(const char *subsysnqn)
2654 struct nvme_subsystem *subsys;
2656 lockdep_assert_held(&nvme_subsystems_lock);
2659 * Fail matches for discovery subsystems. This results
2660 * in each discovery controller bound to a unique subsystem.
2661 * This avoids issues with validating controller values
2662 * that can only be true when there is a single unique subsystem.
2663 * There may be multiple and completely independent entities
2664 * that provide discovery controllers.
2666 if (!strcmp(subsysnqn, NVME_DISC_SUBSYS_NAME))
2669 list_for_each_entry(subsys, &nvme_subsystems, entry) {
2670 if (strcmp(subsys->subnqn, subsysnqn))
2672 if (!kref_get_unless_zero(&subsys->ref))
2680 #define SUBSYS_ATTR_RO(_name, _mode, _show) \
2681 struct device_attribute subsys_attr_##_name = \
2682 __ATTR(_name, _mode, _show, NULL)
2684 static ssize_t nvme_subsys_show_nqn(struct device *dev,
2685 struct device_attribute *attr,
2688 struct nvme_subsystem *subsys =
2689 container_of(dev, struct nvme_subsystem, dev);
2691 return sysfs_emit(buf, "%s\n", subsys->subnqn);
2693 static SUBSYS_ATTR_RO(subsysnqn, S_IRUGO, nvme_subsys_show_nqn);
2695 static ssize_t nvme_subsys_show_type(struct device *dev,
2696 struct device_attribute *attr,
2699 struct nvme_subsystem *subsys =
2700 container_of(dev, struct nvme_subsystem, dev);
2702 switch (subsys->subtype) {
2704 return sysfs_emit(buf, "discovery\n");
2706 return sysfs_emit(buf, "nvm\n");
2708 return sysfs_emit(buf, "reserved\n");
2711 static SUBSYS_ATTR_RO(subsystype, S_IRUGO, nvme_subsys_show_type);
2713 #define nvme_subsys_show_str_function(field) \
2714 static ssize_t subsys_##field##_show(struct device *dev, \
2715 struct device_attribute *attr, char *buf) \
2717 struct nvme_subsystem *subsys = \
2718 container_of(dev, struct nvme_subsystem, dev); \
2719 return sysfs_emit(buf, "%.*s\n", \
2720 (int)sizeof(subsys->field), subsys->field); \
2722 static SUBSYS_ATTR_RO(field, S_IRUGO, subsys_##field##_show);
2724 nvme_subsys_show_str_function(model);
2725 nvme_subsys_show_str_function(serial);
2726 nvme_subsys_show_str_function(firmware_rev);
2728 static struct attribute *nvme_subsys_attrs[] = {
2729 &subsys_attr_model.attr,
2730 &subsys_attr_serial.attr,
2731 &subsys_attr_firmware_rev.attr,
2732 &subsys_attr_subsysnqn.attr,
2733 &subsys_attr_subsystype.attr,
2734 #ifdef CONFIG_NVME_MULTIPATH
2735 &subsys_attr_iopolicy.attr,
2740 static const struct attribute_group nvme_subsys_attrs_group = {
2741 .attrs = nvme_subsys_attrs,
2744 static const struct attribute_group *nvme_subsys_attrs_groups[] = {
2745 &nvme_subsys_attrs_group,
2749 static inline bool nvme_discovery_ctrl(struct nvme_ctrl *ctrl)
2751 return ctrl->opts && ctrl->opts->discovery_nqn;
2754 static bool nvme_validate_cntlid(struct nvme_subsystem *subsys,
2755 struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2757 struct nvme_ctrl *tmp;
2759 lockdep_assert_held(&nvme_subsystems_lock);
2761 list_for_each_entry(tmp, &subsys->ctrls, subsys_entry) {
2762 if (nvme_state_terminal(tmp))
2765 if (tmp->cntlid == ctrl->cntlid) {
2766 dev_err(ctrl->device,
2767 "Duplicate cntlid %u with %s, subsys %s, rejecting\n",
2768 ctrl->cntlid, dev_name(tmp->device),
2773 if ((id->cmic & NVME_CTRL_CMIC_MULTI_CTRL) ||
2774 nvme_discovery_ctrl(ctrl))
2777 dev_err(ctrl->device,
2778 "Subsystem does not support multiple controllers\n");
2785 static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
2787 struct nvme_subsystem *subsys, *found;
2790 subsys = kzalloc(sizeof(*subsys), GFP_KERNEL);
2794 subsys->instance = -1;
2795 mutex_init(&subsys->lock);
2796 kref_init(&subsys->ref);
2797 INIT_LIST_HEAD(&subsys->ctrls);
2798 INIT_LIST_HEAD(&subsys->nsheads);
2799 nvme_init_subnqn(subsys, ctrl, id);
2800 memcpy(subsys->serial, id->sn, sizeof(subsys->serial));
2801 memcpy(subsys->model, id->mn, sizeof(subsys->model));
2802 memcpy(subsys->firmware_rev, id->fr, sizeof(subsys->firmware_rev));
2803 subsys->vendor_id = le16_to_cpu(id->vid);
2804 subsys->cmic = id->cmic;
2806 /* Versions prior to 1.4 don't necessarily report a valid type */
2807 if (id->cntrltype == NVME_CTRL_DISC ||
2808 !strcmp(subsys->subnqn, NVME_DISC_SUBSYS_NAME))
2809 subsys->subtype = NVME_NQN_DISC;
2811 subsys->subtype = NVME_NQN_NVME;
2813 if (nvme_discovery_ctrl(ctrl) && subsys->subtype != NVME_NQN_DISC) {
2814 dev_err(ctrl->device,
2815 "Subsystem %s is not a discovery controller",
2820 subsys->awupf = le16_to_cpu(id->awupf);
2821 nvme_mpath_default_iopolicy(subsys);
2823 subsys->dev.class = nvme_subsys_class;
2824 subsys->dev.release = nvme_release_subsystem;
2825 subsys->dev.groups = nvme_subsys_attrs_groups;
2826 dev_set_name(&subsys->dev, "nvme-subsys%d", ctrl->instance);
2827 device_initialize(&subsys->dev);
2829 mutex_lock(&nvme_subsystems_lock);
2830 found = __nvme_find_get_subsystem(subsys->subnqn);
2832 put_device(&subsys->dev);
2835 if (!nvme_validate_cntlid(subsys, ctrl, id)) {
2837 goto out_put_subsystem;
2840 ret = device_add(&subsys->dev);
2842 dev_err(ctrl->device,
2843 "failed to register subsystem device.\n");
2844 put_device(&subsys->dev);
2847 ida_init(&subsys->ns_ida);
2848 list_add_tail(&subsys->entry, &nvme_subsystems);
2851 ret = sysfs_create_link(&subsys->dev.kobj, &ctrl->device->kobj,
2852 dev_name(ctrl->device));
2854 dev_err(ctrl->device,
2855 "failed to create sysfs link from subsystem.\n");
2856 goto out_put_subsystem;
2860 subsys->instance = ctrl->instance;
2861 ctrl->subsys = subsys;
2862 list_add_tail(&ctrl->subsys_entry, &subsys->ctrls);
2863 mutex_unlock(&nvme_subsystems_lock);
2867 nvme_put_subsystem(subsys);
2869 mutex_unlock(&nvme_subsystems_lock);
2873 int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
2874 void *log, size_t size, u64 offset)
2876 struct nvme_command c = { };
2877 u32 dwlen = nvme_bytes_to_numd(size);
2879 c.get_log_page.opcode = nvme_admin_get_log_page;
2880 c.get_log_page.nsid = cpu_to_le32(nsid);
2881 c.get_log_page.lid = log_page;
2882 c.get_log_page.lsp = lsp;
2883 c.get_log_page.numdl = cpu_to_le16(dwlen & ((1 << 16) - 1));
2884 c.get_log_page.numdu = cpu_to_le16(dwlen >> 16);
2885 c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
2886 c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
2887 c.get_log_page.csi = csi;
2889 return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
2892 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
2893 struct nvme_effects_log **log)
2895 struct nvme_effects_log *cel = xa_load(&ctrl->cels, csi);
2901 cel = kzalloc(sizeof(*cel), GFP_KERNEL);
2905 ret = nvme_get_log(ctrl, 0x00, NVME_LOG_CMD_EFFECTS, 0, csi,
2906 cel, sizeof(*cel), 0);
2912 xa_store(&ctrl->cels, csi, cel, GFP_KERNEL);
2918 static inline u32 nvme_mps_to_sectors(struct nvme_ctrl *ctrl, u32 units)
2920 u32 page_shift = NVME_CAP_MPSMIN(ctrl->cap) + 12, val;
2922 if (check_shl_overflow(1U, units + page_shift - 9, &val))
2927 static int nvme_init_non_mdts_limits(struct nvme_ctrl *ctrl)
2929 struct nvme_command c = { };
2930 struct nvme_id_ctrl_nvm *id;
2933 if (ctrl->oncs & NVME_CTRL_ONCS_DSM) {
2934 ctrl->max_discard_sectors = UINT_MAX;
2935 ctrl->max_discard_segments = NVME_DSM_MAX_RANGES;
2937 ctrl->max_discard_sectors = 0;
2938 ctrl->max_discard_segments = 0;
2942 * Even though NVMe spec explicitly states that MDTS is not applicable
2943 * to the write-zeroes, we are cautious and limit the size to the
2944 * controllers max_hw_sectors value, which is based on the MDTS field
2945 * and possibly other limiting factors.
2947 if ((ctrl->oncs & NVME_CTRL_ONCS_WRITE_ZEROES) &&
2948 !(ctrl->quirks & NVME_QUIRK_DISABLE_WRITE_ZEROES))
2949 ctrl->max_zeroes_sectors = ctrl->max_hw_sectors;
2951 ctrl->max_zeroes_sectors = 0;
2953 if (nvme_ctrl_limited_cns(ctrl))
2956 id = kzalloc(sizeof(*id), GFP_KERNEL);
2960 c.identify.opcode = nvme_admin_identify;
2961 c.identify.cns = NVME_ID_CNS_CS_CTRL;
2962 c.identify.csi = NVME_CSI_NVM;
2964 ret = nvme_submit_sync_cmd(ctrl->admin_q, &c, id, sizeof(*id));
2969 ctrl->max_discard_segments = id->dmrl;
2970 ctrl->dmrsl = le32_to_cpu(id->dmrsl);
2972 ctrl->max_zeroes_sectors = nvme_mps_to_sectors(ctrl, id->wzsl);
2979 static int nvme_init_identify(struct nvme_ctrl *ctrl)
2981 struct nvme_id_ctrl *id;
2983 bool prev_apst_enabled;
2986 ret = nvme_identify_ctrl(ctrl, &id);
2988 dev_err(ctrl->device, "Identify Controller failed (%d)\n", ret);
2992 if (id->lpa & NVME_CTRL_LPA_CMD_EFFECTS_LOG) {
2993 ret = nvme_get_effects_log(ctrl, NVME_CSI_NVM, &ctrl->effects);
2998 if (!(ctrl->ops->flags & NVME_F_FABRICS))
2999 ctrl->cntlid = le16_to_cpu(id->cntlid);
3001 if (!ctrl->identified) {
3004 ret = nvme_init_subsystem(ctrl, id);
3009 * Check for quirks. Quirk can depend on firmware version,
3010 * so, in principle, the set of quirks present can change
3011 * across a reset. As a possible future enhancement, we
3012 * could re-scan for quirks every time we reinitialize
3013 * the device, but we'd have to make sure that the driver
3014 * behaves intelligently if the quirks change.
3016 for (i = 0; i < ARRAY_SIZE(core_quirks); i++) {
3017 if (quirk_matches(id, &core_quirks[i]))
3018 ctrl->quirks |= core_quirks[i].quirks;
3022 if (force_apst && (ctrl->quirks & NVME_QUIRK_NO_DEEPEST_PS)) {
3023 dev_warn(ctrl->device, "forcibly allowing all power states due to nvme_core.force_apst -- use at your own risk\n");
3024 ctrl->quirks &= ~NVME_QUIRK_NO_DEEPEST_PS;
3027 ctrl->crdt[0] = le16_to_cpu(id->crdt1);
3028 ctrl->crdt[1] = le16_to_cpu(id->crdt2);
3029 ctrl->crdt[2] = le16_to_cpu(id->crdt3);
3031 ctrl->oacs = le16_to_cpu(id->oacs);
3032 ctrl->oncs = le16_to_cpu(id->oncs);
3033 ctrl->mtfa = le16_to_cpu(id->mtfa);
3034 ctrl->oaes = le32_to_cpu(id->oaes);
3035 ctrl->wctemp = le16_to_cpu(id->wctemp);
3036 ctrl->cctemp = le16_to_cpu(id->cctemp);
3038 atomic_set(&ctrl->abort_limit, id->acl + 1);
3039 ctrl->vwc = id->vwc;
3041 max_hw_sectors = nvme_mps_to_sectors(ctrl, id->mdts);
3043 max_hw_sectors = UINT_MAX;
3044 ctrl->max_hw_sectors =
3045 min_not_zero(ctrl->max_hw_sectors, max_hw_sectors);
3047 nvme_set_queue_limits(ctrl, ctrl->admin_q);
3048 ctrl->sgls = le32_to_cpu(id->sgls);
3049 ctrl->kas = le16_to_cpu(id->kas);
3050 ctrl->max_namespaces = le32_to_cpu(id->mnan);
3051 ctrl->ctratt = le32_to_cpu(id->ctratt);
3053 ctrl->cntrltype = id->cntrltype;
3054 ctrl->dctype = id->dctype;
3058 u32 transition_time = le32_to_cpu(id->rtd3e) / USEC_PER_SEC;
3060 ctrl->shutdown_timeout = clamp_t(unsigned int, transition_time,
3061 shutdown_timeout, 60);
3063 if (ctrl->shutdown_timeout != shutdown_timeout)
3064 dev_info(ctrl->device,
3065 "Shutdown timeout set to %u seconds\n",
3066 ctrl->shutdown_timeout);
3068 ctrl->shutdown_timeout = shutdown_timeout;
3070 ctrl->npss = id->npss;
3071 ctrl->apsta = id->apsta;
3072 prev_apst_enabled = ctrl->apst_enabled;
3073 if (ctrl->quirks & NVME_QUIRK_NO_APST) {
3074 if (force_apst && id->apsta) {
3075 dev_warn(ctrl->device, "forcibly allowing APST due to nvme_core.force_apst -- use at your own risk\n");
3076 ctrl->apst_enabled = true;
3078 ctrl->apst_enabled = false;
3081 ctrl->apst_enabled = id->apsta;
3083 memcpy(ctrl->psd, id->psd, sizeof(ctrl->psd));
3085 if (ctrl->ops->flags & NVME_F_FABRICS) {
3086 ctrl->icdoff = le16_to_cpu(id->icdoff);
3087 ctrl->ioccsz = le32_to_cpu(id->ioccsz);
3088 ctrl->iorcsz = le32_to_cpu(id->iorcsz);
3089 ctrl->maxcmd = le16_to_cpu(id->maxcmd);
3092 * In fabrics we need to verify the cntlid matches the
3095 if (ctrl->cntlid != le16_to_cpu(id->cntlid)) {
3096 dev_err(ctrl->device,
3097 "Mismatching cntlid: Connect %u vs Identify "
3099 ctrl->cntlid, le16_to_cpu(id->cntlid));
3104 if (!nvme_discovery_ctrl(ctrl) && !ctrl->kas) {
3105 dev_err(ctrl->device,
3106 "keep-alive support is mandatory for fabrics\n");
3111 ctrl->hmpre = le32_to_cpu(id->hmpre);
3112 ctrl->hmmin = le32_to_cpu(id->hmmin);
3113 ctrl->hmminds = le32_to_cpu(id->hmminds);
3114 ctrl->hmmaxd = le16_to_cpu(id->hmmaxd);
3117 ret = nvme_mpath_init_identify(ctrl, id);
3121 if (ctrl->apst_enabled && !prev_apst_enabled)
3122 dev_pm_qos_expose_latency_tolerance(ctrl->device);
3123 else if (!ctrl->apst_enabled && prev_apst_enabled)
3124 dev_pm_qos_hide_latency_tolerance(ctrl->device);
3132 * Initialize the cached copies of the Identify data and various controller
3133 * register in our nvme_ctrl structure. This should be called as soon as
3134 * the admin queue is fully up and running.
3136 int nvme_init_ctrl_finish(struct nvme_ctrl *ctrl)
3140 ret = ctrl->ops->reg_read32(ctrl, NVME_REG_VS, &ctrl->vs);
3142 dev_err(ctrl->device, "Reading VS failed (%d)\n", ret);
3146 ctrl->sqsize = min_t(u16, NVME_CAP_MQES(ctrl->cap), ctrl->sqsize);
3148 if (ctrl->vs >= NVME_VS(1, 1, 0))
3149 ctrl->subsystem = NVME_CAP_NSSRC(ctrl->cap);
3151 ret = nvme_init_identify(ctrl);
3155 ret = nvme_configure_apst(ctrl);
3159 ret = nvme_configure_timestamp(ctrl);
3163 ret = nvme_configure_host_options(ctrl);
3167 if (!ctrl->identified && !nvme_discovery_ctrl(ctrl)) {
3168 ret = nvme_hwmon_init(ctrl);
3173 ctrl->identified = true;
3177 EXPORT_SYMBOL_GPL(nvme_init_ctrl_finish);
3179 static int nvme_dev_open(struct inode *inode, struct file *file)
3181 struct nvme_ctrl *ctrl =
3182 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3184 switch (ctrl->state) {
3185 case NVME_CTRL_LIVE:
3188 return -EWOULDBLOCK;
3191 nvme_get_ctrl(ctrl);
3192 if (!try_module_get(ctrl->ops->module)) {
3193 nvme_put_ctrl(ctrl);
3197 file->private_data = ctrl;
3201 static int nvme_dev_release(struct inode *inode, struct file *file)
3203 struct nvme_ctrl *ctrl =
3204 container_of(inode->i_cdev, struct nvme_ctrl, cdev);
3206 module_put(ctrl->ops->module);
3207 nvme_put_ctrl(ctrl);
3211 static const struct file_operations nvme_dev_fops = {
3212 .owner = THIS_MODULE,
3213 .open = nvme_dev_open,
3214 .release = nvme_dev_release,
3215 .unlocked_ioctl = nvme_dev_ioctl,
3216 .compat_ioctl = compat_ptr_ioctl,
3217 .uring_cmd = nvme_dev_uring_cmd,
3220 static ssize_t nvme_sysfs_reset(struct device *dev,
3221 struct device_attribute *attr, const char *buf,
3224 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3227 ret = nvme_reset_ctrl_sync(ctrl);
3232 static DEVICE_ATTR(reset_controller, S_IWUSR, NULL, nvme_sysfs_reset);
3234 static ssize_t nvme_sysfs_rescan(struct device *dev,
3235 struct device_attribute *attr, const char *buf,
3238 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3240 nvme_queue_scan(ctrl);
3243 static DEVICE_ATTR(rescan_controller, S_IWUSR, NULL, nvme_sysfs_rescan);
3245 static inline struct nvme_ns_head *dev_to_ns_head(struct device *dev)
3247 struct gendisk *disk = dev_to_disk(dev);
3249 if (disk->fops == &nvme_bdev_ops)
3250 return nvme_get_ns_from_dev(dev)->head;
3252 return disk->private_data;
3255 static ssize_t wwid_show(struct device *dev, struct device_attribute *attr,
3258 struct nvme_ns_head *head = dev_to_ns_head(dev);
3259 struct nvme_ns_ids *ids = &head->ids;
3260 struct nvme_subsystem *subsys = head->subsys;
3261 int serial_len = sizeof(subsys->serial);
3262 int model_len = sizeof(subsys->model);
3264 if (!uuid_is_null(&ids->uuid))
3265 return sysfs_emit(buf, "uuid.%pU\n", &ids->uuid);
3267 if (memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3268 return sysfs_emit(buf, "eui.%16phN\n", ids->nguid);
3270 if (memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3271 return sysfs_emit(buf, "eui.%8phN\n", ids->eui64);
3273 while (serial_len > 0 && (subsys->serial[serial_len - 1] == ' ' ||
3274 subsys->serial[serial_len - 1] == '\0'))
3276 while (model_len > 0 && (subsys->model[model_len - 1] == ' ' ||
3277 subsys->model[model_len - 1] == '\0'))
3280 return sysfs_emit(buf, "nvme.%04x-%*phN-%*phN-%08x\n", subsys->vendor_id,
3281 serial_len, subsys->serial, model_len, subsys->model,
3284 static DEVICE_ATTR_RO(wwid);
3286 static ssize_t nguid_show(struct device *dev, struct device_attribute *attr,
3289 return sysfs_emit(buf, "%pU\n", dev_to_ns_head(dev)->ids.nguid);
3291 static DEVICE_ATTR_RO(nguid);
3293 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
3296 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3298 /* For backward compatibility expose the NGUID to userspace if
3299 * we have no UUID set
3301 if (uuid_is_null(&ids->uuid)) {
3302 dev_warn_ratelimited(dev,
3303 "No UUID available providing old NGUID\n");
3304 return sysfs_emit(buf, "%pU\n", ids->nguid);
3306 return sysfs_emit(buf, "%pU\n", &ids->uuid);
3308 static DEVICE_ATTR_RO(uuid);
3310 static ssize_t eui_show(struct device *dev, struct device_attribute *attr,
3313 return sysfs_emit(buf, "%8ph\n", dev_to_ns_head(dev)->ids.eui64);
3315 static DEVICE_ATTR_RO(eui);
3317 static ssize_t nsid_show(struct device *dev, struct device_attribute *attr,
3320 return sysfs_emit(buf, "%d\n", dev_to_ns_head(dev)->ns_id);
3322 static DEVICE_ATTR_RO(nsid);
3324 static struct attribute *nvme_ns_id_attrs[] = {
3325 &dev_attr_wwid.attr,
3326 &dev_attr_uuid.attr,
3327 &dev_attr_nguid.attr,
3329 &dev_attr_nsid.attr,
3330 #ifdef CONFIG_NVME_MULTIPATH
3331 &dev_attr_ana_grpid.attr,
3332 &dev_attr_ana_state.attr,
3337 static umode_t nvme_ns_id_attrs_are_visible(struct kobject *kobj,
3338 struct attribute *a, int n)
3340 struct device *dev = container_of(kobj, struct device, kobj);
3341 struct nvme_ns_ids *ids = &dev_to_ns_head(dev)->ids;
3343 if (a == &dev_attr_uuid.attr) {
3344 if (uuid_is_null(&ids->uuid) &&
3345 !memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3348 if (a == &dev_attr_nguid.attr) {
3349 if (!memchr_inv(ids->nguid, 0, sizeof(ids->nguid)))
3352 if (a == &dev_attr_eui.attr) {
3353 if (!memchr_inv(ids->eui64, 0, sizeof(ids->eui64)))
3356 #ifdef CONFIG_NVME_MULTIPATH
3357 if (a == &dev_attr_ana_grpid.attr || a == &dev_attr_ana_state.attr) {
3358 if (dev_to_disk(dev)->fops != &nvme_bdev_ops) /* per-path attr */
3360 if (!nvme_ctrl_use_ana(nvme_get_ns_from_dev(dev)->ctrl))
3367 static const struct attribute_group nvme_ns_id_attr_group = {
3368 .attrs = nvme_ns_id_attrs,
3369 .is_visible = nvme_ns_id_attrs_are_visible,
3372 const struct attribute_group *nvme_ns_id_attr_groups[] = {
3373 &nvme_ns_id_attr_group,
3377 #define nvme_show_str_function(field) \
3378 static ssize_t field##_show(struct device *dev, \
3379 struct device_attribute *attr, char *buf) \
3381 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3382 return sysfs_emit(buf, "%.*s\n", \
3383 (int)sizeof(ctrl->subsys->field), ctrl->subsys->field); \
3385 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3387 nvme_show_str_function(model);
3388 nvme_show_str_function(serial);
3389 nvme_show_str_function(firmware_rev);
3391 #define nvme_show_int_function(field) \
3392 static ssize_t field##_show(struct device *dev, \
3393 struct device_attribute *attr, char *buf) \
3395 struct nvme_ctrl *ctrl = dev_get_drvdata(dev); \
3396 return sysfs_emit(buf, "%d\n", ctrl->field); \
3398 static DEVICE_ATTR(field, S_IRUGO, field##_show, NULL);
3400 nvme_show_int_function(cntlid);
3401 nvme_show_int_function(numa_node);
3402 nvme_show_int_function(queue_count);
3403 nvme_show_int_function(sqsize);
3404 nvme_show_int_function(kato);
3406 static ssize_t nvme_sysfs_delete(struct device *dev,
3407 struct device_attribute *attr, const char *buf,
3410 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3412 if (device_remove_file_self(dev, attr))
3413 nvme_delete_ctrl_sync(ctrl);
3416 static DEVICE_ATTR(delete_controller, S_IWUSR, NULL, nvme_sysfs_delete);
3418 static ssize_t nvme_sysfs_show_transport(struct device *dev,
3419 struct device_attribute *attr,
3422 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3424 return sysfs_emit(buf, "%s\n", ctrl->ops->name);
3426 static DEVICE_ATTR(transport, S_IRUGO, nvme_sysfs_show_transport, NULL);
3428 static ssize_t nvme_sysfs_show_state(struct device *dev,
3429 struct device_attribute *attr,
3432 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3433 static const char *const state_name[] = {
3434 [NVME_CTRL_NEW] = "new",
3435 [NVME_CTRL_LIVE] = "live",
3436 [NVME_CTRL_RESETTING] = "resetting",
3437 [NVME_CTRL_CONNECTING] = "connecting",
3438 [NVME_CTRL_DELETING] = "deleting",
3439 [NVME_CTRL_DELETING_NOIO]= "deleting (no IO)",
3440 [NVME_CTRL_DEAD] = "dead",
3443 if ((unsigned)ctrl->state < ARRAY_SIZE(state_name) &&
3444 state_name[ctrl->state])
3445 return sysfs_emit(buf, "%s\n", state_name[ctrl->state]);
3447 return sysfs_emit(buf, "unknown state\n");
3450 static DEVICE_ATTR(state, S_IRUGO, nvme_sysfs_show_state, NULL);
3452 static ssize_t nvme_sysfs_show_subsysnqn(struct device *dev,
3453 struct device_attribute *attr,
3456 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3458 return sysfs_emit(buf, "%s\n", ctrl->subsys->subnqn);
3460 static DEVICE_ATTR(subsysnqn, S_IRUGO, nvme_sysfs_show_subsysnqn, NULL);
3462 static ssize_t nvme_sysfs_show_hostnqn(struct device *dev,
3463 struct device_attribute *attr,
3466 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3468 return sysfs_emit(buf, "%s\n", ctrl->opts->host->nqn);
3470 static DEVICE_ATTR(hostnqn, S_IRUGO, nvme_sysfs_show_hostnqn, NULL);
3472 static ssize_t nvme_sysfs_show_hostid(struct device *dev,
3473 struct device_attribute *attr,
3476 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3478 return sysfs_emit(buf, "%pU\n", &ctrl->opts->host->id);
3480 static DEVICE_ATTR(hostid, S_IRUGO, nvme_sysfs_show_hostid, NULL);
3482 static ssize_t nvme_sysfs_show_address(struct device *dev,
3483 struct device_attribute *attr,
3486 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3488 return ctrl->ops->get_address(ctrl, buf, PAGE_SIZE);
3490 static DEVICE_ATTR(address, S_IRUGO, nvme_sysfs_show_address, NULL);
3492 static ssize_t nvme_ctrl_loss_tmo_show(struct device *dev,
3493 struct device_attribute *attr, char *buf)
3495 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3496 struct nvmf_ctrl_options *opts = ctrl->opts;
3498 if (ctrl->opts->max_reconnects == -1)
3499 return sysfs_emit(buf, "off\n");
3500 return sysfs_emit(buf, "%d\n",
3501 opts->max_reconnects * opts->reconnect_delay);
3504 static ssize_t nvme_ctrl_loss_tmo_store(struct device *dev,
3505 struct device_attribute *attr, const char *buf, size_t count)
3507 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3508 struct nvmf_ctrl_options *opts = ctrl->opts;
3509 int ctrl_loss_tmo, err;
3511 err = kstrtoint(buf, 10, &ctrl_loss_tmo);
3515 if (ctrl_loss_tmo < 0)
3516 opts->max_reconnects = -1;
3518 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
3519 opts->reconnect_delay);
3522 static DEVICE_ATTR(ctrl_loss_tmo, S_IRUGO | S_IWUSR,
3523 nvme_ctrl_loss_tmo_show, nvme_ctrl_loss_tmo_store);
3525 static ssize_t nvme_ctrl_reconnect_delay_show(struct device *dev,
3526 struct device_attribute *attr, char *buf)
3528 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3530 if (ctrl->opts->reconnect_delay == -1)
3531 return sysfs_emit(buf, "off\n");
3532 return sysfs_emit(buf, "%d\n", ctrl->opts->reconnect_delay);
3535 static ssize_t nvme_ctrl_reconnect_delay_store(struct device *dev,
3536 struct device_attribute *attr, const char *buf, size_t count)
3538 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3542 err = kstrtou32(buf, 10, &v);
3546 ctrl->opts->reconnect_delay = v;
3549 static DEVICE_ATTR(reconnect_delay, S_IRUGO | S_IWUSR,
3550 nvme_ctrl_reconnect_delay_show, nvme_ctrl_reconnect_delay_store);
3552 static ssize_t nvme_ctrl_fast_io_fail_tmo_show(struct device *dev,
3553 struct device_attribute *attr, char *buf)
3555 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3557 if (ctrl->opts->fast_io_fail_tmo == -1)
3558 return sysfs_emit(buf, "off\n");
3559 return sysfs_emit(buf, "%d\n", ctrl->opts->fast_io_fail_tmo);
3562 static ssize_t nvme_ctrl_fast_io_fail_tmo_store(struct device *dev,
3563 struct device_attribute *attr, const char *buf, size_t count)
3565 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3566 struct nvmf_ctrl_options *opts = ctrl->opts;
3567 int fast_io_fail_tmo, err;
3569 err = kstrtoint(buf, 10, &fast_io_fail_tmo);
3573 if (fast_io_fail_tmo < 0)
3574 opts->fast_io_fail_tmo = -1;
3576 opts->fast_io_fail_tmo = fast_io_fail_tmo;
3579 static DEVICE_ATTR(fast_io_fail_tmo, S_IRUGO | S_IWUSR,
3580 nvme_ctrl_fast_io_fail_tmo_show, nvme_ctrl_fast_io_fail_tmo_store);
3582 static ssize_t cntrltype_show(struct device *dev,
3583 struct device_attribute *attr, char *buf)
3585 static const char * const type[] = {
3586 [NVME_CTRL_IO] = "io\n",
3587 [NVME_CTRL_DISC] = "discovery\n",
3588 [NVME_CTRL_ADMIN] = "admin\n",
3590 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3592 if (ctrl->cntrltype > NVME_CTRL_ADMIN || !type[ctrl->cntrltype])
3593 return sysfs_emit(buf, "reserved\n");
3595 return sysfs_emit(buf, type[ctrl->cntrltype]);
3597 static DEVICE_ATTR_RO(cntrltype);
3599 static ssize_t dctype_show(struct device *dev,
3600 struct device_attribute *attr, char *buf)
3602 static const char * const type[] = {
3603 [NVME_DCTYPE_NOT_REPORTED] = "none\n",
3604 [NVME_DCTYPE_DDC] = "ddc\n",
3605 [NVME_DCTYPE_CDC] = "cdc\n",
3607 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3609 if (ctrl->dctype > NVME_DCTYPE_CDC || !type[ctrl->dctype])
3610 return sysfs_emit(buf, "reserved\n");
3612 return sysfs_emit(buf, type[ctrl->dctype]);
3614 static DEVICE_ATTR_RO(dctype);
3616 static struct attribute *nvme_dev_attrs[] = {
3617 &dev_attr_reset_controller.attr,
3618 &dev_attr_rescan_controller.attr,
3619 &dev_attr_model.attr,
3620 &dev_attr_serial.attr,
3621 &dev_attr_firmware_rev.attr,
3622 &dev_attr_cntlid.attr,
3623 &dev_attr_delete_controller.attr,
3624 &dev_attr_transport.attr,
3625 &dev_attr_subsysnqn.attr,
3626 &dev_attr_address.attr,
3627 &dev_attr_state.attr,
3628 &dev_attr_numa_node.attr,
3629 &dev_attr_queue_count.attr,
3630 &dev_attr_sqsize.attr,
3631 &dev_attr_hostnqn.attr,
3632 &dev_attr_hostid.attr,
3633 &dev_attr_ctrl_loss_tmo.attr,
3634 &dev_attr_reconnect_delay.attr,
3635 &dev_attr_fast_io_fail_tmo.attr,
3636 &dev_attr_kato.attr,
3637 &dev_attr_cntrltype.attr,
3638 &dev_attr_dctype.attr,
3642 static umode_t nvme_dev_attrs_are_visible(struct kobject *kobj,
3643 struct attribute *a, int n)
3645 struct device *dev = container_of(kobj, struct device, kobj);
3646 struct nvme_ctrl *ctrl = dev_get_drvdata(dev);
3648 if (a == &dev_attr_delete_controller.attr && !ctrl->ops->delete_ctrl)
3650 if (a == &dev_attr_address.attr && !ctrl->ops->get_address)
3652 if (a == &dev_attr_hostnqn.attr && !ctrl->opts)
3654 if (a == &dev_attr_hostid.attr && !ctrl->opts)
3656 if (a == &dev_attr_ctrl_loss_tmo.attr && !ctrl->opts)
3658 if (a == &dev_attr_reconnect_delay.attr && !ctrl->opts)
3660 if (a == &dev_attr_fast_io_fail_tmo.attr && !ctrl->opts)
3666 static const struct attribute_group nvme_dev_attrs_group = {
3667 .attrs = nvme_dev_attrs,
3668 .is_visible = nvme_dev_attrs_are_visible,
3671 static const struct attribute_group *nvme_dev_attr_groups[] = {
3672 &nvme_dev_attrs_group,
3676 static struct nvme_ns_head *nvme_find_ns_head(struct nvme_ctrl *ctrl,
3679 struct nvme_ns_head *h;
3681 lockdep_assert_held(&ctrl->subsys->lock);
3683 list_for_each_entry(h, &ctrl->subsys->nsheads, entry) {
3685 * Private namespaces can share NSIDs under some conditions.
3686 * In that case we can't use the same ns_head for namespaces
3687 * with the same NSID.
3689 if (h->ns_id != nsid || !nvme_is_unique_nsid(ctrl, h))
3691 if (!list_empty(&h->list) && nvme_tryget_ns_head(h))
3698 static int nvme_subsys_check_duplicate_ids(struct nvme_subsystem *subsys,
3699 struct nvme_ns_ids *ids)
3701 bool has_uuid = !uuid_is_null(&ids->uuid);
3702 bool has_nguid = memchr_inv(ids->nguid, 0, sizeof(ids->nguid));
3703 bool has_eui64 = memchr_inv(ids->eui64, 0, sizeof(ids->eui64));
3704 struct nvme_ns_head *h;
3706 lockdep_assert_held(&subsys->lock);
3708 list_for_each_entry(h, &subsys->nsheads, entry) {
3709 if (has_uuid && uuid_equal(&ids->uuid, &h->ids.uuid))
3712 memcmp(&ids->nguid, &h->ids.nguid, sizeof(ids->nguid)) == 0)
3715 memcmp(&ids->eui64, &h->ids.eui64, sizeof(ids->eui64)) == 0)
3722 static void nvme_cdev_rel(struct device *dev)
3724 ida_free(&nvme_ns_chr_minor_ida, MINOR(dev->devt));
3727 void nvme_cdev_del(struct cdev *cdev, struct device *cdev_device)
3729 cdev_device_del(cdev, cdev_device);
3730 put_device(cdev_device);
3733 int nvme_cdev_add(struct cdev *cdev, struct device *cdev_device,
3734 const struct file_operations *fops, struct module *owner)
3738 minor = ida_alloc(&nvme_ns_chr_minor_ida, GFP_KERNEL);
3741 cdev_device->devt = MKDEV(MAJOR(nvme_ns_chr_devt), minor);
3742 cdev_device->class = nvme_ns_chr_class;
3743 cdev_device->release = nvme_cdev_rel;
3744 device_initialize(cdev_device);
3745 cdev_init(cdev, fops);
3746 cdev->owner = owner;
3747 ret = cdev_device_add(cdev, cdev_device);
3749 put_device(cdev_device);
3754 static int nvme_ns_chr_open(struct inode *inode, struct file *file)
3756 return nvme_ns_open(container_of(inode->i_cdev, struct nvme_ns, cdev));
3759 static int nvme_ns_chr_release(struct inode *inode, struct file *file)
3761 nvme_ns_release(container_of(inode->i_cdev, struct nvme_ns, cdev));
3765 static const struct file_operations nvme_ns_chr_fops = {
3766 .owner = THIS_MODULE,
3767 .open = nvme_ns_chr_open,
3768 .release = nvme_ns_chr_release,
3769 .unlocked_ioctl = nvme_ns_chr_ioctl,
3770 .compat_ioctl = compat_ptr_ioctl,
3771 .uring_cmd = nvme_ns_chr_uring_cmd,
3774 static int nvme_add_ns_cdev(struct nvme_ns *ns)
3778 ns->cdev_device.parent = ns->ctrl->device;
3779 ret = dev_set_name(&ns->cdev_device, "ng%dn%d",
3780 ns->ctrl->instance, ns->head->instance);
3784 return nvme_cdev_add(&ns->cdev, &ns->cdev_device, &nvme_ns_chr_fops,
3785 ns->ctrl->ops->module);
3788 static struct nvme_ns_head *nvme_alloc_ns_head(struct nvme_ctrl *ctrl,
3789 unsigned nsid, struct nvme_ns_ids *ids)
3791 struct nvme_ns_head *head;
3792 size_t size = sizeof(*head);
3795 #ifdef CONFIG_NVME_MULTIPATH
3796 size += num_possible_nodes() * sizeof(struct nvme_ns *);
3799 head = kzalloc(size, GFP_KERNEL);
3802 ret = ida_alloc_min(&ctrl->subsys->ns_ida, 1, GFP_KERNEL);
3805 head->instance = ret;
3806 INIT_LIST_HEAD(&head->list);
3807 ret = init_srcu_struct(&head->srcu);
3809 goto out_ida_remove;
3810 head->subsys = ctrl->subsys;
3813 kref_init(&head->ref);
3815 if (head->ids.csi) {
3816 ret = nvme_get_effects_log(ctrl, head->ids.csi, &head->effects);
3818 goto out_cleanup_srcu;
3820 head->effects = ctrl->effects;
3822 ret = nvme_mpath_alloc_disk(ctrl, head);
3824 goto out_cleanup_srcu;
3826 list_add_tail(&head->entry, &ctrl->subsys->nsheads);
3828 kref_get(&ctrl->subsys->ref);
3832 cleanup_srcu_struct(&head->srcu);
3834 ida_free(&ctrl->subsys->ns_ida, head->instance);
3839 ret = blk_status_to_errno(nvme_error_status(ret));
3840 return ERR_PTR(ret);
3843 static int nvme_global_check_duplicate_ids(struct nvme_subsystem *this,
3844 struct nvme_ns_ids *ids)
3846 struct nvme_subsystem *s;
3850 * Note that this check is racy as we try to avoid holding the global
3851 * lock over the whole ns_head creation. But it is only intended as
3852 * a sanity check anyway.
3854 mutex_lock(&nvme_subsystems_lock);
3855 list_for_each_entry(s, &nvme_subsystems, entry) {
3858 mutex_lock(&s->lock);
3859 ret = nvme_subsys_check_duplicate_ids(s, ids);
3860 mutex_unlock(&s->lock);
3864 mutex_unlock(&nvme_subsystems_lock);
3869 static int nvme_init_ns_head(struct nvme_ns *ns, unsigned nsid,
3870 struct nvme_ns_ids *ids, bool is_shared)
3872 struct nvme_ctrl *ctrl = ns->ctrl;
3873 struct nvme_ns_head *head = NULL;
3876 ret = nvme_global_check_duplicate_ids(ctrl->subsys, ids);
3878 dev_err(ctrl->device,
3879 "globally duplicate IDs for nsid %d\n", nsid);
3880 nvme_print_device_info(ctrl);
3884 mutex_lock(&ctrl->subsys->lock);
3885 head = nvme_find_ns_head(ctrl, nsid);
3887 ret = nvme_subsys_check_duplicate_ids(ctrl->subsys, ids);
3889 dev_err(ctrl->device,
3890 "duplicate IDs in subsystem for nsid %d\n",
3894 head = nvme_alloc_ns_head(ctrl, nsid, ids);
3896 ret = PTR_ERR(head);
3899 head->shared = is_shared;
3902 if (!is_shared || !head->shared) {
3903 dev_err(ctrl->device,
3904 "Duplicate unshared namespace %d\n", nsid);
3905 goto out_put_ns_head;
3907 if (!nvme_ns_ids_equal(&head->ids, ids)) {
3908 dev_err(ctrl->device,
3909 "IDs don't match for shared namespace %d\n",
3911 goto out_put_ns_head;
3914 if (!multipath && !list_empty(&head->list)) {
3915 dev_warn(ctrl->device,
3916 "Found shared namespace %d, but multipathing not supported.\n",
3918 dev_warn_once(ctrl->device,
3919 "Support for shared namespaces without CONFIG_NVME_MULTIPATH is deprecated and will be removed in Linux 6.0\n.");
3923 list_add_tail_rcu(&ns->siblings, &head->list);
3925 mutex_unlock(&ctrl->subsys->lock);
3929 nvme_put_ns_head(head);
3931 mutex_unlock(&ctrl->subsys->lock);
3935 struct nvme_ns *nvme_find_get_ns(struct nvme_ctrl *ctrl, unsigned nsid)
3937 struct nvme_ns *ns, *ret = NULL;
3939 down_read(&ctrl->namespaces_rwsem);
3940 list_for_each_entry(ns, &ctrl->namespaces, list) {
3941 if (ns->head->ns_id == nsid) {
3942 if (!nvme_get_ns(ns))
3947 if (ns->head->ns_id > nsid)
3950 up_read(&ctrl->namespaces_rwsem);
3953 EXPORT_SYMBOL_NS_GPL(nvme_find_get_ns, NVME_TARGET_PASSTHRU);
3956 * Add the namespace to the controller list while keeping the list ordered.
3958 static void nvme_ns_add_to_ctrl_list(struct nvme_ns *ns)
3960 struct nvme_ns *tmp;
3962 list_for_each_entry_reverse(tmp, &ns->ctrl->namespaces, list) {
3963 if (tmp->head->ns_id < ns->head->ns_id) {
3964 list_add(&ns->list, &tmp->list);
3968 list_add(&ns->list, &ns->ctrl->namespaces);
3971 static void nvme_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid,
3972 struct nvme_ns_ids *ids)
3975 struct gendisk *disk;
3976 struct nvme_id_ns *id;
3977 int node = ctrl->numa_node;
3979 if (nvme_identify_ns(ctrl, nsid, ids, &id))
3982 ns = kzalloc_node(sizeof(*ns), GFP_KERNEL, node);
3986 disk = blk_mq_alloc_disk(ctrl->tagset, ns);
3989 disk->fops = &nvme_bdev_ops;
3990 disk->private_data = ns;
3993 ns->queue = disk->queue;
3995 if (ctrl->opts && ctrl->opts->data_digest)
3996 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, ns->queue);
3998 blk_queue_flag_set(QUEUE_FLAG_NONROT, ns->queue);
3999 if (ctrl->ops->flags & NVME_F_PCI_P2PDMA)
4000 blk_queue_flag_set(QUEUE_FLAG_PCI_P2PDMA, ns->queue);
4003 kref_init(&ns->kref);
4005 if (nvme_init_ns_head(ns, nsid, ids, id->nmic & NVME_NS_NMIC_SHARED))
4006 goto out_cleanup_disk;
4009 * If multipathing is enabled, the device name for all disks and not
4010 * just those that represent shared namespaces needs to be based on the
4011 * subsystem instance. Using the controller instance for private
4012 * namespaces could lead to naming collisions between shared and private
4013 * namespaces if they don't use a common numbering scheme.
4015 * If multipathing is not enabled, disk names must use the controller
4016 * instance as shared namespaces will show up as multiple block
4019 if (ns->head->disk) {
4020 sprintf(disk->disk_name, "nvme%dc%dn%d", ctrl->subsys->instance,
4021 ctrl->instance, ns->head->instance);
4022 disk->flags |= GENHD_FL_HIDDEN;
4023 } else if (multipath) {
4024 sprintf(disk->disk_name, "nvme%dn%d", ctrl->subsys->instance,
4025 ns->head->instance);
4027 sprintf(disk->disk_name, "nvme%dn%d", ctrl->instance,
4028 ns->head->instance);
4031 if (nvme_update_ns_info(ns, id))
4034 down_write(&ctrl->namespaces_rwsem);
4035 nvme_ns_add_to_ctrl_list(ns);
4036 up_write(&ctrl->namespaces_rwsem);
4037 nvme_get_ctrl(ctrl);
4039 if (device_add_disk(ctrl->device, ns->disk, nvme_ns_id_attr_groups))
4040 goto out_cleanup_ns_from_list;
4042 if (!nvme_ns_head_multipath(ns->head))
4043 nvme_add_ns_cdev(ns);
4045 nvme_mpath_add_disk(ns, id);
4046 nvme_fault_inject_init(&ns->fault_inject, ns->disk->disk_name);
4051 out_cleanup_ns_from_list:
4052 nvme_put_ctrl(ctrl);
4053 down_write(&ctrl->namespaces_rwsem);
4054 list_del_init(&ns->list);
4055 up_write(&ctrl->namespaces_rwsem);
4057 mutex_lock(&ctrl->subsys->lock);
4058 list_del_rcu(&ns->siblings);
4059 if (list_empty(&ns->head->list))
4060 list_del_init(&ns->head->entry);
4061 mutex_unlock(&ctrl->subsys->lock);
4062 nvme_put_ns_head(ns->head);
4064 blk_cleanup_disk(disk);
4071 static void nvme_ns_remove(struct nvme_ns *ns)
4073 bool last_path = false;
4075 if (test_and_set_bit(NVME_NS_REMOVING, &ns->flags))
4078 clear_bit(NVME_NS_READY, &ns->flags);
4079 set_capacity(ns->disk, 0);
4080 nvme_fault_inject_fini(&ns->fault_inject);
4083 * Ensure that !NVME_NS_READY is seen by other threads to prevent
4084 * this ns going back into current_path.
4086 synchronize_srcu(&ns->head->srcu);
4088 /* wait for concurrent submissions */
4089 if (nvme_mpath_clear_current_path(ns))
4090 synchronize_srcu(&ns->head->srcu);
4092 mutex_lock(&ns->ctrl->subsys->lock);
4093 list_del_rcu(&ns->siblings);
4094 if (list_empty(&ns->head->list)) {
4095 list_del_init(&ns->head->entry);
4098 mutex_unlock(&ns->ctrl->subsys->lock);
4100 /* guarantee not available in head->list */
4103 if (!nvme_ns_head_multipath(ns->head))
4104 nvme_cdev_del(&ns->cdev, &ns->cdev_device);
4105 del_gendisk(ns->disk);
4106 blk_cleanup_queue(ns->queue);
4108 down_write(&ns->ctrl->namespaces_rwsem);
4109 list_del_init(&ns->list);
4110 up_write(&ns->ctrl->namespaces_rwsem);
4113 nvme_mpath_shutdown_disk(ns->head);
4117 static void nvme_ns_remove_by_nsid(struct nvme_ctrl *ctrl, u32 nsid)
4119 struct nvme_ns *ns = nvme_find_get_ns(ctrl, nsid);
4127 static void nvme_validate_ns(struct nvme_ns *ns, struct nvme_ns_ids *ids)
4129 struct nvme_id_ns *id;
4130 int ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4132 if (test_bit(NVME_NS_DEAD, &ns->flags))
4135 ret = nvme_identify_ns(ns->ctrl, ns->head->ns_id, ids, &id);
4139 ret = NVME_SC_INVALID_NS | NVME_SC_DNR;
4140 if (!nvme_ns_ids_equal(&ns->head->ids, ids)) {
4141 dev_err(ns->ctrl->device,
4142 "identifiers changed for nsid %d\n", ns->head->ns_id);
4146 ret = nvme_update_ns_info(ns, id);
4152 * Only remove the namespace if we got a fatal error back from the
4153 * device, otherwise ignore the error and just move on.
4155 * TODO: we should probably schedule a delayed retry here.
4157 if (ret > 0 && (ret & NVME_SC_DNR))
4161 static void nvme_validate_or_alloc_ns(struct nvme_ctrl *ctrl, unsigned nsid)
4163 struct nvme_ns_ids ids = { };
4164 struct nvme_id_ns_cs_indep *id;
4168 if (nvme_identify_ns_descs(ctrl, nsid, &ids))
4172 * Check if the namespace is ready. If not ignore it, we will get an
4173 * AEN once it becomes ready and restart the scan.
4175 if ((ctrl->cap & NVME_CAP_CRMS_CRIMS) &&
4176 !nvme_identify_ns_cs_indep(ctrl, nsid, &id)) {
4177 ready = id->nstat & NVME_NSTAT_NRDY;
4184 ns = nvme_find_get_ns(ctrl, nsid);
4186 nvme_validate_ns(ns, &ids);
4193 nvme_alloc_ns(ctrl, nsid, &ids);
4196 if (!IS_ENABLED(CONFIG_BLK_DEV_ZONED)) {
4197 dev_warn(ctrl->device,
4198 "nsid %u not supported without CONFIG_BLK_DEV_ZONED\n",
4202 if (!nvme_multi_css(ctrl)) {
4203 dev_warn(ctrl->device,
4204 "command set not reported for nsid: %d\n",
4208 nvme_alloc_ns(ctrl, nsid, &ids);
4211 dev_warn(ctrl->device, "unknown csi %u for nsid %u\n",
4217 static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
4220 struct nvme_ns *ns, *next;
4223 down_write(&ctrl->namespaces_rwsem);
4224 list_for_each_entry_safe(ns, next, &ctrl->namespaces, list) {
4225 if (ns->head->ns_id > nsid || test_bit(NVME_NS_DEAD, &ns->flags))
4226 list_move_tail(&ns->list, &rm_list);
4228 up_write(&ctrl->namespaces_rwsem);
4230 list_for_each_entry_safe(ns, next, &rm_list, list)
4235 static int nvme_scan_ns_list(struct nvme_ctrl *ctrl)
4237 const int nr_entries = NVME_IDENTIFY_DATA_SIZE / sizeof(__le32);
4242 if (nvme_ctrl_limited_cns(ctrl))
4245 ns_list = kzalloc(NVME_IDENTIFY_DATA_SIZE, GFP_KERNEL);
4250 struct nvme_command cmd = {
4251 .identify.opcode = nvme_admin_identify,
4252 .identify.cns = NVME_ID_CNS_NS_ACTIVE_LIST,
4253 .identify.nsid = cpu_to_le32(prev),
4256 ret = nvme_submit_sync_cmd(ctrl->admin_q, &cmd, ns_list,
4257 NVME_IDENTIFY_DATA_SIZE);
4259 dev_warn(ctrl->device,
4260 "Identify NS List failed (status=0x%x)\n", ret);
4264 for (i = 0; i < nr_entries; i++) {
4265 u32 nsid = le32_to_cpu(ns_list[i]);
4267 if (!nsid) /* end of the list? */
4269 nvme_validate_or_alloc_ns(ctrl, nsid);
4270 while (++prev < nsid)
4271 nvme_ns_remove_by_nsid(ctrl, prev);
4275 nvme_remove_invalid_namespaces(ctrl, prev);
4281 static void nvme_scan_ns_sequential(struct nvme_ctrl *ctrl)
4283 struct nvme_id_ctrl *id;
4286 if (nvme_identify_ctrl(ctrl, &id))
4288 nn = le32_to_cpu(id->nn);
4291 for (i = 1; i <= nn; i++)
4292 nvme_validate_or_alloc_ns(ctrl, i);
4294 nvme_remove_invalid_namespaces(ctrl, nn);
4297 static void nvme_clear_changed_ns_log(struct nvme_ctrl *ctrl)
4299 size_t log_size = NVME_MAX_CHANGED_NAMESPACES * sizeof(__le32);
4303 log = kzalloc(log_size, GFP_KERNEL);
4308 * We need to read the log to clear the AEN, but we don't want to rely
4309 * on it for the changed namespace information as userspace could have
4310 * raced with us in reading the log page, which could cause us to miss
4313 error = nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_CHANGED_NS, 0,
4314 NVME_CSI_NVM, log, log_size, 0);
4316 dev_warn(ctrl->device,
4317 "reading changed ns log failed: %d\n", error);
4322 static void nvme_scan_work(struct work_struct *work)
4324 struct nvme_ctrl *ctrl =
4325 container_of(work, struct nvme_ctrl, scan_work);
4328 /* No tagset on a live ctrl means IO queues could not created */
4329 if (ctrl->state != NVME_CTRL_LIVE || !ctrl->tagset)
4333 * Identify controller limits can change at controller reset due to
4334 * new firmware download, even though it is not common we cannot ignore
4335 * such scenario. Controller's non-mdts limits are reported in the unit
4336 * of logical blocks that is dependent on the format of attached
4337 * namespace. Hence re-read the limits at the time of ns allocation.
4339 ret = nvme_init_non_mdts_limits(ctrl);
4341 dev_warn(ctrl->device,
4342 "reading non-mdts-limits failed: %d\n", ret);
4346 if (test_and_clear_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events)) {
4347 dev_info(ctrl->device, "rescanning namespaces.\n");
4348 nvme_clear_changed_ns_log(ctrl);
4351 mutex_lock(&ctrl->scan_lock);
4352 if (nvme_scan_ns_list(ctrl) != 0)
4353 nvme_scan_ns_sequential(ctrl);
4354 mutex_unlock(&ctrl->scan_lock);
4358 * This function iterates the namespace list unlocked to allow recovery from
4359 * controller failure. It is up to the caller to ensure the namespace list is
4360 * not modified by scan work while this function is executing.
4362 void nvme_remove_namespaces(struct nvme_ctrl *ctrl)
4364 struct nvme_ns *ns, *next;
4368 * make sure to requeue I/O to all namespaces as these
4369 * might result from the scan itself and must complete
4370 * for the scan_work to make progress
4372 nvme_mpath_clear_ctrl_paths(ctrl);
4374 /* prevent racing with ns scanning */
4375 flush_work(&ctrl->scan_work);
4378 * The dead states indicates the controller was not gracefully
4379 * disconnected. In that case, we won't be able to flush any data while
4380 * removing the namespaces' disks; fail all the queues now to avoid
4381 * potentially having to clean up the failed sync later.
4383 if (ctrl->state == NVME_CTRL_DEAD)
4384 nvme_kill_queues(ctrl);
4386 /* this is a no-op when called from the controller reset handler */
4387 nvme_change_ctrl_state(ctrl, NVME_CTRL_DELETING_NOIO);
4389 down_write(&ctrl->namespaces_rwsem);
4390 list_splice_init(&ctrl->namespaces, &ns_list);
4391 up_write(&ctrl->namespaces_rwsem);
4393 list_for_each_entry_safe(ns, next, &ns_list, list)
4396 EXPORT_SYMBOL_GPL(nvme_remove_namespaces);
4398 static int nvme_class_uevent(struct device *dev, struct kobj_uevent_env *env)
4400 struct nvme_ctrl *ctrl =
4401 container_of(dev, struct nvme_ctrl, ctrl_device);
4402 struct nvmf_ctrl_options *opts = ctrl->opts;
4405 ret = add_uevent_var(env, "NVME_TRTYPE=%s", ctrl->ops->name);
4410 ret = add_uevent_var(env, "NVME_TRADDR=%s", opts->traddr);
4414 ret = add_uevent_var(env, "NVME_TRSVCID=%s",
4415 opts->trsvcid ?: "none");
4419 ret = add_uevent_var(env, "NVME_HOST_TRADDR=%s",
4420 opts->host_traddr ?: "none");
4424 ret = add_uevent_var(env, "NVME_HOST_IFACE=%s",
4425 opts->host_iface ?: "none");
4430 static void nvme_change_uevent(struct nvme_ctrl *ctrl, char *envdata)
4432 char *envp[2] = { envdata, NULL };
4434 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4437 static void nvme_aen_uevent(struct nvme_ctrl *ctrl)
4439 char *envp[2] = { NULL, NULL };
4440 u32 aen_result = ctrl->aen_result;
4442 ctrl->aen_result = 0;
4446 envp[0] = kasprintf(GFP_KERNEL, "NVME_AEN=%#08x", aen_result);
4449 kobject_uevent_env(&ctrl->device->kobj, KOBJ_CHANGE, envp);
4453 static void nvme_async_event_work(struct work_struct *work)
4455 struct nvme_ctrl *ctrl =
4456 container_of(work, struct nvme_ctrl, async_event_work);
4458 nvme_aen_uevent(ctrl);
4461 * The transport drivers must guarantee AER submission here is safe by
4462 * flushing ctrl async_event_work after changing the controller state
4463 * from LIVE and before freeing the admin queue.
4465 if (ctrl->state == NVME_CTRL_LIVE)
4466 ctrl->ops->submit_async_event(ctrl);
4469 static bool nvme_ctrl_pp_status(struct nvme_ctrl *ctrl)
4474 if (ctrl->ops->reg_read32(ctrl, NVME_REG_CSTS, &csts))
4480 return ((ctrl->ctrl_config & NVME_CC_ENABLE) && (csts & NVME_CSTS_PP));
4483 static void nvme_get_fw_slot_info(struct nvme_ctrl *ctrl)
4485 struct nvme_fw_slot_info_log *log;
4487 log = kmalloc(sizeof(*log), GFP_KERNEL);
4491 if (nvme_get_log(ctrl, NVME_NSID_ALL, NVME_LOG_FW_SLOT, 0, NVME_CSI_NVM,
4492 log, sizeof(*log), 0))
4493 dev_warn(ctrl->device, "Get FW SLOT INFO log error\n");
4497 static void nvme_fw_act_work(struct work_struct *work)
4499 struct nvme_ctrl *ctrl = container_of(work,
4500 struct nvme_ctrl, fw_act_work);
4501 unsigned long fw_act_timeout;
4504 fw_act_timeout = jiffies +
4505 msecs_to_jiffies(ctrl->mtfa * 100);
4507 fw_act_timeout = jiffies +
4508 msecs_to_jiffies(admin_timeout * 1000);
4510 nvme_stop_queues(ctrl);
4511 while (nvme_ctrl_pp_status(ctrl)) {
4512 if (time_after(jiffies, fw_act_timeout)) {
4513 dev_warn(ctrl->device,
4514 "Fw activation timeout, reset controller\n");
4515 nvme_try_sched_reset(ctrl);
4521 if (!nvme_change_ctrl_state(ctrl, NVME_CTRL_LIVE))
4524 nvme_start_queues(ctrl);
4525 /* read FW slot information to clear the AER */
4526 nvme_get_fw_slot_info(ctrl);
4529 static void nvme_handle_aen_notice(struct nvme_ctrl *ctrl, u32 result)
4531 u32 aer_notice_type = (result & 0xff00) >> 8;
4533 trace_nvme_async_event(ctrl, aer_notice_type);
4535 switch (aer_notice_type) {
4536 case NVME_AER_NOTICE_NS_CHANGED:
4537 set_bit(NVME_AER_NOTICE_NS_CHANGED, &ctrl->events);
4538 nvme_queue_scan(ctrl);
4540 case NVME_AER_NOTICE_FW_ACT_STARTING:
4542 * We are (ab)using the RESETTING state to prevent subsequent
4543 * recovery actions from interfering with the controller's
4544 * firmware activation.
4546 if (nvme_change_ctrl_state(ctrl, NVME_CTRL_RESETTING))
4547 queue_work(nvme_wq, &ctrl->fw_act_work);
4549 #ifdef CONFIG_NVME_MULTIPATH
4550 case NVME_AER_NOTICE_ANA:
4551 if (!ctrl->ana_log_buf)
4553 queue_work(nvme_wq, &ctrl->ana_work);
4556 case NVME_AER_NOTICE_DISC_CHANGED:
4557 ctrl->aen_result = result;
4560 dev_warn(ctrl->device, "async event result %08x\n", result);
4564 void nvme_complete_async_event(struct nvme_ctrl *ctrl, __le16 status,
4565 volatile union nvme_result *res)
4567 u32 result = le32_to_cpu(res->u32);
4568 u32 aer_type = result & 0x07;
4570 if (le16_to_cpu(status) >> 1 != NVME_SC_SUCCESS)
4574 case NVME_AER_NOTICE:
4575 nvme_handle_aen_notice(ctrl, result);
4577 case NVME_AER_ERROR:
4578 case NVME_AER_SMART:
4581 trace_nvme_async_event(ctrl, aer_type);
4582 ctrl->aen_result = result;
4587 queue_work(nvme_wq, &ctrl->async_event_work);
4589 EXPORT_SYMBOL_GPL(nvme_complete_async_event);
4591 void nvme_stop_ctrl(struct nvme_ctrl *ctrl)
4593 nvme_mpath_stop(ctrl);
4594 nvme_stop_keep_alive(ctrl);
4595 nvme_stop_failfast_work(ctrl);
4596 flush_work(&ctrl->async_event_work);
4597 cancel_work_sync(&ctrl->fw_act_work);
4598 if (ctrl->ops->stop_ctrl)
4599 ctrl->ops->stop_ctrl(ctrl);
4601 EXPORT_SYMBOL_GPL(nvme_stop_ctrl);
4603 void nvme_start_ctrl(struct nvme_ctrl *ctrl)
4605 nvme_start_keep_alive(ctrl);
4607 nvme_enable_aen(ctrl);
4609 if (ctrl->queue_count > 1) {
4610 nvme_queue_scan(ctrl);
4611 nvme_start_queues(ctrl);
4612 nvme_mpath_update(ctrl);
4615 nvme_change_uevent(ctrl, "NVME_EVENT=connected");
4617 EXPORT_SYMBOL_GPL(nvme_start_ctrl);
4619 void nvme_uninit_ctrl(struct nvme_ctrl *ctrl)
4621 nvme_hwmon_exit(ctrl);
4622 nvme_fault_inject_fini(&ctrl->fault_inject);
4623 dev_pm_qos_hide_latency_tolerance(ctrl->device);
4624 cdev_device_del(&ctrl->cdev, ctrl->device);
4625 nvme_put_ctrl(ctrl);
4627 EXPORT_SYMBOL_GPL(nvme_uninit_ctrl);
4629 static void nvme_free_cels(struct nvme_ctrl *ctrl)
4631 struct nvme_effects_log *cel;
4634 xa_for_each(&ctrl->cels, i, cel) {
4635 xa_erase(&ctrl->cels, i);
4639 xa_destroy(&ctrl->cels);
4642 static void nvme_free_ctrl(struct device *dev)
4644 struct nvme_ctrl *ctrl =
4645 container_of(dev, struct nvme_ctrl, ctrl_device);
4646 struct nvme_subsystem *subsys = ctrl->subsys;
4648 if (!subsys || ctrl->instance != subsys->instance)
4649 ida_free(&nvme_instance_ida, ctrl->instance);
4651 nvme_free_cels(ctrl);
4652 nvme_mpath_uninit(ctrl);
4653 __free_page(ctrl->discard_page);
4656 mutex_lock(&nvme_subsystems_lock);
4657 list_del(&ctrl->subsys_entry);
4658 sysfs_remove_link(&subsys->dev.kobj, dev_name(ctrl->device));
4659 mutex_unlock(&nvme_subsystems_lock);
4662 ctrl->ops->free_ctrl(ctrl);
4665 nvme_put_subsystem(subsys);
4669 * Initialize a NVMe controller structures. This needs to be called during
4670 * earliest initialization so that we have the initialized structured around
4673 int nvme_init_ctrl(struct nvme_ctrl *ctrl, struct device *dev,
4674 const struct nvme_ctrl_ops *ops, unsigned long quirks)
4678 ctrl->state = NVME_CTRL_NEW;
4679 clear_bit(NVME_CTRL_FAILFAST_EXPIRED, &ctrl->flags);
4680 spin_lock_init(&ctrl->lock);
4681 mutex_init(&ctrl->scan_lock);
4682 INIT_LIST_HEAD(&ctrl->namespaces);
4683 xa_init(&ctrl->cels);
4684 init_rwsem(&ctrl->namespaces_rwsem);
4687 ctrl->quirks = quirks;
4688 ctrl->numa_node = NUMA_NO_NODE;
4689 INIT_WORK(&ctrl->scan_work, nvme_scan_work);
4690 INIT_WORK(&ctrl->async_event_work, nvme_async_event_work);
4691 INIT_WORK(&ctrl->fw_act_work, nvme_fw_act_work);
4692 INIT_WORK(&ctrl->delete_work, nvme_delete_ctrl_work);
4693 init_waitqueue_head(&ctrl->state_wq);
4695 INIT_DELAYED_WORK(&ctrl->ka_work, nvme_keep_alive_work);
4696 INIT_DELAYED_WORK(&ctrl->failfast_work, nvme_failfast_work);
4697 memset(&ctrl->ka_cmd, 0, sizeof(ctrl->ka_cmd));
4698 ctrl->ka_cmd.common.opcode = nvme_admin_keep_alive;
4700 BUILD_BUG_ON(NVME_DSM_MAX_RANGES * sizeof(struct nvme_dsm_range) >
4702 ctrl->discard_page = alloc_page(GFP_KERNEL);
4703 if (!ctrl->discard_page) {
4708 ret = ida_alloc(&nvme_instance_ida, GFP_KERNEL);
4711 ctrl->instance = ret;
4713 device_initialize(&ctrl->ctrl_device);
4714 ctrl->device = &ctrl->ctrl_device;
4715 ctrl->device->devt = MKDEV(MAJOR(nvme_ctrl_base_chr_devt),
4717 ctrl->device->class = nvme_class;
4718 ctrl->device->parent = ctrl->dev;
4719 ctrl->device->groups = nvme_dev_attr_groups;
4720 ctrl->device->release = nvme_free_ctrl;
4721 dev_set_drvdata(ctrl->device, ctrl);
4722 ret = dev_set_name(ctrl->device, "nvme%d", ctrl->instance);
4724 goto out_release_instance;
4726 nvme_get_ctrl(ctrl);
4727 cdev_init(&ctrl->cdev, &nvme_dev_fops);
4728 ctrl->cdev.owner = ops->module;
4729 ret = cdev_device_add(&ctrl->cdev, ctrl->device);
4734 * Initialize latency tolerance controls. The sysfs files won't
4735 * be visible to userspace unless the device actually supports APST.
4737 ctrl->device->power.set_latency_tolerance = nvme_set_latency_tolerance;
4738 dev_pm_qos_update_user_latency_tolerance(ctrl->device,
4739 min(default_ps_max_latency_us, (unsigned long)S32_MAX));
4741 nvme_fault_inject_init(&ctrl->fault_inject, dev_name(ctrl->device));
4742 nvme_mpath_init_ctrl(ctrl);
4746 nvme_put_ctrl(ctrl);
4747 kfree_const(ctrl->device->kobj.name);
4748 out_release_instance:
4749 ida_free(&nvme_instance_ida, ctrl->instance);
4751 if (ctrl->discard_page)
4752 __free_page(ctrl->discard_page);
4755 EXPORT_SYMBOL_GPL(nvme_init_ctrl);
4757 static void nvme_start_ns_queue(struct nvme_ns *ns)
4759 if (test_and_clear_bit(NVME_NS_STOPPED, &ns->flags))
4760 blk_mq_unquiesce_queue(ns->queue);
4763 static void nvme_stop_ns_queue(struct nvme_ns *ns)
4765 if (!test_and_set_bit(NVME_NS_STOPPED, &ns->flags))
4766 blk_mq_quiesce_queue(ns->queue);
4768 blk_mq_wait_quiesce_done(ns->queue);
4772 * Prepare a queue for teardown.
4774 * This must forcibly unquiesce queues to avoid blocking dispatch, and only set
4775 * the capacity to 0 after that to avoid blocking dispatchers that may be
4776 * holding bd_butex. This will end buffered writers dirtying pages that can't
4779 static void nvme_set_queue_dying(struct nvme_ns *ns)
4781 if (test_and_set_bit(NVME_NS_DEAD, &ns->flags))
4784 blk_mark_disk_dead(ns->disk);
4785 nvme_start_ns_queue(ns);
4787 set_capacity_and_notify(ns->disk, 0);
4791 * nvme_kill_queues(): Ends all namespace queues
4792 * @ctrl: the dead controller that needs to end
4794 * Call this function when the driver determines it is unable to get the
4795 * controller in a state capable of servicing IO.
4797 void nvme_kill_queues(struct nvme_ctrl *ctrl)
4801 down_read(&ctrl->namespaces_rwsem);
4803 /* Forcibly unquiesce queues to avoid blocking dispatch */
4804 if (ctrl->admin_q && !blk_queue_dying(ctrl->admin_q))
4805 nvme_start_admin_queue(ctrl);
4807 list_for_each_entry(ns, &ctrl->namespaces, list)
4808 nvme_set_queue_dying(ns);
4810 up_read(&ctrl->namespaces_rwsem);
4812 EXPORT_SYMBOL_GPL(nvme_kill_queues);
4814 void nvme_unfreeze(struct nvme_ctrl *ctrl)
4818 down_read(&ctrl->namespaces_rwsem);
4819 list_for_each_entry(ns, &ctrl->namespaces, list)
4820 blk_mq_unfreeze_queue(ns->queue);
4821 up_read(&ctrl->namespaces_rwsem);
4823 EXPORT_SYMBOL_GPL(nvme_unfreeze);
4825 int nvme_wait_freeze_timeout(struct nvme_ctrl *ctrl, long timeout)
4829 down_read(&ctrl->namespaces_rwsem);
4830 list_for_each_entry(ns, &ctrl->namespaces, list) {
4831 timeout = blk_mq_freeze_queue_wait_timeout(ns->queue, timeout);
4835 up_read(&ctrl->namespaces_rwsem);
4838 EXPORT_SYMBOL_GPL(nvme_wait_freeze_timeout);
4840 void nvme_wait_freeze(struct nvme_ctrl *ctrl)
4844 down_read(&ctrl->namespaces_rwsem);
4845 list_for_each_entry(ns, &ctrl->namespaces, list)
4846 blk_mq_freeze_queue_wait(ns->queue);
4847 up_read(&ctrl->namespaces_rwsem);
4849 EXPORT_SYMBOL_GPL(nvme_wait_freeze);
4851 void nvme_start_freeze(struct nvme_ctrl *ctrl)
4855 down_read(&ctrl->namespaces_rwsem);
4856 list_for_each_entry(ns, &ctrl->namespaces, list)
4857 blk_freeze_queue_start(ns->queue);
4858 up_read(&ctrl->namespaces_rwsem);
4860 EXPORT_SYMBOL_GPL(nvme_start_freeze);
4862 void nvme_stop_queues(struct nvme_ctrl *ctrl)
4866 down_read(&ctrl->namespaces_rwsem);
4867 list_for_each_entry(ns, &ctrl->namespaces, list)
4868 nvme_stop_ns_queue(ns);
4869 up_read(&ctrl->namespaces_rwsem);
4871 EXPORT_SYMBOL_GPL(nvme_stop_queues);
4873 void nvme_start_queues(struct nvme_ctrl *ctrl)
4877 down_read(&ctrl->namespaces_rwsem);
4878 list_for_each_entry(ns, &ctrl->namespaces, list)
4879 nvme_start_ns_queue(ns);
4880 up_read(&ctrl->namespaces_rwsem);
4882 EXPORT_SYMBOL_GPL(nvme_start_queues);
4884 void nvme_stop_admin_queue(struct nvme_ctrl *ctrl)
4886 if (!test_and_set_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4887 blk_mq_quiesce_queue(ctrl->admin_q);
4889 blk_mq_wait_quiesce_done(ctrl->admin_q);
4891 EXPORT_SYMBOL_GPL(nvme_stop_admin_queue);
4893 void nvme_start_admin_queue(struct nvme_ctrl *ctrl)
4895 if (test_and_clear_bit(NVME_CTRL_ADMIN_Q_STOPPED, &ctrl->flags))
4896 blk_mq_unquiesce_queue(ctrl->admin_q);
4898 EXPORT_SYMBOL_GPL(nvme_start_admin_queue);
4900 void nvme_sync_io_queues(struct nvme_ctrl *ctrl)
4904 down_read(&ctrl->namespaces_rwsem);
4905 list_for_each_entry(ns, &ctrl->namespaces, list)
4906 blk_sync_queue(ns->queue);
4907 up_read(&ctrl->namespaces_rwsem);
4909 EXPORT_SYMBOL_GPL(nvme_sync_io_queues);
4911 void nvme_sync_queues(struct nvme_ctrl *ctrl)
4913 nvme_sync_io_queues(ctrl);
4915 blk_sync_queue(ctrl->admin_q);
4917 EXPORT_SYMBOL_GPL(nvme_sync_queues);
4919 struct nvme_ctrl *nvme_ctrl_from_file(struct file *file)
4921 if (file->f_op != &nvme_dev_fops)
4923 return file->private_data;
4925 EXPORT_SYMBOL_NS_GPL(nvme_ctrl_from_file, NVME_TARGET_PASSTHRU);
4928 * Check we didn't inadvertently grow the command structure sizes:
4930 static inline void _nvme_check_size(void)
4932 BUILD_BUG_ON(sizeof(struct nvme_common_command) != 64);
4933 BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64);
4934 BUILD_BUG_ON(sizeof(struct nvme_identify) != 64);
4935 BUILD_BUG_ON(sizeof(struct nvme_features) != 64);
4936 BUILD_BUG_ON(sizeof(struct nvme_download_firmware) != 64);
4937 BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64);
4938 BUILD_BUG_ON(sizeof(struct nvme_dsm_cmd) != 64);
4939 BUILD_BUG_ON(sizeof(struct nvme_write_zeroes_cmd) != 64);
4940 BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64);
4941 BUILD_BUG_ON(sizeof(struct nvme_get_log_page_command) != 64);
4942 BUILD_BUG_ON(sizeof(struct nvme_command) != 64);
4943 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE);
4944 BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE);
4945 BUILD_BUG_ON(sizeof(struct nvme_id_ns_cs_indep) !=
4946 NVME_IDENTIFY_DATA_SIZE);
4947 BUILD_BUG_ON(sizeof(struct nvme_id_ns_zns) != NVME_IDENTIFY_DATA_SIZE);
4948 BUILD_BUG_ON(sizeof(struct nvme_id_ns_nvm) != NVME_IDENTIFY_DATA_SIZE);
4949 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_zns) != NVME_IDENTIFY_DATA_SIZE);
4950 BUILD_BUG_ON(sizeof(struct nvme_id_ctrl_nvm) != NVME_IDENTIFY_DATA_SIZE);
4951 BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
4952 BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512);
4953 BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64);
4954 BUILD_BUG_ON(sizeof(struct nvme_directive_cmd) != 64);
4955 BUILD_BUG_ON(sizeof(struct nvme_feat_host_behavior) != 512);
4959 static int __init nvme_core_init(void)
4961 int result = -ENOMEM;
4965 nvme_wq = alloc_workqueue("nvme-wq",
4966 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4970 nvme_reset_wq = alloc_workqueue("nvme-reset-wq",
4971 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4975 nvme_delete_wq = alloc_workqueue("nvme-delete-wq",
4976 WQ_UNBOUND | WQ_MEM_RECLAIM | WQ_SYSFS, 0);
4977 if (!nvme_delete_wq)
4978 goto destroy_reset_wq;
4980 result = alloc_chrdev_region(&nvme_ctrl_base_chr_devt, 0,
4981 NVME_MINORS, "nvme");
4983 goto destroy_delete_wq;
4985 nvme_class = class_create(THIS_MODULE, "nvme");
4986 if (IS_ERR(nvme_class)) {
4987 result = PTR_ERR(nvme_class);
4988 goto unregister_chrdev;
4990 nvme_class->dev_uevent = nvme_class_uevent;
4992 nvme_subsys_class = class_create(THIS_MODULE, "nvme-subsystem");
4993 if (IS_ERR(nvme_subsys_class)) {
4994 result = PTR_ERR(nvme_subsys_class);
4998 result = alloc_chrdev_region(&nvme_ns_chr_devt, 0, NVME_MINORS,
5001 goto destroy_subsys_class;
5003 nvme_ns_chr_class = class_create(THIS_MODULE, "nvme-generic");
5004 if (IS_ERR(nvme_ns_chr_class)) {
5005 result = PTR_ERR(nvme_ns_chr_class);
5006 goto unregister_generic_ns;
5011 unregister_generic_ns:
5012 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5013 destroy_subsys_class:
5014 class_destroy(nvme_subsys_class);
5016 class_destroy(nvme_class);
5018 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5020 destroy_workqueue(nvme_delete_wq);
5022 destroy_workqueue(nvme_reset_wq);
5024 destroy_workqueue(nvme_wq);
5029 static void __exit nvme_core_exit(void)
5031 class_destroy(nvme_ns_chr_class);
5032 class_destroy(nvme_subsys_class);
5033 class_destroy(nvme_class);
5034 unregister_chrdev_region(nvme_ns_chr_devt, NVME_MINORS);
5035 unregister_chrdev_region(nvme_ctrl_base_chr_devt, NVME_MINORS);
5036 destroy_workqueue(nvme_delete_wq);
5037 destroy_workqueue(nvme_reset_wq);
5038 destroy_workqueue(nvme_wq);
5039 ida_destroy(&nvme_ns_chr_minor_ida);
5040 ida_destroy(&nvme_instance_ida);
5043 MODULE_LICENSE("GPL");
5044 MODULE_VERSION("1.0");
5045 module_init(nvme_core_init);
5046 module_exit(nvme_core_exit);